drivers/net/ethernet/ibm/ibmvnic.c

   1 // SPDX-License-Identifier: GPL-2.0-or-later
   2 /**************************************************************************/
   3 /*                                                                        */
   4 /*  IBM System i and System p Virtual NIC Device Driver                   */
   5 /*  Copyright (C) 2014 IBM Corp.                                          */
   6 /*  Santiago Leon ([email protected])                                  */
   7 /*  Thomas Falcon ([email protected])                           */
   8 /*  John Allen ([email protected])                                */
   9 /*                                                                        */
  10 /*                                                                        */
  11 /* This module contains the implementation of a virtual ethernet device   */
  12 /* for use with IBM i/p Series LPAR Linux. It utilizes the logical LAN    */
  13 /* option of the RS/6000 Platform Architecture to interface with virtual  */
  14 /* ethernet NICs that are presented to the partition by the hypervisor.   */
  15 /*                                                                         */
  16 /* Messages are passed between the VNIC driver and the VNIC server using  */
  17 /* Command/Response Queues (CRQs) and sub CRQs (sCRQs). CRQs are used to  */
  18 /* issue and receive commands that initiate communication with the server */
  19 /* on driver initialization. Sub CRQs (sCRQs) are similar to CRQs, but    */
  20 /* are used by the driver to notify the server that a packet is           */
  21 /* ready for transmission or that a buffer has been added to receive a    */
  22 /* packet. Subsequently, sCRQs are used by the server to notify the       */
  23 /* driver that a packet transmission has been completed or that a packet  */
  24 /* has been received and placed in a waiting buffer.                      */
  25 /*                                                                        */
  26 /* In lieu of a more conventional "on-the-fly" DMA mapping strategy in    */
  27 /* which skbs are DMA mapped and immediately unmapped when the transmit   */
  28 /* or receive has been completed, the VNIC driver is required to use      */
  29 /* "long term mapping". This entails that large, continuous DMA mapped    */
  30 /* buffers are allocated on driver initialization and these buffers are   */
  31 /* then continuously reused to pass skbs to and from the VNIC server.     */
  32 /*                                                                        */
  33 /**************************************************************************/
  34
  35 #include <linux/module.h>
  36 #include <linux/moduleparam.h>
  37 #include <linux/types.h>
  38 #include <linux/errno.h>
  39 #include <linux/completion.h>
  40 #include <linux/ioport.h>
  41 #include <linux/dma-mapping.h>
  42 #include <linux/kernel.h>
  43 #include <linux/netdevice.h>
  44 #include <linux/etherdevice.h>
  45 #include <linux/skbuff.h>
  46 #include <linux/init.h>
  47 #include <linux/delay.h>
  48 #include <linux/mm.h>
  49 #include <linux/ethtool.h>
  50 #include <linux/proc_fs.h>
  51 #include <linux/if_arp.h>
  52 #include <linux/in.h>
  53 #include <linux/ip.h>
  54 #include <linux/ipv6.h>
  55 #include <linux/irq.h>
  56 #include <linux/kthread.h>
  57 #include <linux/seq_file.h>
  58 #include <linux/interrupt.h>
  59 #include <net/net_namespace.h>
  60 #include <asm/hvcall.h>
  61 #include <linux/atomic.h>
  62 #include <asm/vio.h>
  63 #include <asm/iommu.h>
  64 #include <linux/uaccess.h>
  65 #include <asm/firmware.h>
  66 #include <linux/workqueue.h>
  67 #include <linux/if_vlan.h>
  68 #include <linux/utsname.h>
  69
  70 #include "ibmvnic.h"
  71
  72 static const char ibmvnic_driver_name[] = "ibmvnic";
  73 static const char ibmvnic_driver_string[] = "IBM System i/p Virtual NIC Driver";
  74
  75 MODULE_AUTHOR("Santiago Leon");
  76 MODULE_DESCRIPTION("IBM System i/p Virtual NIC Driver");
  77 MODULE_LICENSE("GPL");
  78 MODULE_VERSION(IBMVNIC_DRIVER_VERSION);
  79
  80 static int ibmvnic_version = IBMVNIC_INITIAL_VERSION;
  81 static int ibmvnic_remove(struct vio_dev *);
  82 static void release_sub_crqs(struct ibmvnic_adapter *, bool);
  83 static int ibmvnic_reset_crq(struct ibmvnic_adapter *);
  84 static int ibmvnic_send_crq_init(struct ibmvnic_adapter *);
  85 static int ibmvnic_reenable_crq_queue(struct ibmvnic_adapter *);
  86 static int ibmvnic_send_crq(struct ibmvnic_adapter *, union ibmvnic_crq *);
  87 static int send_subcrq_indirect(struct ibmvnic_adapter *, u64, u64, u64);
  88 static irqreturn_t ibmvnic_interrupt_rx(int irq, void *instance);
  89 static int enable_scrq_irq(struct ibmvnic_adapter *,
  90                            struct ibmvnic_sub_crq_queue *);
  91 static int disable_scrq_irq(struct ibmvnic_adapter *,
  92                             struct ibmvnic_sub_crq_queue *);
  93 static int pending_scrq(struct ibmvnic_adapter *,
  94                         struct ibmvnic_sub_crq_queue *);
  95 static union sub_crq *ibmvnic_next_scrq(struct ibmvnic_adapter *,
  96                                         struct ibmvnic_sub_crq_queue *);
  97 static int ibmvnic_poll(struct napi_struct *napi, int data);
  98 static void send_query_map(struct ibmvnic_adapter *adapter);
  99 static int send_request_map(struct ibmvnic_adapter *, dma_addr_t, __be32, u8);
 100 static int send_request_unmap(struct ibmvnic_adapter *, u8);
 101 static int send_login(struct ibmvnic_adapter *adapter);
 102 static void send_query_cap(struct ibmvnic_adapter *adapter);
 103 static int init_sub_crqs(struct ibmvnic_adapter *);
 104 static int init_sub_crq_irqs(struct ibmvnic_adapter *adapter);
 105 static int ibmvnic_reset_init(struct ibmvnic_adapter *, bool reset);
 106 static void release_crq_queue(struct ibmvnic_adapter *);
 107 static int __ibmvnic_set_mac(struct net_device *, u8 *);
 108 static int init_crq_queue(struct ibmvnic_adapter *adapter);
 109 static int send_query_phys_parms(struct ibmvnic_adapter *adapter);
 110
 111 struct ibmvnic_stat {
 112         char name[ETH_GSTRING_LEN];
 113         int offset;
 114 };
 115
 116 #define IBMVNIC_STAT_OFF(stat) (offsetof(struct ibmvnic_adapter, stats) + \
 117                              offsetof(struct ibmvnic_statistics, stat))
 118 #define IBMVNIC_GET_STAT(a, off) (*((u64 *)(((unsigned long)(a)) + off)))
 119
 120 static const struct ibmvnic_stat ibmvnic_stats[] = {
 121         {"rx_packets", IBMVNIC_STAT_OFF(rx_packets)},
 122         {"rx_bytes", IBMVNIC_STAT_OFF(rx_bytes)},
 123         {"tx_packets", IBMVNIC_STAT_OFF(tx_packets)},
 124         {"tx_bytes", IBMVNIC_STAT_OFF(tx_bytes)},
 125         {"ucast_tx_packets", IBMVNIC_STAT_OFF(ucast_tx_packets)},
 126         {"ucast_rx_packets", IBMVNIC_STAT_OFF(ucast_rx_packets)},
 127         {"mcast_tx_packets", IBMVNIC_STAT_OFF(mcast_tx_packets)},
 128         {"mcast_rx_packets", IBMVNIC_STAT_OFF(mcast_rx_packets)},
 129         {"bcast_tx_packets", IBMVNIC_STAT_OFF(bcast_tx_packets)},
 130         {"bcast_rx_packets", IBMVNIC_STAT_OFF(bcast_rx_packets)},
 131         {"align_errors", IBMVNIC_STAT_OFF(align_errors)},
 132         {"fcs_errors", IBMVNIC_STAT_OFF(fcs_errors)},
 133         {"single_collision_frames", IBMVNIC_STAT_OFF(single_collision_frames)},
 134         {"multi_collision_frames", IBMVNIC_STAT_OFF(multi_collision_frames)},
 135         {"sqe_test_errors", IBMVNIC_STAT_OFF(sqe_test_errors)},
 136         {"deferred_tx", IBMVNIC_STAT_OFF(deferred_tx)},
 137         {"late_collisions", IBMVNIC_STAT_OFF(late_collisions)},
 138         {"excess_collisions", IBMVNIC_STAT_OFF(excess_collisions)},
 139         {"internal_mac_tx_errors", IBMVNIC_STAT_OFF(internal_mac_tx_errors)},
 140         {"carrier_sense", IBMVNIC_STAT_OFF(carrier_sense)},
 141         {"too_long_frames", IBMVNIC_STAT_OFF(too_long_frames)},
 142         {"internal_mac_rx_errors", IBMVNIC_STAT_OFF(internal_mac_rx_errors)},
 143 };
 144
 145 static long h_reg_sub_crq(unsigned long unit_address, unsigned long token,
 146                           unsigned long length, unsigned long *number,
 147                           unsigned long *irq)
 148 {
 149         unsigned long retbuf[PLPAR_HCALL_BUFSIZE];
 150         long rc;
 151
 152         rc = plpar_hcall(H_REG_SUB_CRQ, retbuf, unit_address, token, length);
 153         *number = retbuf[0];
 154         *irq = retbuf[1];
 155
 156         return rc;
 157 }
 158
 159 /**
 160  * ibmvnic_wait_for_completion - Check device state and wait for completion
 161  * @adapter: private device data
 162  * @comp_done: completion structure to wait for
 163  * @timeout: time to wait in milliseconds
 164  *
 165  * Wait for a completion signal or until the timeout limit is reached
 166  * while checking that the device is still active.
 167  */
 168 static int ibmvnic_wait_for_completion(struct ibmvnic_adapter *adapter,
 169                                        struct completion *comp_done,
 170                                        unsigned long timeout)
 171 {
 172         struct net_device *netdev;
 173         unsigned long div_timeout;
 174         u8 retry;
 175
 176         netdev = adapter->netdev;
 177         retry = 5;
 178         div_timeout = msecs_to_jiffies(timeout / retry);
 179         while (true) {
 180                 if (!adapter->crq.active) {
 181                         netdev_err(netdev, "Device down!\n");
 182                         return -ENODEV;
 183                 }
 184                 if (!retry--)
 185                         break;
 186                 if (wait_for_completion_timeout(comp_done, div_timeout))
 187                         return 0;
 188         }
 189         netdev_err(netdev, "Operation timed out.\n");
 190         return -ETIMEDOUT;
 191 }
 192
 193 static int alloc_long_term_buff(struct ibmvnic_adapter *adapter,
 194                                 struct ibmvnic_long_term_buff *ltb, int size)
 195 {
 196         struct device *dev = &adapter->vdev->dev;
 197         int rc;
 198
 199         ltb->size = size;
 200         ltb->buff = dma_alloc_coherent(dev, ltb->size, &ltb->addr,
 201                                        GFP_KERNEL);
 202
 203         if (!ltb->buff) {
 204                 dev_err(dev, "Couldn't alloc long term buffer\n");
 205                 return -ENOMEM;
 206         }
 207         ltb->map_id = adapter->map_id;
 208         adapter->map_id++;
 209
 210         mutex_lock(&adapter->fw_lock);
 211         adapter->fw_done_rc = 0;
 212         reinit_completion(&adapter->fw_done);
 213         rc = send_request_map(adapter, ltb->addr,
 214                               ltb->size, ltb->map_id);
 215         if (rc) {
 216                 dma_free_coherent(dev, ltb->size, ltb->buff, ltb->addr);
 217                 mutex_unlock(&adapter->fw_lock);
 218                 return rc;
 219         }
 220
 221         rc = ibmvnic_wait_for_completion(adapter, &adapter->fw_done, 10000);
 222         if (rc) {
 223                 dev_err(dev,
 224                         "Long term map request aborted or timed out,rc = %d\n",
 225                         rc);
 226                 dma_free_coherent(dev, ltb->size, ltb->buff, ltb->addr);
 227                 mutex_unlock(&adapter->fw_lock);
 228                 return rc;
 229         }
 230
 231         if (adapter->fw_done_rc) {
 232                 dev_err(dev, "Couldn't map long term buffer,rc = %d\n",
 233                         adapter->fw_done_rc);
 234                 dma_free_coherent(dev, ltb->size, ltb->buff, ltb->addr);
 235                 mutex_unlock(&adapter->fw_lock);
 236                 return -1;
 237         }
 238         mutex_unlock(&adapter->fw_lock);
 239         return 0;
 240 }
 241
 242 static void free_long_term_buff(struct ibmvnic_adapter *adapter,
 243                                 struct ibmvnic_long_term_buff *ltb)
 244 {
 245         struct device *dev = &adapter->vdev->dev;
 246
 247         if (!ltb->buff)
 248                 return;
 249
 250         if (adapter->reset_reason != VNIC_RESET_FAILOVER &&
 251             adapter->reset_reason != VNIC_RESET_MOBILITY)
 252                 send_request_unmap(adapter, ltb->map_id);
 253         dma_free_coherent(dev, ltb->size, ltb->buff, ltb->addr);
 254 }
 255
 256 static int reset_long_term_buff(struct ibmvnic_adapter *adapter,
 257                                 struct ibmvnic_long_term_buff *ltb)
 258 {
 259         struct device *dev = &adapter->vdev->dev;
 260         int rc;
 261
 262         memset(ltb->buff, 0, ltb->size);
 263
 264         mutex_lock(&adapter->fw_lock);
 265         adapter->fw_done_rc = 0;
 266
 267         reinit_completion(&adapter->fw_done);
 268         rc = send_request_map(adapter, ltb->addr, ltb->size, ltb->map_id);
 269         if (rc) {
 270                 mutex_unlock(&adapter->fw_lock);
 271                 return rc;
 272         }
 273
 274         rc = ibmvnic_wait_for_completion(adapter, &adapter->fw_done, 10000);
 275         if (rc) {
 276                 dev_info(dev,
 277                          "Reset failed, long term map request timed out or aborted\n");
 278                 mutex_unlock(&adapter->fw_lock);
 279                 return rc;
 280         }
 281
 282         if (adapter->fw_done_rc) {
 283                 dev_info(dev,
 284                          "Reset failed, attempting to free and reallocate buffer\n");
 285                 free_long_term_buff(adapter, ltb);
 286                 mutex_unlock(&adapter->fw_lock);
 287                 return alloc_long_term_buff(adapter, ltb, ltb->size);
 288         }
 289         mutex_unlock(&adapter->fw_lock);
 290         return 0;
 291 }
 292
 293 static void deactivate_rx_pools(struct ibmvnic_adapter *adapter)
 294 {
 295         int i;
 296
 297         for (i = 0; i < adapter->num_active_rx_pools; i++)
 298                 adapter->rx_pool[i].active = 0;
 299 }
 300
 301 static void replenish_rx_pool(struct ibmvnic_adapter *adapter,
 302                               struct ibmvnic_rx_pool *pool)
 303 {
 304         int count = pool->size - atomic_read(&pool->available);
 305         u64 handle = adapter->rx_scrq[pool->index]->handle;
 306         struct device *dev = &adapter->vdev->dev;
 307         struct ibmvnic_ind_xmit_queue *ind_bufp;
 308         struct ibmvnic_sub_crq_queue *rx_scrq;
 309         union sub_crq *sub_crq;
 310         int buffers_added = 0;
 311         unsigned long lpar_rc;
 312         struct sk_buff *skb;
 313         unsigned int offset;
 314         dma_addr_t dma_addr;
 315         unsigned char *dst;
 316         int shift = 0;
 317         int index;
 318         int i;
 319
 320         if (!pool->active)
 321                 return;
 322
 323         rx_scrq = adapter->rx_scrq[pool->index];
 324         ind_bufp = &rx_scrq->ind_buf;
 325         for (i = 0; i < count; ++i) {
 326                 skb = netdev_alloc_skb(adapter->netdev, pool->buff_size);
 327                 if (!skb) {
 328                         dev_err(dev, "Couldn't replenish rx buff\n");
 329                         adapter->replenish_no_mem++;
 330                         break;
 331                 }
 332
 333                 index = pool->free_map[pool->next_free];
 334
 335                 if (pool->rx_buff[index].skb)
 336                         dev_err(dev, "Inconsistent free_map!\n");
 337
 338                 /* Copy the skb to the long term mapped DMA buffer */
 339                 offset = index * pool->buff_size;
 340                 dst = pool->long_term_buff.buff + offset;
 341                 memset(dst, 0, pool->buff_size);
 342                 dma_addr = pool->long_term_buff.addr + offset;
 343                 pool->rx_buff[index].data = dst;
 344
 345                 pool->free_map[pool->next_free] = IBMVNIC_INVALID_MAP;
 346                 pool->rx_buff[index].dma = dma_addr;
 347                 pool->rx_buff[index].skb = skb;
 348                 pool->rx_buff[index].pool_index = pool->index;
 349                 pool->rx_buff[index].size = pool->buff_size;
 350
 351                 sub_crq = &ind_bufp->indir_arr[ind_bufp->index++];
 352                 memset(sub_crq, 0, sizeof(*sub_crq));
 353                 sub_crq->rx_add.first = IBMVNIC_CRQ_CMD;
 354                 sub_crq->rx_add.correlator =
 355                     cpu_to_be64((u64)&pool->rx_buff[index]);
 356                 sub_crq->rx_add.ioba = cpu_to_be32(dma_addr);
 357                 sub_crq->rx_add.map_id = pool->long_term_buff.map_id;
 358
 359                 /* The length field of the sCRQ is defined to be 24 bits so the
 360                  * buffer size needs to be left shifted by a byte before it is
 361                  * converted to big endian to prevent the last byte from being
 362                  * truncated.
 363                  */
 364 #ifdef __LITTLE_ENDIAN__
 365                 shift = 8;
 366 #endif
 367                 sub_crq->rx_add.len = cpu_to_be32(pool->buff_size << shift);
 368                 pool->next_free = (pool->next_free + 1) % pool->size;
 369                 if (ind_bufp->index == IBMVNIC_MAX_IND_DESCS ||
 370                     i == count - 1) {
 371                         lpar_rc =
 372                                 send_subcrq_indirect(adapter, handle,
 373                                                      (u64)ind_bufp->indir_dma,
 374                                                      (u64)ind_bufp->index);
 375                         if (lpar_rc != H_SUCCESS)
 376                                 goto failure;
 377                         buffers_added += ind_bufp->index;
 378                         adapter->replenish_add_buff_success += ind_bufp->index;
 379                         ind_bufp->index = 0;
 380                 }
 381         }
 382         atomic_add(buffers_added, &pool->available);
 383         return;
 384
 385 failure:
 386         if (lpar_rc != H_PARAMETER && lpar_rc != H_CLOSED)
 387                 dev_err_ratelimited(dev, "rx: replenish packet buffer failed\n");
 388         for (i = ind_bufp->index - 1; i >= 0; --i) {
 389                 struct ibmvnic_rx_buff *rx_buff;
 390
 391                 pool->next_free = pool->next_free == 0 ?
 392                                   pool->size - 1 : pool->next_free - 1;
 393                 sub_crq = &ind_bufp->indir_arr[i];
 394                 rx_buff = (struct ibmvnic_rx_buff *)
 395                                 be64_to_cpu(sub_crq->rx_add.correlator);
 396                 index = (int)(rx_buff - pool->rx_buff);
 397                 pool->free_map[pool->next_free] = index;
 398                 dev_kfree_skb_any(pool->rx_buff[index].skb);
 399                 pool->rx_buff[index].skb = NULL;
 400         }
 401         adapter->replenish_add_buff_failure += ind_bufp->index;
 402         atomic_add(buffers_added, &pool->available);
 403         ind_bufp->index = 0;
 404         if (lpar_rc == H_CLOSED || adapter->failover_pending) {
 405                 /* Disable buffer pool replenishment and report carrier off if
 406                  * queue is closed or pending failover.
