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                 ibmvnic_napi_disable(adapter);
1153                 release_resources(adapter);
1154                 return rc;
1155         }
1156
1157         netif_tx_start_all_queues(netdev);
1158
1159         if (prev_state == VNIC_CLOSED) {
1160                 for (i = 0; i < adapter->req_rx_queues; i++)
1161                         napi_schedule(&adapter->napi[i]);
1162         }
1163
1164         adapter->state = VNIC_OPEN;
1165         return rc;
1166 }
1167
1168 static int ibmvnic_open(struct net_device *netdev)
1169 {
1170         struct ibmvnic_adapter *adapter = netdev_priv(netdev);
1171         int rc;
1172
1173         ASSERT_RTNL();
1174
1175         /* If device failover is pending or we are about to reset, just set
1176          * device state and return. Device operation will be handled by reset
1177          * routine.
1178          *
1179          * It should be safe to overwrite the adapter->state here. Since
1180          * we hold the rtnl, either the reset has not actually started or
1181          * the rtnl got dropped during the set_link_state() in do_reset().
1182          * In the former case, no one else is changing the state (again we
1183          * have the rtnl) and in the latter case, do_reset() will detect and
1184          * honor our setting below.
1185          */
1186         if (adapter->failover_pending || (test_bit(0, &adapter->resetting))) {
1187                 netdev_dbg(netdev, "[S:%d FOP:%d] Resetting, deferring open\n",
1188                            adapter->state, adapter->failover_pending);
1189                 adapter->state = VNIC_OPEN;
1190                 rc = 0;
1191                 goto out;
1192         }
1193
1194         if (adapter->state != VNIC_CLOSED) {
1195                 rc = ibmvnic_login(netdev);
1196                 if (rc)
1197                         goto out;
1198
1199                 rc = init_resources(adapter);
1200                 if (rc) {
1201                         netdev_err(netdev, "failed to initialize resources\n");
1202                         release_resources(adapter);
1203                         goto out;
1204                 }
1205         }
1206
1207         rc = __ibmvnic_open(netdev);
1208
1209 out:
1210         /* If open failed and there is a pending failover or in-progress reset,
1211          * set device state and return. Device operation will be handled by
1212          * reset routine. See also comments above regarding rtnl.
1213          */
1214         if (rc &&
1215             (adapter->failover_pending || (test_bit(0, &adapter->resetting)))) {
1216                 adapter->state = VNIC_OPEN;
1217                 rc = 0;
1218         }
1219         return rc;
1220 }
1221
1222 static void clean_rx_pools(struct ibmvnic_adapter *adapter)
1223 {
1224         struct ibmvnic_rx_pool *rx_pool;
1225         struct ibmvnic_rx_buff *rx_buff;
1226         u64 rx_entries;
1227         int rx_scrqs;
1228         int i, j;
1229
1230         if (!adapter->rx_pool)
1231                 return;
1232
1233         rx_scrqs = adapter->num_active_rx_pools;
1234         rx_entries = adapter->req_rx_add_entries_per_subcrq;
1235
1236         /* Free any remaining skbs in the rx buffer pools */
1237         for (i = 0; i < rx_scrqs; i++) {
1238                 rx_pool = &adapter->rx_pool[i];
1239                 if (!rx_pool || !rx_pool->rx_buff)
1240                         continue;
1241
1242                 netdev_dbg(adapter->netdev, "Cleaning rx_pool[%d]\n", i);
1243                 for (j = 0; j < rx_entries; j++) {
1244                         rx_buff = &rx_pool->rx_buff[j];
1245                         if (rx_buff && rx_buff->skb) {
1246                                 dev_kfree_skb_any(rx_buff->skb);
1247                                 rx_buff->skb = NULL;
1248                         }
1249                 }
1250         }
1251 }
1252
1253 static void clean_one_tx_pool(struct ibmvnic_adapter *adapter,
1254                               struct ibmvnic_tx_pool *tx_pool)
1255 {
1256         struct ibmvnic_tx_buff *tx_buff;
1257         u64 tx_entries;
1258         int i;
1259
1260         if (!tx_pool || !tx_pool->tx_buff)
1261                 return;
1262
1263         tx_entries = tx_pool->num_buffers;
1264
1265         for (i = 0; i < tx_entries; i++) {
1266                 tx_buff = &tx_pool->tx_buff[i];
1267                 if (tx_buff && tx_buff->skb) {
1268                         dev_kfree_skb_any(tx_buff->skb);
1269                         tx_buff->skb = NULL;
1270                 }
1271         }
1272 }
1273
1274 static void clean_tx_pools(struct ibmvnic_adapter *adapter)
1275 {
1276         int tx_scrqs;
1277         int i;
1278
1279         if (!adapter->tx_pool || !adapter->tso_pool)
1280                 return;
1281
1282         tx_scrqs = adapter->num_active_tx_pools;
1283
1284         /* Free any remaining skbs in the tx buffer pools */
1285         for (i = 0; i < tx_scrqs; i++) {
1286                 netdev_dbg(adapter->netdev, "Cleaning tx_pool[%d]\n", i);
1287                 clean_one_tx_pool(adapter, &adapter->tx_pool[i]);
1288                 clean_one_tx_pool(adapter, &adapter->tso_pool[i]);
1289         }
1290 }
1291
1292 static void ibmvnic_disable_irqs(struct ibmvnic_adapter *adapter)
1293 {
1294         struct net_device *netdev = adapter->netdev;
1295         int i;
1296
1297         if (adapter->tx_scrq) {
1298                 for (i = 0; i < adapter->req_tx_queues; i++)
1299                         if (adapter->tx_scrq[i]->irq) {
1300                                 netdev_dbg(netdev,
1301                                            "Disabling tx_scrq[%d] irq\n", i);
1302                                 disable_scrq_irq(adapter, adapter->tx_scrq[i]);
1303                                 disable_irq(adapter->tx_scrq[i]->irq);
1304                         }
1305         }
1306
1307         if (adapter->rx_scrq) {
1308                 for (i = 0; i < adapter->req_rx_queues; i++) {
1309                         if (adapter->rx_scrq[i]->irq) {
1310                                 netdev_dbg(netdev,
1311                                            "Disabling rx_scrq[%d] irq\n", i);
1312                                 disable_scrq_irq(adapter, adapter->rx_scrq[i]);
1313                                 disable_irq(adapter->rx_scrq[i]->irq);
1314                         }
1315                 }
1316         }
1317 }
1318
1319 static void ibmvnic_cleanup(struct net_device *netdev)
1320 {
1321         struct ibmvnic_adapter *adapter = netdev_priv(netdev);
1322
1323         /* ensure that transmissions are stopped if called by do_reset */
1324         if (test_bit(0, &adapter->resetting))
1325                 netif_tx_disable(netdev);
1326         else
1327                 netif_tx_stop_all_queues(netdev);
1328
1329         ibmvnic_napi_disable(adapter);
1330         ibmvnic_disable_irqs(adapter);
1331
1332         clean_rx_pools(adapter);
1333         clean_tx_pools(adapter);
1334 }
1335
1336 static int __ibmvnic_close(struct net_device *netdev)
1337 {
1338         struct ibmvnic_adapter *adapter = netdev_priv(netdev);
1339         int rc = 0;
1340
1341         adapter->state = VNIC_CLOSING;
1342         rc = set_link_state(adapter, IBMVNIC_LOGICAL_LNK_DN);
1343         adapter->state = VNIC_CLOSED;
1344         return rc;
1345 }
1346
1347 static int ibmvnic_close(struct net_device *netdev)
1348 {
1349         struct ibmvnic_adapter *adapter = netdev_priv(netdev);
1350         int rc;
1351
1352         netdev_dbg(netdev, "[S:%d FOP:%d FRR:%d] Closing\n",
1353                    adapter->state, adapter->failover_pending,
1354                    adapter->force_reset_recovery);
1355
1356         /* If device failover is pending, just set device state and return.
1357          * Device operation will be handled by reset routine.
1358          */
1359         if (adapter->failover_pending) {
1360                 adapter->state = VNIC_CLOSED;
1361                 return 0;
1362         }
1363
1364         rc = __ibmvnic_close(netdev);
1365         ibmvnic_cleanup(netdev);
1366
1367         return rc;
1368 }
1369
1370 /**
1371  * build_hdr_data - creates L2/L3/L4 header data buffer
1372  * @hdr_field: bitfield determining needed headers
1373  * @skb: socket buffer
1374  * @hdr_len: array of header lengths
1375  * @hdr_data: buffer to write the header to
1376  *
1377  * Reads hdr_field to determine which headers are needed by firmware.
1378  * Builds a buffer containing these headers.  Saves individual header
1379  * lengths and total buffer length to be used to build descriptors.
1380  */
1381 static int build_hdr_data(u8 hdr_field, struct sk_buff *skb,
1382                           int *hdr_len, u8 *hdr_data)
1383 {
1384         int len = 0;
1385         u8 *hdr;
1386
1387         if (skb_vlan_tagged(skb) && !skb_vlan_tag_present(skb))
1388                 hdr_len[0] = sizeof(struct vlan_ethhdr);
1389         else
1390                 hdr_len[0] = sizeof(struct ethhdr);
1391
1392         if (skb->protocol == htons(ETH_P_IP)) {
1393                 hdr_len[1] = ip_hdr(skb)->ihl * 4;
1394                 if (ip_hdr(skb)->protocol == IPPROTO_TCP)
1395                         hdr_len[2] = tcp_hdrlen(skb);
1396                 else if (ip_hdr(skb)->protocol == IPPROTO_UDP)
1397                         hdr_len[2] = sizeof(struct udphdr);
1398         } else if (skb->protocol == htons(ETH_P_IPV6)) {
1399                 hdr_len[1] = sizeof(struct ipv6hdr);
1400                 if (ipv6_hdr(skb)->nexthdr == IPPROTO_TCP)
1401                         hdr_len[2] = tcp_hdrlen(skb);
1402                 else if (ipv6_hdr(skb)->nexthdr == IPPROTO_UDP)
1403                         hdr_len[2] = sizeof(struct udphdr);
1404         } else if (skb->protocol == htons(ETH_P_ARP)) {
1405                 hdr_len[1] = arp_hdr_len(skb->dev);
1406                 hdr_len[2] = 0;
1407         }
1408
1409         memset(hdr_data, 0, 120);
1410         if ((hdr_field >> 6) & 1) {
1411                 hdr = skb_mac_header(skb);
1412                 memcpy(hdr_data, hdr, hdr_len[0]);
1413                 len += hdr_len[0];
1414         }
1415
1416         if ((hdr_field >> 5) & 1) {
1417                 hdr = skb_network_header(skb);
1418                 memcpy(hdr_data + len, hdr, hdr_len[1]);
1419                 len += hdr_len[1];
1420         }
1421
1422         if ((hdr_field >> 4) & 1) {
1423                 hdr = skb_transport_header(skb);
1424                 memcpy(hdr_data + len, hdr, hdr_len[2]);
1425                 len += hdr_len[2];
1426         }
1427         return len;
1428 }
1429
1430 /**
1431  * create_hdr_descs - create header and header extension descriptors
1432  * @hdr_field: bitfield determining needed headers
1433  * @hdr_data: buffer containing header data
1434  * @len: length of data buffer
1435  * @hdr_len: array of individual header lengths
1436  * @scrq_arr: descriptor array
1437  *
1438  * Creates header and, if needed, header extension descriptors and
1439  * places them in a descriptor array, scrq_arr
1440  */
1441
1442 static int create_hdr_descs(u8 hdr_field, u8 *hdr_data, int len, int *hdr_len,
1443                             union sub_crq *scrq_arr)
1444 {
1445         union sub_crq hdr_desc;
1446         int tmp_len = len;
1447         int num_descs = 0;
1448         u8 *data, *cur;
1449         int tmp;
1450
1451         while (tmp_len > 0) {
1452                 cur = hdr_data + len - tmp_len;
1453
1454                 memset(&hdr_desc, 0, sizeof(hdr_desc));
1455                 if (cur != hdr_data) {
1456                         data = hdr_desc.hdr_ext.data;
1457                         tmp = tmp_len > 29 ? 29 : tmp_len;
1458                         hdr_desc.hdr_ext.first = IBMVNIC_CRQ_CMD;
1459                         hdr_desc.hdr_ext.type = IBMVNIC_HDR_EXT_DESC;
1460                         hdr_desc.hdr_ext.len = tmp;
1461                 } else {
1462                         data = hdr_desc.hdr.data;
1463                         tmp = tmp_len > 24 ? 24 : tmp_len;
1464                         hdr_desc.hdr.first = IBMVNIC_CRQ_CMD;
1465                         hdr_desc.hdr.type = IBMVNIC_HDR_DESC;
1466                         hdr_desc.hdr.len = tmp;
1467                         hdr_desc.hdr.l2_len = (u8)hdr_len[0];
1468                         hdr_desc.hdr.l3_len = cpu_to_be16((u16)hdr_len[1]);
1469                         hdr_desc.hdr.l4_len = (u8)hdr_len[2];
1470                         hdr_desc.hdr.flag = hdr_field << 1;
1471                 }
1472                 memcpy(data, cur, tmp);
1473                 tmp_len -= tmp;
1474                 *scrq_arr = hdr_desc;
1475                 scrq_arr++;
1476                 num_descs++;
1477         }
1478
1479         return num_descs;
1480 }
1481
1482 /**
1483  * build_hdr_descs_arr - build a header descriptor array
1484  * @txbuff: tx buffer
1485  * @num_entries: number of descriptors to be sent
1486  * @hdr_field: bit field determining which headers will be sent
1487  *
1488  * This function will build a TX descriptor array with applicable
1489  * L2/L3/L4 packet header descriptors to be sent by send_subcrq_indirect.
1490  */
1491
1492 static void build_hdr_descs_arr(struct sk_buff *skb,
1493                                 union sub_crq *indir_arr,
1494                                 int *num_entries, u8 hdr_field)
1495 {
1496         int hdr_len[3] = {0, 0, 0};
1497         u8 hdr_data[140] = {0};
1498         int tot_len;
1499
1500         tot_len = build_hdr_data(hdr_field, skb, hdr_len,
1501                                  hdr_data);
1502         *num_entries += create_hdr_descs(hdr_field, hdr_data, tot_len, hdr_len,
1503                                          indir_arr + 1);
1504 }
1505
1506 static int ibmvnic_xmit_workarounds(struct sk_buff *skb,
1507                                     struct net_device *netdev)
1508 {
1509         /* For some backing devices, mishandling of small packets
1510          * can result in a loss of connection or TX stall. Device
1511          * architects recommend that no packet should be smaller
1512          * than the minimum MTU value provided to the driver, so
1513          * pad any packets to that length
1514          */
1515         if (skb->len < netdev->min_mtu)
1516                 return skb_put_padto(skb, netdev->min_mtu);
1517
1518         return 0;
1519 }
1520
1521 static void ibmvnic_tx_scrq_clean_buffer(struct ibmvnic_adapter *adapter,
1522                                          struct ibmvnic_sub_crq_queue *tx_scrq)
1523 {
1524         struct ibmvnic_ind_xmit_queue *ind_bufp;
1525         struct ibmvnic_tx_buff *tx_buff;
1526         struct ibmvnic_tx_pool *tx_pool;
1527         union sub_crq tx_scrq_entry;
1528         int queue_num;
1529         int entries;
1530         int index;
1531         int i;
1532
1533         ind_bufp = &tx_scrq->ind_buf;
1534         entries = (u64)ind_bufp->index;
1535         queue_num = tx_scrq->pool_index;
1536
1537         for (i = entries - 1; i >= 0; --i) {
1538                 tx_scrq_entry = ind_bufp->indir_arr[i];
1539                 if (tx_scrq_entry.v1.type != IBMVNIC_TX_DESC)
1540                         continue;
1541                 index = be32_to_cpu(tx_scrq_entry.v1.correlator);
1542                 if (index & IBMVNIC_TSO_POOL_MASK) {
1543                         tx_pool = &adapter->tso_pool[queue_num];
1544                         index &= ~IBMVNIC_TSO_POOL_MASK;
1545                 } else {
1546                         tx_pool = &adapter->tx_pool[queue_num];
1547                 }
1548                 tx_pool->free_map[tx_pool->consumer_index] = index;
1549                 tx_pool->consumer_index = tx_pool->consumer_index == 0 ?
1550                                           tx_pool->num_buffers - 1 :
1551                                           tx_pool->consumer_index - 1;
1552                 tx_buff = &tx_pool->tx_buff[index];
1553                 adapter->netdev->stats.tx_packets--;
1554                 adapter->netdev->stats.tx_bytes -= tx_buff->skb->len;
1555                 adapter->tx_stats_buffers[queue_num].packets--;
1556                 adapter->tx_stats_buffers[queue_num].bytes -=
1557                                                 tx_buff->skb->len;
1558                 dev_kfree_skb_any(tx_buff->skb);
1559                 tx_buff->skb = NULL;
1560                 adapter->netdev->stats.tx_dropped++;
1561         }
1562         ind_bufp->index = 0;
1563         if (atomic_sub_return(entries, &tx_scrq->used) <=
1564             (adapter->req_tx_entries_per_subcrq / 2) &&
1565             __netif_subqueue_stopped(adapter->netdev, queue_num)) {
1566                 netif_wake_subqueue(adapter->netdev, queue_num);
1567                 netdev_dbg(adapter->netdev, "Started queue %d\n",
1568                            queue_num);
1569         }
1570 }
1571
1572 static int ibmvnic_tx_scrq_flush(struct ibmvnic_adapter *adapter,
1573                                  struct ibmvnic_sub_crq_queue *tx_scrq)
1574 {
1575         struct ibmvnic_ind_xmit_queue *ind_bufp;
1576         u64 dma_addr;
1577         u64 entries;
1578         u64 handle;
1579         int rc;
1580
1581         ind_bufp = &tx_scrq->ind_buf;
1582         dma_addr = (u64)ind_bufp->indir_dma;
1583         entries = (u64)ind_bufp->index;
1584         handle = tx_scrq->handle;
1585
1586         if (!entries)
1587                 return 0;
1588         rc = send_subcrq_indirect(adapter, handle, dma_addr, entries);
1589         if (rc)
1590                 ibmvnic_tx_scrq_clean_buffer(adapter, tx_scrq);
1591         else
1592                 ind_bufp->index = 0;
1593         return 0;
1594 }
1595
1596 static netdev_tx_t ibmvnic_xmit(struct sk_buff *skb, struct net_device *netdev)
1597 {
1598         struct ibmvnic_adapter *adapter = netdev_priv(netdev);
1599         int queue_num = skb_get_queue_mapping(skb);
1600         u8 *hdrs = (u8 *)&adapter->tx_rx_desc_req;
1601         struct device *dev = &adapter->vdev->dev;
1602         struct ibmvnic_ind_xmit_queue *ind_bufp;
1603         struct ibmvnic_tx_buff *tx_buff = NULL;
1604         struct ibmvnic_sub_crq_queue *tx_scrq;
1605         struct ibmvnic_tx_pool *tx_pool;
1606         unsigned int tx_send_failed = 0;
1607         netdev_tx_t ret = NETDEV_TX_OK;
1608         unsigned int tx_map_failed = 0;
1609         union sub_crq indir_arr[16];
1610         unsigned int tx_dropped = 0;
1611         unsigned int tx_packets = 0;
1612         unsigned int tx_bytes = 0;
1613         dma_addr_t data_dma_addr;
1614         struct netdev_queue *txq;
1615         unsigned long lpar_rc;
1616         union sub_crq tx_crq;
1617         unsigned int offset;
1618         int num_entries = 1;
1619         unsigned char *dst;
1620         int index = 0;
1621         u8 proto = 0;
1622
1623         tx_scrq = adapter->tx_scrq[queue_num];
1624         txq = netdev_get_tx_queue(netdev, queue_num);
1625         ind_bufp = &tx_scrq->ind_buf;
1626
1627         if (test_bit(0, &adapter->resetting)) {
1628                 if (!netif_subqueue_stopped(netdev, skb))
1629                         netif_stop_subqueue(netdev, queue_num);
1630                 dev_kfree_skb_any(skb);
1631
1632                 tx_send_failed++;
1633                 tx_dropped++;
1634                 ret = NETDEV_TX_OK;
1635                 ibmvnic_tx_scrq_flush(adapter, tx_scrq);
1636                 goto out;
1637         }
1638
1639         if (ibmvnic_xmit_workarounds(skb, netdev)) {
1640                 tx_dropped++;
1641                 tx_send_failed++;
1642                 ret = NETDEV_TX_OK;
1643                 ibmvnic_tx_scrq_flush(adapter, tx_scrq);
1644                 goto out;
1645         }
1646         if (skb_is_gso(skb))
1647                 tx_pool = &adapter->tso_pool[queue_num];
1648         else
1649                 tx_pool = &adapter->tx_pool[queue_num];
1650
1651         index = tx_pool->free_map[tx_pool->consumer_index];
1652
1653         if (index == IBMVNIC_INVALID_MAP) {
1654                 dev_kfree_skb_any(skb);
1655                 tx_send_failed++;
1656                 tx_dropped++;
1657                 ret = NETDEV_TX_OK;
1658                 ibmvnic_tx_scrq_flush(adapter, tx_scrq);
1659                 goto out;
1660         }
1661
1662         tx_pool->free_map[tx_pool->consumer_index] = IBMVNIC_INVALID_MAP;
1663
1664         offset = index * tx_pool->buf_size;
1665         dst = tx_pool->long_term_buff.buff + offset;
1666         memset(dst, 0, tx_pool->buf_size);
1667         data_dma_addr = tx_pool->long_term_buff.addr + offset;
1668
1669         if (skb_shinfo(skb)->nr_frags) {
1670                 int cur, i;
1671
1672                 /* Copy the head */
1673                 skb_copy_from_linear_data(skb, dst, skb_headlen(skb));
1674                 cur = skb_headlen(skb);
1675
1676                 /* Copy the frags */
1677                 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
1678                         const skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
1679
1680                         memcpy(dst + cur,
1681                                page_address(skb_frag_page(frag)) +
1682                                skb_frag_off(frag), skb_frag_size(frag));
1683                         cur += skb_frag_size(frag);
1684                 }
1685         } else {
1686                 skb_copy_from_linear_data(skb, dst, skb->len);
1687         }
1688
1689         /* post changes to long_term_buff *dst before VIOS accessing it */
1690         dma_wmb();
1691
1692         tx_pool->consumer_index =
1693             (tx_pool->consumer_index + 1) % tx_pool->num_buffers;
1694
1695         tx_buff = &tx_pool->tx_buff[index];
1696         tx_buff->skb = skb;
1697         tx_buff->index = index;
1698         tx_buff->pool_index = queue_num;
1699
1700         memset(&tx_crq, 0, sizeof(tx_crq));
1701         tx_crq.v1.first = IBMVNIC_CRQ_CMD;
1702         tx_crq.v1.type = IBMVNIC_TX_DESC;
1703         tx_crq.v1.n_crq_elem = 1;
1704         tx_crq.v1.n_sge = 1;
1705         tx_crq.v1.flags1 = IBMVNIC_TX_COMP_NEEDED;
1706
1707         if (skb_is_gso(skb))
1708                 tx_crq.v1.correlator =
1709                         cpu_to_be32(index | IBMVNIC_TSO_POOL_MASK);
1710         else
1711                 tx_crq.v1.correlator = cpu_to_be32(index);
1712         tx_crq.v1.dma_reg = cpu_to_be16(tx_pool->long_term_buff.map_id);
1713         tx_crq.v1.sge_len = cpu_to_be32(skb->len);
1714         tx_crq.v1.ioba = cpu_to_be64(data_dma_addr);
1715
1716         if (adapter->vlan_header_insertion && skb_vlan_tag_present(skb)) {
1717                 tx_crq.v1.flags2 |= IBMVNIC_TX_VLAN_INSERT;
1718                 tx_crq.v1.vlan_id = cpu_to_be16(skb->vlan_tci);
1719         }
1720
1721         if (skb->protocol == htons(ETH_P_IP)) {
1722                 tx_crq.v1.flags1 |= IBMVNIC_TX_PROT_IPV4;
1723                 proto = ip_hdr(skb)->protocol;
1724         } else if (skb->protocol == htons(ETH_P_IPV6)) {
1725                 tx_crq.v1.flags1 |= IBMVNIC_TX_PROT_IPV6;
1726                 proto = ipv6_hdr(skb)->nexthdr;
1727         }
1728
1729         if (proto == IPPROTO_TCP)
1730                 tx_crq.v1.flags1 |= IBMVNIC_TX_PROT_TCP;
1731         else if (proto == IPPROTO_UDP)
1732                 tx_crq.v1.flags1 |= IBMVNIC_TX_PROT_UDP;
1733
1734         if (skb->ip_summed == CHECKSUM_PARTIAL) {
1735                 tx_crq.v1.flags1 |= IBMVNIC_TX_CHKSUM_OFFLOAD;
1736                 hdrs += 2;
1737         }
1738         if (skb_is_gso(skb)) {
1739                 tx_crq.v1.flags1 |= IBMVNIC_TX_LSO;
1740                 tx_crq.v1.mss = cpu_to_be16(skb_shinfo(skb)->gso_size);
1741                 hdrs += 2;
1742         }
1743
1744         if ((*hdrs >> 7) & 1)
1745                 build_hdr_descs_arr(skb, indir_arr, &num_entries, *hdrs);
1746
1747         tx_crq.v1.n_crq_elem = num_entries;
1748         tx_buff->num_entries = num_entries;
1749         /* flush buffer if current entry can not fit */
1750         if (num_entries + ind_bufp->index > IBMVNIC_MAX_IND_DESCS) {
1751                 lpar_rc = ibmvnic_tx_scrq_flush(adapter, tx_scrq);
1752                 if (lpar_rc != H_SUCCESS)
1753                         goto tx_flush_err;
1754         }
1755
1756         indir_arr[0] = tx_crq;
1757         memcpy(&ind_bufp->indir_arr[ind_bufp->index], &indir_arr[0],
1758                num_entries * sizeof(struct ibmvnic_generic_scrq));
1759         ind_bufp->index += num_entries;
1760         if (__netdev_tx_sent_queue(txq, skb->len,
1761                                    netdev_xmit_more() &&
1762                                    ind_bufp->index < IBMVNIC_MAX_IND_DESCS)) {
1763                 lpar_rc = ibmvnic_tx_scrq_flush(adapter, tx_scrq);
1764                 if (lpar_rc != H_SUCCESS)
1765                         goto tx_err;
1766         }
1767
1768         if (atomic_add_return(num_entries, &tx_scrq->used)
1769                                         >= adapter->req_tx_entries_per_subcrq) {
1770                 netdev_dbg(netdev, "Stopping queue %d\n", queue_num);
1771                 netif_stop_subqueue(netdev, queue_num);
1772         }
1773
1774         tx_packets++;
1775         tx_bytes += skb->len;
1776         txq->trans_start = jiffies;
1777         ret = NETDEV_TX_OK;
1778         goto out;
1779
1780 tx_flush_err:
1781         dev_kfree_skb_any(skb);
1782         tx_buff->skb = NULL;
1783         tx_pool->consumer_index = tx_pool->consumer_index == 0 ?