 407                  * Firmware guarantees that a signal will be sent to the
 408                  * driver, triggering a reset.
 409                  */
 410                 deactivate_rx_pools(adapter);
 411                 netif_carrier_off(adapter->netdev);
 412         }
 413 }
 414
 415 static void replenish_pools(struct ibmvnic_adapter *adapter)
 416 {
 417         int i;
 418
 419         adapter->replenish_task_cycles++;
 420         for (i = 0; i < adapter->num_active_rx_pools; i++) {
 421                 if (adapter->rx_pool[i].active)
 422                         replenish_rx_pool(adapter, &adapter->rx_pool[i]);
 423         }
 424
 425         netdev_dbg(adapter->netdev, "Replenished %d pools\n", i);
 426 }
 427
 428 static void release_stats_buffers(struct ibmvnic_adapter *adapter)
 429 {
 430         kfree(adapter->tx_stats_buffers);
 431         kfree(adapter->rx_stats_buffers);
 432         adapter->tx_stats_buffers = NULL;
 433         adapter->rx_stats_buffers = NULL;
 434 }
 435
 436 static int init_stats_buffers(struct ibmvnic_adapter *adapter)
 437 {
 438         adapter->tx_stats_buffers =
 439                                 kcalloc(IBMVNIC_MAX_QUEUES,
 440                                         sizeof(struct ibmvnic_tx_queue_stats),
 441                                         GFP_KERNEL);
 442         if (!adapter->tx_stats_buffers)
 443                 return -ENOMEM;
 444
 445         adapter->rx_stats_buffers =
 446                                 kcalloc(IBMVNIC_MAX_QUEUES,
 447                                         sizeof(struct ibmvnic_rx_queue_stats),
 448                                         GFP_KERNEL);
 449         if (!adapter->rx_stats_buffers)
 450                 return -ENOMEM;
 451
 452         return 0;
 453 }
 454
 455 static void release_stats_token(struct ibmvnic_adapter *adapter)
 456 {
 457         struct device *dev = &adapter->vdev->dev;
 458
 459         if (!adapter->stats_token)
 460                 return;
 461
 462         dma_unmap_single(dev, adapter->stats_token,
 463                          sizeof(struct ibmvnic_statistics),
 464                          DMA_FROM_DEVICE);
 465         adapter->stats_token = 0;
 466 }
 467
 468 static int init_stats_token(struct ibmvnic_adapter *adapter)
 469 {
 470         struct device *dev = &adapter->vdev->dev;
 471         dma_addr_t stok;
 472
 473         stok = dma_map_single(dev, &adapter->stats,
 474                               sizeof(struct ibmvnic_statistics),
 475                               DMA_FROM_DEVICE);
 476         if (dma_mapping_error(dev, stok)) {
 477                 dev_err(dev, "Couldn't map stats buffer\n");
 478                 return -1;
 479         }
 480
 481         adapter->stats_token = stok;
 482         netdev_dbg(adapter->netdev, "Stats token initialized (%llx)\n", stok);
 483         return 0;
 484 }
 485
 486 static int reset_rx_pools(struct ibmvnic_adapter *adapter)
 487 {
 488         struct ibmvnic_rx_pool *rx_pool;
 489         u64 buff_size;
 490         int rx_scrqs;
 491         int i, j, rc;
 492
 493         if (!adapter->rx_pool)
 494                 return -1;
 495
 496         buff_size = adapter->cur_rx_buf_sz;
 497         rx_scrqs = adapter->num_active_rx_pools;
 498         for (i = 0; i < rx_scrqs; i++) {
 499                 rx_pool = &adapter->rx_pool[i];
 500
 501                 netdev_dbg(adapter->netdev, "Re-setting rx_pool[%d]\n", i);
 502
 503                 if (rx_pool->buff_size != buff_size) {
 504                         free_long_term_buff(adapter, &rx_pool->long_term_buff);
 505                         rx_pool->buff_size = ALIGN(buff_size, L1_CACHE_BYTES);
 506                         rc = alloc_long_term_buff(adapter,
 507                                                   &rx_pool->long_term_buff,
 508                                                   rx_pool->size *
 509                                                   rx_pool->buff_size);
 510                 } else {
 511                         rc = reset_long_term_buff(adapter,
 512                                                   &rx_pool->long_term_buff);
 513                 }
 514
 515                 if (rc)
 516                         return rc;
 517
 518                 for (j = 0; j < rx_pool->size; j++)
 519                         rx_pool->free_map[j] = j;
 520
 521                 memset(rx_pool->rx_buff, 0,
 522                        rx_pool->size * sizeof(struct ibmvnic_rx_buff));
 523
 524                 atomic_set(&rx_pool->available, 0);
 525                 rx_pool->next_alloc = 0;
 526                 rx_pool->next_free = 0;
 527                 rx_pool->active = 1;
 528         }
 529
 530         return 0;
 531 }
 532
 533 static void release_rx_pools(struct ibmvnic_adapter *adapter)
 534 {
 535         struct ibmvnic_rx_pool *rx_pool;
 536         int i, j;
 537
 538         if (!adapter->rx_pool)
 539                 return;
 540
 541         for (i = 0; i < adapter->num_active_rx_pools; i++) {
 542                 rx_pool = &adapter->rx_pool[i];
 543
 544                 netdev_dbg(adapter->netdev, "Releasing rx_pool[%d]\n", i);
 545
 546                 kfree(rx_pool->free_map);
 547                 free_long_term_buff(adapter, &rx_pool->long_term_buff);
 548
 549                 if (!rx_pool->rx_buff)
 550                         continue;
 551
 552                 for (j = 0; j < rx_pool->size; j++) {
 553                         if (rx_pool->rx_buff[j].skb) {
 554                                 dev_kfree_skb_any(rx_pool->rx_buff[j].skb);
 555                                 rx_pool->rx_buff[j].skb = NULL;
 556                         }
 557                 }
 558
 559                 kfree(rx_pool->rx_buff);
 560         }
 561
 562         kfree(adapter->rx_pool);
 563         adapter->rx_pool = NULL;
 564         adapter->num_active_rx_pools = 0;
 565 }
 566
 567 static int init_rx_pools(struct net_device *netdev)
 568 {
 569         struct ibmvnic_adapter *adapter = netdev_priv(netdev);
 570         struct device *dev = &adapter->vdev->dev;
 571         struct ibmvnic_rx_pool *rx_pool;
 572         int rxadd_subcrqs;
 573         u64 buff_size;
 574         int i, j;
 575
 576         rxadd_subcrqs = adapter->num_active_rx_scrqs;
 577         buff_size = adapter->cur_rx_buf_sz;
 578
 579         adapter->rx_pool = kcalloc(rxadd_subcrqs,
 580                                    sizeof(struct ibmvnic_rx_pool),
 581                                    GFP_KERNEL);
 582         if (!adapter->rx_pool) {
 583                 dev_err(dev, "Failed to allocate rx pools\n");
 584                 return -1;
 585         }
 586
 587         adapter->num_active_rx_pools = rxadd_subcrqs;
 588
 589         for (i = 0; i < rxadd_subcrqs; i++) {
 590                 rx_pool = &adapter->rx_pool[i];
 591
 592                 netdev_dbg(adapter->netdev,
 593                            "Initializing rx_pool[%d], %lld buffs, %lld bytes each\n",
 594                            i, adapter->req_rx_add_entries_per_subcrq,
 595                            buff_size);
 596
 597                 rx_pool->size = adapter->req_rx_add_entries_per_subcrq;
 598                 rx_pool->index = i;
 599                 rx_pool->buff_size = ALIGN(buff_size, L1_CACHE_BYTES);
 600                 rx_pool->active = 1;
 601
 602                 rx_pool->free_map = kcalloc(rx_pool->size, sizeof(int),
 603                                             GFP_KERNEL);
 604                 if (!rx_pool->free_map) {
 605                         release_rx_pools(adapter);
 606                         return -1;
 607                 }
 608
 609                 rx_pool->rx_buff = kcalloc(rx_pool->size,
 610                                            sizeof(struct ibmvnic_rx_buff),
 611                                            GFP_KERNEL);
 612                 if (!rx_pool->rx_buff) {
 613                         dev_err(dev, "Couldn't alloc rx buffers\n");
 614                         release_rx_pools(adapter);
 615                         return -1;
 616                 }
 617
 618                 if (alloc_long_term_buff(adapter, &rx_pool->long_term_buff,
 619                                          rx_pool->size * rx_pool->buff_size)) {
 620                         release_rx_pools(adapter);
 621                         return -1;
 622                 }
 623
 624                 for (j = 0; j < rx_pool->size; ++j)
 625                         rx_pool->free_map[j] = j;
 626
 627                 atomic_set(&rx_pool->available, 0);
 628                 rx_pool->next_alloc = 0;
 629                 rx_pool->next_free = 0;
 630         }
 631
 632         return 0;
 633 }
 634
 635 static int reset_one_tx_pool(struct ibmvnic_adapter *adapter,
 636                              struct ibmvnic_tx_pool *tx_pool)
 637 {
 638         int rc, i;
 639
 640         rc = reset_long_term_buff(adapter, &tx_pool->long_term_buff);
 641         if (rc)
 642                 return rc;
 643
 644         memset(tx_pool->tx_buff, 0,
 645                tx_pool->num_buffers *
 646                sizeof(struct ibmvnic_tx_buff));
 647
 648         for (i = 0; i < tx_pool->num_buffers; i++)
 649                 tx_pool->free_map[i] = i;
 650
 651         tx_pool->consumer_index = 0;
 652         tx_pool->producer_index = 0;
 653
 654         return 0;
 655 }
 656
 657 static int reset_tx_pools(struct ibmvnic_adapter *adapter)
 658 {
 659         int tx_scrqs;
 660         int i, rc;
 661
 662         if (!adapter->tx_pool)
 663                 return -1;
 664
 665         tx_scrqs = adapter->num_active_tx_pools;
 666         for (i = 0; i < tx_scrqs; i++) {
 667                 rc = reset_one_tx_pool(adapter, &adapter->tso_pool[i]);
 668                 if (rc)
 669                         return rc;
 670                 rc = reset_one_tx_pool(adapter, &adapter->tx_pool[i]);
 671                 if (rc)
 672                         return rc;
 673         }
 674
 675         return 0;
 676 }
 677
 678 static void release_vpd_data(struct ibmvnic_adapter *adapter)
 679 {
 680         if (!adapter->vpd)
 681                 return;
 682
 683         kfree(adapter->vpd->buff);
 684         kfree(adapter->vpd);
 685
 686         adapter->vpd = NULL;
 687 }
 688
 689 static void release_one_tx_pool(struct ibmvnic_adapter *adapter,
 690                                 struct ibmvnic_tx_pool *tx_pool)
 691 {
 692         kfree(tx_pool->tx_buff);
 693         kfree(tx_pool->free_map);
 694         free_long_term_buff(adapter, &tx_pool->long_term_buff);
 695 }
 696
 697 static void release_tx_pools(struct ibmvnic_adapter *adapter)
 698 {
 699         int i;
 700
 701         if (!adapter->tx_pool)
 702                 return;
 703
 704         for (i = 0; i < adapter->num_active_tx_pools; i++) {
 705                 release_one_tx_pool(adapter, &adapter->tx_pool[i]);
 706                 release_one_tx_pool(adapter, &adapter->tso_pool[i]);
 707         }
 708
 709         kfree(adapter->tx_pool);
 710         adapter->tx_pool = NULL;
 711         kfree(adapter->tso_pool);
 712         adapter->tso_pool = NULL;
 713         adapter->num_active_tx_pools = 0;
 714 }
 715
 716 static int init_one_tx_pool(struct net_device *netdev,
 717                             struct ibmvnic_tx_pool *tx_pool,
 718                             int num_entries, int buf_size)
 719 {
 720         struct ibmvnic_adapter *adapter = netdev_priv(netdev);
 721         int i;
 722
 723         tx_pool->tx_buff = kcalloc(num_entries,
 724                                    sizeof(struct ibmvnic_tx_buff),
 725                                    GFP_KERNEL);
 726         if (!tx_pool->tx_buff)
 727                 return -1;
 728
 729         if (alloc_long_term_buff(adapter, &tx_pool->long_term_buff,
 730                                  num_entries * buf_size))
 731                 return -1;
 732
 733         tx_pool->free_map = kcalloc(num_entries, sizeof(int), GFP_KERNEL);
 734         if (!tx_pool->free_map)
 735                 return -1;
 736
 737         for (i = 0; i < num_entries; i++)
 738                 tx_pool->free_map[i] = i;
 739
 740         tx_pool->consumer_index = 0;
 741         tx_pool->producer_index = 0;
 742         tx_pool->num_buffers = num_entries;
 743         tx_pool->buf_size = buf_size;
 744
 745         return 0;
 746 }
 747
 748 static int init_tx_pools(struct net_device *netdev)
 749 {
 750         struct ibmvnic_adapter *adapter = netdev_priv(netdev);
 751         int tx_subcrqs;
 752         u64 buff_size;
 753         int i, rc;
 754
 755         tx_subcrqs = adapter->num_active_tx_scrqs;
 756         adapter->tx_pool = kcalloc(tx_subcrqs,
 757                                    sizeof(struct ibmvnic_tx_pool), GFP_KERNEL);
 758         if (!adapter->tx_pool)
 759                 return -1;
 760
 761         adapter->tso_pool = kcalloc(tx_subcrqs,
 762                                     sizeof(struct ibmvnic_tx_pool), GFP_KERNEL);
 763         if (!adapter->tso_pool)
 764                 return -1;
 765
 766         adapter->num_active_tx_pools = tx_subcrqs;
 767
 768         for (i = 0; i < tx_subcrqs; i++) {
 769                 buff_size = adapter->req_mtu + VLAN_HLEN;
 770                 buff_size = ALIGN(buff_size, L1_CACHE_BYTES);
 771                 rc = init_one_tx_pool(netdev, &adapter->tx_pool[i],
 772                                       adapter->req_tx_entries_per_subcrq,
 773                                       buff_size);
 774                 if (rc) {
 775                         release_tx_pools(adapter);
 776                         return rc;
 777                 }
 778
 779                 rc = init_one_tx_pool(netdev, &adapter->tso_pool[i],
 780                                       IBMVNIC_TSO_BUFS,
 781                                       IBMVNIC_TSO_BUF_SZ);
 782                 if (rc) {
 783                         release_tx_pools(adapter);
 784                         return rc;
 785                 }
 786         }
 787
 788         return 0;
 789 }
 790
 791 static void ibmvnic_napi_enable(struct ibmvnic_adapter *adapter)
 792 {
 793         int i;
 794
 795         if (adapter->napi_enabled)
 796                 return;
 797
 798         for (i = 0; i < adapter->req_rx_queues; i++)
 799                 napi_enable(&adapter->napi[i]);
 800
 801         adapter->napi_enabled = true;
 802 }
 803
 804 static void ibmvnic_napi_disable(struct ibmvnic_adapter *adapter)
 805 {
 806         int i;
 807
 808         if (!adapter->napi_enabled)
 809                 return;
 810
 811         for (i = 0; i < adapter->req_rx_queues; i++) {
 812                 netdev_dbg(adapter->netdev, "Disabling napi[%d]\n", i);
 813                 napi_disable(&adapter->napi[i]);
 814         }
 815
 816         adapter->napi_enabled = false;
 817 }
 818
 819 static int init_napi(struct ibmvnic_adapter *adapter)
 820 {
 821         int i;
 822
 823         adapter->napi = kcalloc(adapter->req_rx_queues,
 824                                 sizeof(struct napi_struct), GFP_KERNEL);
 825         if (!adapter->napi)
 826                 return -ENOMEM;
 827
 828         for (i = 0; i < adapter->req_rx_queues; i++) {
 829                 netdev_dbg(adapter->netdev, "Adding napi[%d]\n", i);
 830                 netif_napi_add(adapter->netdev, &adapter->napi[i],
 831                                ibmvnic_poll, NAPI_POLL_WEIGHT);
 832         }
 833
 834         adapter->num_active_rx_napi = adapter->req_rx_queues;
 835         return 0;
 836 }
 837
 838 static void release_napi(struct ibmvnic_adapter *adapter)
 839 {
 840         int i;
 841
 842         if (!adapter->napi)
 843                 return;
 844
 845         for (i = 0; i < adapter->num_active_rx_napi; i++) {
 846                 netdev_dbg(adapter->netdev, "Releasing napi[%d]\n", i);
 847                 netif_napi_del(&adapter->napi[i]);
 848         }
 849
 850         kfree(adapter->napi);
 851         adapter->napi = NULL;
 852         adapter->num_active_rx_napi = 0;
 853         adapter->napi_enabled = false;
 854 }
 855
 856 static int ibmvnic_login(struct net_device *netdev)
 857 {
 858         struct ibmvnic_adapter *adapter = netdev_priv(netdev);
 859         unsigned long timeout = msecs_to_jiffies(20000);
 860         int retry_count = 0;
 861         int retries = 10;
 862         bool retry;
 863         int rc;
 864
 865         do {
 866                 retry = false;
 867                 if (retry_count > retries) {
 868                         netdev_warn(netdev, "Login attempts exceeded\n");
 869                         return -1;
 870                 }
 871
 872                 adapter->init_done_rc = 0;
 873                 reinit_completion(&adapter->init_done);
 874                 rc = send_login(adapter);
 875                 if (rc)
 876                         return rc;
 877
 878                 if (!wait_for_completion_timeout(&adapter->init_done,
 879                                                  timeout)) {
 880                         netdev_warn(netdev, "Login timed out, retrying...\n");
 881                         retry = true;
 882                         adapter->init_done_rc = 0;
 883                         retry_count++;
 884                         continue;
 885                 }
 886
 887                 if (adapter->init_done_rc == ABORTED) {
 888                         netdev_warn(netdev, "Login aborted, retrying...\n");
 889                         retry = true;
 890                         adapter->init_done_rc = 0;
 891                         retry_count++;
 892                         /* FW or device may be busy, so
 893                          * wait a bit before retrying login
 894                          */
 895                         msleep(500);
 896                 } else if (adapter->init_done_rc == PARTIALSUCCESS) {
 897                         retry_count++;
 898                         release_sub_crqs(adapter, 1);
 899
 900                         retry = true;
 901                         netdev_dbg(netdev,
 902                                    "Received partial success, retrying...\n");
 903                         adapter->init_done_rc = 0;
 904                         reinit_completion(&adapter->init_done);
 905                         send_query_cap(adapter);
 906                         if (!wait_for_completion_timeout(&adapter->init_done,
 907                                                          timeout)) {
 908                                 netdev_warn(netdev,
 909                                             "Capabilities query timed out\n");
 910                                 return -1;
 911                         }
 912
 913                         rc = init_sub_crqs(adapter);
 914                         if (rc) {
 915                                 netdev_warn(netdev,
 916                                             "SCRQ initialization failed\n");
 917                                 return -1;
 918                         }
 919
 920                         rc = init_sub_crq_irqs(adapter);
 921                         if (rc) {
 922                                 netdev_warn(netdev,
 923                                             "SCRQ irq initialization failed\n");
 924                                 return -1;
 925                         }
 926                 } else if (adapter->init_done_rc) {
 927                         netdev_warn(netdev, "Adapter login failed\n");
 928                         return -1;
 929                 }
 930         } while (retry);
 931
 932         __ibmvnic_set_mac(netdev, adapter->mac_addr);
 933
 934         netdev_dbg(netdev, "[S:%d] Login succeeded\n", adapter->state);
 935         return 0;
 936 }
 937
 938 static void release_login_buffer(struct ibmvnic_adapter *adapter)
 939 {
 940         kfree(adapter->login_buf);
 941         adapter->login_buf = NULL;
 942 }
 943
 944 static void release_login_rsp_buffer(struct ibmvnic_adapter *adapter)
 945 {
 946         kfree(adapter->login_rsp_buf);
 947         adapter->login_rsp_buf = NULL;
 948 }
 949
 950 static void release_resources(struct ibmvnic_adapter *adapter)
 951 {
 952         release_vpd_data(adapter);
 953
 954         release_tx_pools(adapter);
 955         release_rx_pools(adapter);
 956
 957         release_napi(adapter);
 958         release_login_buffer(adapter);
 959         release_login_rsp_buffer(adapter);
 960 }
 961
 962 static int set_link_state(struct ibmvnic_adapter *adapter, u8 link_state)
 963 {
 964         struct net_device *netdev = adapter->netdev;
 965         unsigned long timeout = msecs_to_jiffies(20000);
 966         union ibmvnic_crq crq;
 967         bool resend;
 968         int rc;
 969
 970         netdev_dbg(netdev, "setting link state %d\n", link_state);
 971
 972         memset(&crq, 0, sizeof(crq));
 973         crq.logical_link_state.first = IBMVNIC_CRQ_CMD;
 974         crq.logical_link_state.cmd = LOGICAL_LINK_STATE;
 975         crq.logical_link_state.link_state = link_state;
 976
 977         do {
 978                 resend = false;
 979
 980                 reinit_completion(&adapter->init_done);
 981                 rc = ibmvnic_send_crq(adapter, &crq);
 982                 if (rc) {
 983                         netdev_err(netdev, "Failed to set link state\n");
 984                         return rc;
 985                 }
 986
 987                 if (!wait_for_completion_timeout(&adapter->init_done,
 988                                                  timeout)) {
 989                         netdev_err(netdev, "timeout setting link state\n");
 990                         return -1;
 991                 }
 992
 993                 if (adapter->init_done_rc == PARTIALSUCCESS) {
 994                         /* Partuial success, delay and re-send */
 995                         mdelay(1000);
 996                         resend = true;
 997                 } else if (adapter->init_done_rc) {
 998                         netdev_warn(netdev, "Unable to set link state, rc=%d\n",
 999                                     adapter->init_done_rc);
1000                         return adapter->init_done_rc;
1001                 }
1002         } while (resend);
1003
1004         return 0;
1005 }
1006
1007 static int set_real_num_queues(struct net_device *netdev)
1008 {
1009         struct ibmvnic_adapter *adapter = netdev_priv(netdev);
1010         int rc;
1011
1012         netdev_dbg(netdev, "Setting real tx/rx queues (%llx/%llx)\n",
1013                    adapter->req_tx_queues, adapter->req_rx_queues);
1014
1015         rc = netif_set_real_num_tx_queues(netdev, adapter->req_tx_queues);
1016         if (rc) {
1017                 netdev_err(netdev, "failed to set the number of tx queues\n");
1018                 return rc;
1019         }
1020
1021         rc = netif_set_real_num_rx_queues(netdev, adapter->req_rx_queues);
1022         if (rc)
1023                 netdev_err(netdev, "failed to set the number of rx queues\n");
1024
1025         return rc;
1026 }
1027
1028 static int ibmvnic_get_vpd(struct ibmvnic_adapter *adapter)
1029 {
1030         struct device *dev = &adapter->vdev->dev;
1031         union ibmvnic_crq crq;
1032         int len = 0;
1033         int rc;
1034
1035         if (adapter->vpd->buff)
1036                 len = adapter->vpd->len;
1037
1038         mutex_lock(&adapter->fw_lock);
1039         adapter->fw_done_rc = 0;
1040         reinit_completion(&adapter->fw_done);
1041
1042         crq.get_vpd_size.first = IBMVNIC_CRQ_CMD;
1043         crq.get_vpd_size.cmd = GET_VPD_SIZE;
1044         rc = ibmvnic_send_crq(adapter, &crq);
1045         if (rc) {
1046                 mutex_unlock(&adapter->fw_lock);
1047                 return rc;
1048         }
1049
1050         rc = ibmvnic_wait_for_completion(adapter, &adapter->fw_done, 10000);
1051         if (rc) {
1052                 dev_err(dev, "Could not retrieve VPD size, rc = %d\n", rc);
1053                 mutex_unlock(&adapter->fw_lock);
1054                 return rc;
1055         }
1056         mutex_unlock(&adapter->fw_lock);
1057
1058         if (!adapter->vpd->len)
1059                 return -ENODATA;
1060
1061         if (!adapter->vpd->buff)
1062                 adapter->vpd->buff = kzalloc(adapter->vpd->len, GFP_KERNEL);
1063         else if (adapter->vpd->len != len)
1064                 adapter->vpd->buff =
1065                         krealloc(adapter->vpd->buff,
1066                                  adapter->vpd->len, GFP_KERNEL);
1067
1068         if (!adapter->vpd->buff) {
1069                 dev_err(dev, "Could allocate VPD buffer\n");
1070                 return -ENOMEM;
1071         }
1072
1073         adapter->vpd->dma_addr =
1074                 dma_map_single(dev, adapter->vpd->buff, adapter->vpd->len,
1075                                DMA_FROM_DEVICE);
1076         if (dma_mapping_error(dev, adapter->vpd->dma_addr)) {
1077                 dev_err(dev, "Could not map VPD buffer\n");
1078                 kfree(adapter->vpd->buff);
1079                 adapter->vpd->buff = NULL;
1080                 return -ENOMEM;
1081         }
1082
1083         mutex_lock(&adapter->fw_lock);
1084         adapter->fw_done_rc = 0;
1085         reinit_completion(&adapter->fw_done);
1086
1087         crq.get_vpd.first = IBMVNIC_CRQ_CMD;
1088         crq.get_vpd.cmd = GET_VPD;
1089         crq.get_vpd.ioba = cpu_to_be32(adapter->vpd->dma_addr);
1090         crq.get_vpd.len = cpu_to_be32((u32)adapter->vpd->len);
1091         rc = ibmvnic_send_crq(adapter, &crq);
1092         if (rc) {
1093                 kfree(adapter->vpd->buff);
1094                 adapter->vpd->buff = NULL;
1095                 mutex_unlock(&adapter->fw_lock);
1096                 return rc;
1097         }
1098
1099         rc = ibmvnic_wait_for_completion(adapter, &adapter->fw_done, 10000);
1100         if (rc) {
1101                 dev_err(dev, "Unable to retrieve VPD, rc = %d\n", rc);
1102                 kfree(adapter->vpd->buff);
1103                 adapter->vpd->buff = NULL;
1104                 mutex_unlock(&adapter->fw_lock);
1105                 return rc;
1106         }
1107
1108         mutex_unlock(&adapter->fw_lock);
1109         return 0;
1110 }
1111
1112 static int init_resources(struct ibmvnic_adapter *adapter)
1113 {
1114         struct net_device *netdev = adapter->netdev;
1115         int rc;
1116
1117         rc = set_real_num_queues(netdev);
1118         if (rc)
1119                 return rc;
1120
1121         adapter->vpd = kzalloc(sizeof(*adapter->vpd), GFP_KERNEL);
1122         if (!adapter->vpd)
1123                 return -ENOMEM;
1124
1125         /* Vital Product Data (VPD) */
1126         rc = ibmvnic_get_vpd(adapter);
1127         if (rc) {
1128                 netdev_err(netdev, "failed to initialize Vital Product Data (VPD)\n");
1129                 return rc;
1130         }
1131
1132         adapter->map_id = 1;
1133
1134         rc = init_napi(adapter);
1135         if (rc)
1136                 return rc;
1137
1138         send_query_map(adapter);
1139
1140         rc = init_rx_pools(netdev);
1141         if (rc)
1142                 return rc;
1143
1144         rc = init_tx_pools(netdev);
1145         return rc;
1146 }
1147
1148 static int __ibmvnic_open(struct net_device *netdev)
1149 {
1150         struct ibmvnic_adapter *adapter = netdev_priv(netdev);
1151         enum vnic_state prev_state = adapter->state;
1152         int i, rc;
1153
1154         adapter->state = VNIC_OPENING;
1155         replenish_pools(adapter);
1156         ibmvnic_napi_enable(adapter);
1157
1158         /* We're ready to receive frames, enable the sub-crq interrupts and
1159          * set the logical link state to up
1160          */
1161         for (i = 0; i < adapter->req_rx_queues; i++) {
1162                 netdev_dbg(netdev, "Enabling rx_scrq[%d] irq\n", i);
1163                 if (prev_state == VNIC_CLOSED)
1164                         enable_irq(adapter->rx_scrq[i]->irq);
1165                 enable_scrq_irq(adapter, adapter->rx_scrq[i]);
1166         }
1167
1168         for (i = 0; i < adapter->req_tx_queues; i++) {
1169                 netdev_dbg(netdev, "Enabling tx_scrq[%d] irq\n", i);
1170                 if (prev_state == VNIC_CLOSED)
1171                         enable_irq(adapter->tx_scrq[i]->irq);
1172                 enable_scrq_irq(adapter, adapter->tx_scrq[i]);
1173                 netdev_tx_reset_queue(netdev_get_tx_queue(netdev, i));
1174         }
1175
1176         rc = set_link_state(adapter, IBMVNIC_LOGICAL_LNK_UP);
1177         if (rc) {
1178                 for (i = 0; i < adapter->req_rx_queues; i++)
1179                         napi_disable(&adapter->napi[i]);
1180                 release_resources(adapter);
1181                 return rc;
1182         }
1183
1184         netif_tx_start_all_queues(netdev);
1185
1186         if (prev_state == VNIC_CLOSED) {
1187                 for (i = 0; i < adapter->req_rx_queues; i++)
1188                         napi_schedule(&adapter->napi[i]);
1189         }
1190
1191         adapter->state = VNIC_OPEN;
1192         return rc;
1193 }
1194
1195 static int ibmvnic_open(struct net_device *netdev)
1196 {
1197         struct ibmvnic_adapter *adapter = netdev_priv(netdev);
1198         int rc;
1199
1200         /* If device failover is pending, just set device state and return.
1201          * Device operation will be handled by reset routine.
1202          */
1203         if (adapter->failover_pending) {
1204                 adapter->state = VNIC_OPEN;
1205                 return 0;
1206         }
1207
1208         if (adapter->state != VNIC_CLOSED) {
1209                 rc = ibmvnic_login(netdev);
1210                 if (rc)
1211                         goto out;
1212
1213                 rc = init_resources(adapter);
1214                 if (rc) {
1215                         netdev_err(netdev, "failed to initialize resources\n");
1216                         release_resources(adapter);
1217                         goto out;
1218                 }
1219         }
1220
1221         rc = __ibmvnic_open(netdev);
1222
1223 out:
1224         /*
1225          * If open fails due to a pending failover, set device state and
1226          * return. Device operation will be handled by reset routine.
1227          */
1228         if (rc && adapter->failover_pending) {
1229                 adapter->state = VNIC_OPEN;
1230                 rc = 0;
1231         }
1232         return rc;
1233 }
1234
1235 static void clean_rx_pools(struct ibmvnic_adapter *adapter)
1236 {
1237         struct ibmvnic_rx_pool *rx_pool;
1238         struct ibmvnic_rx_buff *rx_buff;
1239         u64 rx_entries;
1240         int rx_scrqs;
1241         int i, j;
1242
1243         if (!adapter->rx_pool)
1244                 return;
1245
1246         rx_scrqs = adapter->num_active_rx_pools;
1247         rx_entries = adapter->req_rx_add_entries_per_subcrq;
1248
1249         /* Free any remaining skbs in the rx buffer pools */
1250         for (i = 0; i < rx_scrqs; i++) {
1251                 rx_pool = &adapter->rx_pool[i];
1252                 if (!rx_pool || !rx_pool->rx_buff)
1253                         continue;
1254
1255                 netdev_dbg(adapter->netdev, "Cleaning rx_pool[%d]\n", i);
1256                 for (j = 0; j < rx_entries; j++) {
1257                         rx_buff = &rx_pool->rx_buff[j];
1258                         if (rx_buff && rx_buff->skb) {
1259                                 dev_kfree_skb_any(rx_buff->skb);
1260                                 rx_buff->skb = NULL;
1261                         }
1262                 }
1263         }
1264 }
1265
1266 static void clean_one_tx_pool(struct ibmvnic_adapter *adapter,
1267                               struct ibmvnic_tx_pool *tx_pool)
1268 {
1269         struct ibmvnic_tx_buff *tx_buff;
1270         u64 tx_entries;
1271         int i;
1272
1273         if (!tx_pool || !tx_pool->tx_buff)
1274                 return;
1275
1276         tx_entries = tx_pool->num_buffers;
1277
1278         for (i = 0; i < tx_entries; i++) {
1279                 tx_buff = &tx_pool->tx_buff[i];
1280                 if (tx_buff && tx_buff->skb) {
1281                         dev_kfree_skb_any(tx_buff->skb);
1282                         tx_buff->skb = NULL;
1283                 }
1284         }
1285 }
1286
1287 static void clean_tx_pools(struct ibmvnic_adapter *adapter)
1288 {
1289         int tx_scrqs;
1290         int i;
1291
1292         if (!adapter->tx_pool || !adapter->tso_pool)
1293                 return;
1294
1295         tx_scrqs = adapter->num_active_tx_pools;
1296
1297         /* Free any remaining skbs in the tx buffer pools */
1298         for (i = 0; i < tx_scrqs; i++) {
1299                 netdev_dbg(adapter->netdev, "Cleaning tx_pool[%d]\n", i);
1300                 clean_one_tx_pool(adapter, &adapter->tx_pool[i]);
1301                 clean_one_tx_pool(adapter, &adapter->tso_pool[i]);
1302         }
1303 }
1304
1305 static void ibmvnic_disable_irqs(struct ibmvnic_adapter *adapter)
1306 {
1307         struct net_device *netdev = adapter->netdev;
1308         int i;
1309
1310         if (adapter->tx_scrq) {
1311                 for (i = 0; i < adapter->req_tx_queues; i++)
1312                         if (adapter->tx_scrq[i]->irq) {
1313                                 netdev_dbg(netdev,
1314                                            "Disabling tx_scrq[%d] irq\n", i);
1315                                 disable_scrq_irq(adapter, adapter->tx_scrq[i]);
1316                                 disable_irq(adapter->tx_scrq[i]->irq);
1317                         }
1318         }
1319
1320         if (adapter->rx_scrq) {
1321                 for (i = 0; i < adapter->req_rx_queues; i++) {
1322                         if (adapter->rx_scrq[i]->irq) {
1323                                 netdev_dbg(netdev,
1324                                            "Disabling rx_scrq[%d] irq\n", i);
1325                                 disable_scrq_irq(adapter, adapter->rx_scrq[i]);
1326                                 disable_irq(adapter->rx_scrq[i]->irq);
1327                         }
1328                 }
1329         }
1330 }
1331
1332 static void ibmvnic_cleanup(struct net_device *netdev)
1333 {
1334         struct ibmvnic_adapter *adapter = netdev_priv(netdev);
1335
1336         /* ensure that transmissions are stopped if called by do_reset */
1337         if (test_bit(0, &adapter->resetting))
1338                 netif_tx_disable(netdev);
1339         else
1340                 netif_tx_stop_all_queues(netdev);
1341
1342         ibmvnic_napi_disable(adapter);
1343         ibmvnic_disable_irqs(adapter);
1344
1345         clean_rx_pools(adapter);
1346         clean_tx_pools(adapter);
1347 }
1348
1349 static int __ibmvnic_close(struct net_device *netdev)
1350 {
1351         struct ibmvnic_adapter *adapter = netdev_priv(netdev);
1352         int rc = 0;
1353
1354         adapter->state = VNIC_CLOSING;
1355         rc = set_link_state(adapter, IBMVNIC_LOGICAL_LNK_DN);
1356         if (rc)
1357                 return rc;
1358         adapter->state = VNIC_CLOSED;
1359         return 0;
1360 }
1361
1362 static int ibmvnic_close(struct net_device *netdev)
1363 {
1364         struct ibmvnic_adapter *adapter = netdev_priv(netdev);
1365         int rc;
1366
1367         netdev_dbg(netdev, "[S:%d FOP:%d FRR:%d] Closing\n",
1368                    adapter->state, adapter->failover_pending,
1369                    adapter->force_reset_recovery);
1370
1371         /* If device failover is pending, just set device state and return.
1372          * Device operation will be handled by reset routine.
1373          */
1374         if (adapter->failover_pending) {
1375                 adapter->state = VNIC_CLOSED;
1376                 return 0;
1377         }
1378
1379         rc = __ibmvnic_close(netdev);
1380         ibmvnic_cleanup(netdev);
1381
1382         return rc;
1383 }
1384
1385 /**
1386  * build_hdr_data - creates L2/L3/L4 header data buffer
1387  * @hdr_field - bitfield determining needed headers
1388  * @skb - socket buffer
1389  * @hdr_len - array of header lengths
1390  * @tot_len - total length of data
1391  *
1392  * Reads hdr_field to determine which headers are needed by firmware.
1393  * Builds a buffer containing these headers.  Saves individual header
1394  * lengths and total buffer length to be used to build descriptors.