1784                                   tx_pool->num_buffers - 1 :
1785                                   tx_pool->consumer_index - 1;
1786         tx_dropped++;
1787 tx_err:
1788         if (lpar_rc != H_CLOSED && lpar_rc != H_PARAMETER)
1789                 dev_err_ratelimited(dev, "tx: send failed\n");
1790
1791         if (lpar_rc == H_CLOSED || adapter->failover_pending) {
1792                 /* Disable TX and report carrier off if queue is closed
1793                  * or pending failover.
1794                  * Firmware guarantees that a signal will be sent to the
1795                  * driver, triggering a reset or some other action.
1796                  */
1797                 netif_tx_stop_all_queues(netdev);
1798                 netif_carrier_off(netdev);
1799         }
1800 out:
1801         netdev->stats.tx_dropped += tx_dropped;
1802         netdev->stats.tx_bytes += tx_bytes;
1803         netdev->stats.tx_packets += tx_packets;
1804         adapter->tx_send_failed += tx_send_failed;
1805         adapter->tx_map_failed += tx_map_failed;
1806         adapter->tx_stats_buffers[queue_num].packets += tx_packets;
1807         adapter->tx_stats_buffers[queue_num].bytes += tx_bytes;
1808         adapter->tx_stats_buffers[queue_num].dropped_packets += tx_dropped;
1809
1810         return ret;
1811 }
1812
1813 static void ibmvnic_set_multi(struct net_device *netdev)
1814 {
1815         struct ibmvnic_adapter *adapter = netdev_priv(netdev);
1816         struct netdev_hw_addr *ha;
1817         union ibmvnic_crq crq;
1818
1819         memset(&crq, 0, sizeof(crq));
1820         crq.request_capability.first = IBMVNIC_CRQ_CMD;
1821         crq.request_capability.cmd = REQUEST_CAPABILITY;
1822
1823         if (netdev->flags & IFF_PROMISC) {
1824                 if (!adapter->promisc_supported)
1825                         return;
1826         } else {
1827                 if (netdev->flags & IFF_ALLMULTI) {
1828                         /* Accept all multicast */
1829                         memset(&crq, 0, sizeof(crq));
1830                         crq.multicast_ctrl.first = IBMVNIC_CRQ_CMD;
1831                         crq.multicast_ctrl.cmd = MULTICAST_CTRL;
1832                         crq.multicast_ctrl.flags = IBMVNIC_ENABLE_ALL;
1833                         ibmvnic_send_crq(adapter, &crq);
1834                 } else if (netdev_mc_empty(netdev)) {
1835                         /* Reject all multicast */
1836                         memset(&crq, 0, sizeof(crq));
1837                         crq.multicast_ctrl.first = IBMVNIC_CRQ_CMD;
1838                         crq.multicast_ctrl.cmd = MULTICAST_CTRL;
1839                         crq.multicast_ctrl.flags = IBMVNIC_DISABLE_ALL;
1840                         ibmvnic_send_crq(adapter, &crq);
1841                 } else {
1842                         /* Accept one or more multicast(s) */
1843                         netdev_for_each_mc_addr(ha, netdev) {
1844                                 memset(&crq, 0, sizeof(crq));
1845                                 crq.multicast_ctrl.first = IBMVNIC_CRQ_CMD;
1846                                 crq.multicast_ctrl.cmd = MULTICAST_CTRL;
1847                                 crq.multicast_ctrl.flags = IBMVNIC_ENABLE_MC;
1848                                 ether_addr_copy(&crq.multicast_ctrl.mac_addr[0],
1849                                                 ha->addr);
1850                                 ibmvnic_send_crq(adapter, &crq);
1851                         }
1852                 }
1853         }
1854 }
1855
1856 static int __ibmvnic_set_mac(struct net_device *netdev, u8 *dev_addr)
1857 {
1858         struct ibmvnic_adapter *adapter = netdev_priv(netdev);
1859         union ibmvnic_crq crq;
1860         int rc;
1861
1862         if (!is_valid_ether_addr(dev_addr)) {
1863                 rc = -EADDRNOTAVAIL;
1864                 goto err;
1865         }
1866
1867         memset(&crq, 0, sizeof(crq));
1868         crq.change_mac_addr.first = IBMVNIC_CRQ_CMD;
1869         crq.change_mac_addr.cmd = CHANGE_MAC_ADDR;
1870         ether_addr_copy(&crq.change_mac_addr.mac_addr[0], dev_addr);
1871
1872         mutex_lock(&adapter->fw_lock);
1873         adapter->fw_done_rc = 0;
1874         reinit_completion(&adapter->fw_done);
1875
1876         rc = ibmvnic_send_crq(adapter, &crq);
1877         if (rc) {
1878                 rc = -EIO;
1879                 mutex_unlock(&adapter->fw_lock);
1880                 goto err;
1881         }
1882
1883         rc = ibmvnic_wait_for_completion(adapter, &adapter->fw_done, 10000);
1884         /* netdev->dev_addr is changed in handle_change_mac_rsp function */
1885         if (rc || adapter->fw_done_rc) {
1886                 rc = -EIO;
1887                 mutex_unlock(&adapter->fw_lock);
1888                 goto err;
1889         }
1890         mutex_unlock(&adapter->fw_lock);
1891         return 0;
1892 err:
1893         ether_addr_copy(adapter->mac_addr, netdev->dev_addr);
1894         return rc;
1895 }
1896
1897 static int ibmvnic_set_mac(struct net_device *netdev, void *p)
1898 {
1899         struct ibmvnic_adapter *adapter = netdev_priv(netdev);
1900         struct sockaddr *addr = p;
1901         int rc;
1902
1903         rc = 0;
1904         if (!is_valid_ether_addr(addr->sa_data))
1905                 return -EADDRNOTAVAIL;
1906
1907         ether_addr_copy(adapter->mac_addr, addr->sa_data);
1908         if (adapter->state != VNIC_PROBED)
1909                 rc = __ibmvnic_set_mac(netdev, addr->sa_data);
1910
1911         return rc;
1912 }
1913
1914 /*
1915  * do_reset returns zero if we are able to keep processing reset events, or
1916  * non-zero if we hit a fatal error and must halt.
1917  */
1918 static int do_reset(struct ibmvnic_adapter *adapter,
1919                     struct ibmvnic_rwi *rwi, u32 reset_state)
1920 {
1921         u64 old_num_rx_queues, old_num_tx_queues;
1922         u64 old_num_rx_slots, old_num_tx_slots;
1923         struct net_device *netdev = adapter->netdev;
1924         int rc;
1925
1926         netdev_dbg(adapter->netdev,
1927                    "[S:%d FOP:%d] Reset reason %d, reset_state %d\n",
1928                    adapter->state, adapter->failover_pending,
1929                    rwi->reset_reason, reset_state);
1930
1931         adapter->reset_reason = rwi->reset_reason;
1932         /* requestor of VNIC_RESET_CHANGE_PARAM already has the rtnl lock */
1933         if (!(adapter->reset_reason == VNIC_RESET_CHANGE_PARAM))
1934                 rtnl_lock();
1935
1936         /* Now that we have the rtnl lock, clear any pending failover.
1937          * This will ensure ibmvnic_open() has either completed or will
1938          * block until failover is complete.
1939          */
1940         if (rwi->reset_reason == VNIC_RESET_FAILOVER)
1941                 adapter->failover_pending = false;
1942
1943         /* read the state and check (again) after getting rtnl */
1944         reset_state = adapter->state;
1945
1946         if (reset_state == VNIC_REMOVING || reset_state == VNIC_REMOVED) {
1947                 rc = -EBUSY;
1948                 goto out;
1949         }
1950
1951         netif_carrier_off(netdev);
1952
1953         old_num_rx_queues = adapter->req_rx_queues;
1954         old_num_tx_queues = adapter->req_tx_queues;
1955         old_num_rx_slots = adapter->req_rx_add_entries_per_subcrq;
1956         old_num_tx_slots = adapter->req_tx_entries_per_subcrq;
1957
1958         ibmvnic_cleanup(netdev);
1959
1960         if (reset_state == VNIC_OPEN &&
1961             adapter->reset_reason != VNIC_RESET_MOBILITY &&
1962             adapter->reset_reason != VNIC_RESET_FAILOVER) {
1963                 if (adapter->reset_reason == VNIC_RESET_CHANGE_PARAM) {
1964                         rc = __ibmvnic_close(netdev);
1965                         if (rc)
1966                                 goto out;
1967                 } else {
1968                         adapter->state = VNIC_CLOSING;
1969
1970                         /* Release the RTNL lock before link state change and
1971                          * re-acquire after the link state change to allow
1972                          * linkwatch_event to grab the RTNL lock and run during
1973                          * a reset.
1974                          */
1975                         rtnl_unlock();
1976                         rc = set_link_state(adapter, IBMVNIC_LOGICAL_LNK_DN);
1977                         rtnl_lock();
1978                         if (rc)
1979                                 goto out;
1980
1981                         if (adapter->state == VNIC_OPEN) {
1982                                 /* When we dropped rtnl, ibmvnic_open() got
1983                                  * it and noticed that we are resetting and
1984                                  * set the adapter state to OPEN. Update our
1985                                  * new "target" state, and resume the reset
1986                                  * from VNIC_CLOSING state.
1987                                  */
1988                                 netdev_dbg(netdev,
1989                                            "Open changed state from %d, updating.\n",
1990                                            reset_state);
1991                                 reset_state = VNIC_OPEN;
1992                                 adapter->state = VNIC_CLOSING;
1993                         }
1994
1995                         if (adapter->state != VNIC_CLOSING) {
1996                                 /* If someone else changed the adapter state
1997                                  * when we dropped the rtnl, fail the reset
1998                                  */
1999                                 rc = -1;
2000                                 goto out;
2001                         }
2002                         adapter->state = VNIC_CLOSED;
2003                 }
2004         }
2005
2006         if (adapter->reset_reason == VNIC_RESET_CHANGE_PARAM) {
2007                 release_resources(adapter);
2008                 release_sub_crqs(adapter, 1);
2009                 release_crq_queue(adapter);
2010         }
2011
2012         if (adapter->reset_reason != VNIC_RESET_NON_FATAL) {
2013                 /* remove the closed state so when we call open it appears
2014                  * we are coming from the probed state.
2015                  */
2016                 adapter->state = VNIC_PROBED;
2017
2018                 if (adapter->reset_reason == VNIC_RESET_CHANGE_PARAM) {
2019                         rc = init_crq_queue(adapter);
2020                 } else if (adapter->reset_reason == VNIC_RESET_MOBILITY) {
2021                         rc = ibmvnic_reenable_crq_queue(adapter);
2022                         release_sub_crqs(adapter, 1);
2023                 } else {
2024                         rc = ibmvnic_reset_crq(adapter);
2025                         if (rc == H_CLOSED || rc == H_SUCCESS) {
2026                                 rc = vio_enable_interrupts(adapter->vdev);
2027                                 if (rc)
2028                                         netdev_err(adapter->netdev,
2029                                                    "Reset failed to enable interrupts. rc=%d\n",
2030                                                    rc);
2031                         }
2032                 }
2033
2034                 if (rc) {
2035                         netdev_err(adapter->netdev,
2036                                    "Reset couldn't initialize crq. rc=%d\n", rc);
2037                         goto out;
2038                 }
2039
2040                 rc = ibmvnic_reset_init(adapter, true);
2041                 if (rc) {
2042                         rc = IBMVNIC_INIT_FAILED;
2043                         goto out;
2044                 }
2045
2046                 /* If the adapter was in PROBE state prior to the reset,
2047                  * exit here.
2048                  */
2049                 if (reset_state == VNIC_PROBED) {
2050                         rc = 0;
2051                         goto out;
2052                 }
2053
2054                 rc = ibmvnic_login(netdev);
2055                 if (rc)
2056                         goto out;
2057
2058                 if (adapter->reset_reason == VNIC_RESET_CHANGE_PARAM) {
2059                         rc = init_resources(adapter);
2060                         if (rc)
2061                                 goto out;
2062                 } else if (adapter->req_rx_queues != old_num_rx_queues ||
2063                     adapter->req_tx_queues != old_num_tx_queues ||
2064                     adapter->req_rx_add_entries_per_subcrq !=
2065                     old_num_rx_slots ||
2066                     adapter->req_tx_entries_per_subcrq !=
2067                     old_num_tx_slots ||
2068                     !adapter->rx_pool ||
2069                     !adapter->tso_pool ||
2070                     !adapter->tx_pool) {
2071                         release_rx_pools(adapter);
2072                         release_tx_pools(adapter);
2073                         release_napi(adapter);
2074                         release_vpd_data(adapter);
2075
2076                         rc = init_resources(adapter);
2077                         if (rc)
2078                                 goto out;
2079
2080                 } else {
2081                         rc = reset_tx_pools(adapter);
2082                         if (rc) {
2083                                 netdev_dbg(adapter->netdev, "reset tx pools failed (%d)\n",
2084                                            rc);
2085                                 goto out;
2086                         }
2087
2088                         rc = reset_rx_pools(adapter);
2089                         if (rc) {
2090                                 netdev_dbg(adapter->netdev, "reset rx pools failed (%d)\n",
2091                                            rc);
2092                                 goto out;
2093                         }
2094                 }
2095                 ibmvnic_disable_irqs(adapter);
2096         }
2097         adapter->state = VNIC_CLOSED;
2098
2099         if (reset_state == VNIC_CLOSED) {
2100                 rc = 0;
2101                 goto out;
2102         }
2103
2104         rc = __ibmvnic_open(netdev);
2105         if (rc) {
2106                 rc = IBMVNIC_OPEN_FAILED;
2107                 goto out;
2108         }
2109
2110         /* refresh device's multicast list */
2111         ibmvnic_set_multi(netdev);
2112
2113         if (adapter->reset_reason == VNIC_RESET_FAILOVER ||
2114             adapter->reset_reason == VNIC_RESET_MOBILITY)
2115                 __netdev_notify_peers(netdev);
2116
2117         rc = 0;
2118
2119 out:
2120         /* restore the adapter state if reset failed */
2121         if (rc)
2122                 adapter->state = reset_state;
2123         /* requestor of VNIC_RESET_CHANGE_PARAM should still hold the rtnl lock */
2124         if (!(adapter->reset_reason == VNIC_RESET_CHANGE_PARAM))
2125                 rtnl_unlock();
2126
2127         netdev_dbg(adapter->netdev, "[S:%d FOP:%d] Reset done, rc %d\n",
2128                    adapter->state, adapter->failover_pending, rc);
2129         return rc;
2130 }
2131
2132 static int do_hard_reset(struct ibmvnic_adapter *adapter,
2133                          struct ibmvnic_rwi *rwi, u32 reset_state)
2134 {
2135         struct net_device *netdev = adapter->netdev;
2136         int rc;
2137
2138         netdev_dbg(adapter->netdev, "Hard resetting driver (%d)\n",
2139                    rwi->reset_reason);
2140
2141         /* read the state and check (again) after getting rtnl */
2142         reset_state = adapter->state;
2143
2144         if (reset_state == VNIC_REMOVING || reset_state == VNIC_REMOVED) {
2145                 rc = -EBUSY;
2146                 goto out;
2147         }
2148
2149         netif_carrier_off(netdev);
2150         adapter->reset_reason = rwi->reset_reason;
2151
2152         ibmvnic_cleanup(netdev);
2153         release_resources(adapter);
2154         release_sub_crqs(adapter, 0);
2155         release_crq_queue(adapter);
2156
2157         /* remove the closed state so when we call open it appears
2158          * we are coming from the probed state.
2159          */
2160         adapter->state = VNIC_PROBED;
2161
2162         reinit_completion(&adapter->init_done);
2163         rc = init_crq_queue(adapter);
2164         if (rc) {
2165                 netdev_err(adapter->netdev,
2166                            "Couldn't initialize crq. rc=%d\n", rc);
2167                 goto out;
2168         }
2169
2170         rc = ibmvnic_reset_init(adapter, false);
2171         if (rc)
2172                 goto out;
2173
2174         /* If the adapter was in PROBE state prior to the reset,
2175          * exit here.
2176          */
2177         if (reset_state == VNIC_PROBED)
2178                 goto out;
2179
2180         rc = ibmvnic_login(netdev);
2181         if (rc)
2182                 goto out;
2183
2184         rc = init_resources(adapter);
2185         if (rc)
2186                 goto out;
2187
2188         ibmvnic_disable_irqs(adapter);
2189         adapter->state = VNIC_CLOSED;
2190
2191         if (reset_state == VNIC_CLOSED)
2192                 goto out;
2193
2194         rc = __ibmvnic_open(netdev);
2195         if (rc) {
2196                 rc = IBMVNIC_OPEN_FAILED;
2197                 goto out;
2198         }
2199
2200         __netdev_notify_peers(netdev);
2201 out:
2202         /* restore adapter state if reset failed */
2203         if (rc)
2204                 adapter->state = reset_state;
2205         netdev_dbg(adapter->netdev, "[S:%d FOP:%d] Hard reset done, rc %d\n",
2206                    adapter->state, adapter->failover_pending, rc);
2207         return rc;
2208 }
2209
2210 static struct ibmvnic_rwi *get_next_rwi(struct ibmvnic_adapter *adapter)
2211 {
2212         struct ibmvnic_rwi *rwi;
2213         unsigned long flags;
2214
2215         spin_lock_irqsave(&adapter->rwi_lock, flags);
2216
2217         if (!list_empty(&adapter->rwi_list)) {
2218                 rwi = list_first_entry(&adapter->rwi_list, struct ibmvnic_rwi,
2219                                        list);
2220                 list_del(&rwi->list);
2221         } else {
2222                 rwi = NULL;
2223         }
2224
2225         spin_unlock_irqrestore(&adapter->rwi_lock, flags);
2226         return rwi;
2227 }
2228
2229 static void __ibmvnic_reset(struct work_struct *work)
2230 {
2231         struct ibmvnic_rwi *rwi;
2232         struct ibmvnic_adapter *adapter;
2233         bool saved_state = false;
2234         unsigned long flags;
2235         u32 reset_state;
2236         int rc = 0;
2237
2238         adapter = container_of(work, struct ibmvnic_adapter, ibmvnic_reset);
2239
2240         if (test_and_set_bit_lock(0, &adapter->resetting)) {
2241                 schedule_delayed_work(&adapter->ibmvnic_delayed_reset,
2242                                       IBMVNIC_RESET_DELAY);
2243                 return;
2244         }
2245
2246         rwi = get_next_rwi(adapter);
2247         while (rwi) {
2248                 spin_lock_irqsave(&adapter->state_lock, flags);
2249
2250                 if (adapter->state == VNIC_REMOVING ||
2251                     adapter->state == VNIC_REMOVED) {
2252                         spin_unlock_irqrestore(&adapter->state_lock, flags);
2253                         kfree(rwi);
2254                         rc = EBUSY;
2255                         break;
2256                 }
2257
2258                 if (!saved_state) {
2259                         reset_state = adapter->state;
2260                         saved_state = true;
2261                 }
2262                 spin_unlock_irqrestore(&adapter->state_lock, flags);
2263
2264                 if (adapter->force_reset_recovery) {
2265                         /* Since we are doing a hard reset now, clear the
2266                          * failover_pending flag so we don't ignore any
2267                          * future MOBILITY or other resets.