1395  */
1396 static int build_hdr_data(u8 hdr_field, struct sk_buff *skb,
1397                           int *hdr_len, u8 *hdr_data)
1398 {
1399         int len = 0;
1400         u8 *hdr;
1401
1402         if (skb_vlan_tagged(skb) && !skb_vlan_tag_present(skb))
1403                 hdr_len[0] = sizeof(struct vlan_ethhdr);
1404         else
1405                 hdr_len[0] = sizeof(struct ethhdr);
1406
1407         if (skb->protocol == htons(ETH_P_IP)) {
1408                 hdr_len[1] = ip_hdr(skb)->ihl * 4;
1409                 if (ip_hdr(skb)->protocol == IPPROTO_TCP)
1410                         hdr_len[2] = tcp_hdrlen(skb);
1411                 else if (ip_hdr(skb)->protocol == IPPROTO_UDP)
1412                         hdr_len[2] = sizeof(struct udphdr);
1413         } else if (skb->protocol == htons(ETH_P_IPV6)) {
1414                 hdr_len[1] = sizeof(struct ipv6hdr);
1415                 if (ipv6_hdr(skb)->nexthdr == IPPROTO_TCP)
1416                         hdr_len[2] = tcp_hdrlen(skb);
1417                 else if (ipv6_hdr(skb)->nexthdr == IPPROTO_UDP)
1418                         hdr_len[2] = sizeof(struct udphdr);
1419         } else if (skb->protocol == htons(ETH_P_ARP)) {
1420                 hdr_len[1] = arp_hdr_len(skb->dev);
1421                 hdr_len[2] = 0;
1422         }
1423
1424         memset(hdr_data, 0, 120);
1425         if ((hdr_field >> 6) & 1) {
1426                 hdr = skb_mac_header(skb);
1427                 memcpy(hdr_data, hdr, hdr_len[0]);
1428                 len += hdr_len[0];
1429         }
1430
1431         if ((hdr_field >> 5) & 1) {
1432                 hdr = skb_network_header(skb);
1433                 memcpy(hdr_data + len, hdr, hdr_len[1]);
1434                 len += hdr_len[1];
1435         }
1436
1437         if ((hdr_field >> 4) & 1) {
1438                 hdr = skb_transport_header(skb);
1439                 memcpy(hdr_data + len, hdr, hdr_len[2]);
1440                 len += hdr_len[2];
1441         }
1442         return len;
1443 }
1444
1445 /**
1446  * create_hdr_descs - create header and header extension descriptors
1447  * @hdr_field - bitfield determining needed headers
1448  * @data - buffer containing header data
1449  * @len - length of data buffer
1450  * @hdr_len - array of individual header lengths
1451  * @scrq_arr - descriptor array
1452  *
1453  * Creates header and, if needed, header extension descriptors and
1454  * places them in a descriptor array, scrq_arr
1455  */
1456
1457 static int create_hdr_descs(u8 hdr_field, u8 *hdr_data, int len, int *hdr_len,
1458                             union sub_crq *scrq_arr)
1459 {
1460         union sub_crq hdr_desc;
1461         int tmp_len = len;
1462         int num_descs = 0;
1463         u8 *data, *cur;
1464         int tmp;
1465
1466         while (tmp_len > 0) {
1467                 cur = hdr_data + len - tmp_len;
1468
1469                 memset(&hdr_desc, 0, sizeof(hdr_desc));
1470                 if (cur != hdr_data) {
1471                         data = hdr_desc.hdr_ext.data;
1472                         tmp = tmp_len > 29 ? 29 : tmp_len;
1473                         hdr_desc.hdr_ext.first = IBMVNIC_CRQ_CMD;
1474                         hdr_desc.hdr_ext.type = IBMVNIC_HDR_EXT_DESC;
1475                         hdr_desc.hdr_ext.len = tmp;
1476                 } else {
1477                         data = hdr_desc.hdr.data;
1478                         tmp = tmp_len > 24 ? 24 : tmp_len;
1479                         hdr_desc.hdr.first = IBMVNIC_CRQ_CMD;
1480                         hdr_desc.hdr.type = IBMVNIC_HDR_DESC;
1481                         hdr_desc.hdr.len = tmp;
1482                         hdr_desc.hdr.l2_len = (u8)hdr_len[0];
1483                         hdr_desc.hdr.l3_len = cpu_to_be16((u16)hdr_len[1]);
1484                         hdr_desc.hdr.l4_len = (u8)hdr_len[2];
1485                         hdr_desc.hdr.flag = hdr_field << 1;
1486                 }
1487                 memcpy(data, cur, tmp);
1488                 tmp_len -= tmp;
1489                 *scrq_arr = hdr_desc;
1490                 scrq_arr++;
1491                 num_descs++;
1492         }
1493
1494         return num_descs;
1495 }
1496
1497 /**
1498  * build_hdr_descs_arr - build a header descriptor array
1499  * @skb - socket buffer
1500  * @num_entries - number of descriptors to be sent
1501  * @subcrq - first TX descriptor
1502  * @hdr_field - bit field determining which headers will be sent
1503  *
1504  * This function will build a TX descriptor array with applicable
1505  * L2/L3/L4 packet header descriptors to be sent by send_subcrq_indirect.
1506  */
1507
1508 static void build_hdr_descs_arr(struct sk_buff *skb,
1509                                 union sub_crq *indir_arr,
1510                                 int *num_entries, u8 hdr_field)
1511 {
1512         int hdr_len[3] = {0, 0, 0};
1513         u8 hdr_data[140] = {0};
1514         int tot_len;
1515
1516         tot_len = build_hdr_data(hdr_field, skb, hdr_len,
1517                                  hdr_data);
1518         *num_entries += create_hdr_descs(hdr_field, hdr_data, tot_len, hdr_len,
1519                                          indir_arr + 1);
1520 }
1521
1522 static int ibmvnic_xmit_workarounds(struct sk_buff *skb,
1523                                     struct net_device *netdev)
1524 {
1525         /* For some backing devices, mishandling of small packets
1526          * can result in a loss of connection or TX stall. Device
1527          * architects recommend that no packet should be smaller
1528          * than the minimum MTU value provided to the driver, so
1529          * pad any packets to that length
1530          */
1531         if (skb->len < netdev->min_mtu)
1532                 return skb_put_padto(skb, netdev->min_mtu);
1533
1534         return 0;
1535 }
1536
1537 static void ibmvnic_tx_scrq_clean_buffer(struct ibmvnic_adapter *adapter,
1538                                          struct ibmvnic_sub_crq_queue *tx_scrq)
1539 {
1540         struct ibmvnic_ind_xmit_queue *ind_bufp;
1541         struct ibmvnic_tx_buff *tx_buff;
1542         struct ibmvnic_tx_pool *tx_pool;
1543         union sub_crq tx_scrq_entry;
1544         int queue_num;
1545         int entries;
1546         int index;
1547         int i;
1548
1549         ind_bufp = &tx_scrq->ind_buf;
1550         entries = (u64)ind_bufp->index;
1551         queue_num = tx_scrq->pool_index;
1552
1553         for (i = entries - 1; i >= 0; --i) {
1554                 tx_scrq_entry = ind_bufp->indir_arr[i];
1555                 if (tx_scrq_entry.v1.type != IBMVNIC_TX_DESC)
1556                         continue;
1557                 index = be32_to_cpu(tx_scrq_entry.v1.correlator);
1558                 if (index & IBMVNIC_TSO_POOL_MASK) {
1559                         tx_pool = &adapter->tso_pool[queue_num];
1560                         index &= ~IBMVNIC_TSO_POOL_MASK;
1561                 } else {
1562                         tx_pool = &adapter->tx_pool[queue_num];
1563                 }
1564                 tx_pool->free_map[tx_pool->consumer_index] = index;
1565                 tx_pool->consumer_index = tx_pool->consumer_index == 0 ?
1566                                           tx_pool->num_buffers - 1 :
1567                                           tx_pool->consumer_index - 1;
1568                 tx_buff = &tx_pool->tx_buff[index];
1569                 adapter->netdev->stats.tx_packets--;
1570                 adapter->netdev->stats.tx_bytes -= tx_buff->skb->len;
1571                 adapter->tx_stats_buffers[queue_num].packets--;
1572                 adapter->tx_stats_buffers[queue_num].bytes -=
1573                                                 tx_buff->skb->len;
1574                 dev_kfree_skb_any(tx_buff->skb);
1575                 tx_buff->skb = NULL;
1576                 adapter->netdev->stats.tx_dropped++;
1577         }
1578         ind_bufp->index = 0;
1579         if (atomic_sub_return(entries, &tx_scrq->used) <=
1580             (adapter->req_tx_entries_per_subcrq / 2) &&
1581             __netif_subqueue_stopped(adapter->netdev, queue_num)) {
1582                 netif_wake_subqueue(adapter->netdev, queue_num);
1583                 netdev_dbg(adapter->netdev, "Started queue %d\n",
1584                            queue_num);
1585         }
1586 }
1587
1588 static int ibmvnic_tx_scrq_flush(struct ibmvnic_adapter *adapter,
1589                                  struct ibmvnic_sub_crq_queue *tx_scrq)
1590 {
1591         struct ibmvnic_ind_xmit_queue *ind_bufp;
1592         u64 dma_addr;
1593         u64 entries;
1594         u64 handle;
1595         int rc;
1596
1597         ind_bufp = &tx_scrq->ind_buf;
1598         dma_addr = (u64)ind_bufp->indir_dma;
1599         entries = (u64)ind_bufp->index;
1600         handle = tx_scrq->handle;
1601
1602         if (!entries)
1603                 return 0;
1604         rc = send_subcrq_indirect(adapter, handle, dma_addr, entries);
1605         if (rc)
1606                 ibmvnic_tx_scrq_clean_buffer(adapter, tx_scrq);
1607         else
1608                 ind_bufp->index = 0;
1609         return 0;
1610 }
1611
1612 static netdev_tx_t ibmvnic_xmit(struct sk_buff *skb, struct net_device *netdev)
1613 {
1614         struct ibmvnic_adapter *adapter = netdev_priv(netdev);
1615         int queue_num = skb_get_queue_mapping(skb);
1616         u8 *hdrs = (u8 *)&adapter->tx_rx_desc_req;
1617         struct device *dev = &adapter->vdev->dev;
1618         struct ibmvnic_ind_xmit_queue *ind_bufp;
1619         struct ibmvnic_tx_buff *tx_buff = NULL;
1620         struct ibmvnic_sub_crq_queue *tx_scrq;
1621         struct ibmvnic_tx_pool *tx_pool;
1622         unsigned int tx_send_failed = 0;
1623         netdev_tx_t ret = NETDEV_TX_OK;
1624         unsigned int tx_map_failed = 0;
1625         union sub_crq indir_arr[16];
1626         unsigned int tx_dropped = 0;
1627         unsigned int tx_packets = 0;
1628         unsigned int tx_bytes = 0;
1629         dma_addr_t data_dma_addr;
1630         struct netdev_queue *txq;
1631         unsigned long lpar_rc;
1632         union sub_crq tx_crq;
1633         unsigned int offset;
1634         int num_entries = 1;
1635         unsigned char *dst;
1636         int index = 0;
1637         u8 proto = 0;
1638
1639         tx_scrq = adapter->tx_scrq[queue_num];
1640         txq = netdev_get_tx_queue(netdev, queue_num);
1641         ind_bufp = &tx_scrq->ind_buf;
1642
1643         if (test_bit(0, &adapter->resetting)) {
1644                 if (!netif_subqueue_stopped(netdev, skb))
1645                         netif_stop_subqueue(netdev, queue_num);
1646                 dev_kfree_skb_any(skb);
1647
1648                 tx_send_failed++;
1649                 tx_dropped++;
1650                 ret = NETDEV_TX_OK;
1651                 ibmvnic_tx_scrq_flush(adapter, tx_scrq);
1652                 goto out;
1653         }
1654
1655         if (ibmvnic_xmit_workarounds(skb, netdev)) {
1656                 tx_dropped++;
1657                 tx_send_failed++;
1658                 ret = NETDEV_TX_OK;
1659                 ibmvnic_tx_scrq_flush(adapter, tx_scrq);
1660                 goto out;
1661         }
1662         if (skb_is_gso(skb))
1663                 tx_pool = &adapter->tso_pool[queue_num];
1664         else
1665                 tx_pool = &adapter->tx_pool[queue_num];
1666
1667         index = tx_pool->free_map[tx_pool->consumer_index];
1668
1669         if (index == IBMVNIC_INVALID_MAP) {
1670                 dev_kfree_skb_any(skb);
1671                 tx_send_failed++;
1672                 tx_dropped++;
1673                 ret = NETDEV_TX_OK;
1674                 ibmvnic_tx_scrq_flush(adapter, tx_scrq);
1675                 goto out;
1676         }
1677
1678         tx_pool->free_map[tx_pool->consumer_index] = IBMVNIC_INVALID_MAP;
1679
1680         offset = index * tx_pool->buf_size;
1681         dst = tx_pool->long_term_buff.buff + offset;
1682         memset(dst, 0, tx_pool->buf_size);
1683         data_dma_addr = tx_pool->long_term_buff.addr + offset;
1684
1685         if (skb_shinfo(skb)->nr_frags) {
1686                 int cur, i;
1687
1688                 /* Copy the head */
1689                 skb_copy_from_linear_data(skb, dst, skb_headlen(skb));
1690                 cur = skb_headlen(skb);
1691
1692                 /* Copy the frags */
1693                 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
1694                         const skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
1695
1696                         memcpy(dst + cur,
1697                                page_address(skb_frag_page(frag)) +
1698                                skb_frag_off(frag), skb_frag_size(frag));
1699                         cur += skb_frag_size(frag);
1700                 }
1701         } else {
1702                 skb_copy_from_linear_data(skb, dst, skb->len);
1703         }
1704
1705         tx_pool->consumer_index =
1706             (tx_pool->consumer_index + 1) % tx_pool->num_buffers;
1707
1708         tx_buff = &tx_pool->tx_buff[index];
1709         tx_buff->skb = skb;
1710         tx_buff->index = index;
1711         tx_buff->pool_index = queue_num;
1712
1713         memset(&tx_crq, 0, sizeof(tx_crq));
1714         tx_crq.v1.first = IBMVNIC_CRQ_CMD;
1715         tx_crq.v1.type = IBMVNIC_TX_DESC;
1716         tx_crq.v1.n_crq_elem = 1;
1717         tx_crq.v1.n_sge = 1;
1718         tx_crq.v1.flags1 = IBMVNIC_TX_COMP_NEEDED;
1719
1720         if (skb_is_gso(skb))
1721                 tx_crq.v1.correlator =
1722                         cpu_to_be32(index | IBMVNIC_TSO_POOL_MASK);
1723         else
1724                 tx_crq.v1.correlator = cpu_to_be32(index);
1725         tx_crq.v1.dma_reg = cpu_to_be16(tx_pool->long_term_buff.map_id);
1726         tx_crq.v1.sge_len = cpu_to_be32(skb->len);
1727         tx_crq.v1.ioba = cpu_to_be64(data_dma_addr);
1728
1729         if (adapter->vlan_header_insertion && skb_vlan_tag_present(skb)) {
1730                 tx_crq.v1.flags2 |= IBMVNIC_TX_VLAN_INSERT;
1731                 tx_crq.v1.vlan_id = cpu_to_be16(skb->vlan_tci);
1732         }
1733
1734         if (skb->protocol == htons(ETH_P_IP)) {
1735                 tx_crq.v1.flags1 |= IBMVNIC_TX_PROT_IPV4;
1736                 proto = ip_hdr(skb)->protocol;
1737         } else if (skb->protocol == htons(ETH_P_IPV6)) {
1738                 tx_crq.v1.flags1 |= IBMVNIC_TX_PROT_IPV6;
1739                 proto = ipv6_hdr(skb)->nexthdr;
1740         }
1741
1742         if (proto == IPPROTO_TCP)
1743                 tx_crq.v1.flags1 |= IBMVNIC_TX_PROT_TCP;
1744         else if (proto == IPPROTO_UDP)
1745                 tx_crq.v1.flags1 |= IBMVNIC_TX_PROT_UDP;
1746
1747         if (skb->ip_summed == CHECKSUM_PARTIAL) {
1748                 tx_crq.v1.flags1 |= IBMVNIC_TX_CHKSUM_OFFLOAD;
1749                 hdrs += 2;
1750         }
1751         if (skb_is_gso(skb)) {
1752                 tx_crq.v1.flags1 |= IBMVNIC_TX_LSO;
1753                 tx_crq.v1.mss = cpu_to_be16(skb_shinfo(skb)->gso_size);
1754                 hdrs += 2;
1755         }
1756
1757         if ((*hdrs >> 7) & 1)
1758                 build_hdr_descs_arr(skb, indir_arr, &num_entries, *hdrs);
1759
1760         tx_crq.v1.n_crq_elem = num_entries;
1761         tx_buff->num_entries = num_entries;
1762         /* flush buffer if current entry can not fit */
1763         if (num_entries + ind_bufp->index > IBMVNIC_MAX_IND_DESCS) {
1764                 lpar_rc = ibmvnic_tx_scrq_flush(adapter, tx_scrq);
1765                 if (lpar_rc != H_SUCCESS)
1766                         goto tx_flush_err;
1767         }
1768
1769         indir_arr[0] = tx_crq;
1770         memcpy(&ind_bufp->indir_arr[ind_bufp->index], &indir_arr[0],
1771                num_entries * sizeof(struct ibmvnic_generic_scrq));
1772         ind_bufp->index += num_entries;
1773         if (__netdev_tx_sent_queue(txq, skb->len,
1774                                    netdev_xmit_more() &&
1775                                    ind_bufp->index < IBMVNIC_MAX_IND_DESCS)) {
1776                 lpar_rc = ibmvnic_tx_scrq_flush(adapter, tx_scrq);
1777                 if (lpar_rc != H_SUCCESS)
1778                         goto tx_err;
1779         }
1780
1781         if (atomic_add_return(num_entries, &tx_scrq->used)
1782                                         >= adapter->req_tx_entries_per_subcrq) {
1783                 netdev_dbg(netdev, "Stopping queue %d\n", queue_num);
1784                 netif_stop_subqueue(netdev, queue_num);
1785         }
1786
1787         tx_packets++;
1788         tx_bytes += skb->len;
1789         txq->trans_start = jiffies;
1790         ret = NETDEV_TX_OK;
1791         goto out;
1792
1793 tx_flush_err:
1794         dev_kfree_skb_any(skb);
1795         tx_buff->skb = NULL;
1796         tx_pool->consumer_index = tx_pool->consumer_index == 0 ?
1797                                   tx_pool->num_buffers - 1 :
1798                                   tx_pool->consumer_index - 1;
1799         tx_dropped++;
1800 tx_err:
1801         if (lpar_rc != H_CLOSED && lpar_rc != H_PARAMETER)
1802                 dev_err_ratelimited(dev, "tx: send failed\n");
1803
1804         if (lpar_rc == H_CLOSED || adapter->failover_pending) {
1805                 /* Disable TX and report carrier off if queue is closed
1806                  * or pending failover.
1807                  * Firmware guarantees that a signal will be sent to the
1808                  * driver, triggering a reset or some other action.