2268                          */
2269                         adapter->failover_pending = false;
2270
2271                         /* Transport event occurred during previous reset */
2272                         if (adapter->wait_for_reset) {
2273                                 /* Previous was CHANGE_PARAM; caller locked */
2274                                 adapter->force_reset_recovery = false;
2275                                 rc = do_hard_reset(adapter, rwi, reset_state);
2276                         } else {
2277                                 rtnl_lock();
2278                                 adapter->force_reset_recovery = false;
2279                                 rc = do_hard_reset(adapter, rwi, reset_state);
2280                                 rtnl_unlock();
2281                         }
2282                         if (rc) {
2283                                 /* give backing device time to settle down */
2284                                 netdev_dbg(adapter->netdev,
2285                                            "[S:%d] Hard reset failed, waiting 60 secs\n",
2286                                            adapter->state);
2287                                 set_current_state(TASK_UNINTERRUPTIBLE);
2288                                 schedule_timeout(60 * HZ);
2289                         }
2290                 } else {
2291                         rc = do_reset(adapter, rwi, reset_state);
2292                 }
2293                 kfree(rwi);
2294                 adapter->last_reset_time = jiffies;
2295
2296                 if (rc)
2297                         netdev_dbg(adapter->netdev, "Reset failed, rc=%d\n", rc);
2298
2299                 rwi = get_next_rwi(adapter);
2300
2301                 if (rwi && (rwi->reset_reason == VNIC_RESET_FAILOVER ||
2302                             rwi->reset_reason == VNIC_RESET_MOBILITY))
2303                         adapter->force_reset_recovery = true;
2304         }
2305
2306         if (adapter->wait_for_reset) {
2307                 adapter->reset_done_rc = rc;
2308                 complete(&adapter->reset_done);
2309         }
2310
2311         clear_bit_unlock(0, &adapter->resetting);
2312
2313         netdev_dbg(adapter->netdev,
2314                    "[S:%d FRR:%d WFR:%d] Done processing resets\n",
2315                    adapter->state, adapter->force_reset_recovery,
2316                    adapter->wait_for_reset);
2317 }
2318
2319 static void __ibmvnic_delayed_reset(struct work_struct *work)
2320 {
2321         struct ibmvnic_adapter *adapter;
2322
2323         adapter = container_of(work, struct ibmvnic_adapter,
2324                                ibmvnic_delayed_reset.work);
2325         __ibmvnic_reset(&adapter->ibmvnic_reset);
2326 }
2327
2328 static int ibmvnic_reset(struct ibmvnic_adapter *adapter,
2329                          enum ibmvnic_reset_reason reason)
2330 {
2331         struct list_head *entry, *tmp_entry;
2332         struct ibmvnic_rwi *rwi, *tmp;
2333         struct net_device *netdev = adapter->netdev;
2334         unsigned long flags;
2335         int ret;
2336
2337         spin_lock_irqsave(&adapter->rwi_lock, flags);
2338
2339         /* If failover is pending don't schedule any other reset.
2340          * Instead let the failover complete. If there is already a
2341          * a failover reset scheduled, we will detect and drop the
2342          * duplicate reset when walking the ->rwi_list below.
2343          */
2344         if (adapter->state == VNIC_REMOVING ||
2345             adapter->state == VNIC_REMOVED ||
2346             (adapter->failover_pending && reason != VNIC_RESET_FAILOVER)) {
2347                 ret = EBUSY;
2348                 netdev_dbg(netdev, "Adapter removing or pending failover, skipping reset\n");
2349                 goto err;
2350         }
2351
2352         if (adapter->state == VNIC_PROBING) {
2353                 netdev_warn(netdev, "Adapter reset during probe\n");
2354                 adapter->init_done_rc = EAGAIN;
2355                 ret = EAGAIN;
2356                 goto err;
2357         }
2358
2359         list_for_each(entry, &adapter->rwi_list) {
2360                 tmp = list_entry(entry, struct ibmvnic_rwi, list);
2361                 if (tmp->reset_reason == reason) {
2362                         netdev_dbg(netdev, "Skipping matching reset, reason=%d\n",
2363                                    reason);
2364                         ret = EBUSY;
2365                         goto err;
2366                 }
2367         }
2368
2369         rwi = kzalloc(sizeof(*rwi), GFP_ATOMIC);
2370         if (!rwi) {
2371                 ret = ENOMEM;
2372                 goto err;
2373         }
2374         /* if we just received a transport event,
2375          * flush reset queue and process this reset
2376          */
2377         if (adapter->force_reset_recovery && !list_empty(&adapter->rwi_list)) {
2378                 list_for_each_safe(entry, tmp_entry, &adapter->rwi_list)
2379                         list_del(entry);
2380         }
2381         rwi->reset_reason = reason;
2382         list_add_tail(&rwi->list, &adapter->rwi_list);
2383         netdev_dbg(adapter->netdev, "Scheduling reset (reason %d)\n", reason);
2384         schedule_work(&adapter->ibmvnic_reset);
2385
2386         ret = 0;
2387 err:
2388         /* ibmvnic_close() below can block, so drop the lock first */
2389         spin_unlock_irqrestore(&adapter->rwi_lock, flags);
2390
2391         if (ret == ENOMEM)
2392                 ibmvnic_close(netdev);
2393
2394         return -ret;
2395 }
2396
2397 static void ibmvnic_tx_timeout(struct net_device *dev, unsigned int txqueue)
2398 {
2399         struct ibmvnic_adapter *adapter = netdev_priv(dev);
2400
2401         if (test_bit(0, &adapter->resetting)) {
2402                 netdev_err(adapter->netdev,
2403                            "Adapter is resetting, skip timeout reset\n");
2404                 return;
2405         }
2406         /* No queuing up reset until at least 5 seconds (default watchdog val)
2407          * after last reset
2408          */
2409         if (time_before(jiffies, (adapter->last_reset_time + dev->watchdog_timeo))) {
2410                 netdev_dbg(dev, "Not yet time to tx timeout.\n");
2411                 return;
2412         }
2413         ibmvnic_reset(adapter, VNIC_RESET_TIMEOUT);
2414 }
2415
2416 static void remove_buff_from_pool(struct ibmvnic_adapter *adapter,
2417                                   struct ibmvnic_rx_buff *rx_buff)
2418 {
2419         struct ibmvnic_rx_pool *pool = &adapter->rx_pool[rx_buff->pool_index];
2420
2421         rx_buff->skb = NULL;
2422
2423         pool->free_map[pool->next_alloc] = (int)(rx_buff - pool->rx_buff);
2424         pool->next_alloc = (pool->next_alloc + 1) % pool->size;
2425
2426         atomic_dec(&pool->available);
2427 }
2428
2429 static int ibmvnic_poll(struct napi_struct *napi, int budget)
2430 {
2431         struct ibmvnic_sub_crq_queue *rx_scrq;
2432         struct ibmvnic_adapter *adapter;
2433         struct net_device *netdev;
2434         int frames_processed;
2435         int scrq_num;
2436
2437         netdev = napi->dev;
2438         adapter = netdev_priv(netdev);
2439         scrq_num = (int)(napi - adapter->napi);
2440         frames_processed = 0;
2441         rx_scrq = adapter->rx_scrq[scrq_num];
2442
2443 restart_poll:
2444         while (frames_processed < budget) {
2445                 struct sk_buff *skb;
2446                 struct ibmvnic_rx_buff *rx_buff;
2447                 union sub_crq *next;
2448                 u32 length;
2449                 u16 offset;
2450                 u8 flags = 0;
2451
2452                 if (unlikely(test_bit(0, &adapter->resetting) &&
2453                              adapter->reset_reason != VNIC_RESET_NON_FATAL)) {
2454                         enable_scrq_irq(adapter, rx_scrq);
2455                         napi_complete_done(napi, frames_processed);
2456                         return frames_processed;
2457                 }
2458
2459                 if (!pending_scrq(adapter, rx_scrq))
2460                         break;
2461                 next = ibmvnic_next_scrq(adapter, rx_scrq);
2462                 rx_buff = (struct ibmvnic_rx_buff *)
2463                           be64_to_cpu(next->rx_comp.correlator);
2464                 /* do error checking */
2465                 if (next->rx_comp.rc) {
2466                         netdev_dbg(netdev, "rx buffer returned with rc %x\n",
2467                                    be16_to_cpu(next->rx_comp.rc));
2468                         /* free the entry */
2469                         next->rx_comp.first = 0;
2470                         dev_kfree_skb_any(rx_buff->skb);
2471                         remove_buff_from_pool(adapter, rx_buff);
2472                         continue;
2473                 } else if (!rx_buff->skb) {
2474                         /* free the entry */
2475                         next->rx_comp.first = 0;
2476                         remove_buff_from_pool(adapter, rx_buff);
2477                         continue;
2478                 }
2479
2480                 length = be32_to_cpu(next->rx_comp.len);
2481                 offset = be16_to_cpu(next->rx_comp.off_frame_data);
2482                 flags = next->rx_comp.flags;
2483                 skb = rx_buff->skb;
2484                 /* load long_term_buff before copying to skb */
2485                 dma_rmb();
2486                 skb_copy_to_linear_data(skb, rx_buff->data + offset,
2487                                         length);
2488
2489                 /* VLAN Header has been stripped by the system firmware and
2490                  * needs to be inserted by the driver
2491                  */
2492                 if (adapter->rx_vlan_header_insertion &&
2493                     (flags & IBMVNIC_VLAN_STRIPPED))
2494                         __vlan_hwaccel_put_tag(skb, htons(ETH_P_8021Q),
2495                                                ntohs(next->rx_comp.vlan_tci));
2496
2497                 /* free the entry */
2498                 next->rx_comp.first = 0;
2499                 remove_buff_from_pool(adapter, rx_buff);
2500
2501                 skb_put(skb, length);
2502                 skb->protocol = eth_type_trans(skb, netdev);
2503                 skb_record_rx_queue(skb, scrq_num);
2504
2505                 if (flags & IBMVNIC_IP_CHKSUM_GOOD &&
2506                     flags & IBMVNIC_TCP_UDP_CHKSUM_GOOD) {
2507                         skb->ip_summed = CHECKSUM_UNNECESSARY;
2508                 }
2509
2510                 length = skb->len;
2511                 napi_gro_receive(napi, skb); /* send it up */
2512                 netdev->stats.rx_packets++;
2513                 netdev->stats.rx_bytes += length;
2514                 adapter->rx_stats_buffers[scrq_num].packets++;
2515                 adapter->rx_stats_buffers[scrq_num].bytes += length;
2516                 frames_processed++;
2517         }
2518
2519         if (adapter->state != VNIC_CLOSING &&
2520             ((atomic_read(&adapter->rx_pool[scrq_num].available) <
2521               adapter->req_rx_add_entries_per_subcrq / 2) ||
2522               frames_processed < budget))
2523                 replenish_rx_pool(adapter, &adapter->rx_pool[scrq_num]);
2524         if (frames_processed < budget) {
2525                 if (napi_complete_done(napi, frames_processed)) {
2526                         enable_scrq_irq(adapter, rx_scrq);
2527                         if (pending_scrq(adapter, rx_scrq)) {
2528                                 if (napi_reschedule(napi)) {
2529                                         disable_scrq_irq(adapter, rx_scrq);
2530                                         goto restart_poll;
2531                                 }
2532                         }
2533                 }
2534         }
2535         return frames_processed;
2536 }
2537
2538 static int wait_for_reset(struct ibmvnic_adapter *adapter)
2539 {
2540         int rc, ret;
2541
2542         adapter->fallback.mtu = adapter->req_mtu;
2543         adapter->fallback.rx_queues = adapter->req_rx_queues;
2544         adapter->fallback.tx_queues = adapter->req_tx_queues;
2545         adapter->fallback.rx_entries = adapter->req_rx_add_entries_per_subcrq;
2546         adapter->fallback.tx_entries = adapter->req_tx_entries_per_subcrq;
2547
2548         reinit_completion(&adapter->reset_done);
2549         adapter->wait_for_reset = true;
2550         rc = ibmvnic_reset(adapter, VNIC_RESET_CHANGE_PARAM);
2551
2552         if (rc) {
2553                 ret = rc;
2554                 goto out;
2555         }
2556         rc = ibmvnic_wait_for_completion(adapter, &adapter->reset_done, 60000);
2557         if (rc) {
2558                 ret = -ENODEV;
2559                 goto out;
2560         }
2561
2562         ret = 0;
2563         if (adapter->reset_done_rc) {
2564                 ret = -EIO;
2565                 adapter->desired.mtu = adapter->fallback.mtu;
2566                 adapter->desired.rx_queues = adapter->fallback.rx_queues;
2567                 adapter->desired.tx_queues = adapter->fallback.tx_queues;
2568                 adapter->desired.rx_entries = adapter->fallback.rx_entries;
2569                 adapter->desired.tx_entries = adapter->fallback.tx_entries;
2570
2571                 reinit_completion(&adapter->reset_done);
2572                 adapter->wait_for_reset = true;
2573                 rc = ibmvnic_reset(adapter, VNIC_RESET_CHANGE_PARAM);
2574                 if (rc) {
2575                         ret = rc;
2576                         goto out;
2577                 }
2578                 rc = ibmvnic_wait_for_completion(adapter, &adapter->reset_done,
2579                                                  60000);
2580                 if (rc) {
2581                         ret = -ENODEV;
2582                         goto out;
2583                 }
2584         }
2585 out:
2586         adapter->wait_for_reset = false;
2587
2588         return ret;
2589 }
2590
2591 static int ibmvnic_change_mtu(struct net_device *netdev, int new_mtu)
2592 {
2593         struct ibmvnic_adapter *adapter = netdev_priv(netdev);
2594
2595         adapter->desired.mtu = new_mtu + ETH_HLEN;
2596
2597         return wait_for_reset(adapter);
2598 }
2599
2600 static netdev_features_t ibmvnic_features_check(struct sk_buff *skb,
2601                                                 struct net_device *dev,
2602                                                 netdev_features_t features)
2603 {
2604         /* Some backing hardware adapters can not
2605          * handle packets with a MSS less than 224
2606          * or with only one segment.
2607          */
2608         if (skb_is_gso(skb)) {
2609                 if (skb_shinfo(skb)->gso_size < 224 ||
2610                     skb_shinfo(skb)->gso_segs == 1)
2611                         features &= ~NETIF_F_GSO_MASK;
2612         }
2613
2614         return features;
2615 }
2616
2617 static const struct net_device_ops ibmvnic_netdev_ops = {
2618         .ndo_open               = ibmvnic_open,
2619         .ndo_stop               = ibmvnic_close,
2620         .ndo_start_xmit         = ibmvnic_xmit,
2621         .ndo_set_rx_mode        = ibmvnic_set_multi,
2622         .ndo_set_mac_address    = ibmvnic_set_mac,
2623         .ndo_validate_addr      = eth_validate_addr,
2624         .ndo_tx_timeout         = ibmvnic_tx_timeout,
2625         .ndo_change_mtu         = ibmvnic_change_mtu,
2626         .ndo_features_check     = ibmvnic_features_check,
2627 };
2628
2629 /* ethtool functions */
2630
2631 static int ibmvnic_get_link_ksettings(struct net_device *netdev,
2632                                       struct ethtool_link_ksettings *cmd)
2633 {
2634         struct ibmvnic_adapter *adapter = netdev_priv(netdev);
2635         int rc;
2636
2637         rc = send_query_phys_parms(adapter);
2638         if (rc) {
2639                 adapter->speed = SPEED_UNKNOWN;
2640                 adapter->duplex = DUPLEX_UNKNOWN;
2641         }
2642         cmd->base.speed = adapter->speed;
2643         cmd->base.duplex = adapter->duplex;
2644         cmd->base.port = PORT_FIBRE;
2645         cmd->base.phy_address = 0;
2646         cmd->base.autoneg = AUTONEG_ENABLE;
2647
2648         return 0;
2649 }
2650
2651 static void ibmvnic_get_drvinfo(struct net_device *netdev,
2652                                 struct ethtool_drvinfo *info)
2653 {
2654         struct ibmvnic_adapter *adapter = netdev_priv(netdev);
2655
2656         strscpy(info->driver, ibmvnic_driver_name, sizeof(info->driver));
2657         strscpy(info->version, IBMVNIC_DRIVER_VERSION, sizeof(info->version));
2658         strscpy(info->fw_version, adapter->fw_version,
2659                 sizeof(info->fw_version));
2660 }
2661
2662 static u32 ibmvnic_get_msglevel(struct net_device *netdev)
2663 {
2664         struct ibmvnic_adapter *adapter = netdev_priv(netdev);
2665
2666         return adapter->msg_enable;
2667 }
2668
2669 static void ibmvnic_set_msglevel(struct net_device *netdev, u32 data)
2670 {
2671         struct ibmvnic_adapter *adapter = netdev_priv(netdev);
2672
2673         adapter->msg_enable = data;
2674 }
2675
2676 static u32 ibmvnic_get_link(struct net_device *netdev)
2677 {
2678         struct ibmvnic_adapter *adapter = netdev_priv(netdev);
2679
2680         /* Don't need to send a query because we request a logical link up at
2681          * init and then we wait for link state indications
2682          */
2683         return adapter->logical_link_state;
2684 }
2685
2686 static void ibmvnic_get_ringparam(struct net_device *netdev,
2687                                   struct ethtool_ringparam *ring)
2688 {
2689         struct ibmvnic_adapter *adapter = netdev_priv(netdev);
2690
2691         if (adapter->priv_flags & IBMVNIC_USE_SERVER_MAXES) {
2692                 ring->rx_max_pending = adapter->max_rx_add_entries_per_subcrq;
2693                 ring->tx_max_pending = adapter->max_tx_entries_per_subcrq;
2694         } else {
2695                 ring->rx_max_pending = IBMVNIC_MAX_QUEUE_SZ;
2696                 ring->tx_max_pending = IBMVNIC_MAX_QUEUE_SZ;
2697         }
2698         ring->rx_mini_max_pending = 0;
2699         ring->rx_jumbo_max_pending = 0;
2700         ring->rx_pending = adapter->req_rx_add_entries_per_subcrq;
2701         ring->tx_pending = adapter->req_tx_entries_per_subcrq;
2702         ring->rx_mini_pending = 0;
2703         ring->rx_jumbo_pending = 0;
2704 }
2705
2706 static int ibmvnic_set_ringparam(struct net_device *netdev,
2707                                  struct ethtool_ringparam *ring)
2708 {
2709         struct ibmvnic_adapter *adapter = netdev_priv(netdev);
2710         int ret;
2711
2712         ret = 0;
2713         adapter->desired.rx_entries = ring->rx_pending;
2714         adapter->desired.tx_entries = ring->tx_pending;
2715
2716         ret = wait_for_reset(adapter);
2717
2718         if (!ret &&
2719             (adapter->req_rx_add_entries_per_subcrq != ring->rx_pending ||
2720              adapter->req_tx_entries_per_subcrq != ring->tx_pending))
2721                 netdev_info(netdev,
2722                             "Could not match full ringsize request. Requested: RX %d, TX %d; Allowed: RX %llu, TX %llu\n",
2723                             ring->rx_pending, ring->tx_pending,
2724                             adapter->req_rx_add_entries_per_subcrq,
2725                             adapter->req_tx_entries_per_subcrq);
2726         return ret;
2727 }
2728
2729 static void ibmvnic_get_channels(struct net_device *netdev,
2730                                  struct ethtool_channels *channels)
2731 {
2732         struct ibmvnic_adapter *adapter = netdev_priv(netdev);
2733
2734         if (adapter->priv_flags & IBMVNIC_USE_SERVER_MAXES) {
2735                 channels->max_rx = adapter->max_rx_queues;
2736                 channels->max_tx = adapter->max_tx_queues;
2737         } else {
2738                 channels->max_rx = IBMVNIC_MAX_QUEUES;
2739                 channels->max_tx = IBMVNIC_MAX_QUEUES;
2740         }
2741
2742         channels->max_other = 0;
2743         channels->max_combined = 0;
2744         channels->rx_count = adapter->req_rx_queues;
2745         channels->tx_count = adapter->req_tx_queues;
2746         channels->other_count = 0;
2747         channels->combined_count = 0;
2748 }
2749
2750 static int ibmvnic_set_channels(struct net_device *netdev,
2751                                 struct ethtool_channels *channels)
2752 {
2753         struct ibmvnic_adapter *adapter = netdev_priv(netdev);
2754         int ret;
2755
2756         ret = 0;
2757         adapter->desired.rx_queues = channels->rx_count;
2758         adapter->desired.tx_queues = channels->tx_count;
2759
2760         ret = wait_for_reset(adapter);
2761
2762         if (!ret &&
2763             (adapter->req_rx_queues != channels->rx_count ||
2764              adapter->req_tx_queues != channels->tx_count))
2765                 netdev_info(netdev,
2766                             "Could not match full channels request. Requested: RX %d, TX %d; Allowed: RX %llu, TX %llu\n",