1809                  */
1810                 netif_tx_stop_all_queues(netdev);
1811                 netif_carrier_off(netdev);
1812         }
1813 out:
1814         netdev->stats.tx_dropped += tx_dropped;
1815         netdev->stats.tx_bytes += tx_bytes;
1816         netdev->stats.tx_packets += tx_packets;
1817         adapter->tx_send_failed += tx_send_failed;
1818         adapter->tx_map_failed += tx_map_failed;
1819         adapter->tx_stats_buffers[queue_num].packets += tx_packets;
1820         adapter->tx_stats_buffers[queue_num].bytes += tx_bytes;
1821         adapter->tx_stats_buffers[queue_num].dropped_packets += tx_dropped;
1822
1823         return ret;
1824 }
1825
1826 static void ibmvnic_set_multi(struct net_device *netdev)
1827 {
1828         struct ibmvnic_adapter *adapter = netdev_priv(netdev);
1829         struct netdev_hw_addr *ha;
1830         union ibmvnic_crq crq;
1831
1832         memset(&crq, 0, sizeof(crq));
1833         crq.request_capability.first = IBMVNIC_CRQ_CMD;
1834         crq.request_capability.cmd = REQUEST_CAPABILITY;
1835
1836         if (netdev->flags & IFF_PROMISC) {
1837                 if (!adapter->promisc_supported)
1838                         return;
1839         } else {
1840                 if (netdev->flags & IFF_ALLMULTI) {
1841                         /* Accept all multicast */
1842                         memset(&crq, 0, sizeof(crq));
1843                         crq.multicast_ctrl.first = IBMVNIC_CRQ_CMD;
1844                         crq.multicast_ctrl.cmd = MULTICAST_CTRL;
1845                         crq.multicast_ctrl.flags = IBMVNIC_ENABLE_ALL;
1846                         ibmvnic_send_crq(adapter, &crq);
1847                 } else if (netdev_mc_empty(netdev)) {
1848                         /* Reject all multicast */
1849                         memset(&crq, 0, sizeof(crq));
1850                         crq.multicast_ctrl.first = IBMVNIC_CRQ_CMD;
1851                         crq.multicast_ctrl.cmd = MULTICAST_CTRL;
1852                         crq.multicast_ctrl.flags = IBMVNIC_DISABLE_ALL;
1853                         ibmvnic_send_crq(adapter, &crq);
1854                 } else {
1855                         /* Accept one or more multicast(s) */
1856                         netdev_for_each_mc_addr(ha, netdev) {
1857                                 memset(&crq, 0, sizeof(crq));
1858                                 crq.multicast_ctrl.first = IBMVNIC_CRQ_CMD;
1859                                 crq.multicast_ctrl.cmd = MULTICAST_CTRL;
1860                                 crq.multicast_ctrl.flags = IBMVNIC_ENABLE_MC;
1861                                 ether_addr_copy(&crq.multicast_ctrl.mac_addr[0],
1862                                                 ha->addr);
1863                                 ibmvnic_send_crq(adapter, &crq);
1864                         }
1865                 }
1866         }
1867 }
1868
1869 static int __ibmvnic_set_mac(struct net_device *netdev, u8 *dev_addr)
1870 {
1871         struct ibmvnic_adapter *adapter = netdev_priv(netdev);
1872         union ibmvnic_crq crq;
1873         int rc;
1874
1875         if (!is_valid_ether_addr(dev_addr)) {
1876                 rc = -EADDRNOTAVAIL;
1877                 goto err;
1878         }
1879
1880         memset(&crq, 0, sizeof(crq));
1881         crq.change_mac_addr.first = IBMVNIC_CRQ_CMD;
1882         crq.change_mac_addr.cmd = CHANGE_MAC_ADDR;
1883         ether_addr_copy(&crq.change_mac_addr.mac_addr[0], dev_addr);
1884
1885         mutex_lock(&adapter->fw_lock);
1886         adapter->fw_done_rc = 0;
1887         reinit_completion(&adapter->fw_done);
1888
1889         rc = ibmvnic_send_crq(adapter, &crq);
1890         if (rc) {
1891                 rc = -EIO;
1892                 mutex_unlock(&adapter->fw_lock);
1893                 goto err;
1894         }
1895
1896         rc = ibmvnic_wait_for_completion(adapter, &adapter->fw_done, 10000);
1897         /* netdev->dev_addr is changed in handle_change_mac_rsp function */
1898         if (rc || adapter->fw_done_rc) {
1899                 rc = -EIO;
1900                 mutex_unlock(&adapter->fw_lock);
1901                 goto err;
1902         }
1903         mutex_unlock(&adapter->fw_lock);
1904         return 0;
1905 err:
1906         ether_addr_copy(adapter->mac_addr, netdev->dev_addr);
1907         return rc;
1908 }
1909
1910 static int ibmvnic_set_mac(struct net_device *netdev, void *p)
1911 {
1912         struct ibmvnic_adapter *adapter = netdev_priv(netdev);
1913         struct sockaddr *addr = p;
1914         int rc;
1915
1916         rc = 0;
1917         if (!is_valid_ether_addr(addr->sa_data))
1918                 return -EADDRNOTAVAIL;
1919
1920         if (adapter->state != VNIC_PROBED) {
1921                 ether_addr_copy(adapter->mac_addr, addr->sa_data);
1922                 rc = __ibmvnic_set_mac(netdev, addr->sa_data);
1923         }
1924
1925         return rc;
1926 }
1927
1928 /**
1929  * do_change_param_reset returns zero if we are able to keep processing reset
1930  * events, or non-zero if we hit a fatal error and must halt.
1931  */
1932 static int do_change_param_reset(struct ibmvnic_adapter *adapter,
1933                                  struct ibmvnic_rwi *rwi,
1934                                  u32 reset_state)
1935 {
1936         struct net_device *netdev = adapter->netdev;
1937         int i, rc;
1938
1939         netdev_dbg(adapter->netdev, "Change param resetting driver (%d)\n",
1940                    rwi->reset_reason);
1941
1942         netif_carrier_off(netdev);
1943         adapter->reset_reason = rwi->reset_reason;
1944
1945         ibmvnic_cleanup(netdev);
1946
1947         if (reset_state == VNIC_OPEN) {
1948                 rc = __ibmvnic_close(netdev);
1949                 if (rc)
1950                         goto out;
1951         }
1952
1953         release_resources(adapter);
1954         release_sub_crqs(adapter, 1);
1955         release_crq_queue(adapter);
1956
1957         adapter->state = VNIC_PROBED;
1958
1959         rc = init_crq_queue(adapter);
1960
1961         if (rc) {
1962                 netdev_err(adapter->netdev,
1963                            "Couldn't initialize crq. rc=%d\n", rc);
1964                 return rc;
1965         }
1966
1967         rc = ibmvnic_reset_init(adapter, true);
1968         if (rc) {
1969                 rc = IBMVNIC_INIT_FAILED;
1970                 goto out;
1971         }
1972
1973         /* If the adapter was in PROBE state prior to the reset,
1974          * exit here.
1975          */
1976         if (reset_state == VNIC_PROBED)
1977                 goto out;
1978
1979         rc = ibmvnic_login(netdev);
1980         if (rc) {
1981                 goto out;
1982         }
1983
1984         rc = init_resources(adapter);
1985         if (rc)
1986                 goto out;
1987
1988         ibmvnic_disable_irqs(adapter);
1989
1990         adapter->state = VNIC_CLOSED;
1991
1992         if (reset_state == VNIC_CLOSED)
1993                 return 0;
1994
1995         rc = __ibmvnic_open(netdev);
1996         if (rc) {
1997                 rc = IBMVNIC_OPEN_FAILED;
1998                 goto out;
1999         }
2000
2001         /* refresh device's multicast list */
2002         ibmvnic_set_multi(netdev);
2003
2004         /* kick napi */
2005         for (i = 0; i < adapter->req_rx_queues; i++)
2006                 napi_schedule(&adapter->napi[i]);
2007
2008 out:
2009         if (rc)
2010                 adapter->state = reset_state;
2011         return rc;
2012 }
2013
2014 /**
2015  * do_reset returns zero if we are able to keep processing reset events, or
2016  * non-zero if we hit a fatal error and must halt.
2017  */
2018 static int do_reset(struct ibmvnic_adapter *adapter,
2019                     struct ibmvnic_rwi *rwi, u32 reset_state)
2020 {
2021         u64 old_num_rx_queues, old_num_tx_queues;
2022         u64 old_num_rx_slots, old_num_tx_slots;
2023         struct net_device *netdev = adapter->netdev;
2024         int i, rc;
2025
2026         netdev_dbg(adapter->netdev,
2027                    "[S:%d FOP:%d] Reset reason %d, reset_state %d\n",
2028                    adapter->state, adapter->failover_pending,
2029                    rwi->reset_reason, reset_state);
2030
2031         rtnl_lock();
2032         /*
2033          * Now that we have the rtnl lock, clear any pending failover.
2034          * This will ensure ibmvnic_open() has either completed or will
2035          * block until failover is complete.
2036          */
2037         if (rwi->reset_reason == VNIC_RESET_FAILOVER)
2038                 adapter->failover_pending = false;
2039
2040         netif_carrier_off(netdev);
2041         adapter->reset_reason = rwi->reset_reason;
2042
2043         old_num_rx_queues = adapter->req_rx_queues;
2044         old_num_tx_queues = adapter->req_tx_queues;
2045         old_num_rx_slots = adapter->req_rx_add_entries_per_subcrq;
2046         old_num_tx_slots = adapter->req_tx_entries_per_subcrq;
2047
2048         ibmvnic_cleanup(netdev);
2049
2050         if (reset_state == VNIC_OPEN &&
2051             adapter->reset_reason != VNIC_RESET_MOBILITY &&
2052             adapter->reset_reason != VNIC_RESET_FAILOVER) {
2053                 adapter->state = VNIC_CLOSING;
2054
2055                 /* Release the RTNL lock before link state change and
2056                  * re-acquire after the link state change to allow
2057                  * linkwatch_event to grab the RTNL lock and run during
2058                  * a reset.
2059                  */
2060                 rtnl_unlock();
2061                 rc = set_link_state(adapter, IBMVNIC_LOGICAL_LNK_DN);
2062                 rtnl_lock();
2063                 if (rc)
2064                         goto out;
2065
2066                 if (adapter->state != VNIC_CLOSING) {
2067                         rc = -1;
2068                         goto out;
2069                 }
2070
2071                 adapter->state = VNIC_CLOSED;
2072         }
2073
2074         if (adapter->reset_reason != VNIC_RESET_NON_FATAL) {
2075                 /* remove the closed state so when we call open it appears
2076                  * we are coming from the probed state.
2077                  */
2078                 adapter->state = VNIC_PROBED;
2079
2080                 if (adapter->reset_reason == VNIC_RESET_MOBILITY) {
2081                         rc = ibmvnic_reenable_crq_queue(adapter);
2082                         release_sub_crqs(adapter, 1);
2083                 } else {
2084                         rc = ibmvnic_reset_crq(adapter);
2085                         if (rc == H_CLOSED || rc == H_SUCCESS) {
2086                                 rc = vio_enable_interrupts(adapter->vdev);
2087                                 if (rc)
2088                                         netdev_err(adapter->netdev,
2089                                                    "Reset failed to enable interrupts. rc=%d\n",
2090                                                    rc);
2091                         }
2092                 }
2093
2094                 if (rc) {
2095                         netdev_err(adapter->netdev,
2096                                    "Reset couldn't initialize crq. rc=%d\n", rc);
2097                         goto out;
2098                 }
2099
2100                 rc = ibmvnic_reset_init(adapter, true);
2101                 if (rc) {
2102                         rc = IBMVNIC_INIT_FAILED;
2103                         goto out;
2104                 }
2105
2106                 /* If the adapter was in PROBE state prior to the reset,
2107                  * exit here.
2108                  */
2109                 if (reset_state == VNIC_PROBED) {
2110                         rc = 0;
2111                         goto out;
2112                 }
2113
2114                 rc = ibmvnic_login(netdev);
2115                 if (rc) {
2116                         goto out;
2117                 }
2118
2119                 if (adapter->req_rx_queues != old_num_rx_queues ||
2120                     adapter->req_tx_queues != old_num_tx_queues ||
2121                     adapter->req_rx_add_entries_per_subcrq !=
2122                     old_num_rx_slots ||
2123                     adapter->req_tx_entries_per_subcrq !=
2124                     old_num_tx_slots ||
2125                     !adapter->rx_pool ||
2126                     !adapter->tso_pool ||
2127                     !adapter->tx_pool) {
2128                         release_rx_pools(adapter);
2129                         release_tx_pools(adapter);
2130                         release_napi(adapter);
2131                         release_vpd_data(adapter);
2132
2133                         rc = init_resources(adapter);
2134                         if (rc)
2135                                 goto out;
2136
2137                 } else {
2138                         rc = reset_tx_pools(adapter);
2139                         if (rc) {
2140                                 netdev_dbg(adapter->netdev, "reset tx pools failed (%d)\n",
2141                                                 rc);
2142                                 goto out;
2143                         }
2144
2145                         rc = reset_rx_pools(adapter);
2146                         if (rc) {
2147                                 netdev_dbg(adapter->netdev, "reset rx pools failed (%d)\n",
2148                                                 rc);
2149                                 goto out;
2150                         }
2151                 }
2152                 ibmvnic_disable_irqs(adapter);
2153         }
2154         adapter->state = VNIC_CLOSED;
2155
2156         if (reset_state == VNIC_CLOSED) {
2157                 rc = 0;
2158                 goto out;
2159         }
2160
2161         rc = __ibmvnic_open(netdev);
2162         if (rc) {
2163                 rc = IBMVNIC_OPEN_FAILED;
2164                 goto out;
2165         }
2166
2167         /* refresh device's multicast list */
2168         ibmvnic_set_multi(netdev);
2169
2170         /* kick napi */
2171         for (i = 0; i < adapter->req_rx_queues; i++)
2172                 napi_schedule(&adapter->napi[i]);
2173
2174         if (adapter->reset_reason == VNIC_RESET_FAILOVER ||
2175             adapter->reset_reason == VNIC_RESET_MOBILITY)
2176                 __netdev_notify_peers(netdev);
2177
2178         rc = 0;
2179
2180 out:
2181         /* restore the adapter state if reset failed */
2182         if (rc)
2183                 adapter->state = reset_state;
2184         rtnl_unlock();
2185
2186         netdev_dbg(adapter->netdev, "[S:%d FOP:%d] Reset done, rc %d\n",
2187                    adapter->state, adapter->failover_pending, rc);
2188         return rc;
2189 }
2190
2191 static int do_hard_reset(struct ibmvnic_adapter *adapter,
2192                          struct ibmvnic_rwi *rwi, u32 reset_state)
2193 {
2194         struct net_device *netdev = adapter->netdev;
2195         int rc;
2196
2197         netdev_dbg(adapter->netdev, "Hard resetting driver (%d)\n",
2198                    rwi->reset_reason);
2199
2200         netif_carrier_off(netdev);
2201         adapter->reset_reason = rwi->reset_reason;
2202
2203         ibmvnic_cleanup(netdev);
2204         release_resources(adapter);
2205         release_sub_crqs(adapter, 0);
2206         release_crq_queue(adapter);
2207
2208         /* remove the closed state so when we call open it appears
2209          * we are coming from the probed state.
2210          */
2211         adapter->state = VNIC_PROBED;
2212
2213         reinit_completion(&adapter->init_done);
2214         rc = init_crq_queue(adapter);
2215         if (rc) {
2216                 netdev_err(adapter->netdev,
2217                            "Couldn't initialize crq. rc=%d\n", rc);
2218                 goto out;
2219         }
2220
2221         rc = ibmvnic_reset_init(adapter, false);
2222         if (rc)
2223                 goto out;
2224
2225         /* If the adapter was in PROBE state prior to the reset,
2226          * exit here.
2227          */
2228         if (reset_state == VNIC_PROBED)
2229                 goto out;
2230
2231         rc = ibmvnic_login(netdev);
2232         if (rc)
2233                 goto out;
2234
2235         rc = init_resources(adapter);
2236         if (rc)
2237                 goto out;
2238
2239         ibmvnic_disable_irqs(adapter);
2240         adapter->state = VNIC_CLOSED;
2241
2242         if (reset_state == VNIC_CLOSED)
2243                 goto out;
2244
2245         rc = __ibmvnic_open(netdev);
2246         if (rc) {
2247                 rc = IBMVNIC_OPEN_FAILED;
2248                 goto out;
2249         }
2250
2251         __netdev_notify_peers(netdev);
2252 out:
2253         /* restore adapter state if reset failed */
2254         if (rc)
2255                 adapter->state = reset_state;
2256         netdev_dbg(adapter->netdev, "[S:%d FOP:%d] Hard reset done, rc %d\n",
2257                    adapter->state, adapter->failover_pending, rc);
2258         return rc;
2259 }
2260
2261 static struct ibmvnic_rwi *get_next_rwi(struct ibmvnic_adapter *adapter)
2262 {
2263         struct ibmvnic_rwi *rwi;
2264         unsigned long flags;
2265
2266         spin_lock_irqsave(&adapter->rwi_lock, flags);
2267
2268         if (!list_empty(&adapter->rwi_list)) {
2269                 rwi = list_first_entry(&adapter->rwi_list, struct ibmvnic_rwi,
2270                                        list);
2271                 list_del(&rwi->list);
2272         } else {
2273                 rwi = NULL;
2274         }
2275
2276         spin_unlock_irqrestore(&adapter->rwi_lock, flags);
2277         return rwi;
2278 }
2279
2280 static void __ibmvnic_reset(struct work_struct *work)
2281 {
2282         struct ibmvnic_rwi *rwi;
2283         struct ibmvnic_adapter *adapter;
2284         bool saved_state = false;
2285         unsigned long flags;
2286         u32 reset_state;
2287         int rc = 0;
2288
2289         adapter = container_of(work, struct ibmvnic_adapter, ibmvnic_reset);
2290
2291         if (test_and_set_bit_lock(0, &adapter->resetting)) {
2292                 schedule_delayed_work(&adapter->ibmvnic_delayed_reset,
2293                                       IBMVNIC_RESET_DELAY);
2294                 return;
2295         }
2296
2297         rwi = get_next_rwi(adapter);
2298         while (rwi) {
2299                 spin_lock_irqsave(&adapter->state_lock, flags);
2300
2301                 if (adapter->state == VNIC_REMOVING ||
2302                     adapter->state == VNIC_REMOVED) {
2303                         spin_unlock_irqrestore(&adapter->state_lock, flags);
2304                         kfree(rwi);
2305                         rc = EBUSY;
2306                         break;
2307                 }
2308
2309                 if (!saved_state) {
2310                         reset_state = adapter->state;
2311                         saved_state = true;
2312                 }
2313                 spin_unlock_irqrestore(&adapter->state_lock, flags);
2314
2315                 if (rwi->reset_reason == VNIC_RESET_CHANGE_PARAM) {
2316                         /* CHANGE_PARAM requestor holds rtnl_lock */
2317                         rc = do_change_param_reset(adapter, rwi, reset_state);
2318                 } else if (adapter->force_reset_recovery) {
2319                         /*
2320                          * Since we are doing a hard reset now, clear the
2321                          * failover_pending flag so we don't ignore any
2322                          * future MOBILITY or other resets.
2323                          */
2324                         adapter->failover_pending = false;
2325
2326                         /* Transport event occurred during previous reset */
2327                         if (adapter->wait_for_reset) {
2328                                 /* Previous was CHANGE_PARAM; caller locked */
2329                                 adapter->force_reset_recovery = false;
2330                                 rc = do_hard_reset(adapter, rwi, reset_state);
2331                         } else {
2332                                 rtnl_lock();
2333                                 adapter->force_reset_recovery = false;
2334                                 rc = do_hard_reset(adapter, rwi, reset_state);
2335                                 rtnl_unlock();
2336                         }
2337                         if (rc) {
2338                                 /* give backing device time to settle down */
2339                                 netdev_dbg(adapter->netdev,
2340                                            "[S:%d] Hard reset failed, waiting 60 secs\n",
2341                                            adapter->state);
2342                                 set_current_state(TASK_UNINTERRUPTIBLE);
2343                                 schedule_timeout(60 * HZ);
2344                         }
2345                 } else {
2346                         rc = do_reset(adapter, rwi, reset_state);
2347                 }
2348                 kfree(rwi);
2349                 adapter->last_reset_time = jiffies;
2350
2351                 if (rc)
2352                         netdev_dbg(adapter->netdev, "Reset failed, rc=%d\n", rc);
2353
2354                 rwi = get_next_rwi(adapter);
2355
2356                 if (rwi && (rwi->reset_reason == VNIC_RESET_FAILOVER ||
2357                             rwi->reset_reason == VNIC_RESET_MOBILITY))
2358                         adapter->force_reset_recovery = true;
2359         }
2360
2361         if (adapter->wait_for_reset) {
2362                 adapter->reset_done_rc = rc;
2363                 complete(&adapter->reset_done);
2364         }
2365
2366         clear_bit_unlock(0, &adapter->resetting);
2367
2368         netdev_dbg(adapter->netdev,
2369                    "[S:%d FRR:%d WFR:%d] Done processing resets\n",
2370                    adapter->state, adapter->force_reset_recovery,
2371                    adapter->wait_for_reset);
2372 }
2373
2374 static void __ibmvnic_delayed_reset(struct work_struct *work)
2375 {
2376         struct ibmvnic_adapter *adapter;
2377
2378         adapter = container_of(work, struct ibmvnic_adapter,
2379                                ibmvnic_delayed_reset.work);
2380         __ibmvnic_reset(&adapter->ibmvnic_reset);
2381 }
2382
2383 static int ibmvnic_reset(struct ibmvnic_adapter *adapter,
2384                          enum ibmvnic_reset_reason reason)
2385 {
2386         struct list_head *entry, *tmp_entry;
2387         struct ibmvnic_rwi *rwi, *tmp;
2388         struct net_device *netdev = adapter->netdev;
2389         unsigned long flags;
2390         int ret;
2391
2392         /*
2393          * If failover is pending don't schedule any other reset.