2767                             channels->rx_count, channels->tx_count,
2768                             adapter->req_rx_queues, adapter->req_tx_queues);
2769         return ret;
2770 }
2771
2772 static void ibmvnic_get_strings(struct net_device *dev, u32 stringset, u8 *data)
2773 {
2774         struct ibmvnic_adapter *adapter = netdev_priv(dev);
2775         int i;
2776
2777         switch (stringset) {
2778         case ETH_SS_STATS:
2779                 for (i = 0; i < ARRAY_SIZE(ibmvnic_stats);
2780                                 i++, data += ETH_GSTRING_LEN)
2781                         memcpy(data, ibmvnic_stats[i].name, ETH_GSTRING_LEN);
2782
2783                 for (i = 0; i < adapter->req_tx_queues; i++) {
2784                         snprintf(data, ETH_GSTRING_LEN, "tx%d_packets", i);
2785                         data += ETH_GSTRING_LEN;
2786
2787                         snprintf(data, ETH_GSTRING_LEN, "tx%d_bytes", i);
2788                         data += ETH_GSTRING_LEN;
2789
2790                         snprintf(data, ETH_GSTRING_LEN,
2791                                  "tx%d_dropped_packets", i);
2792                         data += ETH_GSTRING_LEN;
2793                 }
2794
2795                 for (i = 0; i < adapter->req_rx_queues; i++) {
2796                         snprintf(data, ETH_GSTRING_LEN, "rx%d_packets", i);
2797                         data += ETH_GSTRING_LEN;
2798
2799                         snprintf(data, ETH_GSTRING_LEN, "rx%d_bytes", i);
2800                         data += ETH_GSTRING_LEN;
2801
2802                         snprintf(data, ETH_GSTRING_LEN, "rx%d_interrupts", i);
2803                         data += ETH_GSTRING_LEN;
2804                 }
2805                 break;
2806
2807         case ETH_SS_PRIV_FLAGS:
2808                 for (i = 0; i < ARRAY_SIZE(ibmvnic_priv_flags); i++)
2809                         strcpy(data + i * ETH_GSTRING_LEN,
2810                                ibmvnic_priv_flags[i]);
2811                 break;
2812         default:
2813                 return;
2814         }
2815 }
2816
2817 static int ibmvnic_get_sset_count(struct net_device *dev, int sset)
2818 {
2819         struct ibmvnic_adapter *adapter = netdev_priv(dev);
2820
2821         switch (sset) {
2822         case ETH_SS_STATS:
2823                 return ARRAY_SIZE(ibmvnic_stats) +
2824                        adapter->req_tx_queues * NUM_TX_STATS +
2825                        adapter->req_rx_queues * NUM_RX_STATS;
2826         case ETH_SS_PRIV_FLAGS:
2827                 return ARRAY_SIZE(ibmvnic_priv_flags);
2828         default:
2829                 return -EOPNOTSUPP;
2830         }
2831 }
2832
2833 static void ibmvnic_get_ethtool_stats(struct net_device *dev,
2834                                       struct ethtool_stats *stats, u64 *data)
2835 {
2836         struct ibmvnic_adapter *adapter = netdev_priv(dev);
2837         union ibmvnic_crq crq;
2838         int i, j;
2839         int rc;
2840
2841         memset(&crq, 0, sizeof(crq));
2842         crq.request_statistics.first = IBMVNIC_CRQ_CMD;
2843         crq.request_statistics.cmd = REQUEST_STATISTICS;
2844         crq.request_statistics.ioba = cpu_to_be32(adapter->stats_token);
2845         crq.request_statistics.len =
2846             cpu_to_be32(sizeof(struct ibmvnic_statistics));
2847
2848         /* Wait for data to be written */
2849         reinit_completion(&adapter->stats_done);
2850         rc = ibmvnic_send_crq(adapter, &crq);
2851         if (rc)
2852                 return;
2853         rc = ibmvnic_wait_for_completion(adapter, &adapter->stats_done, 10000);
2854         if (rc)
2855                 return;
2856
2857         for (i = 0; i < ARRAY_SIZE(ibmvnic_stats); i++)
2858                 data[i] = be64_to_cpu(IBMVNIC_GET_STAT
2859                                       (adapter, ibmvnic_stats[i].offset));
2860
2861         for (j = 0; j < adapter->req_tx_queues; j++) {
2862                 data[i] = adapter->tx_stats_buffers[j].packets;
2863                 i++;
2864                 data[i] = adapter->tx_stats_buffers[j].bytes;
2865                 i++;
2866                 data[i] = adapter->tx_stats_buffers[j].dropped_packets;
2867                 i++;
2868         }
2869
2870         for (j = 0; j < adapter->req_rx_queues; j++) {
2871                 data[i] = adapter->rx_stats_buffers[j].packets;
2872                 i++;
2873                 data[i] = adapter->rx_stats_buffers[j].bytes;
2874                 i++;
2875                 data[i] = adapter->rx_stats_buffers[j].interrupts;
2876                 i++;
2877         }
2878 }
2879
2880 static u32 ibmvnic_get_priv_flags(struct net_device *netdev)
2881 {
2882         struct ibmvnic_adapter *adapter = netdev_priv(netdev);
2883
2884         return adapter->priv_flags;
2885 }
2886
2887 static int ibmvnic_set_priv_flags(struct net_device *netdev, u32 flags)
2888 {
2889         struct ibmvnic_adapter *adapter = netdev_priv(netdev);
2890         bool which_maxes = !!(flags & IBMVNIC_USE_SERVER_MAXES);
2891
2892         if (which_maxes)
2893                 adapter->priv_flags |= IBMVNIC_USE_SERVER_MAXES;
2894         else
2895                 adapter->priv_flags &= ~IBMVNIC_USE_SERVER_MAXES;
2896
2897         return 0;
2898 }
2899
2900 static const struct ethtool_ops ibmvnic_ethtool_ops = {
2901         .get_drvinfo            = ibmvnic_get_drvinfo,
2902         .get_msglevel           = ibmvnic_get_msglevel,
2903         .set_msglevel           = ibmvnic_set_msglevel,
2904         .get_link               = ibmvnic_get_link,
2905         .get_ringparam          = ibmvnic_get_ringparam,
2906         .set_ringparam          = ibmvnic_set_ringparam,
2907         .get_channels           = ibmvnic_get_channels,
2908         .set_channels           = ibmvnic_set_channels,
2909         .get_strings            = ibmvnic_get_strings,
2910         .get_sset_count         = ibmvnic_get_sset_count,
2911         .get_ethtool_stats      = ibmvnic_get_ethtool_stats,
2912         .get_link_ksettings     = ibmvnic_get_link_ksettings,
2913         .get_priv_flags         = ibmvnic_get_priv_flags,
2914         .set_priv_flags         = ibmvnic_set_priv_flags,
2915 };
2916
2917 /* Routines for managing CRQs/sCRQs  */
2918
2919 static int reset_one_sub_crq_queue(struct ibmvnic_adapter *adapter,
2920                                    struct ibmvnic_sub_crq_queue *scrq)
2921 {
2922         int rc;
2923
2924         if (!scrq) {
2925                 netdev_dbg(adapter->netdev, "Invalid scrq reset.\n");
2926                 return -EINVAL;
2927         }
2928
2929         if (scrq->irq) {
2930                 free_irq(scrq->irq, scrq);
2931                 irq_dispose_mapping(scrq->irq);
2932                 scrq->irq = 0;
2933         }
2934
2935         if (scrq->msgs) {
2936                 memset(scrq->msgs, 0, 4 * PAGE_SIZE);
2937                 atomic_set(&scrq->used, 0);
2938                 scrq->cur = 0;
2939                 scrq->ind_buf.index = 0;
2940         } else {
2941                 netdev_dbg(adapter->netdev, "Invalid scrq reset\n");
2942                 return -EINVAL;
2943         }
2944
2945         rc = h_reg_sub_crq(adapter->vdev->unit_address, scrq->msg_token,
2946                            4 * PAGE_SIZE, &scrq->crq_num, &scrq->hw_irq);
2947         return rc;
2948 }
2949
2950 static int reset_sub_crq_queues(struct ibmvnic_adapter *adapter)
2951 {
2952         int i, rc;
2953
2954         if (!adapter->tx_scrq || !adapter->rx_scrq)
2955                 return -EINVAL;
2956
2957         for (i = 0; i < adapter->req_tx_queues; i++) {
2958                 netdev_dbg(adapter->netdev, "Re-setting tx_scrq[%d]\n", i);
2959                 rc = reset_one_sub_crq_queue(adapter, adapter->tx_scrq[i]);
2960                 if (rc)
2961                         return rc;
2962         }
2963
2964         for (i = 0; i < adapter->req_rx_queues; i++) {
2965                 netdev_dbg(adapter->netdev, "Re-setting rx_scrq[%d]\n", i);
2966                 rc = reset_one_sub_crq_queue(adapter, adapter->rx_scrq[i]);
2967                 if (rc)
2968                         return rc;
2969         }
2970
2971         return rc;
2972 }
2973
2974 static void release_sub_crq_queue(struct ibmvnic_adapter *adapter,
2975                                   struct ibmvnic_sub_crq_queue *scrq,
2976                                   bool do_h_free)
2977 {
2978         struct device *dev = &adapter->vdev->dev;
2979         long rc;
2980
2981         netdev_dbg(adapter->netdev, "Releasing sub-CRQ\n");
2982
2983         if (do_h_free) {
2984                 /* Close the sub-crqs */
2985                 do {
2986                         rc = plpar_hcall_norets(H_FREE_SUB_CRQ,
2987                                                 adapter->vdev->unit_address,
2988                                                 scrq->crq_num);
2989                 } while (rc == H_BUSY || H_IS_LONG_BUSY(rc));
2990
2991                 if (rc) {
2992                         netdev_err(adapter->netdev,
2993                                    "Failed to release sub-CRQ %16lx, rc = %ld\n",
2994                                    scrq->crq_num, rc);
2995                 }
2996         }
2997
2998         dma_free_coherent(dev,
2999                           IBMVNIC_IND_ARR_SZ,
3000                           scrq->ind_buf.indir_arr,
3001                           scrq->ind_buf.indir_dma);
3002
3003         dma_unmap_single(dev, scrq->msg_token, 4 * PAGE_SIZE,
3004                          DMA_BIDIRECTIONAL);
3005         free_pages((unsigned long)scrq->msgs, 2);
3006         kfree(scrq);
3007 }
3008
3009 static struct ibmvnic_sub_crq_queue *init_sub_crq_queue(struct ibmvnic_adapter
3010                                                         *adapter)
3011 {
3012         struct device *dev = &adapter->vdev->dev;
3013         struct ibmvnic_sub_crq_queue *scrq;
3014         int rc;
3015
3016         scrq = kzalloc(sizeof(*scrq), GFP_KERNEL);
3017         if (!scrq)
3018                 return NULL;
3019
3020         scrq->msgs =
3021                 (union sub_crq *)__get_free_pages(GFP_KERNEL | __GFP_ZERO, 2);
3022         if (!scrq->msgs) {
3023                 dev_warn(dev, "Couldn't allocate crq queue messages page\n");
3024                 goto zero_page_failed;
3025         }
3026
3027         scrq->msg_token = dma_map_single(dev, scrq->msgs, 4 * PAGE_SIZE,
3028                                          DMA_BIDIRECTIONAL);
3029         if (dma_mapping_error(dev, scrq->msg_token)) {
3030                 dev_warn(dev, "Couldn't map crq queue messages page\n");
3031                 goto map_failed;
3032         }
3033
3034         rc = h_reg_sub_crq(adapter->vdev->unit_address, scrq->msg_token,
3035                            4 * PAGE_SIZE, &scrq->crq_num, &scrq->hw_irq);
3036
3037         if (rc == H_RESOURCE)
3038                 rc = ibmvnic_reset_crq(adapter);
3039
3040         if (rc == H_CLOSED) {
3041                 dev_warn(dev, "Partner adapter not ready, waiting.\n");
3042         } else if (rc) {
3043                 dev_warn(dev, "Error %d registering sub-crq\n", rc);
3044                 goto reg_failed;
3045         }
3046
3047         scrq->adapter = adapter;
3048         scrq->size = 4 * PAGE_SIZE / sizeof(*scrq->msgs);
3049         scrq->ind_buf.index = 0;
3050
3051         scrq->ind_buf.indir_arr =
3052                 dma_alloc_coherent(dev,
3053                                    IBMVNIC_IND_ARR_SZ,
3054                                    &scrq->ind_buf.indir_dma,
3055                                    GFP_KERNEL);
3056
3057         if (!scrq->ind_buf.indir_arr)
3058                 goto indir_failed;
3059
3060         spin_lock_init(&scrq->lock);
3061
3062         netdev_dbg(adapter->netdev,
3063                    "sub-crq initialized, num %lx, hw_irq=%lx, irq=%x\n",
3064                    scrq->crq_num, scrq->hw_irq, scrq->irq);
3065
3066         return scrq;
3067
3068 indir_failed:
3069         do {
3070                 rc = plpar_hcall_norets(H_FREE_SUB_CRQ,
3071                                         adapter->vdev->unit_address,
3072                                         scrq->crq_num);
3073         } while (rc == H_BUSY || rc == H_IS_LONG_BUSY(rc));
3074 reg_failed:
3075         dma_unmap_single(dev, scrq->msg_token, 4 * PAGE_SIZE,
3076                          DMA_BIDIRECTIONAL);
3077 map_failed:
3078         free_pages((unsigned long)scrq->msgs, 2);
3079 zero_page_failed:
3080         kfree(scrq);
3081
3082         return NULL;
3083 }
3084
3085 static void release_sub_crqs(struct ibmvnic_adapter *adapter, bool do_h_free)
3086 {
3087         int i;
3088
3089         if (adapter->tx_scrq) {
3090                 for (i = 0; i < adapter->num_active_tx_scrqs; i++) {
3091                         if (!adapter->tx_scrq[i])
3092                                 continue;
3093
3094                         netdev_dbg(adapter->netdev, "Releasing tx_scrq[%d]\n",
3095                                    i);
3096                         if (adapter->tx_scrq[i]->irq) {
3097                                 free_irq(adapter->tx_scrq[i]->irq,
3098                                          adapter->tx_scrq[i]);
3099                                 irq_dispose_mapping(adapter->tx_scrq[i]->irq);
3100                                 adapter->tx_scrq[i]->irq = 0;
3101                         }
3102
3103                         release_sub_crq_queue(adapter, adapter->tx_scrq[i],
3104                                               do_h_free);
3105                 }
3106
3107                 kfree(adapter->tx_scrq);
3108                 adapter->tx_scrq = NULL;
3109                 adapter->num_active_tx_scrqs = 0;
3110         }
3111
3112         if (adapter->rx_scrq) {
3113                 for (i = 0; i < adapter->num_active_rx_scrqs; i++) {
3114                         if (!adapter->rx_scrq[i])
3115                                 continue;
3116
3117                         netdev_dbg(adapter->netdev, "Releasing rx_scrq[%d]\n",
3118                                    i);
3119                         if (adapter->rx_scrq[i]->irq) {
3120                                 free_irq(adapter->rx_scrq[i]->irq,
3121                                          adapter->rx_scrq[i]);
3122                                 irq_dispose_mapping(adapter->rx_scrq[i]->irq);
3123                                 adapter->rx_scrq[i]->irq = 0;
3124                         }
3125
3126                         release_sub_crq_queue(adapter, adapter->rx_scrq[i],
3127                                               do_h_free);
3128                 }
3129
3130                 kfree(adapter->rx_scrq);
3131                 adapter->rx_scrq = NULL;
3132                 adapter->num_active_rx_scrqs = 0;
3133         }
3134 }
3135
3136 static int disable_scrq_irq(struct ibmvnic_adapter *adapter,
3137                             struct ibmvnic_sub_crq_queue *scrq)
3138 {
3139         struct device *dev = &adapter->vdev->dev;
3140         unsigned long rc;
3141
3142         rc = plpar_hcall_norets(H_VIOCTL, adapter->vdev->unit_address,
3143                                 H_DISABLE_VIO_INTERRUPT, scrq->hw_irq, 0, 0);
3144         if (rc)
3145                 dev_err(dev, "Couldn't disable scrq irq 0x%lx. rc=%ld\n",
3146                         scrq->hw_irq, rc);
3147         return rc;
3148 }
3149
3150 static int enable_scrq_irq(struct ibmvnic_adapter *adapter,
3151                            struct ibmvnic_sub_crq_queue *scrq)
3152 {
3153         struct device *dev = &adapter->vdev->dev;
3154         unsigned long rc;
3155
3156         if (scrq->hw_irq > 0x100000000ULL) {
3157                 dev_err(dev, "bad hw_irq = %lx\n", scrq->hw_irq);
3158                 return 1;
3159         }
3160
3161         if (test_bit(0, &adapter->resetting) &&
3162             adapter->reset_reason == VNIC_RESET_MOBILITY) {
3163                 u64 val = (0xff000000) | scrq->hw_irq;
3164
3165                 rc = plpar_hcall_norets(H_EOI, val);
3166                 /* H_EOI would fail with rc = H_FUNCTION when running
3167                  * in XIVE mode which is expected, but not an error.
3168                  */
3169                 if (rc && rc != H_FUNCTION)
3170                         dev_err(dev, "H_EOI FAILED irq 0x%llx. rc=%ld\n",
3171                                 val, rc);
3172         }
3173
3174         rc = plpar_hcall_norets(H_VIOCTL, adapter->vdev->unit_address,
3175                                 H_ENABLE_VIO_INTERRUPT, scrq->hw_irq, 0, 0);
3176         if (rc)
3177                 dev_err(dev, "Couldn't enable scrq irq 0x%lx. rc=%ld\n",
3178                         scrq->hw_irq, rc);
3179         return rc;
3180 }
3181
3182 static int ibmvnic_complete_tx(struct ibmvnic_adapter *adapter,
3183                                struct ibmvnic_sub_crq_queue *scrq)
3184 {
3185         struct device *dev = &adapter->vdev->dev;
3186         struct ibmvnic_tx_pool *tx_pool;
3187         struct ibmvnic_tx_buff *txbuff;
3188         struct netdev_queue *txq;
3189         union sub_crq *next;
3190         int index;
3191         int i;
3192
3193 restart_loop:
3194         while (pending_scrq(adapter, scrq)) {
3195                 unsigned int pool = scrq->pool_index;
3196                 int num_entries = 0;
3197                 int total_bytes = 0;
3198                 int num_packets = 0;
3199
3200                 next = ibmvnic_next_scrq(adapter, scrq);
3201                 for (i = 0; i < next->tx_comp.num_comps; i++) {
3202                         index = be32_to_cpu(next->tx_comp.correlators[i]);
3203                         if (index & IBMVNIC_TSO_POOL_MASK) {
3204                                 tx_pool = &adapter->tso_pool[pool];
3205                                 index &= ~IBMVNIC_TSO_POOL_MASK;
3206                         } else {
3207                                 tx_pool = &adapter->tx_pool[pool];
3208                         }
3209
3210                         txbuff = &tx_pool->tx_buff[index];
3211                         num_packets++;
3212                         num_entries += txbuff->num_entries;
3213                         if (txbuff->skb) {
3214                                 total_bytes += txbuff->skb->len;
3215                                 if (next->tx_comp.rcs[i]) {
3216                                         dev_err(dev, "tx error %x\n",
3217                                                 next->tx_comp.rcs[i]);
3218                                         dev_kfree_skb_irq(txbuff->skb);
3219                                 } else {
3220                                         dev_consume_skb_irq(txbuff->skb);
3221                                 }
3222                                 txbuff->skb = NULL;
3223                         } else {
3224                                 netdev_warn(adapter->netdev,
3225                                             "TX completion received with NULL socket buffer\n");
3226                         }
3227                         tx_pool->free_map[tx_pool->producer_index] = index;
3228                         tx_pool->producer_index =
3229                                 (tx_pool->producer_index + 1) %
3230                                         tx_pool->num_buffers;
3231                 }
3232                 /* remove tx_comp scrq*/
3233                 next->tx_comp.first = 0;
3234
3235                 txq = netdev_get_tx_queue(adapter->netdev, scrq->pool_index);
3236                 netdev_tx_completed_queue(txq, num_packets, total_bytes);
3237
3238                 if (atomic_sub_return(num_entries, &scrq->used) <=
3239                     (adapter->req_tx_entries_per_subcrq / 2) &&
3240                     __netif_subqueue_stopped(adapter->netdev,
3241                                              scrq->pool_index)) {
3242                         netif_wake_subqueue(adapter->netdev, scrq->pool_index);
3243                         netdev_dbg(adapter->netdev, "Started queue %d\n",
3244                                    scrq->pool_index);
3245                 }
3246         }
3247
3248         enable_scrq_irq(adapter, scrq);
3249
3250         if (pending_scrq(adapter, scrq)) {
3251                 disable_scrq_irq(adapter, scrq);
3252                 goto restart_loop;
3253         }
3254
3255         return 0;
3256 }
3257
3258 static irqreturn_t ibmvnic_interrupt_tx(int irq, void *instance)
3259 {
3260         struct ibmvnic_sub_crq_queue *scrq = instance;
3261         struct ibmvnic_adapter *adapter = scrq->adapter;
3262
3263         disable_scrq_irq(adapter, scrq);
3264         ibmvnic_complete_tx(adapter, scrq);
3265
3266         return IRQ_HANDLED;
3267 }
3268
3269 static irqreturn_t ibmvnic_interrupt_rx(int irq, void *instance)
3270 {
3271         struct ibmvnic_sub_crq_queue *scrq = instance;
3272         struct ibmvnic_adapter *adapter = scrq->adapter;
3273
3274         /* When booting a kdump kernel we can hit pending interrupts
3275          * prior to completing driver initialization.
3276          */
3277         if (unlikely(adapter->state != VNIC_OPEN))
3278                 return IRQ_NONE;
3279
3280         adapter->rx_stats_buffers[scrq->scrq_num].interrupts++;
3281
3282         if (napi_schedule_prep(&adapter->napi[scrq->scrq_num])) {
3283                 disable_scrq_irq(adapter, scrq);
3284                 __napi_schedule(&adapter->napi[scrq->scrq_num]);
3285         }
3286
3287         return IRQ_HANDLED;
3288 }
3289
3290 static int init_sub_crq_irqs(struct ibmvnic_adapter *adapter)
3291 {
3292         struct device *dev = &adapter->vdev->dev;
3293         struct ibmvnic_sub_crq_queue *scrq;
3294         int i = 0, j = 0;
3295         int rc = 0;
3296
3297         for (i = 0; i < adapter->req_tx_queues; i++) {
3298                 netdev_dbg(adapter->netdev, "Initializing tx_scrq[%d] irq\n",
3299                            i);
3300                 scrq = adapter->tx_scrq[i];
3301                 scrq->irq = irq_create_mapping(NULL, scrq->hw_irq);
3302
3303                 if (!scrq->irq) {
3304                         rc = -EINVAL;
3305                         dev_err(dev, "Error mapping irq\n");
3306                         goto req_tx_irq_failed;
3307                 }
3308
3309                 snprintf(scrq->name, sizeof(scrq->name), "ibmvnic-%x-tx%d",
3310                          adapter->vdev->unit_address, i);
3311                 rc = request_irq(scrq->irq, ibmvnic_interrupt_tx,
3312                                  0, scrq->name, scrq);
3313
3314                 if (rc) {
3315                         dev_err(dev, "Couldn't register tx irq 0x%x. rc=%d\n",
3316                                 scrq->irq, rc);
3317                         irq_dispose_mapping(scrq->irq);
3318                         goto req_tx_irq_failed;
3319                 }
3320         }
3321
3322         for (i = 0; i < adapter->req_rx_queues; i++) {
3323                 netdev_dbg(adapter->netdev, "Initializing rx_scrq[%d] irq\n",
3324                            i);
3325                 scrq = adapter->rx_scrq[i];
3326                 scrq->irq = irq_create_mapping(NULL, scrq->hw_irq);
3327                 if (!scrq->irq) {
3328                         rc = -EINVAL;
3329                         dev_err(dev, "Error mapping irq\n");
3330                         goto req_rx_irq_failed;
3331                 }
3332                 snprintf(scrq->name, sizeof(scrq->name), "ibmvnic-%x-rx%d",
3333                          adapter->vdev->unit_address, i);
3334                 rc = request_irq(scrq->irq, ibmvnic_interrupt_rx,
3335                                  0, scrq->name, scrq);
3336                 if (rc) {
3337                         dev_err(dev, "Couldn't register rx irq 0x%x. rc=%d\n",
3338                                 scrq->irq, rc);
3339                         irq_dispose_mapping(scrq->irq);
3340                         goto req_rx_irq_failed;
3341                 }
3342         }
3343         return rc;
3344
3345 req_rx_irq_failed:
3346         for (j = 0; j < i; j++) {
3347                 free_irq(adapter->rx_scrq[j]->irq, adapter->rx_scrq[j]);
3348                 irq_dispose_mapping(adapter->rx_scrq[j]->irq);
3349         }
3350         i = adapter->req_tx_queues;
3351 req_tx_irq_failed:
3352         for (j = 0; j < i; j++) {
3353                 free_irq(adapter->tx_scrq[j]->irq, adapter->tx_scrq[j]);
3354                 irq_dispose_mapping(adapter->tx_scrq[j]->irq);
3355         }
3356         release_sub_crqs(adapter, 1);
3357         return rc;
3358 }
3359
3360 static int init_sub_crqs(struct ibmvnic_adapter *adapter)
3361 {
3362         struct device *dev = &adapter->vdev->dev;
3363         struct ibmvnic_sub_crq_queue **allqueues;
3364         int registered_queues = 0;
3365         int total_queues;
3366         int more = 0;
3367         int i;
3368
3369         total_queues = adapter->req_tx_queues + adapter->req_rx_queues;
3370
3371         allqueues = kcalloc(total_queues, sizeof(*allqueues), GFP_KERNEL);
3372         if (!allqueues)
3373                 return -1;
3374
3375         for (i = 0; i < total_queues; i++) {
3376                 allqueues[i] = init_sub_crq_queue(adapter);
3377                 if (!allqueues[i]) {
3378                         dev_warn(dev, "Couldn't allocate all sub-crqs\n");
3379                         break;
3380                 }
3381                 registered_queues++;
3382         }
3383
3384         /* Make sure we were able to register the minimum number of queues */
3385         if (registered_queues <
3386             adapter->min_tx_queues + adapter->min_rx_queues) {
3387                 dev_err(dev, "Fatal: Couldn't init  min number of sub-crqs\n");
3388                 goto tx_failed;
3389         }
3390
3391         /* Distribute the failed allocated queues*/
3392         for (i = 0; i < total_queues - registered_queues + more ; i++) {
3393                 netdev_dbg(adapter->netdev, "Reducing number of queues\n");
3394                 switch (i % 3) {
3395                 case 0:
3396                         if (adapter->req_rx_queues > adapter->min_rx_queues)
3397                                 adapter->req_rx_queues--;
3398                         else
3399                                 more++;
3400                         break;
3401                 case 1:
3402                         if (adapter->req_tx_queues > adapter->min_tx_queues)
3403                                 adapter->req_tx_queues--;
3404                         else
3405                                 more++;
3406                         break;
3407                 }
3408         }
3409
3410         adapter->tx_scrq = kcalloc(adapter->req_tx_queues,
3411                                    sizeof(*adapter->tx_scrq), GFP_KERNEL);
3412         if (!adapter->tx_scrq)
3413                 goto tx_failed;
3414
3415         for (i = 0; i < adapter->req_tx_queues; i++) {
3416                 adapter->tx_scrq[i] = allqueues[i];
3417                 adapter->tx_scrq[i]->pool_index = i;
3418                 adapter->num_active_tx_scrqs++;
3419         }
3420
3421         adapter->rx_scrq = kcalloc(adapter->req_rx_queues,
3422                                    sizeof(*adapter->rx_scrq), GFP_KERNEL);
3423         if (!adapter->rx_scrq)
3424                 goto rx_failed;
3425
3426         for (i = 0; i < adapter->req_rx_queues; i++) {
3427                 adapter->rx_scrq[i] = allqueues[i + adapter->req_tx_queues];
3428                 adapter->rx_scrq[i]->scrq_num = i;
3429                 adapter->num_active_rx_scrqs++;
3430         }
3431
3432         kfree(allqueues);
3433         return 0;
3434
3435 rx_failed:
3436         kfree(adapter->tx_scrq);
3437         adapter->tx_scrq = NULL;
3438 tx_failed:
3439         for (i = 0; i < registered_queues; i++)
3440                 release_sub_crq_queue(adapter, allqueues[i], 1);
3441         kfree(allqueues);
3442         return -1;
3443 }
3444
3445 static void send_request_cap(struct ibmvnic_adapter *adapter, int retry)
3446 {
3447         struct device *dev = &adapter->vdev->dev;
3448         union ibmvnic_crq crq;
3449         int max_entries;
3450
3451         if (!retry) {
3452                 /* Sub-CRQ entries are 32 byte long */
3453                 int entries_page = 4 * PAGE_SIZE / (sizeof(u64) * 4);
3454
3455                 if (adapter->min_tx_entries_per_subcrq > entries_page ||
3456                     adapter->min_rx_add_entries_per_subcrq > entries_page) {
3457                         dev_err(dev, "Fatal, invalid entries per sub-crq\n");
3458                         return;
3459                 }
3460
3461                 if (adapter->desired.mtu)
3462                         adapter->req_mtu = adapter->desired.mtu;
3463                 else
3464                         adapter->req_mtu = adapter->netdev->mtu + ETH_HLEN;
3465
3466                 if (!adapter->desired.tx_entries)
3467                         adapter->desired.tx_entries =
3468                                         adapter->max_tx_entries_per_subcrq;
3469                 if (!adapter->desired.rx_entries)
3470                         adapter->desired.rx_entries =
3471                                         adapter->max_rx_add_entries_per_subcrq;
3472
3473                 max_entries = IBMVNIC_MAX_LTB_SIZE /
3474                               (adapter->req_mtu + IBMVNIC_BUFFER_HLEN);
3475
3476                 if ((adapter->req_mtu + IBMVNIC_BUFFER_HLEN) *
3477                         adapter->desired.tx_entries > IBMVNIC_MAX_LTB_SIZE) {
3478                         adapter->desired.tx_entries = max_entries;
3479                 }
3480
3481                 if ((adapter->req_mtu + IBMVNIC_BUFFER_HLEN) *
3482                         adapter->desired.rx_entries > IBMVNIC_MAX_LTB_SIZE) {
3483                         adapter->desired.rx_entries = max_entries;
3484                 }
3485
3486                 if (adapter->desired.tx_entries)
3487                         adapter->req_tx_entries_per_subcrq =
3488                                         adapter->desired.tx_entries;
3489                 else
3490                         adapter->req_tx_entries_per_subcrq =
3491                                         adapter->max_tx_entries_per_subcrq;
3492
3493                 if (adapter->desired.rx_entries)
3494                         adapter->req_rx_add_entries_per_subcrq =
3495                                         adapter->desired.rx_entries;
3496                 else
3497                         adapter->req_rx_add_entries_per_subcrq =
3498                                         adapter->max_rx_add_entries_per_subcrq;
3499
3500                 if (adapter->desired.tx_queues)
3501                         adapter->req_tx_queues =
3502                                         adapter->desired.tx_queues;
3503                 else
3504                         adapter->req_tx_queues =
3505                                         adapter->opt_tx_comp_sub_queues;
3506
3507                 if (adapter->desired.rx_queues)
3508                         adapter->req_rx_queues =
3509                                         adapter->desired.rx_queues;
3510                 else
3511                         adapter->req_rx_queues =
3512                                         adapter->opt_rx_comp_queues;
3513
3514                 adapter->req_rx_add_queues = adapter->max_rx_add_queues;
3515         }
3516
3517         memset(&crq, 0, sizeof(crq));
3518         crq.request_capability.first = IBMVNIC_CRQ_CMD;
3519         crq.request_capability.cmd = REQUEST_CAPABILITY;
3520
3521         crq.request_capability.capability = cpu_to_be16(REQ_TX_QUEUES);
3522         crq.request_capability.number = cpu_to_be64(adapter->req_tx_queues);
3523         atomic_inc(&adapter->running_cap_crqs);
3524         ibmvnic_send_crq(adapter, &crq);
3525
3526         crq.request_capability.capability = cpu_to_be16(REQ_RX_QUEUES);
3527         crq.request_capability.number = cpu_to_be64(adapter->req_rx_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_ADD_QUEUES);
3532         crq.request_capability.number = cpu_to_be64(adapter->req_rx_add_queues);
3533         atomic_inc(&adapter->running_cap_crqs);
3534         ibmvnic_send_crq(adapter, &crq);
3535
3536         crq.request_capability.capability =
3537             cpu_to_be16(REQ_TX_ENTRIES_PER_SUBCRQ);
3538         crq.request_capability.number =
3539             cpu_to_be64(adapter->req_tx_entries_per_subcrq);
3540         atomic_inc(&adapter->running_cap_crqs);
3541         ibmvnic_send_crq(adapter, &crq);
3542
3543         crq.request_capability.capability =
3544             cpu_to_be16(REQ_RX_ADD_ENTRIES_PER_SUBCRQ);
3545         crq.request_capability.number =
3546             cpu_to_be64(adapter->req_rx_add_entries_per_subcrq);
3547         atomic_inc(&adapter->running_cap_crqs);
3548         ibmvnic_send_crq(adapter, &crq);
3549
3550         crq.request_capability.capability = cpu_to_be16(REQ_MTU);
3551         crq.request_capability.number = cpu_to_be64(adapter->req_mtu);
3552         atomic_inc(&adapter->running_cap_crqs);
3553         ibmvnic_send_crq(adapter, &crq);
3554
3555         if (adapter->netdev->flags & IFF_PROMISC) {
3556                 if (adapter->promisc_supported) {
3557                         crq.request_capability.capability =
3558                             cpu_to_be16(PROMISC_REQUESTED);
3559                         crq.request_capability.number = cpu_to_be64(1);
3560                         atomic_inc(&adapter->running_cap_crqs);
3561                         ibmvnic_send_crq(adapter, &crq);
3562                 }
3563         } else {
3564                 crq.request_capability.capability =
3565                     cpu_to_be16(PROMISC_REQUESTED);
3566                 crq.request_capability.number = cpu_to_be64(0);
3567                 atomic_inc(&adapter->running_cap_crqs);
3568                 ibmvnic_send_crq(adapter, &crq);
3569         }
3570 }
3571
3572 static int pending_scrq(struct ibmvnic_adapter *adapter,
3573                         struct ibmvnic_sub_crq_queue *scrq)
3574 {
3575         union sub_crq *entry = &scrq->msgs[scrq->cur];
3576         int rc;
3577
3578         rc = !!(entry->generic.first & IBMVNIC_CRQ_CMD_RSP);
3579
3580         /* Ensure that the SCRQ valid flag is loaded prior to loading the
3581          * contents of the SCRQ descriptor
3582          */
3583         dma_rmb();
3584
3585         return rc;
3586 }
3587
3588 static union sub_crq *ibmvnic_next_scrq(struct ibmvnic_adapter *adapter,
3589                                         struct ibmvnic_sub_crq_queue *scrq)
3590 {
3591         union sub_crq *entry;
3592         unsigned long flags;
3593
3594         spin_lock_irqsave(&scrq->lock, flags);
3595         entry = &scrq->msgs[scrq->cur];
3596         if (entry->generic.first & IBMVNIC_CRQ_CMD_RSP) {
3597                 if (++scrq->cur == scrq->size)
3598                         scrq->cur = 0;
3599         } else {
3600                 entry = NULL;
3601         }
3602         spin_unlock_irqrestore(&scrq->lock, flags);
3603
3604         /* Ensure that the SCRQ valid flag is loaded prior to loading the
3605          * contents of the SCRQ descriptor
3606          */
3607         dma_rmb();
3608
3609         return entry;
3610 }
3611
3612 static union ibmvnic_crq *ibmvnic_next_crq(struct ibmvnic_adapter *adapter)
3613 {
3614         struct ibmvnic_crq_queue *queue = &adapter->crq;
3615         union ibmvnic_crq *crq;
3616
3617         crq = &queue->msgs[queue->cur];
3618         if (crq->generic.first & IBMVNIC_CRQ_CMD_RSP) {
3619                 if (++queue->cur == queue->size)
3620                         queue->cur = 0;
3621         } else {
3622                 crq = NULL;
3623         }
3624
3625         return crq;
3626 }
3627
3628 static void print_subcrq_error(struct device *dev, int rc, const char *func)
3629 {
3630         switch (rc) {
3631         case H_PARAMETER:
3632                 dev_warn_ratelimited(dev,
3633                                      "%s failed: Send request is malformed or adapter failover pending. (rc=%d)\n",
3634                                      func, rc);
3635                 break;
3636         case H_CLOSED:
3637                 dev_warn_ratelimited(dev,
3638                                      "%s failed: Backing queue closed. Adapter is down or failover pending. (rc=%d)\n",
3639                                      func, rc);
3640                 break;
3641         default:
3642                 dev_err_ratelimited(dev, "%s failed: (rc=%d)\n", func, rc);
3643                 break;
3644         }
3645 }
3646
3647 static int send_subcrq_indirect(struct ibmvnic_adapter *adapter,
3648                                 u64 remote_handle, u64 ioba, u64 num_entries)
3649 {
3650         unsigned int ua = adapter->vdev->unit_address;
3651         struct device *dev = &adapter->vdev->dev;
3652         int rc;
3653
3654         /* Make sure the hypervisor sees the complete request */
3655         dma_wmb();
3656         rc = plpar_hcall_norets(H_SEND_SUB_CRQ_INDIRECT, ua,
3657                                 cpu_to_be64(remote_handle),
3658                                 ioba, num_entries);
3659
3660         if (rc)
3661                 print_subcrq_error(dev, rc, __func__);
3662
3663         return rc;
3664 }
3665
3666 static int ibmvnic_send_crq(struct ibmvnic_adapter *adapter,
3667                             union ibmvnic_crq *crq)
3668 {
3669         unsigned int ua = adapter->vdev->unit_address;
3670         struct device *dev = &adapter->vdev->dev;
3671         u64 *u64_crq = (u64 *)crq;
3672         int rc;
3673
3674         netdev_dbg(adapter->netdev, "Sending CRQ: %016lx %016lx\n",
3675                    (unsigned long)cpu_to_be64(u64_crq[0]),
3676                    (unsigned long)cpu_to_be64(u64_crq[1]));
3677
3678         if (!adapter->crq.active &&
3679             crq->generic.first != IBMVNIC_CRQ_INIT_CMD) {
3680                 dev_warn(dev, "Invalid request detected while CRQ is inactive, possible device state change during reset\n");
3681                 return -EINVAL;
3682         }
3683
3684         /* Make sure the hypervisor sees the complete request */
3685         dma_wmb();
3686
3687         rc = plpar_hcall_norets(H_SEND_CRQ, ua,
3688                                 cpu_to_be64(u64_crq[0]),
3689                                 cpu_to_be64(u64_crq[1]));
3690
3691         if (rc) {
3692                 if (rc == H_CLOSED) {
3693                         dev_warn(dev, "CRQ Queue closed\n");
3694                         /* do not reset, report the fail, wait for passive init from server */
3695                 }
3696
3697                 dev_warn(dev, "Send error (rc=%d)\n", rc);
3698         }
3699
3700         return rc;
3701 }
3702
3703 static int ibmvnic_send_crq_init(struct ibmvnic_adapter *adapter)
3704 {
3705         struct device *dev = &adapter->vdev->dev;
3706         union ibmvnic_crq crq;
3707         int retries = 100;
3708         int rc;
3709
3710         memset(&crq, 0, sizeof(crq));
3711         crq.generic.first = IBMVNIC_CRQ_INIT_CMD;
3712         crq.generic.cmd = IBMVNIC_CRQ_INIT;
3713         netdev_dbg(adapter->netdev, "Sending CRQ init\n");
3714
3715         do {
3716                 rc = ibmvnic_send_crq(adapter, &crq);
3717                 if (rc != H_CLOSED)
3718                         break;
3719                 retries--;
3720                 msleep(50);
3721
3722         } while (retries > 0);
3723
3724         if (rc) {
3725                 dev_err(dev, "Failed to send init request, rc = %d\n", rc);
3726                 return rc;
3727         }
3728
3729         return 0;
3730 }
3731
3732 static int send_version_xchg(struct ibmvnic_adapter *adapter)
3733 {
3734         union ibmvnic_crq crq;
3735
3736         memset(&crq, 0, sizeof(crq));
3737         crq.version_exchange.first = IBMVNIC_CRQ_CMD;
3738         crq.version_exchange.cmd = VERSION_EXCHANGE;
3739         crq.version_exchange.version = cpu_to_be16(ibmvnic_version);
3740
3741         return ibmvnic_send_crq(adapter, &crq);
3742 }
3743
3744 struct vnic_login_client_data {
3745         u8      type;
3746         __be16  len;
3747         char    name[];
3748 } __packed;
3749
3750 static int vnic_client_data_len(struct ibmvnic_adapter *adapter)
3751 {
3752         int len;
3753
3754         /* Calculate the amount of buffer space needed for the
3755          * vnic client data in the login buffer. There are four entries,
3756          * OS name, LPAR name, device name, and a null last entry.
3757          */
3758         len = 4 * sizeof(struct vnic_login_client_data);
3759         len += 6; /* "Linux" plus NULL */
3760         len += strlen(utsname()->nodename) + 1;
3761         len += strlen(adapter->netdev->name) + 1;
3762
3763         return len;
3764 }
3765
3766 static void vnic_add_client_data(struct ibmvnic_adapter *adapter,
3767                                  struct vnic_login_client_data *vlcd)
3768 {
3769         const char *os_name = "Linux";
3770         int len;
3771
3772         /* Type 1 - LPAR OS */
3773         vlcd->type = 1;
3774         len = strlen(os_name) + 1;
3775         vlcd->len = cpu_to_be16(len);
3776         strncpy(vlcd->name, os_name, len);
3777         vlcd = (struct vnic_login_client_data *)(vlcd->name + len);
3778
3779         /* Type 2 - LPAR name */
3780         vlcd->type = 2;
3781         len = strlen(utsname()->nodename) + 1;
3782         vlcd->len = cpu_to_be16(len);
3783         strncpy(vlcd->name, utsname()->nodename, len);
3784         vlcd = (struct vnic_login_client_data *)(vlcd->name + len);
3785
3786         /* Type 3 - device name */
3787         vlcd->type = 3;
3788         len = strlen(adapter->netdev->name) + 1;
3789         vlcd->len = cpu_to_be16(len);
3790         strncpy(vlcd->name, adapter->netdev->name, len);
3791 }
3792
3793 static int send_login(struct ibmvnic_adapter *adapter)
3794 {
3795         struct ibmvnic_login_rsp_buffer *login_rsp_buffer;
3796         struct ibmvnic_login_buffer *login_buffer;
3797         struct device *dev = &adapter->vdev->dev;
3798         struct vnic_login_client_data *vlcd;
3799         dma_addr_t rsp_buffer_token;
3800         dma_addr_t buffer_token;
3801         size_t rsp_buffer_size;
3802         union ibmvnic_crq crq;
3803         int client_data_len;
3804         size_t buffer_size;
3805         __be64 *tx_list_p;
3806         __be64 *rx_list_p;
3807         int rc;
3808         int i;
3809
3810         if (!adapter->tx_scrq || !adapter->rx_scrq) {
3811                 netdev_err(adapter->netdev,
3812                            "RX or TX queues are not allocated, device login failed\n");
3813                 return -1;
3814         }
3815
3816         release_login_buffer(adapter);
3817         release_login_rsp_buffer(adapter);
3818
3819         client_data_len = vnic_client_data_len(adapter);
3820
3821         buffer_size =
3822             sizeof(struct ibmvnic_login_buffer) +
3823             sizeof(u64) * (adapter->req_tx_queues + adapter->req_rx_queues) +
3824             client_data_len;
3825
3826         login_buffer = kzalloc(buffer_size, GFP_ATOMIC);
3827         if (!login_buffer)
3828                 goto buf_alloc_failed;
3829
3830         buffer_token = dma_map_single(dev, login_buffer, buffer_size,
3831                                       DMA_TO_DEVICE);
3832         if (dma_mapping_error(dev, buffer_token)) {
3833                 dev_err(dev, "Couldn't map login buffer\n");
3834                 goto buf_map_failed;
3835         }
3836
3837         rsp_buffer_size = sizeof(struct ibmvnic_login_rsp_buffer) +
3838                           sizeof(u64) * adapter->req_tx_queues +
3839                           sizeof(u64) * adapter->req_rx_queues +
3840                           sizeof(u64) * adapter->req_rx_queues +
3841                           sizeof(u8) * IBMVNIC_TX_DESC_VERSIONS;
3842
3843         login_rsp_buffer = kmalloc(rsp_buffer_size, GFP_ATOMIC);
3844         if (!login_rsp_buffer)
3845                 goto buf_rsp_alloc_failed;
3846
3847         rsp_buffer_token = dma_map_single(dev, login_rsp_buffer,
3848                                           rsp_buffer_size, DMA_FROM_DEVICE);
3849         if (dma_mapping_error(dev, rsp_buffer_token)) {
3850                 dev_err(dev, "Couldn't map login rsp buffer\n");
3851                 goto buf_rsp_map_failed;
3852         }
3853
3854         adapter->login_buf = login_buffer;
3855         adapter->login_buf_token = buffer_token;
3856         adapter->login_buf_sz = buffer_size;
3857         adapter->login_rsp_buf = login_rsp_buffer;
3858         adapter->login_rsp_buf_token = rsp_buffer_token;
3859         adapter->login_rsp_buf_sz = rsp_buffer_size;
3860
3861         login_buffer->len = cpu_to_be32(buffer_size);
3862         login_buffer->version = cpu_to_be32(INITIAL_VERSION_LB);
3863         login_buffer->num_txcomp_subcrqs = cpu_to_be32(adapter->req_tx_queues);
3864         login_buffer->off_txcomp_subcrqs =
3865             cpu_to_be32(sizeof(struct ibmvnic_login_buffer));
3866         login_buffer->num_rxcomp_subcrqs = cpu_to_be32(adapter->req_rx_queues);
3867         login_buffer->off_rxcomp_subcrqs =
3868             cpu_to_be32(sizeof(struct ibmvnic_login_buffer) +
3869                         sizeof(u64) * adapter->req_tx_queues);
3870         login_buffer->login_rsp_ioba = cpu_to_be32(rsp_buffer_token);
3871         login_buffer->login_rsp_len = cpu_to_be32(rsp_buffer_size);
3872
3873         tx_list_p = (__be64 *)((char *)login_buffer +
3874                                       sizeof(struct ibmvnic_login_buffer));
3875         rx_list_p = (__be64 *)((char *)login_buffer +
3876                                       sizeof(struct ibmvnic_login_buffer) +
3877                                       sizeof(u64) * adapter->req_tx_queues);
3878
3879         for (i = 0; i < adapter->req_tx_queues; i++) {
3880                 if (adapter->tx_scrq[i]) {
3881                         tx_list_p[i] =
3882                                 cpu_to_be64(adapter->tx_scrq[i]->crq_num);
3883                 }
3884         }
3885
3886         for (i = 0; i < adapter->req_rx_queues; i++) {
3887                 if (adapter->rx_scrq[i]) {