2394          * Instead let the failover complete. If there is already a
2395          * a failover reset scheduled, we will detect and drop the
2396          * duplicate reset when walking the ->rwi_list below.
2397          */
2398         if (adapter->state == VNIC_REMOVING ||
2399             adapter->state == VNIC_REMOVED ||
2400             (adapter->failover_pending && reason != VNIC_RESET_FAILOVER)) {
2401                 ret = EBUSY;
2402                 netdev_dbg(netdev, "Adapter removing or pending failover, skipping reset\n");
2403                 goto err;
2404         }
2405
2406         if (adapter->state == VNIC_PROBING) {
2407                 netdev_warn(netdev, "Adapter reset during probe\n");
2408                 ret = adapter->init_done_rc = EAGAIN;
2409                 goto err;
2410         }
2411
2412         spin_lock_irqsave(&adapter->rwi_lock, flags);
2413
2414         list_for_each(entry, &adapter->rwi_list) {
2415                 tmp = list_entry(entry, struct ibmvnic_rwi, list);
2416                 if (tmp->reset_reason == reason) {
2417                         netdev_dbg(netdev, "Skipping matching reset, reason=%d\n",
2418                                    reason);
2419                         spin_unlock_irqrestore(&adapter->rwi_lock, flags);
2420                         ret = EBUSY;
2421                         goto err;
2422                 }
2423         }
2424
2425         rwi = kzalloc(sizeof(*rwi), GFP_ATOMIC);
2426         if (!rwi) {
2427                 spin_unlock_irqrestore(&adapter->rwi_lock, flags);
2428                 ibmvnic_close(netdev);
2429                 ret = ENOMEM;
2430                 goto err;
2431         }
2432         /* if we just received a transport event,
2433          * flush reset queue and process this reset
2434          */
2435         if (adapter->force_reset_recovery && !list_empty(&adapter->rwi_list)) {
2436                 list_for_each_safe(entry, tmp_entry, &adapter->rwi_list)
2437                         list_del(entry);
2438         }
2439         rwi->reset_reason = reason;
2440         list_add_tail(&rwi->list, &adapter->rwi_list);
2441         spin_unlock_irqrestore(&adapter->rwi_lock, flags);
2442         netdev_dbg(adapter->netdev, "Scheduling reset (reason %d)\n", reason);
2443         schedule_work(&adapter->ibmvnic_reset);
2444
2445         return 0;
2446 err:
2447         return -ret;
2448 }
2449
2450 static void ibmvnic_tx_timeout(struct net_device *dev, unsigned int txqueue)
2451 {
2452         struct ibmvnic_adapter *adapter = netdev_priv(dev);
2453
2454         if (test_bit(0, &adapter->resetting)) {
2455                 netdev_err(adapter->netdev,
2456                            "Adapter is resetting, skip timeout reset\n");
2457                 return;
2458         }
2459         /* No queuing up reset until at least 5 seconds (default watchdog val)
2460          * after last reset
2461          */
2462         if (time_before(jiffies, (adapter->last_reset_time + dev->watchdog_timeo))) {
2463                 netdev_dbg(dev, "Not yet time to tx timeout.\n");
2464                 return;
2465         }
2466         ibmvnic_reset(adapter, VNIC_RESET_TIMEOUT);
2467 }
2468
2469 static void remove_buff_from_pool(struct ibmvnic_adapter *adapter,
2470                                   struct ibmvnic_rx_buff *rx_buff)
2471 {
2472         struct ibmvnic_rx_pool *pool = &adapter->rx_pool[rx_buff->pool_index];
2473
2474         rx_buff->skb = NULL;
2475
2476         pool->free_map[pool->next_alloc] = (int)(rx_buff - pool->rx_buff);
2477         pool->next_alloc = (pool->next_alloc + 1) % pool->size;
2478
2479         atomic_dec(&pool->available);
2480 }
2481
2482 static int ibmvnic_poll(struct napi_struct *napi, int budget)
2483 {
2484         struct ibmvnic_sub_crq_queue *rx_scrq;
2485         struct ibmvnic_adapter *adapter;
2486         struct net_device *netdev;
2487         int frames_processed;
2488         int scrq_num;
2489
2490         netdev = napi->dev;
2491         adapter = netdev_priv(netdev);
2492         scrq_num = (int)(napi - adapter->napi);
2493         frames_processed = 0;
2494         rx_scrq = adapter->rx_scrq[scrq_num];
2495
2496 restart_poll:
2497         while (frames_processed < budget) {
2498                 struct sk_buff *skb;
2499                 struct ibmvnic_rx_buff *rx_buff;
2500                 union sub_crq *next;
2501                 u32 length;
2502                 u16 offset;
2503                 u8 flags = 0;
2504
2505                 if (unlikely(test_bit(0, &adapter->resetting) &&
2506                              adapter->reset_reason != VNIC_RESET_NON_FATAL)) {
2507                         enable_scrq_irq(adapter, rx_scrq);
2508                         napi_complete_done(napi, frames_processed);
2509                         return frames_processed;
2510                 }
2511
2512                 if (!pending_scrq(adapter, rx_scrq))
2513                         break;
2514                 /* The queue entry at the current index is peeked at above
2515                  * to determine that there is a valid descriptor awaiting
2516                  * processing. We want to be sure that the current slot
2517                  * holds a valid descriptor before reading its contents.
2518                  */
2519                 dma_rmb();
2520                 next = ibmvnic_next_scrq(adapter, rx_scrq);
2521                 rx_buff =
2522                     (struct ibmvnic_rx_buff *)be64_to_cpu(next->
2523                                                           rx_comp.correlator);
2524                 /* do error checking */
2525                 if (next->rx_comp.rc) {
2526                         netdev_dbg(netdev, "rx buffer returned with rc %x\n",
2527                                    be16_to_cpu(next->rx_comp.rc));
2528                         /* free the entry */
2529                         next->rx_comp.first = 0;
2530                         dev_kfree_skb_any(rx_buff->skb);
2531                         remove_buff_from_pool(adapter, rx_buff);
2532                         continue;
2533                 } else if (!rx_buff->skb) {
2534                         /* free the entry */
2535                         next->rx_comp.first = 0;
2536                         remove_buff_from_pool(adapter, rx_buff);
2537                         continue;
2538                 }
2539
2540                 length = be32_to_cpu(next->rx_comp.len);
2541                 offset = be16_to_cpu(next->rx_comp.off_frame_data);
2542                 flags = next->rx_comp.flags;
2543                 skb = rx_buff->skb;
2544                 skb_copy_to_linear_data(skb, rx_buff->data + offset,
2545                                         length);
2546
2547                 /* VLAN Header has been stripped by the system firmware and
2548                  * needs to be inserted by the driver
2549                  */
2550                 if (adapter->rx_vlan_header_insertion &&
2551                     (flags & IBMVNIC_VLAN_STRIPPED))
2552                         __vlan_hwaccel_put_tag(skb, htons(ETH_P_8021Q),
2553                                                ntohs(next->rx_comp.vlan_tci));
2554
2555                 /* free the entry */
2556                 next->rx_comp.first = 0;
2557                 remove_buff_from_pool(adapter, rx_buff);
2558
2559                 skb_put(skb, length);
2560                 skb->protocol = eth_type_trans(skb, netdev);
2561                 skb_record_rx_queue(skb, scrq_num);
2562
2563                 if (flags & IBMVNIC_IP_CHKSUM_GOOD &&
2564                     flags & IBMVNIC_TCP_UDP_CHKSUM_GOOD) {
2565                         skb->ip_summed = CHECKSUM_UNNECESSARY;
2566                 }
2567
2568                 length = skb->len;
2569                 napi_gro_receive(napi, skb); /* send it up */
2570                 netdev->stats.rx_packets++;
2571                 netdev->stats.rx_bytes += length;
2572                 adapter->rx_stats_buffers[scrq_num].packets++;
2573                 adapter->rx_stats_buffers[scrq_num].bytes += length;
2574                 frames_processed++;
2575         }
2576
2577         if (adapter->state != VNIC_CLOSING &&
2578             ((atomic_read(&adapter->rx_pool[scrq_num].available) <
2579               adapter->req_rx_add_entries_per_subcrq / 2) ||
2580               frames_processed < budget))
2581                 replenish_rx_pool(adapter, &adapter->rx_pool[scrq_num]);
2582         if (frames_processed < budget) {
2583                 if (napi_complete_done(napi, frames_processed)) {
2584                         enable_scrq_irq(adapter, rx_scrq);
2585                         if (pending_scrq(adapter, rx_scrq)) {
2586                                 rmb();
2587                                 if (napi_reschedule(napi)) {
2588                                         disable_scrq_irq(adapter, rx_scrq);
2589                                         goto restart_poll;
2590                                 }
2591                         }
2592                 }
2593         }
2594         return frames_processed;
2595 }
2596
2597 static int wait_for_reset(struct ibmvnic_adapter *adapter)
2598 {
2599         int rc, ret;
2600
2601         adapter->fallback.mtu = adapter->req_mtu;
2602         adapter->fallback.rx_queues = adapter->req_rx_queues;
2603         adapter->fallback.tx_queues = adapter->req_tx_queues;
2604         adapter->fallback.rx_entries = adapter->req_rx_add_entries_per_subcrq;
2605         adapter->fallback.tx_entries = adapter->req_tx_entries_per_subcrq;
2606
2607         reinit_completion(&adapter->reset_done);
2608         adapter->wait_for_reset = true;
2609         rc = ibmvnic_reset(adapter, VNIC_RESET_CHANGE_PARAM);
2610
2611         if (rc) {
2612                 ret = rc;
2613                 goto out;
2614         }
2615         rc = ibmvnic_wait_for_completion(adapter, &adapter->reset_done, 60000);
2616         if (rc) {
2617                 ret = -ENODEV;
2618                 goto out;
2619         }
2620
2621         ret = 0;
2622         if (adapter->reset_done_rc) {
2623                 ret = -EIO;
2624                 adapter->desired.mtu = adapter->fallback.mtu;
2625                 adapter->desired.rx_queues = adapter->fallback.rx_queues;
2626                 adapter->desired.tx_queues = adapter->fallback.tx_queues;
2627                 adapter->desired.rx_entries = adapter->fallback.rx_entries;
2628                 adapter->desired.tx_entries = adapter->fallback.tx_entries;
2629
2630                 reinit_completion(&adapter->reset_done);
2631                 adapter->wait_for_reset = true;
2632                 rc = ibmvnic_reset(adapter, VNIC_RESET_CHANGE_PARAM);
2633                 if (rc) {
2634                         ret = rc;
2635                         goto out;
2636                 }
2637                 rc = ibmvnic_wait_for_completion(adapter, &adapter->reset_done,
2638                                                  60000);
2639                 if (rc) {
2640                         ret = -ENODEV;
2641                         goto out;
2642                 }
2643         }
2644 out:
2645         adapter->wait_for_reset = false;
2646
2647         return ret;
2648 }
2649
2650 static int ibmvnic_change_mtu(struct net_device *netdev, int new_mtu)
2651 {
2652         struct ibmvnic_adapter *adapter = netdev_priv(netdev);
2653
2654         adapter->desired.mtu = new_mtu + ETH_HLEN;
2655
2656         return wait_for_reset(adapter);
2657 }
2658
2659 static netdev_features_t ibmvnic_features_check(struct sk_buff *skb,
2660                                                 struct net_device *dev,
2661                                                 netdev_features_t features)
2662 {
2663         /* Some backing hardware adapters can not
2664          * handle packets with a MSS less than 224
2665          * or with only one segment.
2666          */
2667         if (skb_is_gso(skb)) {
2668                 if (skb_shinfo(skb)->gso_size < 224 ||
2669                     skb_shinfo(skb)->gso_segs == 1)
2670                         features &= ~NETIF_F_GSO_MASK;
2671         }
2672
2673         return features;
2674 }
2675
2676 static const struct net_device_ops ibmvnic_netdev_ops = {
2677         .ndo_open               = ibmvnic_open,
2678         .ndo_stop               = ibmvnic_close,
2679         .ndo_start_xmit         = ibmvnic_xmit,
2680         .ndo_set_rx_mode        = ibmvnic_set_multi,
2681         .ndo_set_mac_address    = ibmvnic_set_mac,
2682         .ndo_validate_addr      = eth_validate_addr,
2683         .ndo_tx_timeout         = ibmvnic_tx_timeout,
2684         .ndo_change_mtu         = ibmvnic_change_mtu,
2685         .ndo_features_check     = ibmvnic_features_check,
2686 };
2687
2688 /* ethtool functions */
2689
2690 static int ibmvnic_get_link_ksettings(struct net_device *netdev,
2691                                       struct ethtool_link_ksettings *cmd)
2692 {
2693         struct ibmvnic_adapter *adapter = netdev_priv(netdev);
2694         int rc;
2695
2696         rc = send_query_phys_parms(adapter);
2697         if (rc) {
2698                 adapter->speed = SPEED_UNKNOWN;
2699                 adapter->duplex = DUPLEX_UNKNOWN;
2700         }
2701         cmd->base.speed = adapter->speed;
2702         cmd->base.duplex = adapter->duplex;
2703         cmd->base.port = PORT_FIBRE;
2704         cmd->base.phy_address = 0;
2705         cmd->base.autoneg = AUTONEG_ENABLE;
2706
2707         return 0;
2708 }
2709
2710 static void ibmvnic_get_drvinfo(struct net_device *netdev,
2711                                 struct ethtool_drvinfo *info)
2712 {
2713         struct ibmvnic_adapter *adapter = netdev_priv(netdev);
2714
2715         strlcpy(info->driver, ibmvnic_driver_name, sizeof(info->driver));
2716         strlcpy(info->version, IBMVNIC_DRIVER_VERSION, sizeof(info->version));
2717         strlcpy(info->fw_version, adapter->fw_version,
2718                 sizeof(info->fw_version));
2719 }
2720
2721 static u32 ibmvnic_get_msglevel(struct net_device *netdev)
2722 {
2723         struct ibmvnic_adapter *adapter = netdev_priv(netdev);
2724
2725         return adapter->msg_enable;
2726 }
2727
2728 static void ibmvnic_set_msglevel(struct net_device *netdev, u32 data)
2729 {
2730         struct ibmvnic_adapter *adapter = netdev_priv(netdev);
2731
2732         adapter->msg_enable = data;
2733 }
2734
2735 static u32 ibmvnic_get_link(struct net_device *netdev)
2736 {
2737         struct ibmvnic_adapter *adapter = netdev_priv(netdev);
2738
2739         /* Don't need to send a query because we request a logical link up at
2740          * init and then we wait for link state indications
2741          */
2742         return adapter->logical_link_state;
2743 }
2744
2745 static void ibmvnic_get_ringparam(struct net_device *netdev,
2746                                   struct ethtool_ringparam *ring)
2747 {
2748         struct ibmvnic_adapter *adapter = netdev_priv(netdev);
2749
2750         if (adapter->priv_flags & IBMVNIC_USE_SERVER_MAXES) {
2751                 ring->rx_max_pending = adapter->max_rx_add_entries_per_subcrq;
2752                 ring->tx_max_pending = adapter->max_tx_entries_per_subcrq;
2753         } else {
2754                 ring->rx_max_pending = IBMVNIC_MAX_QUEUE_SZ;
2755                 ring->tx_max_pending = IBMVNIC_MAX_QUEUE_SZ;
2756         }
2757         ring->rx_mini_max_pending = 0;
2758         ring->rx_jumbo_max_pending = 0;
2759         ring->rx_pending = adapter->req_rx_add_entries_per_subcrq;
2760         ring->tx_pending = adapter->req_tx_entries_per_subcrq;
2761         ring->rx_mini_pending = 0;
2762         ring->rx_jumbo_pending = 0;
2763 }
2764
2765 static int ibmvnic_set_ringparam(struct net_device *netdev,
2766                                  struct ethtool_ringparam *ring)
2767 {
2768         struct ibmvnic_adapter *adapter = netdev_priv(netdev);
2769         int ret;
2770
2771         ret = 0;
2772         adapter->desired.rx_entries = ring->rx_pending;
2773         adapter->desired.tx_entries = ring->tx_pending;
2774
2775         ret = wait_for_reset(adapter);
2776
2777         if (!ret &&
2778             (adapter->req_rx_add_entries_per_subcrq != ring->rx_pending ||
2779              adapter->req_tx_entries_per_subcrq != ring->tx_pending))
2780                 netdev_info(netdev,
2781                             "Could not match full ringsize request. Requested: RX %d, TX %d; Allowed: RX %llu, TX %llu\n",
2782                             ring->rx_pending, ring->tx_pending,
2783                             adapter->req_rx_add_entries_per_subcrq,
2784                             adapter->req_tx_entries_per_subcrq);
2785         return ret;
2786 }
2787
2788 static void ibmvnic_get_channels(struct net_device *netdev,
2789                                  struct ethtool_channels *channels)
2790 {
2791         struct ibmvnic_adapter *adapter = netdev_priv(netdev);
2792
2793         if (adapter->priv_flags & IBMVNIC_USE_SERVER_MAXES) {
2794                 channels->max_rx = adapter->max_rx_queues;
2795                 channels->max_tx = adapter->max_tx_queues;
2796         } else {
2797                 channels->max_rx = IBMVNIC_MAX_QUEUES;
2798                 channels->max_tx = IBMVNIC_MAX_QUEUES;
2799         }
2800
2801         channels->max_other = 0;
2802         channels->max_combined = 0;
2803         channels->rx_count = adapter->req_rx_queues;
2804         channels->tx_count = adapter->req_tx_queues;
2805         channels->other_count = 0;
2806         channels->combined_count = 0;
2807 }
2808
2809 static int ibmvnic_set_channels(struct net_device *netdev,
2810                                 struct ethtool_channels *channels)
2811 {
2812         struct ibmvnic_adapter *adapter = netdev_priv(netdev);
2813         int ret;
2814
2815         ret = 0;
2816         adapter->desired.rx_queues = channels->rx_count;
2817         adapter->desired.tx_queues = channels->tx_count;
2818
2819         ret = wait_for_reset(adapter);
2820
2821         if (!ret &&
2822             (adapter->req_rx_queues != channels->rx_count ||
2823              adapter->req_tx_queues != channels->tx_count))
2824                 netdev_info(netdev,
2825                             "Could not match full channels request. Requested: RX %d, TX %d; Allowed: RX %llu, TX %llu\n",
2826                             channels->rx_count, channels->tx_count,
2827                             adapter->req_rx_queues, adapter->req_tx_queues);
2828         return ret;
2829
2830 }
2831
2832 static void ibmvnic_get_strings(struct net_device *dev, u32 stringset, u8 *data)
2833 {
2834         struct ibmvnic_adapter *adapter = netdev_priv(dev);
2835         int i;
2836
2837         switch (stringset) {
2838         case ETH_SS_STATS:
2839                 for (i = 0; i < ARRAY_SIZE(ibmvnic_stats);
2840                                 i++, data += ETH_GSTRING_LEN)
2841                         memcpy(data, ibmvnic_stats[i].name, ETH_GSTRING_LEN);
2842
2843                 for (i = 0; i < adapter->req_tx_queues; i++) {
2844                         snprintf(data, ETH_GSTRING_LEN, "tx%d_packets", i);
2845                         data += ETH_GSTRING_LEN;
2846
2847                         snprintf(data, ETH_GSTRING_LEN, "tx%d_bytes", i);
2848                         data += ETH_GSTRING_LEN;
2849
2850                         snprintf(data, ETH_GSTRING_LEN,
2851                                  "tx%d_dropped_packets", i);
2852                         data += ETH_GSTRING_LEN;
2853                 }
2854
2855                 for (i = 0; i < adapter->req_rx_queues; i++) {
2856                         snprintf(data, ETH_GSTRING_LEN, "rx%d_packets", i);
2857                         data += ETH_GSTRING_LEN;
2858