3888                         rx_list_p[i] =
3889                                 cpu_to_be64(adapter->rx_scrq[i]->crq_num);
3890                 }
3891         }
3892
3893         /* Insert vNIC login client data */
3894         vlcd = (struct vnic_login_client_data *)
3895                 ((char *)rx_list_p + (sizeof(u64) * adapter->req_rx_queues));
3896         login_buffer->client_data_offset =
3897                         cpu_to_be32((char *)vlcd - (char *)login_buffer);
3898         login_buffer->client_data_len = cpu_to_be32(client_data_len);
3899
3900         vnic_add_client_data(adapter, vlcd);
3901
3902         netdev_dbg(adapter->netdev, "Login Buffer:\n");
3903         for (i = 0; i < (adapter->login_buf_sz - 1) / 8 + 1; i++) {
3904                 netdev_dbg(adapter->netdev, "%016lx\n",
3905                            ((unsigned long *)(adapter->login_buf))[i]);
3906         }
3907
3908         memset(&crq, 0, sizeof(crq));
3909         crq.login.first = IBMVNIC_CRQ_CMD;
3910         crq.login.cmd = LOGIN;
3911         crq.login.ioba = cpu_to_be32(buffer_token);
3912         crq.login.len = cpu_to_be32(buffer_size);
3913
3914         adapter->login_pending = true;
3915         rc = ibmvnic_send_crq(adapter, &crq);
3916         if (rc) {
3917                 adapter->login_pending = false;
3918                 netdev_err(adapter->netdev, "Failed to send login, rc=%d\n", rc);
3919                 goto buf_rsp_map_failed;
3920         }
3921
3922         return 0;
3923
3924 buf_rsp_map_failed:
3925         kfree(login_rsp_buffer);
3926         adapter->login_rsp_buf = NULL;
3927 buf_rsp_alloc_failed:
3928         dma_unmap_single(dev, buffer_token, buffer_size, DMA_TO_DEVICE);
3929 buf_map_failed:
3930         kfree(login_buffer);
3931         adapter->login_buf = NULL;
3932 buf_alloc_failed:
3933         return -1;
3934 }
3935
3936 static int send_request_map(struct ibmvnic_adapter *adapter, dma_addr_t addr,
3937                             u32 len, u8 map_id)
3938 {
3939         union ibmvnic_crq crq;
3940
3941         memset(&crq, 0, sizeof(crq));
3942         crq.request_map.first = IBMVNIC_CRQ_CMD;
3943         crq.request_map.cmd = REQUEST_MAP;
3944         crq.request_map.map_id = map_id;
3945         crq.request_map.ioba = cpu_to_be32(addr);
3946         crq.request_map.len = cpu_to_be32(len);
3947         return ibmvnic_send_crq(adapter, &crq);
3948 }
3949
3950 static int send_request_unmap(struct ibmvnic_adapter *adapter, u8 map_id)
3951 {
3952         union ibmvnic_crq crq;
3953
3954         memset(&crq, 0, sizeof(crq));
3955         crq.request_unmap.first = IBMVNIC_CRQ_CMD;
3956         crq.request_unmap.cmd = REQUEST_UNMAP;
3957         crq.request_unmap.map_id = map_id;
3958         return ibmvnic_send_crq(adapter, &crq);
3959 }
3960
3961 static void send_query_map(struct ibmvnic_adapter *adapter)
3962 {
3963         union ibmvnic_crq crq;
3964
3965         memset(&crq, 0, sizeof(crq));
3966         crq.query_map.first = IBMVNIC_CRQ_CMD;
3967         crq.query_map.cmd = QUERY_MAP;
3968         ibmvnic_send_crq(adapter, &crq);
3969 }
3970
3971 /* Send a series of CRQs requesting various capabilities of the VNIC server */
3972 static void send_query_cap(struct ibmvnic_adapter *adapter)
3973 {
3974         union ibmvnic_crq crq;
3975
3976         atomic_set(&adapter->running_cap_crqs, 0);
3977         memset(&crq, 0, sizeof(crq));
3978         crq.query_capability.first = IBMVNIC_CRQ_CMD;
3979         crq.query_capability.cmd = QUERY_CAPABILITY;
3980
3981         crq.query_capability.capability = cpu_to_be16(MIN_TX_QUEUES);
3982         atomic_inc(&adapter->running_cap_crqs);
3983         ibmvnic_send_crq(adapter, &crq);
3984
3985         crq.query_capability.capability = cpu_to_be16(MIN_RX_QUEUES);
3986         atomic_inc(&adapter->running_cap_crqs);
3987         ibmvnic_send_crq(adapter, &crq);
3988
3989         crq.query_capability.capability = cpu_to_be16(MIN_RX_ADD_QUEUES);
3990         atomic_inc(&adapter->running_cap_crqs);
3991         ibmvnic_send_crq(adapter, &crq);
3992
3993         crq.query_capability.capability = cpu_to_be16(MAX_TX_QUEUES);
3994         atomic_inc(&adapter->running_cap_crqs);
3995         ibmvnic_send_crq(adapter, &crq);
3996
3997         crq.query_capability.capability = cpu_to_be16(MAX_RX_QUEUES);
3998         atomic_inc(&adapter->running_cap_crqs);
3999         ibmvnic_send_crq(adapter, &crq);
4000
4001         crq.query_capability.capability = cpu_to_be16(MAX_RX_ADD_QUEUES);
4002         atomic_inc(&adapter->running_cap_crqs);
4003         ibmvnic_send_crq(adapter, &crq);
4004
4005         crq.query_capability.capability =
4006             cpu_to_be16(MIN_TX_ENTRIES_PER_SUBCRQ);
4007         atomic_inc(&adapter->running_cap_crqs);
4008         ibmvnic_send_crq(adapter, &crq);
4009
4010         crq.query_capability.capability =
4011             cpu_to_be16(MIN_RX_ADD_ENTRIES_PER_SUBCRQ);
4012         atomic_inc(&adapter->running_cap_crqs);
4013         ibmvnic_send_crq(adapter, &crq);
4014
4015         crq.query_capability.capability =
4016             cpu_to_be16(MAX_TX_ENTRIES_PER_SUBCRQ);
4017         atomic_inc(&adapter->running_cap_crqs);
4018         ibmvnic_send_crq(adapter, &crq);
4019
4020         crq.query_capability.capability =
4021             cpu_to_be16(MAX_RX_ADD_ENTRIES_PER_SUBCRQ);
4022         atomic_inc(&adapter->running_cap_crqs);
4023         ibmvnic_send_crq(adapter, &crq);
4024
4025         crq.query_capability.capability = cpu_to_be16(TCP_IP_OFFLOAD);
4026         atomic_inc(&adapter->running_cap_crqs);
4027         ibmvnic_send_crq(adapter, &crq);
4028
4029         crq.query_capability.capability = cpu_to_be16(PROMISC_SUPPORTED);
4030         atomic_inc(&adapter->running_cap_crqs);
4031         ibmvnic_send_crq(adapter, &crq);
4032
4033         crq.query_capability.capability = cpu_to_be16(MIN_MTU);
4034         atomic_inc(&adapter->running_cap_crqs);
4035         ibmvnic_send_crq(adapter, &crq);
4036
4037         crq.query_capability.capability = cpu_to_be16(MAX_MTU);
4038         atomic_inc(&adapter->running_cap_crqs);
4039         ibmvnic_send_crq(adapter, &crq);
4040
4041         crq.query_capability.capability = cpu_to_be16(MAX_MULTICAST_FILTERS);
4042         atomic_inc(&adapter->running_cap_crqs);
4043         ibmvnic_send_crq(adapter, &crq);
4044
4045         crq.query_capability.capability = cpu_to_be16(VLAN_HEADER_INSERTION);
4046         atomic_inc(&adapter->running_cap_crqs);
4047         ibmvnic_send_crq(adapter, &crq);
4048
4049         crq.query_capability.capability = cpu_to_be16(RX_VLAN_HEADER_INSERTION);
4050         atomic_inc(&adapter->running_cap_crqs);
4051         ibmvnic_send_crq(adapter, &crq);
4052
4053         crq.query_capability.capability = cpu_to_be16(MAX_TX_SG_ENTRIES);
4054         atomic_inc(&adapter->running_cap_crqs);
4055         ibmvnic_send_crq(adapter, &crq);
4056
4057         crq.query_capability.capability = cpu_to_be16(RX_SG_SUPPORTED);
4058         atomic_inc(&adapter->running_cap_crqs);
4059         ibmvnic_send_crq(adapter, &crq);
4060
4061         crq.query_capability.capability = cpu_to_be16(OPT_TX_COMP_SUB_QUEUES);
4062         atomic_inc(&adapter->running_cap_crqs);
4063         ibmvnic_send_crq(adapter, &crq);
4064
4065         crq.query_capability.capability = cpu_to_be16(OPT_RX_COMP_QUEUES);
4066         atomic_inc(&adapter->running_cap_crqs);
4067         ibmvnic_send_crq(adapter, &crq);
4068
4069         crq.query_capability.capability =
4070                         cpu_to_be16(OPT_RX_BUFADD_Q_PER_RX_COMP_Q);
4071         atomic_inc(&adapter->running_cap_crqs);
4072         ibmvnic_send_crq(adapter, &crq);
4073
4074         crq.query_capability.capability =
4075                         cpu_to_be16(OPT_TX_ENTRIES_PER_SUBCRQ);
4076         atomic_inc(&adapter->running_cap_crqs);
4077         ibmvnic_send_crq(adapter, &crq);
4078
4079         crq.query_capability.capability =
4080                         cpu_to_be16(OPT_RXBA_ENTRIES_PER_SUBCRQ);
4081         atomic_inc(&adapter->running_cap_crqs);
4082         ibmvnic_send_crq(adapter, &crq);
4083
4084         crq.query_capability.capability = cpu_to_be16(TX_RX_DESC_REQ);
4085         atomic_inc(&adapter->running_cap_crqs);
4086         ibmvnic_send_crq(adapter, &crq);
4087 }
4088
4089 static void send_query_ip_offload(struct ibmvnic_adapter *adapter)
4090 {
4091         int buf_sz = sizeof(struct ibmvnic_query_ip_offload_buffer);
4092         struct device *dev = &adapter->vdev->dev;
4093         union ibmvnic_crq crq;
4094
4095         adapter->ip_offload_tok =
4096                 dma_map_single(dev,
4097                                &adapter->ip_offload_buf,
4098                                buf_sz,
4099                                DMA_FROM_DEVICE);
4100
4101         if (dma_mapping_error(dev, adapter->ip_offload_tok)) {
4102                 if (!firmware_has_feature(FW_FEATURE_CMO))
4103                         dev_err(dev, "Couldn't map offload buffer\n");
4104                 return;
4105         }
4106
4107         memset(&crq, 0, sizeof(crq));
4108         crq.query_ip_offload.first = IBMVNIC_CRQ_CMD;
4109         crq.query_ip_offload.cmd = QUERY_IP_OFFLOAD;
4110         crq.query_ip_offload.len = cpu_to_be32(buf_sz);
4111         crq.query_ip_offload.ioba =
4112             cpu_to_be32(adapter->ip_offload_tok);
4113
4114         ibmvnic_send_crq(adapter, &crq);
4115 }
4116
4117 static void send_control_ip_offload(struct ibmvnic_adapter *adapter)
4118 {
4119         struct ibmvnic_control_ip_offload_buffer *ctrl_buf = &adapter->ip_offload_ctrl;
4120         struct ibmvnic_query_ip_offload_buffer *buf = &adapter->ip_offload_buf;
4121         struct device *dev = &adapter->vdev->dev;
4122         netdev_features_t old_hw_features = 0;
4123         union ibmvnic_crq crq;
4124
4125         adapter->ip_offload_ctrl_tok =
4126                 dma_map_single(dev,
4127                                ctrl_buf,
4128                                sizeof(adapter->ip_offload_ctrl),
4129                                DMA_TO_DEVICE);
4130
4131         if (dma_mapping_error(dev, adapter->ip_offload_ctrl_tok)) {
4132                 dev_err(dev, "Couldn't map ip offload control buffer\n");
4133                 return;
4134         }
4135
4136         ctrl_buf->len = cpu_to_be32(sizeof(adapter->ip_offload_ctrl));
4137         ctrl_buf->version = cpu_to_be32(INITIAL_VERSION_IOB);
4138         ctrl_buf->ipv4_chksum = buf->ipv4_chksum;
4139         ctrl_buf->ipv6_chksum = buf->ipv6_chksum;
4140         ctrl_buf->tcp_ipv4_chksum = buf->tcp_ipv4_chksum;
4141         ctrl_buf->udp_ipv4_chksum = buf->udp_ipv4_chksum;
4142         ctrl_buf->tcp_ipv6_chksum = buf->tcp_ipv6_chksum;
4143         ctrl_buf->udp_ipv6_chksum = buf->udp_ipv6_chksum;
4144         ctrl_buf->large_tx_ipv4 = buf->large_tx_ipv4;
4145         ctrl_buf->large_tx_ipv6 = buf->large_tx_ipv6;
4146
4147         /* large_rx disabled for now, additional features needed */
4148         ctrl_buf->large_rx_ipv4 = 0;
4149         ctrl_buf->large_rx_ipv6 = 0;
4150
4151         if (adapter->state != VNIC_PROBING) {
4152                 old_hw_features = adapter->netdev->hw_features;
4153                 adapter->netdev->hw_features = 0;
4154         }
4155
4156         adapter->netdev->hw_features = NETIF_F_SG | NETIF_F_GSO | NETIF_F_GRO;
4157
4158         if (buf->tcp_ipv4_chksum || buf->udp_ipv4_chksum)
4159                 adapter->netdev->hw_features |= NETIF_F_IP_CSUM;
4160
4161         if (buf->tcp_ipv6_chksum || buf->udp_ipv6_chksum)
4162                 adapter->netdev->hw_features |= NETIF_F_IPV6_CSUM;
4163
4164         if ((adapter->netdev->features &
4165             (NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM)))
4166                 adapter->netdev->hw_features |= NETIF_F_RXCSUM;
4167
4168         if (buf->large_tx_ipv4)
4169                 adapter->netdev->hw_features |= NETIF_F_TSO;
4170         if (buf->large_tx_ipv6)
4171                 adapter->netdev->hw_features |= NETIF_F_TSO6;
4172
4173         if (adapter->state == VNIC_PROBING) {
4174                 adapter->netdev->features |= adapter->netdev->hw_features;
4175         } else if (old_hw_features != adapter->netdev->hw_features) {
4176                 netdev_features_t tmp = 0;
4177
4178                 /* disable features no longer supported */
4179                 adapter->netdev->features &= adapter->netdev->hw_features;
4180                 /* turn on features now supported if previously enabled */
4181                 tmp = (old_hw_features ^ adapter->netdev->hw_features) &
4182                         adapter->netdev->hw_features;
4183                 adapter->netdev->features |=
4184                                 tmp & adapter->netdev->wanted_features;
4185         }
4186
4187         memset(&crq, 0, sizeof(crq));
4188         crq.control_ip_offload.first = IBMVNIC_CRQ_CMD;
4189         crq.control_ip_offload.cmd = CONTROL_IP_OFFLOAD;
4190         crq.control_ip_offload.len =
4191             cpu_to_be32(sizeof(adapter->ip_offload_ctrl));
4192         crq.control_ip_offload.ioba = cpu_to_be32(adapter->ip_offload_ctrl_tok);
4193         ibmvnic_send_crq(adapter, &crq);
4194 }
4195
4196 static void handle_vpd_size_rsp(union ibmvnic_crq *crq,
4197                                 struct ibmvnic_adapter *adapter)
4198 {
4199         struct device *dev = &adapter->vdev->dev;
4200
4201         if (crq->get_vpd_size_rsp.rc.code) {
4202                 dev_err(dev, "Error retrieving VPD size, rc=%x\n",
4203                         crq->get_vpd_size_rsp.rc.code);
4204                 complete(&adapter->fw_done);
4205                 return;
4206         }
4207
4208         adapter->vpd->len = be64_to_cpu(crq->get_vpd_size_rsp.len);
4209         complete(&adapter->fw_done);
4210 }
4211
4212 static void handle_vpd_rsp(union ibmvnic_crq *crq,
4213                            struct ibmvnic_adapter *adapter)
4214 {
4215         struct device *dev = &adapter->vdev->dev;
4216         unsigned char *substr = NULL;
4217         u8 fw_level_len = 0;
4218
4219         memset(adapter->fw_version, 0, 32);
4220
4221         dma_unmap_single(dev, adapter->vpd->dma_addr, adapter->vpd->len,
4222                          DMA_FROM_DEVICE);
4223
4224         if (crq->get_vpd_rsp.rc.code) {
4225                 dev_err(dev, "Error retrieving VPD from device, rc=%x\n",
4226                         crq->get_vpd_rsp.rc.code);
4227                 goto complete;
4228         }
4229
4230         /* get the position of the firmware version info
4231          * located after the ASCII 'RM' substring in the buffer
4232          */
4233         substr = strnstr(adapter->vpd->buff, "RM", adapter->vpd->len);
4234         if (!substr) {
4235                 dev_info(dev, "Warning - No FW level has been provided in the VPD buffer by the VIOS Server\n");
4236                 goto complete;
4237         }
4238
4239         /* get length of firmware level ASCII substring */
4240         if ((substr + 2) < (adapter->vpd->buff + adapter->vpd->len)) {
4241                 fw_level_len = *(substr + 2);
4242         } else {
4243                 dev_info(dev, "Length of FW substr extrapolated VDP buff\n");
4244                 goto complete;
4245         }
4246
4247         /* copy firmware version string from vpd into adapter */
4248         if ((substr + 3 + fw_level_len) <
4249             (adapter->vpd->buff + adapter->vpd->len)) {
4250                 strncpy((char *)adapter->fw_version, substr + 3, fw_level_len);
4251         } else {
4252                 dev_info(dev, "FW substr extrapolated VPD buff\n");
4253         }
4254
4255 complete:
4256         if (adapter->fw_version[0] == '\0')
4257                 strncpy((char *)adapter->fw_version, "N/A", 3 * sizeof(char));
4258         complete(&adapter->fw_done);
4259 }
4260
4261 static void handle_query_ip_offload_rsp(struct ibmvnic_adapter *adapter)
4262 {
4263         struct device *dev = &adapter->vdev->dev;
4264         struct ibmvnic_query_ip_offload_buffer *buf = &adapter->ip_offload_buf;
4265         int i;
4266
4267         dma_unmap_single(dev, adapter->ip_offload_tok,
4268                          sizeof(adapter->ip_offload_buf), DMA_FROM_DEVICE);
4269
4270         netdev_dbg(adapter->netdev, "Query IP Offload Buffer:\n");
4271         for (i = 0; i < (sizeof(adapter->ip_offload_buf) - 1) / 8 + 1; i++)
4272                 netdev_dbg(adapter->netdev, "%016lx\n",
4273                            ((unsigned long *)(buf))[i]);
4274
4275         netdev_dbg(adapter->netdev, "ipv4_chksum = %d\n", buf->ipv4_chksum);
4276         netdev_dbg(adapter->netdev, "ipv6_chksum = %d\n", buf->ipv6_chksum);
4277         netdev_dbg(adapter->netdev, "tcp_ipv4_chksum = %d\n",
4278                    buf->tcp_ipv4_chksum);
4279         netdev_dbg(adapter->netdev, "tcp_ipv6_chksum = %d\n",
4280                    buf->tcp_ipv6_chksum);
4281         netdev_dbg(adapter->netdev, "udp_ipv4_chksum = %d\n",
4282                    buf->udp_ipv4_chksum);
4283         netdev_dbg(adapter->netdev, "udp_ipv6_chksum = %d\n",
4284                    buf->udp_ipv6_chksum);
4285         netdev_dbg(adapter->netdev, "large_tx_ipv4 = %d\n",
4286                    buf->large_tx_ipv4);
4287         netdev_dbg(adapter->netdev, "large_tx_ipv6 = %d\n",
4288                    buf->large_tx_ipv6);
4289         netdev_dbg(adapter->netdev, "large_rx_ipv4 = %d\n",
4290                    buf->large_rx_ipv4);
4291         netdev_dbg(adapter->netdev, "large_rx_ipv6 = %d\n",
4292                    buf->large_rx_ipv6);
4293         netdev_dbg(adapter->netdev, "max_ipv4_hdr_sz = %d\n",
4294                    buf->max_ipv4_header_size);
4295         netdev_dbg(adapter->netdev, "max_ipv6_hdr_sz = %d\n",
4296                    buf->max_ipv6_header_size);
4297         netdev_dbg(adapter->netdev, "max_tcp_hdr_size = %d\n",
4298                    buf->max_tcp_header_size);
4299         netdev_dbg(adapter->netdev, "max_udp_hdr_size = %d\n",
4300                    buf->max_udp_header_size);
4301         netdev_dbg(adapter->netdev, "max_large_tx_size = %d\n",
4302                    buf->max_large_tx_size);
4303         netdev_dbg(adapter->netdev, "max_large_rx_size = %d\n",
4304                    buf->max_large_rx_size);
4305         netdev_dbg(adapter->netdev, "ipv6_ext_hdr = %d\n",
4306                    buf->ipv6_extension_header);
4307         netdev_dbg(adapter->netdev, "tcp_pseudosum_req = %d\n",
4308                    buf->tcp_pseudosum_req);
4309         netdev_dbg(adapter->netdev, "num_ipv6_ext_hd = %d\n",
4310                    buf->num_ipv6_ext_headers);
4311         netdev_dbg(adapter->netdev, "off_ipv6_ext_hd = %d\n",
4312                    buf->off_ipv6_ext_headers);
4313
4314         send_control_ip_offload(adapter);
4315 }
4316
4317 static const char *ibmvnic_fw_err_cause(u16 cause)
4318 {
4319         switch (cause) {
4320         case ADAPTER_PROBLEM:
4321                 return "adapter problem";
4322         case BUS_PROBLEM:
4323                 return "bus problem";
4324         case FW_PROBLEM:
4325                 return "firmware problem";
4326         case DD_PROBLEM:
4327                 return "device driver problem";
4328         case EEH_RECOVERY:
4329                 return "EEH recovery";
4330         case FW_UPDATED:
4331                 return "firmware updated";
4332         case LOW_MEMORY:
4333                 return "low Memory";
4334         default:
4335                 return "unknown";
4336         }
4337 }
4338
4339 static void handle_error_indication(union ibmvnic_crq *crq,
4340                                     struct ibmvnic_adapter *adapter)
4341 {
4342         struct device *dev = &adapter->vdev->dev;
4343         u16 cause;
4344
4345         cause = be16_to_cpu(crq->error_indication.error_cause);
4346
4347         dev_warn_ratelimited(dev,
4348                              "Firmware reports %serror, cause: %s. Starting recovery...\n",
4349                              crq->error_indication.flags
4350                                 & IBMVNIC_FATAL_ERROR ? "FATAL " : "",
4351                              ibmvnic_fw_err_cause(cause));
4352
4353         if (crq->error_indication.flags & IBMVNIC_FATAL_ERROR)
4354                 ibmvnic_reset(adapter, VNIC_RESET_FATAL);
4355         else
4356                 ibmvnic_reset(adapter, VNIC_RESET_NON_FATAL);
4357 }
4358
4359 static int handle_change_mac_rsp(union ibmvnic_crq *crq,
4360                                  struct ibmvnic_adapter *adapter)
4361 {
4362         struct net_device *netdev = adapter->netdev;
4363         struct device *dev = &adapter->vdev->dev;
4364         long rc;
4365
4366         rc = crq->change_mac_addr_rsp.rc.code;
4367         if (rc) {
4368                 dev_err(dev, "Error %ld in CHANGE_MAC_ADDR_RSP\n", rc);
4369                 goto out;
4370         }
4371         /* crq->change_mac_addr.mac_addr is the requested one
4372          * crq->change_mac_addr_rsp.mac_addr is the returned valid one.
4373          */
4374         ether_addr_copy(netdev->dev_addr,
4375                         &crq->change_mac_addr_rsp.mac_addr[0]);
4376         ether_addr_copy(adapter->mac_addr,
4377                         &crq->change_mac_addr_rsp.mac_addr[0]);
4378 out:
4379         complete(&adapter->fw_done);
4380         return rc;
4381 }
4382
4383 static void handle_request_cap_rsp(union ibmvnic_crq *crq,
4384                                    struct ibmvnic_adapter *adapter)
4385 {
4386         struct device *dev = &adapter->vdev->dev;
4387         u64 *req_value;
4388         char *name;
4389
4390         atomic_dec(&adapter->running_cap_crqs);
4391         switch (be16_to_cpu(crq->request_capability_rsp.capability)) {
4392         case REQ_TX_QUEUES:
4393                 req_value = &adapter->req_tx_queues;
4394                 name = "tx";
4395                 break;
4396         case REQ_RX_QUEUES:
4397                 req_value = &adapter->req_rx_queues;
4398                 name = "rx";
4399                 break;
4400         case REQ_RX_ADD_QUEUES:
4401                 req_value = &adapter->req_rx_add_queues;
4402                 name = "rx_add";
4403                 break;
4404         case REQ_TX_ENTRIES_PER_SUBCRQ:
4405                 req_value = &adapter->req_tx_entries_per_subcrq;
4406                 name = "tx_entries_per_subcrq";
4407                 break;
4408         case REQ_RX_ADD_ENTRIES_PER_SUBCRQ:
4409                 req_value = &adapter->req_rx_add_entries_per_subcrq;
4410                 name = "rx_add_entries_per_subcrq";
4411                 break;
4412         case REQ_MTU:
4413                 req_value = &adapter->req_mtu;
4414                 name = "mtu";
4415                 break;
4416         case PROMISC_REQUESTED:
4417                 req_value = &adapter->promisc;
4418                 name = "promisc";
4419                 break;
4420         default:
4421                 dev_err(dev, "Got invalid cap request rsp %d\n",
4422                         crq->request_capability.capability);
4423                 return;
4424         }
4425
4426         switch (crq->request_capability_rsp.rc.code) {
4427         case SUCCESS:
4428                 break;
4429         case PARTIALSUCCESS:
4430                 dev_info(dev, "req=%lld, rsp=%ld in %s queue, retrying.\n",
4431                          *req_value,
4432                          (long)be64_to_cpu(crq->request_capability_rsp.number),
4433                          name);
4434
4435                 if (be16_to_cpu(crq->request_capability_rsp.capability) ==
4436                     REQ_MTU) {
4437                         pr_err("mtu of %llu is not supported. Reverting.\n",
4438                                *req_value);
4439                         *req_value = adapter->fallback.mtu;
4440                 } else {
4441                         *req_value =
4442                                 be64_to_cpu(crq->request_capability_rsp.number);
4443                 }
4444
4445                 send_request_cap(adapter, 1);
4446                 return;
4447         default:
4448                 dev_err(dev, "Error %d in request cap rsp\n",
4449                         crq->request_capability_rsp.rc.code);
4450                 return;
4451         }
4452
4453         /* Done receiving requested capabilities, query IP offload support */
4454         if (atomic_read(&adapter->running_cap_crqs) == 0) {
4455                 adapter->wait_capability = false;
4456                 send_query_ip_offload(adapter);
4457         }
4458 }
4459
4460 static int handle_login_rsp(union ibmvnic_crq *login_rsp_crq,
4461                             struct ibmvnic_adapter *adapter)
4462 {
4463         struct device *dev = &adapter->vdev->dev;
4464         struct net_device *netdev = adapter->netdev;
4465         struct ibmvnic_login_rsp_buffer *login_rsp = adapter->login_rsp_buf;
4466         struct ibmvnic_login_buffer *login = adapter->login_buf;
4467         u64 *tx_handle_array;
4468         u64 *rx_handle_array;
4469         int num_tx_pools;
4470         int num_rx_pools;
4471         u64 *size_array;
4472         int i;
4473
4474         /* CHECK: Test/set of login_pending does not need to be atomic
4475          * because only ibmvnic_tasklet tests/clears this.
4476          */
4477         if (!adapter->login_pending) {
4478                 netdev_warn(netdev, "Ignoring unexpected login response\n");
4479                 return 0;
4480         }
4481         adapter->login_pending = false;
4482
4483         dma_unmap_single(dev, adapter->login_buf_token, adapter->login_buf_sz,
4484                          DMA_TO_DEVICE);
4485         dma_unmap_single(dev, adapter->login_rsp_buf_token,
4486                          adapter->login_rsp_buf_sz, DMA_FROM_DEVICE);
4487
4488         /* If the number of queues requested can't be allocated by the
4489          * server, the login response will return with code 1. We will need
4490          * to resend the login buffer with fewer queues requested.
4491          */
4492         if (login_rsp_crq->generic.rc.code) {
4493                 adapter->init_done_rc = login_rsp_crq->generic.rc.code;
4494                 complete(&adapter->init_done);
4495                 return 0;
4496         }
4497
4498         netdev->mtu = adapter->req_mtu - ETH_HLEN;
4499
4500         netdev_dbg(adapter->netdev, "Login Response Buffer:\n");
4501         for (i = 0; i < (adapter->login_rsp_buf_sz - 1) / 8 + 1; i++) {
4502                 netdev_dbg(adapter->netdev, "%016lx\n",
4503                            ((unsigned long *)(adapter->login_rsp_buf))[i]);
4504         }
4505
4506         /* Sanity checks */
4507         if (login->num_txcomp_subcrqs != login_rsp->num_txsubm_subcrqs ||
4508             (be32_to_cpu(login->num_rxcomp_subcrqs) *
4509              adapter->req_rx_add_queues !=
4510              be32_to_cpu(login_rsp->num_rxadd_subcrqs))) {
4511                 dev_err(dev, "FATAL: Inconsistent login and login rsp\n");
4512                 ibmvnic_reset(adapter, VNIC_RESET_FATAL);
4513                 return -EIO;
4514         }
4515         size_array = (u64 *)((u8 *)(adapter->login_rsp_buf) +
4516                 be32_to_cpu(adapter->login_rsp_buf->off_rxadd_buff_size));
4517         /* variable buffer sizes are not supported, so just read the
4518          * first entry.
4519          */
4520         adapter->cur_rx_buf_sz = be64_to_cpu(size_array[0]);
4521
4522         num_tx_pools = be32_to_cpu(adapter->login_rsp_buf->num_txsubm_subcrqs);
4523         num_rx_pools = be32_to_cpu(adapter->login_rsp_buf->num_rxadd_subcrqs);
4524
4525         tx_handle_array = (u64 *)((u8 *)(adapter->login_rsp_buf) +
4526                                   be32_to_cpu(adapter->login_rsp_buf->off_txsubm_subcrqs));
4527         rx_handle_array = (u64 *)((u8 *)(adapter->login_rsp_buf) +
4528                                   be32_to_cpu(adapter->login_rsp_buf->off_rxadd_subcrqs));
4529
4530         for (i = 0; i < num_tx_pools; i++)
4531                 adapter->tx_scrq[i]->handle = tx_handle_array[i];
4532
4533         for (i = 0; i < num_rx_pools; i++)
4534                 adapter->rx_scrq[i]->handle = rx_handle_array[i];
4535
4536         adapter->num_active_tx_scrqs = num_tx_pools;
4537         adapter->num_active_rx_scrqs = num_rx_pools;
4538         release_login_rsp_buffer(adapter);
4539         release_login_buffer(adapter);
4540         complete(&adapter->init_done);
4541
4542         return 0;
4543 }
4544
4545 static void handle_request_unmap_rsp(union ibmvnic_crq *crq,
4546                                      struct ibmvnic_adapter *adapter)
4547 {
4548         struct device *dev = &adapter->vdev->dev;
4549         long rc;
4550
4551         rc = crq->request_unmap_rsp.rc.code;
4552         if (rc)
4553                 dev_err(dev, "Error %ld in REQUEST_UNMAP_RSP\n", rc);
4554 }
4555
4556 static void handle_query_map_rsp(union ibmvnic_crq *crq,
4557                                  struct ibmvnic_adapter *adapter)
4558 {
4559         struct net_device *netdev = adapter->netdev;
4560         struct device *dev = &adapter->vdev->dev;
4561         long rc;
4562
4563         rc = crq->query_map_rsp.rc.code;
4564         if (rc) {
4565                 dev_err(dev, "Error %ld in QUERY_MAP_RSP\n", rc);
4566                 return;
4567         }
4568         netdev_dbg(netdev, "page_size = %d\ntot_pages = %d\nfree_pages = %d\n",
4569                    crq->query_map_rsp.page_size, crq->query_map_rsp.tot_pages,
4570                    crq->query_map_rsp.free_pages);
4571 }
4572
4573 static void handle_query_cap_rsp(union ibmvnic_crq *crq,
4574                                  struct ibmvnic_adapter *adapter)
4575 {
4576         struct net_device *netdev = adapter->netdev;
4577         struct device *dev = &adapter->vdev->dev;
4578         long rc;
4579
4580         atomic_dec(&adapter->running_cap_crqs);
4581         netdev_dbg(netdev, "Outstanding queries: %d\n",
4582                    atomic_read(&adapter->running_cap_crqs));
4583         rc = crq->query_capability.rc.code;
4584         if (rc) {
4585                 dev_err(dev, "Error %ld in QUERY_CAP_RSP\n", rc);
4586                 goto out;
4587         }
4588
4589         switch (be16_to_cpu(crq->query_capability.capability)) {
4590         case MIN_TX_QUEUES:
4591                 adapter->min_tx_queues =
4592                     be64_to_cpu(crq->query_capability.number);
4593                 netdev_dbg(netdev, "min_tx_queues = %lld\n",
4594                            adapter->min_tx_queues);
4595                 break;
4596         case MIN_RX_QUEUES:
4597                 adapter->min_rx_queues =
4598                     be64_to_cpu(crq->query_capability.number);
4599                 netdev_dbg(netdev, "min_rx_queues = %lld\n",
4600                            adapter->min_rx_queues);
4601                 break;
4602         case MIN_RX_ADD_QUEUES:
4603                 adapter->min_rx_add_queues =
4604                     be64_to_cpu(crq->query_capability.number);
4605                 netdev_dbg(netdev, "min_rx_add_queues = %lld\n",
4606                            adapter->min_rx_add_queues);
4607                 break;
4608         case MAX_TX_QUEUES:
4609                 adapter->max_tx_queues =
4610                     be64_to_cpu(crq->query_capability.number);
4611                 netdev_dbg(netdev, "max_tx_queues = %lld\n",
4612                            adapter->max_tx_queues);
4613                 break;
4614         case MAX_RX_QUEUES:
4615                 adapter->max_rx_queues =
4616                     be64_to_cpu(crq->query_capability.number);
4617                 netdev_dbg(netdev, "max_rx_queues = %lld\n",
4618                            adapter->max_rx_queues);
4619                 break;
4620         case MAX_RX_ADD_QUEUES:
4621                 adapter->max_rx_add_queues =
4622                     be64_to_cpu(crq->query_capability.number);
4623                 netdev_dbg(netdev, "max_rx_add_queues = %lld\n",
4624                            adapter->max_rx_add_queues);
4625                 break;
4626         case MIN_TX_ENTRIES_PER_SUBCRQ:
4627                 adapter->min_tx_entries_per_subcrq =
4628                     be64_to_cpu(crq->query_capability.number);
4629                 netdev_dbg(netdev, "min_tx_entries_per_subcrq = %lld\n",
4630                            adapter->min_tx_entries_per_subcrq);
4631                 break;
4632         case MIN_RX_ADD_ENTRIES_PER_SUBCRQ:
4633                 adapter->min_rx_add_entries_per_subcrq =
4634                     be64_to_cpu(crq->query_capability.number);
4635                 netdev_dbg(netdev, "min_rx_add_entrs_per_subcrq = %lld\n",
4636                            adapter->min_rx_add_entries_per_subcrq);
4637                 break;
4638         case MAX_TX_ENTRIES_PER_SUBCRQ:
4639                 adapter->max_tx_entries_per_subcrq =
4640                     be64_to_cpu(crq->query_capability.number);
4641                 netdev_dbg(netdev, "max_tx_entries_per_subcrq = %lld\n",
4642                            adapter->max_tx_entries_per_subcrq);
4643                 break;
4644         case MAX_RX_ADD_ENTRIES_PER_SUBCRQ:
4645                 adapter->max_rx_add_entries_per_subcrq =
4646                     be64_to_cpu(crq->query_capability.number);
4647                 netdev_dbg(netdev, "max_rx_add_entrs_per_subcrq = %lld\n",
4648                            adapter->max_rx_add_entries_per_subcrq);
4649                 break;
4650         case TCP_IP_OFFLOAD:
4651                 adapter->tcp_ip_offload =
4652                     be64_to_cpu(crq->query_capability.number);
4653                 netdev_dbg(netdev, "tcp_ip_offload = %lld\n",
4654                            adapter->tcp_ip_offload);
4655                 break;
4656         case PROMISC_SUPPORTED:
4657                 adapter->promisc_supported =
4658                     be64_to_cpu(crq->query_capability.number);
4659                 netdev_dbg(netdev, "promisc_supported = %lld\n",
4660                            adapter->promisc_supported);
4661                 break;
4662         case MIN_MTU:
4663                 adapter->min_mtu = be64_to_cpu(crq->query_capability.number);
4664                 netdev->min_mtu = adapter->min_mtu - ETH_HLEN;
4665                 netdev_dbg(netdev, "min_mtu = %lld\n", adapter->min_mtu);
4666                 break;
4667         case MAX_MTU:
4668                 adapter->max_mtu = be64_to_cpu(crq->query_capability.number);
4669                 netdev->max_mtu = adapter->max_mtu - ETH_HLEN;
4670                 netdev_dbg(netdev, "max_mtu = %lld\n", adapter->max_mtu);
4671                 break;
4672         case MAX_MULTICAST_FILTERS:
4673                 adapter->max_multicast_filters =
4674                     be64_to_cpu(crq->query_capability.number);
4675                 netdev_dbg(netdev, "max_multicast_filters = %lld\n",
4676                            adapter->max_multicast_filters);
4677                 break;
4678         case VLAN_HEADER_INSERTION:
4679                 adapter->vlan_header_insertion =
4680                     be64_to_cpu(crq->query_capability.number);
4681                 if (adapter->vlan_header_insertion)
4682                         netdev->features |= NETIF_F_HW_VLAN_STAG_TX;
4683                 netdev_dbg(netdev, "vlan_header_insertion = %lld\n",
4684                            adapter->vlan_header_insertion);
4685                 break;
4686         case RX_VLAN_HEADER_INSERTION:
4687                 adapter->rx_vlan_header_insertion =
4688                     be64_to_cpu(crq->query_capability.number);
4689                 netdev_dbg(netdev, "rx_vlan_header_insertion = %lld\n",
4690                            adapter->rx_vlan_header_insertion);
4691                 break;
4692         case MAX_TX_SG_ENTRIES:
4693                 adapter->max_tx_sg_entries =
4694                     be64_to_cpu(crq->query_capability.number);
4695                 netdev_dbg(netdev, "max_tx_sg_entries = %lld\n",
4696                            adapter->max_tx_sg_entries);
4697                 break;
4698         case RX_SG_SUPPORTED:
4699                 adapter->rx_sg_supported =
4700                     be64_to_cpu(crq->query_capability.number);
4701                 netdev_dbg(netdev, "rx_sg_supported = %lld\n",
4702                            adapter->rx_sg_supported);
4703                 break;
4704         case OPT_TX_COMP_SUB_QUEUES:
4705                 adapter->opt_tx_comp_sub_queues =
4706                     be64_to_cpu(crq->query_capability.number);
4707                 netdev_dbg(netdev, "opt_tx_comp_sub_queues = %lld\n",
4708                            adapter->opt_tx_comp_sub_queues);
4709                 break;
4710         case OPT_RX_COMP_QUEUES:
4711                 adapter->opt_rx_comp_queues =
4712                     be64_to_cpu(crq->query_capability.number);
4713                 netdev_dbg(netdev, "opt_rx_comp_queues = %lld\n",
4714                            adapter->opt_rx_comp_queues);
4715                 break;
4716         case OPT_RX_BUFADD_Q_PER_RX_COMP_Q:
4717                 adapter->opt_rx_bufadd_q_per_rx_comp_q =
4718                     be64_to_cpu(crq->query_capability.number);
4719                 netdev_dbg(netdev, "opt_rx_bufadd_q_per_rx_comp_q = %lld\n",
4720                            adapter->opt_rx_bufadd_q_per_rx_comp_q);
4721                 break;
4722         case OPT_TX_ENTRIES_PER_SUBCRQ:
4723                 adapter->opt_tx_entries_per_subcrq =
4724                     be64_to_cpu(crq->query_capability.number);
4725                 netdev_dbg(netdev, "opt_tx_entries_per_subcrq = %lld\n",
4726                            adapter->opt_tx_entries_per_subcrq);
4727                 break;
4728         case OPT_RXBA_ENTRIES_PER_SUBCRQ:
4729                 adapter->opt_rxba_entries_per_subcrq =
4730                     be64_to_cpu(crq->query_capability.number);
4731                 netdev_dbg(netdev, "opt_rxba_entries_per_subcrq = %lld\n",
4732                            adapter->opt_rxba_entries_per_subcrq);
4733                 break;
4734         case TX_RX_DESC_REQ:
4735                 adapter->tx_rx_desc_req = crq->query_capability.number;
4736                 netdev_dbg(netdev, "tx_rx_desc_req = %llx\n",
4737                            adapter->tx_rx_desc_req);
4738                 break;
4739
4740         default:
4741                 netdev_err(netdev, "Got invalid cap rsp %d\n",
4742                            crq->query_capability.capability);
4743         }
4744
4745 out:
4746         if (atomic_read(&adapter->running_cap_crqs) == 0) {
4747                 adapter->wait_capability = false;
4748                 send_request_cap(adapter, 0);
4749         }
4750 }
4751
4752 static int send_query_phys_parms(struct ibmvnic_adapter *adapter)
4753 {
4754         union ibmvnic_crq crq;
4755         int rc;
4756
4757         memset(&crq, 0, sizeof(crq));
4758         crq.query_phys_parms.first = IBMVNIC_CRQ_CMD;
4759         crq.query_phys_parms.cmd = QUERY_PHYS_PARMS;
4760
4761         mutex_lock(&adapter->fw_lock);
4762         adapter->fw_done_rc = 0;
4763         reinit_completion(&adapter->fw_done);
4764
4765         rc = ibmvnic_send_crq(adapter, &crq);
4766         if (rc) {
4767                 mutex_unlock(&adapter->fw_lock);
4768                 return rc;
4769         }
4770
4771         rc = ibmvnic_wait_for_completion(adapter, &adapter->fw_done, 10000);
4772         if (rc) {
4773                 mutex_unlock(&adapter->fw_lock);
4774                 return rc;
4775         }
4776
4777         mutex_unlock(&adapter->fw_lock);
4778         return adapter->fw_done_rc ? -EIO : 0;
4779 }
4780
4781 static int handle_query_phys_parms_rsp(union ibmvnic_crq *crq,
4782                                        struct ibmvnic_adapter *adapter)
4783 {
4784         struct net_device *netdev = adapter->netdev;
4785         int rc;
4786         __be32 rspeed = cpu_to_be32(crq->query_phys_parms_rsp.speed);
4787
4788         rc = crq->query_phys_parms_rsp.rc.code;
4789         if (rc) {
4790                 netdev_err(netdev, "Error %d in QUERY_PHYS_PARMS\n", rc);
4791                 return rc;
4792         }
4793         switch (rspeed) {
4794         case IBMVNIC_10MBPS:
4795                 adapter->speed = SPEED_10;
4796                 break;
4797         case IBMVNIC_100MBPS:
4798                 adapter->speed = SPEED_100;
4799                 break;
4800         case IBMVNIC_1GBPS:
4801                 adapter->speed = SPEED_1000;
4802                 break;
4803         case IBMVNIC_10GBPS:
4804                 adapter->speed = SPEED_10000;
4805                 break;
4806         case IBMVNIC_25GBPS:
4807                 adapter->speed = SPEED_25000;
4808                 break;
4809         case IBMVNIC_40GBPS:
4810                 adapter->speed = SPEED_40000;
4811                 break;
4812         case IBMVNIC_50GBPS:
4813                 adapter->speed = SPEED_50000;
4814                 break;
4815         case IBMVNIC_100GBPS:
4816                 adapter->speed = SPEED_100000;
4817                 break;
4818         case IBMVNIC_200GBPS:
4819                 adapter->speed = SPEED_200000;
4820                 break;
4821         default:
4822                 if (netif_carrier_ok(netdev))
4823                         netdev_warn(netdev, "Unknown speed 0x%08x\n", rspeed);
4824                 adapter->speed = SPEED_UNKNOWN;
4825         }
4826         if (crq->query_phys_parms_rsp.flags1 & IBMVNIC_FULL_DUPLEX)
4827                 adapter->duplex = DUPLEX_FULL;
4828         else if (crq->query_phys_parms_rsp.flags1 & IBMVNIC_HALF_DUPLEX)
4829                 adapter->duplex = DUPLEX_HALF;
4830         else
4831                 adapter->duplex = DUPLEX_UNKNOWN;
4832
4833         return rc;
4834 }
4835
4836 static void ibmvnic_handle_crq(union ibmvnic_crq *crq,
4837                                struct ibmvnic_adapter *adapter)
4838 {
4839         struct ibmvnic_generic_crq *gen_crq = &crq->generic;
4840         struct net_device *netdev = adapter->netdev;
4841         struct device *dev = &adapter->vdev->dev;
4842         u64 *u64_crq = (u64 *)crq;
4843         long rc;
4844
4845         netdev_dbg(netdev, "Handling CRQ: %016lx %016lx\n",
4846                    (unsigned long)cpu_to_be64(u64_crq[0]),
4847                    (unsigned long)cpu_to_be64(u64_crq[1]));
4848         switch (gen_crq->first) {
4849         case IBMVNIC_CRQ_INIT_RSP:
4850                 switch (gen_crq->cmd) {
4851                 case IBMVNIC_CRQ_INIT:
4852                         dev_info(dev, "Partner initialized\n");
4853                         adapter->from_passive_init = true;
4854                         /* Discard any stale login responses from prev reset.
4855                          * CHECK: should we clear even on INIT_COMPLETE?
4856                          */
4857                         adapter->login_pending = false;
4858
4859                         if (!completion_done(&adapter->init_done)) {
4860                                 complete(&adapter->init_done);
4861                                 adapter->init_done_rc = -EIO;
4862                         }
4863                         rc = ibmvnic_reset(adapter, VNIC_RESET_FAILOVER);
4864                         if (rc && rc != -EBUSY) {
4865                                 /* We were unable to schedule the failover
4866                                  * reset either because the adapter was still
4867                                  * probing (eg: during kexec) or we could not
4868                                  * allocate memory. Clear the failover_pending
4869                                  * flag since no one else will. We ignore
4870                                  * EBUSY because it means either FAILOVER reset
4871                                  * is already scheduled or the adapter is
4872                                  * being removed.
4873                                  */
4874                                 netdev_err(netdev,
4875                                            "Error %ld scheduling failover reset\n",
4876                                            rc);
4877                                 adapter->failover_pending = false;
4878                         }
4879                         break;
4880                 case IBMVNIC_CRQ_INIT_COMPLETE:
4881                         dev_info(dev, "Partner initialization complete\n");
4882                         adapter->crq.active = true;
4883                         send_version_xchg(adapter);
4884                         break;
4885                 default:
4886                         dev_err(dev, "Unknown crq cmd: %d\n", gen_crq->cmd);
4887                 }
4888                 return;
4889         case IBMVNIC_CRQ_XPORT_EVENT:
4890                 netif_carrier_off(netdev);
4891                 adapter->crq.active = false;
4892                 /* terminate any thread waiting for a response
4893                  * from the device
4894                  */
4895                 if (!completion_done(&adapter->fw_done)) {
4896                         adapter->fw_done_rc = -EIO;
4897                         complete(&adapter->fw_done);
4898                 }
4899                 if (!completion_done(&adapter->stats_done))
4900                         complete(&adapter->stats_done);
4901                 if (test_bit(0, &adapter->resetting))
4902                         adapter->force_reset_recovery = true;
4903                 if (gen_crq->cmd == IBMVNIC_PARTITION_MIGRATED) {
4904                         dev_info(dev, "Migrated, re-enabling adapter\n");
4905                         ibmvnic_reset(adapter, VNIC_RESET_MOBILITY);
4906                 } else if (gen_crq->cmd == IBMVNIC_DEVICE_FAILOVER) {
4907                         dev_info(dev, "Backing device failover detected\n");
4908                         adapter->failover_pending = true;
4909                 } else {
4910                         /* The adapter lost the connection */
4911                         dev_err(dev, "Virtual Adapter failed (rc=%d)\n",
4912                                 gen_crq->cmd);
4913                         ibmvnic_reset(adapter, VNIC_RESET_FATAL);
4914                 }
4915                 return;
4916         case IBMVNIC_CRQ_CMD_RSP:
4917                 break;
4918         default:
4919                 dev_err(dev, "Got an invalid msg type 0x%02x\n",
4920                         gen_crq->first);
4921                 return;
4922         }
4923
4924         switch (gen_crq->cmd) {
4925         case VERSION_EXCHANGE_RSP:
4926                 rc = crq->version_exchange_rsp.rc.code;
4927                 if (rc) {
4928                         dev_err(dev, "Error %ld in VERSION_EXCHG_RSP\n", rc);
4929                         break;
4930                 }
4931                 ibmvnic_version =
4932                             be16_to_cpu(crq->version_exchange_rsp.version);
4933                 dev_info(dev, "Partner protocol version is %d\n",
4934                          ibmvnic_version);
4935                 send_query_cap(adapter);
4936                 break;
4937         case QUERY_CAPABILITY_RSP:
4938                 handle_query_cap_rsp(crq, adapter);
4939                 break;
4940         case QUERY_MAP_RSP:
4941                 handle_query_map_rsp(crq, adapter);
4942                 break;
4943         case REQUEST_MAP_RSP:
4944                 adapter->fw_done_rc = crq->request_map_rsp.rc.code;
4945                 complete(&adapter->fw_done);
4946                 break;
4947         case REQUEST_UNMAP_RSP:
4948                 handle_request_unmap_rsp(crq, adapter);
4949                 break;
4950         case REQUEST_CAPABILITY_RSP:
4951                 handle_request_cap_rsp(crq, adapter);
4952                 break;
4953         case LOGIN_RSP:
4954                 netdev_dbg(netdev, "Got Login Response\n");
4955                 handle_login_rsp(crq, adapter);
4956                 break;
4957         case LOGICAL_LINK_STATE_RSP:
4958                 netdev_dbg(netdev,
4959                            "Got Logical Link State Response, state: %d rc: %d\n",
4960                            crq->logical_link_state_rsp.link_state,
4961                            crq->logical_link_state_rsp.rc.code);
4962                 adapter->logical_link_state =
4963                     crq->logical_link_state_rsp.link_state;
4964                 adapter->init_done_rc = crq->logical_link_state_rsp.rc.code;
4965                 complete(&adapter->init_done);
4966                 break;
4967         case LINK_STATE_INDICATION:
4968                 netdev_dbg(netdev, "Got Logical Link State Indication\n");
4969                 adapter->phys_link_state =
4970                     crq->link_state_indication.phys_link_state;
4971                 adapter->logical_link_state =
4972                     crq->link_state_indication.logical_link_state;
4973                 if (adapter->phys_link_state && adapter->logical_link_state)
4974                         netif_carrier_on(netdev);
4975                 else
4976                         netif_carrier_off(netdev);
4977                 break;
4978         case CHANGE_MAC_ADDR_RSP:
4979                 netdev_dbg(netdev, "Got MAC address change Response\n");
4980                 adapter->fw_done_rc = handle_change_mac_rsp(crq, adapter);
4981                 break;
4982         case ERROR_INDICATION:
4983                 netdev_dbg(netdev, "Got Error Indication\n");
4984                 handle_error_indication(crq, adapter);
4985                 break;
4986         case REQUEST_STATISTICS_RSP:
4987                 netdev_dbg(netdev, "Got Statistics Response\n");
4988                 complete(&adapter->stats_done);
4989                 break;
4990         case QUERY_IP_OFFLOAD_RSP:
4991                 netdev_dbg(netdev, "Got Query IP offload Response\n");
4992                 handle_query_ip_offload_rsp(adapter);
4993                 break;
4994         case MULTICAST_CTRL_RSP:
4995                 netdev_dbg(netdev, "Got multicast control Response\n");
4996                 break;
4997         case CONTROL_IP_OFFLOAD_RSP:
4998                 netdev_dbg(netdev, "Got Control IP offload Response\n");
4999                 dma_unmap_single(dev, adapter->ip_offload_ctrl_tok,
5000                                  sizeof(adapter->ip_offload_ctrl),
5001                                  DMA_TO_DEVICE);
5002                 complete(&adapter->init_done);
5003                 break;
5004         case COLLECT_FW_TRACE_RSP:
5005                 netdev_dbg(netdev, "Got Collect firmware trace Response\n");
5006                 complete(&adapter->fw_done);
5007                 break;
5008         case GET_VPD_SIZE_RSP:
5009                 handle_vpd_size_rsp(crq, adapter);
5010                 break;
5011         case GET_VPD_RSP:
5012                 handle_vpd_rsp(crq, adapter);
5013                 break;
5014         case QUERY_PHYS_PARMS_RSP:
5015                 adapter->fw_done_rc = handle_query_phys_parms_rsp(crq, adapter);
5016                 complete(&adapter->fw_done);
5017                 break;
5018         default:
5019                 netdev_err(netdev, "Got an invalid cmd type 0x%02x\n",
5020                            gen_crq->cmd);
5021         }
5022 }
5023
5024 static irqreturn_t ibmvnic_interrupt(int irq, void *instance)
5025 {
5026         struct ibmvnic_adapter *adapter = instance;
5027
5028         tasklet_schedule(&adapter->tasklet);
5029         return IRQ_HANDLED;
5030 }
5031
5032 static void ibmvnic_tasklet(struct tasklet_struct *t)
5033 {
5034         struct ibmvnic_adapter *adapter = from_tasklet(adapter, t, tasklet);
5035         struct ibmvnic_crq_queue *queue = &adapter->crq;
5036         union ibmvnic_crq *crq;
5037         unsigned long flags;
5038         bool done = false;
5039
5040         spin_lock_irqsave(&queue->lock, flags);
5041         while (!done) {
5042                 /* Pull all the valid messages off the CRQ */
5043                 while ((crq = ibmvnic_next_crq(adapter)) != NULL) {
5044                         /* This barrier makes sure ibmvnic_next_crq()'s
5045                          * crq->generic.first & IBMVNIC_CRQ_CMD_RSP is loaded
5046                          * before ibmvnic_handle_crq()'s
5047                          * switch(gen_crq->first) and switch(gen_crq->cmd).