2859                         snprintf(data, ETH_GSTRING_LEN, "rx%d_bytes", i);
2860                         data += ETH_GSTRING_LEN;
2861
2862                         snprintf(data, ETH_GSTRING_LEN, "rx%d_interrupts", i);
2863                         data += ETH_GSTRING_LEN;
2864                 }
2865                 break;
2866
2867         case ETH_SS_PRIV_FLAGS:
2868                 for (i = 0; i < ARRAY_SIZE(ibmvnic_priv_flags); i++)
2869                         strcpy(data + i * ETH_GSTRING_LEN,
2870                                ibmvnic_priv_flags[i]);
2871                 break;
2872         default:
2873                 return;
2874         }
2875 }
2876
2877 static int ibmvnic_get_sset_count(struct net_device *dev, int sset)
2878 {
2879         struct ibmvnic_adapter *adapter = netdev_priv(dev);
2880
2881         switch (sset) {
2882         case ETH_SS_STATS:
2883                 return ARRAY_SIZE(ibmvnic_stats) +
2884                        adapter->req_tx_queues * NUM_TX_STATS +
2885                        adapter->req_rx_queues * NUM_RX_STATS;
2886         case ETH_SS_PRIV_FLAGS:
2887                 return ARRAY_SIZE(ibmvnic_priv_flags);
2888         default:
2889                 return -EOPNOTSUPP;
2890         }
2891 }
2892
2893 static void ibmvnic_get_ethtool_stats(struct net_device *dev,
2894                                       struct ethtool_stats *stats, u64 *data)
2895 {
2896         struct ibmvnic_adapter *adapter = netdev_priv(dev);
2897         union ibmvnic_crq crq;
2898         int i, j;
2899         int rc;
2900
2901         memset(&crq, 0, sizeof(crq));
2902         crq.request_statistics.first = IBMVNIC_CRQ_CMD;
2903         crq.request_statistics.cmd = REQUEST_STATISTICS;
2904         crq.request_statistics.ioba = cpu_to_be32(adapter->stats_token);
2905         crq.request_statistics.len =
2906             cpu_to_be32(sizeof(struct ibmvnic_statistics));
2907
2908         /* Wait for data to be written */
2909         reinit_completion(&adapter->stats_done);
2910         rc = ibmvnic_send_crq(adapter, &crq);
2911         if (rc)
2912                 return;
2913         rc = ibmvnic_wait_for_completion(adapter, &adapter->stats_done, 10000);
2914         if (rc)
2915                 return;
2916
2917         for (i = 0; i < ARRAY_SIZE(ibmvnic_stats); i++)
2918                 data[i] = be64_to_cpu(IBMVNIC_GET_STAT(adapter,
2919                                                 ibmvnic_stats[i].offset));
2920
2921         for (j = 0; j < adapter->req_tx_queues; j++) {
2922                 data[i] = adapter->tx_stats_buffers[j].packets;
2923                 i++;
2924                 data[i] = adapter->tx_stats_buffers[j].bytes;
2925                 i++;
2926                 data[i] = adapter->tx_stats_buffers[j].dropped_packets;
2927                 i++;
2928         }
2929
2930         for (j = 0; j < adapter->req_rx_queues; j++) {
2931                 data[i] = adapter->rx_stats_buffers[j].packets;
2932                 i++;
2933                 data[i] = adapter->rx_stats_buffers[j].bytes;
2934                 i++;
2935                 data[i] = adapter->rx_stats_buffers[j].interrupts;
2936                 i++;
2937         }
2938 }
2939
2940 static u32 ibmvnic_get_priv_flags(struct net_device *netdev)
2941 {
2942         struct ibmvnic_adapter *adapter = netdev_priv(netdev);
2943
2944         return adapter->priv_flags;
2945 }
2946
2947 static int ibmvnic_set_priv_flags(struct net_device *netdev, u32 flags)
2948 {
2949         struct ibmvnic_adapter *adapter = netdev_priv(netdev);
2950         bool which_maxes = !!(flags & IBMVNIC_USE_SERVER_MAXES);
2951
2952         if (which_maxes)
2953                 adapter->priv_flags |= IBMVNIC_USE_SERVER_MAXES;
2954         else
2955                 adapter->priv_flags &= ~IBMVNIC_USE_SERVER_MAXES;
2956
2957         return 0;
2958 }
2959 static const struct ethtool_ops ibmvnic_ethtool_ops = {
2960         .get_drvinfo            = ibmvnic_get_drvinfo,
2961         .get_msglevel           = ibmvnic_get_msglevel,
2962         .set_msglevel           = ibmvnic_set_msglevel,
2963         .get_link               = ibmvnic_get_link,
2964         .get_ringparam          = ibmvnic_get_ringparam,
2965         .set_ringparam          = ibmvnic_set_ringparam,
2966         .get_channels           = ibmvnic_get_channels,
2967         .set_channels           = ibmvnic_set_channels,
2968         .get_strings            = ibmvnic_get_strings,
2969         .get_sset_count         = ibmvnic_get_sset_count,
2970         .get_ethtool_stats      = ibmvnic_get_ethtool_stats,
2971         .get_link_ksettings     = ibmvnic_get_link_ksettings,
2972         .get_priv_flags         = ibmvnic_get_priv_flags,
2973         .set_priv_flags         = ibmvnic_set_priv_flags,
2974 };
2975
2976 /* Routines for managing CRQs/sCRQs  */
2977
2978 static int reset_one_sub_crq_queue(struct ibmvnic_adapter *adapter,
2979                                    struct ibmvnic_sub_crq_queue *scrq)
2980 {
2981         int rc;
2982
2983         if (!scrq) {
2984                 netdev_dbg(adapter->netdev, "Invalid scrq reset.\n");
2985                 return -EINVAL;
2986         }
2987
2988         if (scrq->irq) {
2989                 free_irq(scrq->irq, scrq);
2990                 irq_dispose_mapping(scrq->irq);
2991                 scrq->irq = 0;
2992         }
2993
2994         if (scrq->msgs) {
2995                 memset(scrq->msgs, 0, 4 * PAGE_SIZE);
2996                 atomic_set(&scrq->used, 0);
2997                 scrq->cur = 0;
2998                 scrq->ind_buf.index = 0;
2999         } else {
3000                 netdev_dbg(adapter->netdev, "Invalid scrq reset\n");
3001                 return -EINVAL;
3002         }
3003
3004         rc = h_reg_sub_crq(adapter->vdev->unit_address, scrq->msg_token,
3005                            4 * PAGE_SIZE, &scrq->crq_num, &scrq->hw_irq);
3006         return rc;
3007 }
3008
3009 static int reset_sub_crq_queues(struct ibmvnic_adapter *adapter)
3010 {
3011         int i, rc;
3012
3013         if (!adapter->tx_scrq || !adapter->rx_scrq)
3014                 return -EINVAL;
3015
3016         for (i = 0; i < adapter->req_tx_queues; i++) {
3017                 netdev_dbg(adapter->netdev, "Re-setting tx_scrq[%d]\n", i);
3018                 rc = reset_one_sub_crq_queue(adapter, adapter->tx_scrq[i]);
3019                 if (rc)
3020                         return rc;
3021         }
3022
3023         for (i = 0; i < adapter->req_rx_queues; i++) {
3024                 netdev_dbg(adapter->netdev, "Re-setting rx_scrq[%d]\n", i);
3025                 rc = reset_one_sub_crq_queue(adapter, adapter->rx_scrq[i]);
3026                 if (rc)
3027                         return rc;
3028         }
3029
3030         return rc;
3031 }
3032
3033 static void release_sub_crq_queue(struct ibmvnic_adapter *adapter,
3034                                   struct ibmvnic_sub_crq_queue *scrq,
3035                                   bool do_h_free)
3036 {
3037         struct device *dev = &adapter->vdev->dev;
3038         long rc;
3039
3040         netdev_dbg(adapter->netdev, "Releasing sub-CRQ\n");
3041
3042         if (do_h_free) {
3043                 /* Close the sub-crqs */
3044                 do {
3045                         rc = plpar_hcall_norets(H_FREE_SUB_CRQ,
3046                                                 adapter->vdev->unit_address,
3047                                                 scrq->crq_num);
3048                 } while (rc == H_BUSY || H_IS_LONG_BUSY(rc));
3049
3050                 if (rc) {
3051                         netdev_err(adapter->netdev,
3052                                    "Failed to release sub-CRQ %16lx, rc = %ld\n",
3053                                    scrq->crq_num, rc);
3054                 }
3055         }
3056
3057         dma_free_coherent(dev,
3058                           IBMVNIC_IND_ARR_SZ,
3059                           scrq->ind_buf.indir_arr,
3060                           scrq->ind_buf.indir_dma);
3061
3062         dma_unmap_single(dev, scrq->msg_token, 4 * PAGE_SIZE,
3063                          DMA_BIDIRECTIONAL);
3064         free_pages((unsigned long)scrq->msgs, 2);
3065         kfree(scrq);
3066 }
3067
3068 static struct ibmvnic_sub_crq_queue *init_sub_crq_queue(struct ibmvnic_adapter
3069                                                         *adapter)
3070 {
3071         struct device *dev = &adapter->vdev->dev;
3072         struct ibmvnic_sub_crq_queue *scrq;
3073         int rc;
3074
3075         scrq = kzalloc(sizeof(*scrq), GFP_KERNEL);
3076         if (!scrq)
3077                 return NULL;
3078
3079         scrq->msgs =
3080                 (union sub_crq *)__get_free_pages(GFP_KERNEL | __GFP_ZERO, 2);
3081         if (!scrq->msgs) {
3082                 dev_warn(dev, "Couldn't allocate crq queue messages page\n");
3083                 goto zero_page_failed;
3084         }
3085
3086         scrq->msg_token = dma_map_single(dev, scrq->msgs, 4 * PAGE_SIZE,
3087                                          DMA_BIDIRECTIONAL);
3088         if (dma_mapping_error(dev, scrq->msg_token)) {
3089                 dev_warn(dev, "Couldn't map crq queue messages page\n");
3090                 goto map_failed;
3091         }
3092
3093         rc = h_reg_sub_crq(adapter->vdev->unit_address, scrq->msg_token,
3094                            4 * PAGE_SIZE, &scrq->crq_num, &scrq->hw_irq);
3095
3096         if (rc == H_RESOURCE)
3097                 rc = ibmvnic_reset_crq(adapter);
3098
3099         if (rc == H_CLOSED) {
3100                 dev_warn(dev, "Partner adapter not ready, waiting.\n");
3101         } else if (rc) {
3102                 dev_warn(dev, "Error %d registering sub-crq\n", rc);
3103                 goto reg_failed;
3104         }
3105
3106         scrq->adapter = adapter;
3107         scrq->size = 4 * PAGE_SIZE / sizeof(*scrq->msgs);
3108         scrq->ind_buf.index = 0;
3109
3110         scrq->ind_buf.indir_arr =
3111                 dma_alloc_coherent(dev,
3112                                    IBMVNIC_IND_ARR_SZ,
3113                                    &scrq->ind_buf.indir_dma,
3114                                    GFP_KERNEL);
3115
3116         if (!scrq->ind_buf.indir_arr)
3117                 goto indir_failed;
3118
3119         spin_lock_init(&scrq->lock);
3120
3121         netdev_dbg(adapter->netdev,
3122                    "sub-crq initialized, num %lx, hw_irq=%lx, irq=%x\n",
3123                    scrq->crq_num, scrq->hw_irq, scrq->irq);
3124
3125         return scrq;
3126
3127 indir_failed:
3128         do {
3129                 rc = plpar_hcall_norets(H_FREE_SUB_CRQ,
3130                                         adapter->vdev->unit_address,
3131                                         scrq->crq_num);
3132         } while (rc == H_BUSY || rc == H_IS_LONG_BUSY(rc));
3133 reg_failed:
3134         dma_unmap_single(dev, scrq->msg_token, 4 * PAGE_SIZE,
3135                          DMA_BIDIRECTIONAL);
3136 map_failed:
3137         free_pages((unsigned long)scrq->msgs, 2);
3138 zero_page_failed:
3139         kfree(scrq);
3140
3141         return NULL;
3142 }
3143
3144 static void release_sub_crqs(struct ibmvnic_adapter *adapter, bool do_h_free)
3145 {
3146         int i;
3147
3148         if (adapter->tx_scrq) {
3149                 for (i = 0; i < adapter->num_active_tx_scrqs; i++) {
3150                         if (!adapter->tx_scrq[i])
3151                                 continue;
3152
3153                         netdev_dbg(adapter->netdev, "Releasing tx_scrq[%d]\n",
3154                                    i);
3155                         if (adapter->tx_scrq[i]->irq) {
3156                                 free_irq(adapter->tx_scrq[i]->irq,
3157                                          adapter->tx_scrq[i]);
3158                                 irq_dispose_mapping(adapter->tx_scrq[i]->irq);
3159                                 adapter->tx_scrq[i]->irq = 0;
3160                         }
3161
3162                         release_sub_crq_queue(adapter, adapter->tx_scrq[i],
3163                                               do_h_free);
3164                 }
3165
3166                 kfree(adapter->tx_scrq);
3167                 adapter->tx_scrq = NULL;
3168                 adapter->num_active_tx_scrqs = 0;
3169         }
3170
3171         if (adapter->rx_scrq) {
3172                 for (i = 0; i < adapter->num_active_rx_scrqs; i++) {
3173                         if (!adapter->rx_scrq[i])
3174                                 continue;
3175
3176                         netdev_dbg(adapter->netdev, "Releasing rx_scrq[%d]\n",
3177                                    i);
3178                         if (adapter->rx_scrq[i]->irq) {
3179                                 free_irq(adapter->rx_scrq[i]->irq,
3180                                          adapter->rx_scrq[i]);
3181                                 irq_dispose_mapping(adapter->rx_scrq[i]->irq);
3182                                 adapter->rx_scrq[i]->irq = 0;
3183                         }
3184
3185                         release_sub_crq_queue(adapter, adapter->rx_scrq[i],
3186                                               do_h_free);
3187                 }
3188
3189                 kfree(adapter->rx_scrq);
3190                 adapter->rx_scrq = NULL;
3191                 adapter->num_active_rx_scrqs = 0;
3192         }
3193 }
3194
3195 static int disable_scrq_irq(struct ibmvnic_adapter *adapter,
3196                             struct ibmvnic_sub_crq_queue *scrq)
3197 {
3198         struct device *dev = &adapter->vdev->dev;
3199         unsigned long rc;
3200
3201         rc = plpar_hcall_norets(H_VIOCTL, adapter->vdev->unit_address,
3202                                 H_DISABLE_VIO_INTERRUPT, scrq->hw_irq, 0, 0);
3203         if (rc)
3204                 dev_err(dev, "Couldn't disable scrq irq 0x%lx. rc=%ld\n",
3205                         scrq->hw_irq, rc);
3206         return rc;
3207 }
3208
3209 static int enable_scrq_irq(struct ibmvnic_adapter *adapter,
3210                            struct ibmvnic_sub_crq_queue *scrq)
3211 {
3212         struct device *dev = &adapter->vdev->dev;
3213         unsigned long rc;
3214
3215         if (scrq->hw_irq > 0x100000000ULL) {
3216                 dev_err(dev, "bad hw_irq = %lx\n", scrq->hw_irq);
3217                 return 1;
3218         }
3219
3220         if (test_bit(0, &adapter->resetting) &&
3221             adapter->reset_reason == VNIC_RESET_MOBILITY) {
3222                 u64 val = (0xff000000) | scrq->hw_irq;
3223
3224                 rc = plpar_hcall_norets(H_EOI, val);
3225                 /* H_EOI would fail with rc = H_FUNCTION when running
3226                  * in XIVE mode which is expected, but not an error.
3227                  */
3228                 if (rc && (rc != H_FUNCTION))
3229                         dev_err(dev, "H_EOI FAILED irq 0x%llx. rc=%ld\n",
3230                                 val, rc);
3231         }
3232
3233         rc = plpar_hcall_norets(H_VIOCTL, adapter->vdev->unit_address,
3234                                 H_ENABLE_VIO_INTERRUPT, scrq->hw_irq, 0, 0);
3235         if (rc)
3236                 dev_err(dev, "Couldn't enable scrq irq 0x%lx. rc=%ld\n",
3237                         scrq->hw_irq, rc);
3238         return rc;
3239 }
3240
3241 static int ibmvnic_complete_tx(struct ibmvnic_adapter *adapter,
3242                                struct ibmvnic_sub_crq_queue *scrq)
3243 {
3244         struct device *dev = &adapter->vdev->dev;
3245         struct ibmvnic_tx_pool *tx_pool;
3246         struct ibmvnic_tx_buff *txbuff;
3247         struct netdev_queue *txq;
3248         union sub_crq *next;
3249         int index;
3250         int i;
3251
3252 restart_loop:
3253         while (pending_scrq(adapter, scrq)) {
3254                 unsigned int pool = scrq->pool_index;
3255                 int num_entries = 0;
3256                 int total_bytes = 0;
3257                 int num_packets = 0;
3258
3259                 /* The queue entry at the current index is peeked at above
3260                  * to determine that there is a valid descriptor awaiting
3261                  * processing. We want to be sure that the current slot
3262                  * holds a valid descriptor before reading its contents.
3263                  */
3264                 dma_rmb();
3265
3266                 next = ibmvnic_next_scrq(adapter, scrq);
3267                 for (i = 0; i < next->tx_comp.num_comps; i++) {
3268                         if (next->tx_comp.rcs[i])
3269                                 dev_err(dev, "tx error %x\n",
3270                                         next->tx_comp.rcs[i]);
3271                         index = be32_to_cpu(next->tx_comp.correlators[i]);
3272                         if (index & IBMVNIC_TSO_POOL_MASK) {
3273                                 tx_pool = &adapter->tso_pool[pool];
3274                                 index &= ~IBMVNIC_TSO_POOL_MASK;
3275                         } else {
3276                                 tx_pool = &adapter->tx_pool[pool];
3277                         }
3278
3279                         txbuff = &tx_pool->tx_buff[index];
3280                         num_packets++;
3281                         num_entries += txbuff->num_entries;
3282                         if (txbuff->skb) {
3283                                 total_bytes += txbuff->skb->len;
3284                                 dev_consume_skb_irq(txbuff->skb);
3285                                 txbuff->skb = NULL;
3286                         } else {
3287                                 netdev_warn(adapter->netdev,
3288                                             "TX completion received with NULL socket buffer\n");
3289                         }
3290                         tx_pool->free_map[tx_pool->producer_index] = index;
3291                         tx_pool->producer_index =
3292                                 (tx_pool->producer_index + 1) %
3293                                         tx_pool->num_buffers;
3294                 }
3295                 /* remove tx_comp scrq*/
3296                 next->tx_comp.first = 0;
3297
3298                 txq = netdev_get_tx_queue(adapter->netdev, scrq->pool_index);
3299                 netdev_tx_completed_queue(txq, num_packets, total_bytes);
3300
3301                 if (atomic_sub_return(num_entries, &scrq->used) <=
3302                     (adapter->req_tx_entries_per_subcrq / 2) &&
3303                     __netif_subqueue_stopped(adapter->netdev,
3304                                              scrq->pool_index)) {
3305                         netif_wake_subqueue(adapter->netdev, scrq->pool_index);
3306                         netdev_dbg(adapter->netdev, "Started queue %d\n",
3307                                    scrq->pool_index);
3308                 }
3309         }
3310
3311         enable_scrq_irq(adapter, scrq);
3312
3313         if (pending_scrq(adapter, scrq)) {
3314                 disable_scrq_irq(adapter, scrq);
3315                 goto restart_loop;
3316         }
3317
3318         return 0;
3319 }
3320
3321 static irqreturn_t ibmvnic_interrupt_tx(int irq, void *instance)
3322 {
3323         struct ibmvnic_sub_crq_queue *scrq = instance;
3324         struct ibmvnic_adapter *adapter = scrq->adapter;
3325
3326         disable_scrq_irq(adapter, scrq);
3327         ibmvnic_complete_tx(adapter, scrq);
3328
3329         return IRQ_HANDLED;
3330 }
3331
3332 static irqreturn_t ibmvnic_interrupt_rx(int irq, void *instance)
3333 {
3334         struct ibmvnic_sub_crq_queue *scrq = instance;
3335         struct ibmvnic_adapter *adapter = scrq->adapter;
3336
3337         /* When booting a kdump kernel we can hit pending interrupts
3338          * prior to completing driver initialization.