5048                          */
5049                         dma_rmb();
5050                         ibmvnic_handle_crq(crq, adapter);
5051                         crq->generic.first = 0;
5052                 }
5053
5054                 /* remain in tasklet until all
5055                  * capabilities responses are received
5056                  */
5057                 if (!adapter->wait_capability)
5058                         done = true;
5059         }
5060         /* if capabilities CRQ's were sent in this tasklet, the following
5061          * tasklet must wait until all responses are received
5062          */
5063         if (atomic_read(&adapter->running_cap_crqs) != 0)
5064                 adapter->wait_capability = true;
5065         spin_unlock_irqrestore(&queue->lock, flags);
5066 }
5067
5068 static int ibmvnic_reenable_crq_queue(struct ibmvnic_adapter *adapter)
5069 {
5070         struct vio_dev *vdev = adapter->vdev;
5071         int rc;
5072
5073         do {
5074                 rc = plpar_hcall_norets(H_ENABLE_CRQ, vdev->unit_address);
5075         } while (rc == H_IN_PROGRESS || rc == H_BUSY || H_IS_LONG_BUSY(rc));
5076
5077         if (rc)
5078                 dev_err(&vdev->dev, "Error enabling adapter (rc=%d)\n", rc);
5079
5080         return rc;
5081 }
5082
5083 static int ibmvnic_reset_crq(struct ibmvnic_adapter *adapter)
5084 {
5085         struct ibmvnic_crq_queue *crq = &adapter->crq;
5086         struct device *dev = &adapter->vdev->dev;
5087         struct vio_dev *vdev = adapter->vdev;
5088         int rc;
5089
5090         /* Close the CRQ */
5091         do {
5092                 rc = plpar_hcall_norets(H_FREE_CRQ, vdev->unit_address);
5093         } while (rc == H_BUSY || H_IS_LONG_BUSY(rc));
5094
5095         /* Clean out the queue */
5096         if (!crq->msgs)
5097                 return -EINVAL;
5098
5099         memset(crq->msgs, 0, PAGE_SIZE);
5100         crq->cur = 0;
5101         crq->active = false;
5102
5103         /* And re-open it again */
5104         rc = plpar_hcall_norets(H_REG_CRQ, vdev->unit_address,
5105                                 crq->msg_token, PAGE_SIZE);
5106
5107         if (rc == H_CLOSED)
5108                 /* Adapter is good, but other end is not ready */
5109                 dev_warn(dev, "Partner adapter not ready\n");
5110         else if (rc != 0)
5111                 dev_warn(dev, "Couldn't register crq (rc=%d)\n", rc);
5112
5113         return rc;
5114 }
5115
5116 static void release_crq_queue(struct ibmvnic_adapter *adapter)
5117 {
5118         struct ibmvnic_crq_queue *crq = &adapter->crq;
5119         struct vio_dev *vdev = adapter->vdev;
5120         long rc;
5121
5122         if (!crq->msgs)
5123                 return;
5124
5125         netdev_dbg(adapter->netdev, "Releasing CRQ\n");
5126         free_irq(vdev->irq, adapter);
5127         tasklet_kill(&adapter->tasklet);
5128         do {
5129                 rc = plpar_hcall_norets(H_FREE_CRQ, vdev->unit_address);
5130         } while (rc == H_BUSY || H_IS_LONG_BUSY(rc));
5131
5132         dma_unmap_single(&vdev->dev, crq->msg_token, PAGE_SIZE,
5133                          DMA_BIDIRECTIONAL);
5134         free_page((unsigned long)crq->msgs);
5135         crq->msgs = NULL;
5136         crq->active = false;
5137 }
5138
5139 static int init_crq_queue(struct ibmvnic_adapter *adapter)
5140 {
5141         struct ibmvnic_crq_queue *crq = &adapter->crq;
5142         struct device *dev = &adapter->vdev->dev;
5143         struct vio_dev *vdev = adapter->vdev;
5144         int rc, retrc = -ENOMEM;
5145
5146         if (crq->msgs)
5147                 return 0;
5148
5149         crq->msgs = (union ibmvnic_crq *)get_zeroed_page(GFP_KERNEL);
5150         /* Should we allocate more than one page? */
5151
5152         if (!crq->msgs)
5153                 return -ENOMEM;
5154
5155         crq->size = PAGE_SIZE / sizeof(*crq->msgs);
5156         crq->msg_token = dma_map_single(dev, crq->msgs, PAGE_SIZE,
5157                                         DMA_BIDIRECTIONAL);
5158         if (dma_mapping_error(dev, crq->msg_token))
5159                 goto map_failed;
5160
5161         rc = plpar_hcall_norets(H_REG_CRQ, vdev->unit_address,
5162                                 crq->msg_token, PAGE_SIZE);
5163
5164         if (rc == H_RESOURCE)
5165                 /* maybe kexecing and resource is busy. try a reset */
5166                 rc = ibmvnic_reset_crq(adapter);
5167         retrc = rc;
5168
5169         if (rc == H_CLOSED) {
5170                 dev_warn(dev, "Partner adapter not ready\n");
5171         } else if (rc) {
5172                 dev_warn(dev, "Error %d opening adapter\n", rc);
5173                 goto reg_crq_failed;
5174         }
5175
5176         retrc = 0;
5177
5178         tasklet_setup(&adapter->tasklet, (void *)ibmvnic_tasklet);
5179
5180         netdev_dbg(adapter->netdev, "registering irq 0x%x\n", vdev->irq);
5181         snprintf(crq->name, sizeof(crq->name), "ibmvnic-%x",
5182                  adapter->vdev->unit_address);
5183         rc = request_irq(vdev->irq, ibmvnic_interrupt, 0, crq->name, adapter);
5184         if (rc) {
5185                 dev_err(dev, "Couldn't register irq 0x%x. rc=%d\n",
5186                         vdev->irq, rc);
5187                 goto req_irq_failed;
5188         }
5189
5190         rc = vio_enable_interrupts(vdev);
5191         if (rc) {
5192                 dev_err(dev, "Error %d enabling interrupts\n", rc);
5193                 goto req_irq_failed;
5194         }
5195
5196         crq->cur = 0;
5197         spin_lock_init(&crq->lock);
5198
5199         return retrc;
5200
5201 req_irq_failed:
5202         tasklet_kill(&adapter->tasklet);
5203         do {
5204                 rc = plpar_hcall_norets(H_FREE_CRQ, vdev->unit_address);
5205         } while (rc == H_BUSY || H_IS_LONG_BUSY(rc));
5206 reg_crq_failed:
5207         dma_unmap_single(dev, crq->msg_token, PAGE_SIZE, DMA_BIDIRECTIONAL);
5208 map_failed:
5209         free_page((unsigned long)crq->msgs);
5210         crq->msgs = NULL;
5211         return retrc;
5212 }
5213
5214 static int ibmvnic_reset_init(struct ibmvnic_adapter *adapter, bool reset)
5215 {
5216         struct device *dev = &adapter->vdev->dev;
5217         unsigned long timeout = msecs_to_jiffies(20000);
5218         u64 old_num_rx_queues = adapter->req_rx_queues;
5219         u64 old_num_tx_queues = adapter->req_tx_queues;
5220         int rc;
5221
5222         adapter->from_passive_init = false;
5223
5224         if (reset)
5225                 reinit_completion(&adapter->init_done);
5226
5227         adapter->init_done_rc = 0;
5228         rc = ibmvnic_send_crq_init(adapter);
5229         if (rc) {
5230                 dev_err(dev, "Send crq init failed with error %d\n", rc);
5231                 return rc;
5232         }
5233
5234         if (!wait_for_completion_timeout(&adapter->init_done, timeout)) {
5235                 dev_err(dev, "Initialization sequence timed out\n");
5236                 return -1;
5237         }
5238
5239         if (adapter->init_done_rc) {
5240                 release_crq_queue(adapter);
5241                 return adapter->init_done_rc;
5242         }
5243
5244         if (adapter->from_passive_init) {
5245                 adapter->state = VNIC_OPEN;
5246                 adapter->from_passive_init = false;
5247                 return -1;
5248         }
5249
5250         if (reset &&
5251             test_bit(0, &adapter->resetting) && !adapter->wait_for_reset &&
5252             adapter->reset_reason != VNIC_RESET_MOBILITY) {
5253                 if (adapter->req_rx_queues != old_num_rx_queues ||
5254                     adapter->req_tx_queues != old_num_tx_queues) {
5255                         release_sub_crqs(adapter, 0);
5256                         rc = init_sub_crqs(adapter);
5257                 } else {
5258                         rc = reset_sub_crq_queues(adapter);
5259                 }
5260         } else {
5261                 rc = init_sub_crqs(adapter);
5262         }
5263
5264         if (rc) {
5265                 dev_err(dev, "Initialization of sub crqs failed\n");
5266                 release_crq_queue(adapter);
5267                 return rc;
5268         }
5269
5270         rc = init_sub_crq_irqs(adapter);
5271         if (rc) {
5272                 dev_err(dev, "Failed to initialize sub crq irqs\n");
5273                 release_crq_queue(adapter);
5274         }
5275
5276         return rc;
5277 }
5278
5279 static struct device_attribute dev_attr_failover;
5280
5281 static int ibmvnic_probe(struct vio_dev *dev, const struct vio_device_id *id)
5282 {
5283         struct ibmvnic_adapter *adapter;
5284         struct net_device *netdev;
5285         unsigned char *mac_addr_p;
5286         int rc;
5287
5288         dev_dbg(&dev->dev, "entering ibmvnic_probe for UA 0x%x\n",
5289                 dev->unit_address);
5290
5291         mac_addr_p = (unsigned char *)vio_get_attribute(dev,
5292                                                         VETH_MAC_ADDR, NULL);
5293         if (!mac_addr_p) {
5294                 dev_err(&dev->dev,
5295                         "(%s:%3.3d) ERROR: Can't find MAC_ADDR attribute\n",
5296                         __FILE__, __LINE__);
5297                 return 0;
5298         }
5299
5300         netdev = alloc_etherdev_mq(sizeof(struct ibmvnic_adapter),
5301                                    IBMVNIC_MAX_QUEUES);
5302         if (!netdev)
5303                 return -ENOMEM;
5304
5305         adapter = netdev_priv(netdev);
5306         adapter->state = VNIC_PROBING;
5307         dev_set_drvdata(&dev->dev, netdev);
5308         adapter->vdev = dev;
5309         adapter->netdev = netdev;
5310         adapter->login_pending = false;
5311
5312         ether_addr_copy(adapter->mac_addr, mac_addr_p);
5313         ether_addr_copy(netdev->dev_addr, adapter->mac_addr);
5314         netdev->irq = dev->irq;
5315         netdev->netdev_ops = &ibmvnic_netdev_ops;
5316         netdev->ethtool_ops = &ibmvnic_ethtool_ops;
5317         SET_NETDEV_DEV(netdev, &dev->dev);
5318
5319         INIT_WORK(&adapter->ibmvnic_reset, __ibmvnic_reset);
5320         INIT_DELAYED_WORK(&adapter->ibmvnic_delayed_reset,
5321                           __ibmvnic_delayed_reset);
5322         INIT_LIST_HEAD(&adapter->rwi_list);
5323         spin_lock_init(&adapter->rwi_lock);
5324         spin_lock_init(&adapter->state_lock);
5325         mutex_init(&adapter->fw_lock);
5326         init_completion(&adapter->init_done);
5327         init_completion(&adapter->fw_done);
5328         init_completion(&adapter->reset_done);
5329         init_completion(&adapter->stats_done);
5330         clear_bit(0, &adapter->resetting);
5331
5332         do {
5333                 rc = init_crq_queue(adapter);
5334                 if (rc) {
5335                         dev_err(&dev->dev, "Couldn't initialize crq. rc=%d\n",
5336                                 rc);
5337                         goto ibmvnic_init_fail;
5338                 }
5339
5340                 rc = ibmvnic_reset_init(adapter, false);
5341                 if (rc && rc != EAGAIN)
5342                         goto ibmvnic_init_fail;
5343         } while (rc == EAGAIN);
5344
5345         rc = init_stats_buffers(adapter);
5346         if (rc)
5347                 goto ibmvnic_init_fail;
5348
5349         rc = init_stats_token(adapter);
5350         if (rc)
5351                 goto ibmvnic_stats_fail;
5352
5353         netdev->mtu = adapter->req_mtu - ETH_HLEN;
5354         netdev->min_mtu = adapter->min_mtu - ETH_HLEN;
5355         netdev->max_mtu = adapter->max_mtu - ETH_HLEN;
5356
5357         rc = device_create_file(&dev->dev, &dev_attr_failover);
5358         if (rc)
5359                 goto ibmvnic_dev_file_err;
5360
5361         netif_carrier_off(netdev);
5362         rc = register_netdev(netdev);
5363         if (rc) {
5364                 dev_err(&dev->dev, "failed to register netdev rc=%d\n", rc);
5365                 goto ibmvnic_register_fail;
5366         }
5367         dev_info(&dev->dev, "ibmvnic registered\n");
5368
5369         adapter->state = VNIC_PROBED;
5370
5371         adapter->wait_for_reset = false;
5372         adapter->last_reset_time = jiffies;
5373         return 0;
5374
5375 ibmvnic_register_fail:
5376         device_remove_file(&dev->dev, &dev_attr_failover);
5377
5378 ibmvnic_dev_file_err:
5379         release_stats_token(adapter);
5380
5381 ibmvnic_stats_fail:
5382         release_stats_buffers(adapter);
5383
5384 ibmvnic_init_fail:
5385         release_sub_crqs(adapter, 1);
5386         release_crq_queue(adapter);
5387         mutex_destroy(&adapter->fw_lock);
5388         free_netdev(netdev);
5389
5390         return rc;
5391 }
5392
5393 static void ibmvnic_remove(struct vio_dev *dev)
5394 {
5395         struct net_device *netdev = dev_get_drvdata(&dev->dev);
5396         struct ibmvnic_adapter *adapter = netdev_priv(netdev);
5397         unsigned long flags;
5398
5399         spin_lock_irqsave(&adapter->state_lock, flags);
5400
5401         /* If ibmvnic_reset() is scheduling a reset, wait for it to
5402          * finish. Then, set the state to REMOVING to prevent it from
5403          * scheduling any more work and to have reset functions ignore
5404          * any resets that have already been scheduled. Drop the lock
5405          * after setting state, so __ibmvnic_reset() which is called
5406          * from the flush_work() below, can make progress.
5407          */
5408         spin_lock(&adapter->rwi_lock);
5409         adapter->state = VNIC_REMOVING;
5410         spin_unlock(&adapter->rwi_lock);
5411
5412         spin_unlock_irqrestore(&adapter->state_lock, flags);
5413
5414         flush_work(&adapter->ibmvnic_reset);
5415         flush_delayed_work(&adapter->ibmvnic_delayed_reset);
5416
5417         rtnl_lock();
5418         unregister_netdevice(netdev);
5419
5420         release_resources(adapter);
5421         release_sub_crqs(adapter, 1);
5422         release_crq_queue(adapter);
5423
5424         release_stats_token(adapter);
5425         release_stats_buffers(adapter);
5426
5427         adapter->state = VNIC_REMOVED;
5428
5429         rtnl_unlock();
5430         mutex_destroy(&adapter->fw_lock);
5431         device_remove_file(&dev->dev, &dev_attr_failover);
5432         free_netdev(netdev);
5433         dev_set_drvdata(&dev->dev, NULL);
5434 }
5435
5436 static ssize_t failover_store(struct device *dev, struct device_attribute *attr,
5437                               const char *buf, size_t count)
5438 {
5439         struct net_device *netdev = dev_get_drvdata(dev);
5440         struct ibmvnic_adapter *adapter = netdev_priv(netdev);
5441         unsigned long retbuf[PLPAR_HCALL_BUFSIZE];
5442         __be64 session_token;
5443         long rc;
5444
5445         if (!sysfs_streq(buf, "1"))
5446                 return -EINVAL;
5447
5448         rc = plpar_hcall(H_VIOCTL, retbuf, adapter->vdev->unit_address,
5449                          H_GET_SESSION_TOKEN, 0, 0, 0);
5450         if (rc) {
5451                 netdev_err(netdev, "Couldn't retrieve session token, rc %ld\n",
5452                            rc);
5453                 return -EINVAL;
5454         }
5455
5456         session_token = (__be64)retbuf[0];
5457         netdev_dbg(netdev, "Initiating client failover, session id %llx\n",
5458                    be64_to_cpu(session_token));
5459         rc = plpar_hcall_norets(H_VIOCTL, adapter->vdev->unit_address,
5460                                 H_SESSION_ERR_DETECTED, session_token, 0, 0);
5461         if (rc) {
5462                 netdev_err(netdev, "Client initiated failover failed, rc %ld\n",
5463                            rc);
5464                 return -EINVAL;
5465         }
5466
5467         return count;
5468 }
5469
5470 static DEVICE_ATTR_WO(failover);
5471
5472 static unsigned long ibmvnic_get_desired_dma(struct vio_dev *vdev)
5473 {
5474         struct net_device *netdev = dev_get_drvdata(&vdev->dev);
5475         struct ibmvnic_adapter *adapter;
5476         struct iommu_table *tbl;
5477         unsigned long ret = 0;
5478         int i;
5479
5480         tbl = get_iommu_table_base(&vdev->dev);
5481
5482         /* netdev inits at probe time along with the structures we need below*/
5483         if (!netdev)
5484                 return IOMMU_PAGE_ALIGN(IBMVNIC_IO_ENTITLEMENT_DEFAULT, tbl);
5485
5486         adapter = netdev_priv(netdev);
5487
5488         ret += PAGE_SIZE; /* the crq message queue */
5489         ret += IOMMU_PAGE_ALIGN(sizeof(struct ibmvnic_statistics), tbl);
5490
5491         for (i = 0; i < adapter->req_tx_queues + adapter->req_rx_queues; i++)
5492                 ret += 4 * PAGE_SIZE; /* the scrq message queue */
5493
5494         for (i = 0; i < adapter->num_active_rx_pools; i++)
5495                 ret += adapter->rx_pool[i].size *
5496                     IOMMU_PAGE_ALIGN(adapter->rx_pool[i].buff_size, tbl);
5497
5498         return ret;
5499 }
5500
5501 static int ibmvnic_resume(struct device *dev)
5502 {
5503         struct net_device *netdev = dev_get_drvdata(dev);
5504         struct ibmvnic_adapter *adapter = netdev_priv(netdev);
5505
5506         if (adapter->state != VNIC_OPEN)
5507                 return 0;
5508
5509         tasklet_schedule(&adapter->tasklet);
5510
5511         return 0;
5512 }
5513
5514 static const struct vio_device_id ibmvnic_device_table[] = {
5515         {"network", "IBM,vnic"},
5516         {"", "" }
5517 };
5518 MODULE_DEVICE_TABLE(vio, ibmvnic_device_table);
5519
5520 static const struct dev_pm_ops ibmvnic_pm_ops = {
5521         .resume = ibmvnic_resume
5522 };
5523
5524 static struct vio_driver ibmvnic_driver = {
5525         .id_table       = ibmvnic_device_table,
5526         .probe          = ibmvnic_probe,
5527         .remove         = ibmvnic_remove,
5528         .get_desired_dma = ibmvnic_get_desired_dma,
5529         .name           = ibmvnic_driver_name,
5530         .pm             = &ibmvnic_pm_ops,
5531 };
5532
5533 /* module functions */
5534 static int __init ibmvnic_module_init(void)
5535 {
5536         pr_info("%s: %s %s\n", ibmvnic_driver_name, ibmvnic_driver_string,
5537                 IBMVNIC_DRIVER_VERSION);
5538
5539         return vio_register_driver(&ibmvnic_driver);
5540 }
5541
5542 static void __exit ibmvnic_module_exit(void)
5543 {
5544         vio_unregister_driver(&ibmvnic_driver);
5545 }
5546
5547 module_init(ibmvnic_module_init);
5548 module_exit(ibmvnic_module_exit);