3339          */
3340         if (unlikely(adapter->state != VNIC_OPEN))
3341                 return IRQ_NONE;
3342
3343         adapter->rx_stats_buffers[scrq->scrq_num].interrupts++;
3344
3345         if (napi_schedule_prep(&adapter->napi[scrq->scrq_num])) {
3346                 disable_scrq_irq(adapter, scrq);
3347                 __napi_schedule(&adapter->napi[scrq->scrq_num]);
3348         }
3349
3350         return IRQ_HANDLED;
3351 }
3352
3353 static int init_sub_crq_irqs(struct ibmvnic_adapter *adapter)
3354 {
3355         struct device *dev = &adapter->vdev->dev;
3356         struct ibmvnic_sub_crq_queue *scrq;
3357         int i = 0, j = 0;
3358         int rc = 0;
3359
3360         for (i = 0; i < adapter->req_tx_queues; i++) {
3361                 netdev_dbg(adapter->netdev, "Initializing tx_scrq[%d] irq\n",
3362                            i);
3363                 scrq = adapter->tx_scrq[i];
3364                 scrq->irq = irq_create_mapping(NULL, scrq->hw_irq);
3365
3366                 if (!scrq->irq) {
3367                         rc = -EINVAL;
3368                         dev_err(dev, "Error mapping irq\n");
3369                         goto req_tx_irq_failed;
3370                 }
3371
3372                 snprintf(scrq->name, sizeof(scrq->name), "ibmvnic-%x-tx%d",
3373                          adapter->vdev->unit_address, i);
3374                 rc = request_irq(scrq->irq, ibmvnic_interrupt_tx,
3375                                  0, scrq->name, scrq);
3376
3377                 if (rc) {
3378                         dev_err(dev, "Couldn't register tx irq 0x%x. rc=%d\n",
3379                                 scrq->irq, rc);
3380                         irq_dispose_mapping(scrq->irq);
3381                         goto req_tx_irq_failed;
3382                 }
3383         }
3384
3385         for (i = 0; i < adapter->req_rx_queues; i++) {
3386                 netdev_dbg(adapter->netdev, "Initializing rx_scrq[%d] irq\n",
3387                            i);
3388                 scrq = adapter->rx_scrq[i];
3389                 scrq->irq = irq_create_mapping(NULL, scrq->hw_irq);
3390                 if (!scrq->irq) {
3391                         rc = -EINVAL;
3392                         dev_err(dev, "Error mapping irq\n");
3393                         goto req_rx_irq_failed;
3394                 }
3395                 snprintf(scrq->name, sizeof(scrq->name), "ibmvnic-%x-rx%d",
3396                          adapter->vdev->unit_address, i);
3397                 rc = request_irq(scrq->irq, ibmvnic_interrupt_rx,
3398                                  0, scrq->name, scrq);
3399                 if (rc) {
3400                         dev_err(dev, "Couldn't register rx irq 0x%x. rc=%d\n",
3401                                 scrq->irq, rc);
3402                         irq_dispose_mapping(scrq->irq);
3403                         goto req_rx_irq_failed;
3404                 }
3405         }
3406         return rc;
3407
3408 req_rx_irq_failed:
3409         for (j = 0; j < i; j++) {
3410                 free_irq(adapter->rx_scrq[j]->irq, adapter->rx_scrq[j]);
3411                 irq_dispose_mapping(adapter->rx_scrq[j]->irq);
3412         }
3413         i = adapter->req_tx_queues;
3414 req_tx_irq_failed:
3415         for (j = 0; j < i; j++) {
3416                 free_irq(adapter->tx_scrq[j]->irq, adapter->tx_scrq[j]);
3417                 irq_dispose_mapping(adapter->tx_scrq[j]->irq);
3418         }
3419         release_sub_crqs(adapter, 1);
3420         return rc;
3421 }
3422
3423 static int init_sub_crqs(struct ibmvnic_adapter *adapter)
3424 {
3425         struct device *dev = &adapter->vdev->dev;
3426         struct ibmvnic_sub_crq_queue **allqueues;
3427         int registered_queues = 0;
3428         int total_queues;
3429         int more = 0;
3430         int i;
3431
3432         total_queues = adapter->req_tx_queues + adapter->req_rx_queues;
3433
3434         allqueues = kcalloc(total_queues, sizeof(*allqueues), GFP_KERNEL);
3435         if (!allqueues)
3436                 return -1;
3437
3438         for (i = 0; i < total_queues; i++) {
3439                 allqueues[i] = init_sub_crq_queue(adapter);
3440                 if (!allqueues[i]) {
3441                         dev_warn(dev, "Couldn't allocate all sub-crqs\n");
3442                         break;
3443                 }
3444                 registered_queues++;
3445         }
3446
3447         /* Make sure we were able to register the minimum number of queues */
3448         if (registered_queues <
3449             adapter->min_tx_queues + adapter->min_rx_queues) {
3450                 dev_err(dev, "Fatal: Couldn't init  min number of sub-crqs\n");
3451                 goto tx_failed;
3452         }
3453
3454         /* Distribute the failed allocated queues*/
3455         for (i = 0; i < total_queues - registered_queues + more ; i++) {
3456                 netdev_dbg(adapter->netdev, "Reducing number of queues\n");
3457                 switch (i % 3) {
3458                 case 0:
3459                         if (adapter->req_rx_queues > adapter->min_rx_queues)
3460                                 adapter->req_rx_queues--;
3461                         else
3462                                 more++;
3463                         break;
3464                 case 1:
3465                         if (adapter->req_tx_queues > adapter->min_tx_queues)
3466                                 adapter->req_tx_queues--;
3467                         else
3468                                 more++;
3469                         break;
3470                 }
3471         }
3472
3473         adapter->tx_scrq = kcalloc(adapter->req_tx_queues,
3474                                    sizeof(*adapter->tx_scrq), GFP_KERNEL);
3475         if (!adapter->tx_scrq)
3476                 goto tx_failed;
3477
3478         for (i = 0; i < adapter->req_tx_queues; i++) {
3479                 adapter->tx_scrq[i] = allqueues[i];
3480                 adapter->tx_scrq[i]->pool_index = i;
3481                 adapter->num_active_tx_scrqs++;
3482         }
3483
3484         adapter->rx_scrq = kcalloc(adapter->req_rx_queues,
3485                                    sizeof(*adapter->rx_scrq), GFP_KERNEL);
3486         if (!adapter->rx_scrq)
3487                 goto rx_failed;
3488
3489         for (i = 0; i < adapter->req_rx_queues; i++) {
3490                 adapter->rx_scrq[i] = allqueues[i + adapter->req_tx_queues];
3491                 adapter->rx_scrq[i]->scrq_num = i;
3492                 adapter->num_active_rx_scrqs++;
3493         }
3494
3495         kfree(allqueues);
3496         return 0;
3497
3498 rx_failed:
3499         kfree(adapter->tx_scrq);
3500         adapter->tx_scrq = NULL;
3501 tx_failed:
3502         for (i = 0; i < registered_queues; i++)
3503                 release_sub_crq_queue(adapter, allqueues[i], 1);
3504         kfree(allqueues);
3505         return -1;
3506 }
3507
3508 static void send_request_cap(struct ibmvnic_adapter *adapter, int retry)
3509 {
3510         struct device *dev = &adapter->vdev->dev;
3511         union ibmvnic_crq crq;
3512         int max_entries;
3513
3514         if (!retry) {
3515                 /* Sub-CRQ entries are 32 byte long */
3516                 int entries_page = 4 * PAGE_SIZE / (sizeof(u64) * 4);
3517
3518                 if (adapter->min_tx_entries_per_subcrq > entries_page ||
3519                     adapter->min_rx_add_entries_per_subcrq > entries_page) {
3520                         dev_err(dev, "Fatal, invalid entries per sub-crq\n");
3521                         return;
3522                 }
3523
3524                 if (adapter->desired.mtu)
3525                         adapter->req_mtu = adapter->desired.mtu;
3526                 else
3527                         adapter->req_mtu = adapter->netdev->mtu + ETH_HLEN;
3528
3529                 if (!adapter->desired.tx_entries)
3530                         adapter->desired.tx_entries =
3531                                         adapter->max_tx_entries_per_subcrq;
3532                 if (!adapter->desired.rx_entries)
3533                         adapter->desired.rx_entries =
3534                                         adapter->max_rx_add_entries_per_subcrq;
3535
3536                 max_entries = IBMVNIC_MAX_LTB_SIZE /
3537                               (adapter->req_mtu + IBMVNIC_BUFFER_HLEN);
3538
3539                 if ((adapter->req_mtu + IBMVNIC_BUFFER_HLEN) *
3540                         adapter->desired.tx_entries > IBMVNIC_MAX_LTB_SIZE) {
3541                         adapter->desired.tx_entries = max_entries;
3542                 }
3543
3544                 if ((adapter->req_mtu + IBMVNIC_BUFFER_HLEN) *
3545                         adapter->desired.rx_entries > IBMVNIC_MAX_LTB_SIZE) {
3546                         adapter->desired.rx_entries = max_entries;
3547                 }
3548
3549                 if (adapter->desired.tx_entries)
3550                         adapter->req_tx_entries_per_subcrq =
3551                                         adapter->desired.tx_entries;
3552                 else
3553                         adapter->req_tx_entries_per_subcrq =
3554                                         adapter->max_tx_entries_per_subcrq;
3555
3556                 if (adapter->desired.rx_entries)
3557                         adapter->req_rx_add_entries_per_subcrq =
3558                                         adapter->desired.rx_entries;
3559                 else
3560                         adapter->req_rx_add_entries_per_subcrq =
3561                                         adapter->max_rx_add_entries_per_subcrq;
3562
3563                 if (adapter->desired.tx_queues)
3564                         adapter->req_tx_queues =
3565                                         adapter->desired.tx_queues;
3566                 else
3567                         adapter->req_tx_queues =
3568                                         adapter->opt_tx_comp_sub_queues;
3569
3570                 if (adapter->desired.rx_queues)
3571                         adapter->req_rx_queues =
3572                                         adapter->desired.rx_queues;
3573                 else
3574                         adapter->req_rx_queues =
3575                                         adapter->opt_rx_comp_queues;
3576
3577                 adapter->req_rx_add_queues = adapter->max_rx_add_queues;
3578         }
3579
3580         memset(&crq, 0, sizeof(crq));
3581         crq.request_capability.first = IBMVNIC_CRQ_CMD;
3582         crq.request_capability.cmd = REQUEST_CAPABILITY;
3583
3584         crq.request_capability.capability = cpu_to_be16(REQ_TX_QUEUES);
3585         crq.request_capability.number = cpu_to_be64(adapter->req_tx_queues);
3586         atomic_inc(&adapter->running_cap_crqs);
3587         ibmvnic_send_crq(adapter, &crq);
3588
3589         crq.request_capability.capability = cpu_to_be16(REQ_RX_QUEUES);
3590         crq.request_capability.number = cpu_to_be64(adapter->req_rx_queues);
3591         atomic_inc(&adapter->running_cap_crqs);
3592         ibmvnic_send_crq(adapter, &crq);
3593
3594         crq.request_capability.capability = cpu_to_be16(REQ_RX_ADD_QUEUES);
3595         crq.request_capability.number = cpu_to_be64(adapter->req_rx_add_queues);
3596         atomic_inc(&adapter->running_cap_crqs);
3597         ibmvnic_send_crq(adapter, &crq);
3598
3599         crq.request_capability.capability =
3600             cpu_to_be16(REQ_TX_ENTRIES_PER_SUBCRQ);
3601         crq.request_capability.number =
3602             cpu_to_be64(adapter->req_tx_entries_per_subcrq);
3603         atomic_inc(&adapter->running_cap_crqs);
3604         ibmvnic_send_crq(adapter, &crq);
3605
3606         crq.request_capability.capability =
3607             cpu_to_be16(REQ_RX_ADD_ENTRIES_PER_SUBCRQ);
3608         crq.request_capability.number =
3609             cpu_to_be64(adapter->req_rx_add_entries_per_subcrq);
3610         atomic_inc(&adapter->running_cap_crqs);
3611         ibmvnic_send_crq(adapter, &crq);
3612
3613         crq.request_capability.capability = cpu_to_be16(REQ_MTU);
3614         crq.request_capability.number = cpu_to_be64(adapter->req_mtu);
3615         atomic_inc(&adapter->running_cap_crqs);
3616         ibmvnic_send_crq(adapter, &crq);
3617
3618         if (adapter->netdev->flags & IFF_PROMISC) {
3619                 if (adapter->promisc_supported) {
3620                         crq.request_capability.capability =
3621                             cpu_to_be16(PROMISC_REQUESTED);
3622                         crq.request_capability.number = cpu_to_be64(1);
3623                         atomic_inc(&adapter->running_cap_crqs);
3624                         ibmvnic_send_crq(adapter, &crq);
3625                 }
3626         } else {
3627                 crq.request_capability.capability =
3628                     cpu_to_be16(PROMISC_REQUESTED);
3629                 crq.request_capability.number = cpu_to_be64(0);
3630                 atomic_inc(&adapter->running_cap_crqs);
3631                 ibmvnic_send_crq(adapter, &crq);
3632         }
3633 }
3634
3635 static int pending_scrq(struct ibmvnic_adapter *adapter,
3636                         struct ibmvnic_sub_crq_queue *scrq)
3637 {
3638         union sub_crq *entry = &scrq->msgs[scrq->cur];
3639
3640         if (entry->generic.first & IBMVNIC_CRQ_CMD_RSP)
3641                 return 1;
3642         else
3643                 return 0;
3644 }
3645
3646 static union sub_crq *ibmvnic_next_scrq(struct ibmvnic_adapter *adapter,
3647                                         struct ibmvnic_sub_crq_queue *scrq)
3648 {
3649         union sub_crq *entry;
3650         unsigned long flags;
3651
3652         spin_lock_irqsave(&scrq->lock, flags);
3653         entry = &scrq->msgs[scrq->cur];
3654         if (entry->generic.first & IBMVNIC_CRQ_CMD_RSP) {
3655                 if (++scrq->cur == scrq->size)
3656                         scrq->cur = 0;
3657         } else {
3658                 entry = NULL;
3659         }
3660         spin_unlock_irqrestore(&scrq->lock, flags);
3661
3662         /* Ensure that the entire buffer descriptor has been
3663          * loaded before reading its contents
3664          */
3665         dma_rmb();
3666
3667         return entry;
3668 }
3669
3670 static union ibmvnic_crq *ibmvnic_next_crq(struct ibmvnic_adapter *adapter)
3671 {
3672         struct ibmvnic_crq_queue *queue = &adapter->crq;
3673         union ibmvnic_crq *crq;
3674
3675         crq = &queue->msgs[queue->cur];
3676         if (crq->generic.first & IBMVNIC_CRQ_CMD_RSP) {
3677                 if (++queue->cur == queue->size)
3678                         queue->cur = 0;
3679         } else {
3680                 crq = NULL;
3681         }
3682
3683         return crq;
3684 }
3685
3686 static void print_subcrq_error(struct device *dev, int rc, const char *func)
3687 {
3688         switch (rc) {
3689         case H_PARAMETER:
3690                 dev_warn_ratelimited(dev,
3691                                      "%s failed: Send request is malformed or adapter failover pending. (rc=%d)\n",
3692                                      func, rc);
3693                 break;
3694         case H_CLOSED:
3695                 dev_warn_ratelimited(dev,
3696                                      "%s failed: Backing queue closed. Adapter is down or failover pending. (rc=%d)\n",
3697                                      func, rc);
3698                 break;
3699         default:
3700                 dev_err_ratelimited(dev, "%s failed: (rc=%d)\n", func, rc);
3701                 break;
3702         }
3703 }
3704
3705 static int send_subcrq_indirect(struct ibmvnic_adapter *adapter,
3706                                 u64 remote_handle, u64 ioba, u64 num_entries)
3707 {
3708         unsigned int ua = adapter->vdev->unit_address;
3709         struct device *dev = &adapter->vdev->dev;
3710         int rc;
3711
3712         /* Make sure the hypervisor sees the complete request */
3713         mb();
3714         rc = plpar_hcall_norets(H_SEND_SUB_CRQ_INDIRECT, ua,
3715                                 cpu_to_be64(remote_handle),
3716                                 ioba, num_entries);
3717
3718         if (rc)
3719                 print_subcrq_error(dev, rc, __func__);
3720
3721         return rc;
3722 }
3723
3724 static int ibmvnic_send_crq(struct ibmvnic_adapter *adapter,
3725                             union ibmvnic_crq *crq)
3726 {
3727         unsigned int ua = adapter->vdev->unit_address;
3728         struct device *dev = &adapter->vdev->dev;
3729         u64 *u64_crq = (u64 *)crq;
3730         int rc;
3731
3732         netdev_dbg(adapter->netdev, "Sending CRQ: %016lx %016lx\n",
3733                    (unsigned long int)cpu_to_be64(u64_crq[0]),
3734                    (unsigned long int)cpu_to_be64(u64_crq[1]));
3735
3736         if (!adapter->crq.active &&
3737             crq->generic.first != IBMVNIC_CRQ_INIT_CMD) {
3738                 dev_warn(dev, "Invalid request detected while CRQ is inactive, possible device state change during reset\n");
3739                 return -EINVAL;
3740         }
3741
3742         /* Make sure the hypervisor sees the complete request */
3743         mb();
3744
3745         rc = plpar_hcall_norets(H_SEND_CRQ, ua,
3746                                 cpu_to_be64(u64_crq[0]),
3747                                 cpu_to_be64(u64_crq[1]));
3748
3749         if (rc) {
3750                 if (rc == H_CLOSED) {
3751                         dev_warn(dev, "CRQ Queue closed\n");
3752                         /* do not reset, report the fail, wait for passive init from server */
3753                 }
3754
3755                 dev_warn(dev, "Send error (rc=%d)\n", rc);
3756         }
3757
3758         return rc;
3759 }
3760
3761 static int ibmvnic_send_crq_init(struct ibmvnic_adapter *adapter)
3762 {
3763         struct device *dev = &adapter->vdev->dev;
3764         union ibmvnic_crq crq;
3765         int retries = 100;
3766         int rc;
3767
3768         memset(&crq, 0, sizeof(crq));
3769         crq.generic.first = IBMVNIC_CRQ_INIT_CMD;
3770         crq.generic.cmd = IBMVNIC_CRQ_INIT;
3771         netdev_dbg(adapter->netdev, "Sending CRQ init\n");
3772
3773         do {
3774                 rc = ibmvnic_send_crq(adapter, &crq);
3775                 if (rc != H_CLOSED)
3776                         break;
3777                 retries--;
3778                 msleep(50);
3779
3780         } while (retries > 0);
3781
3782         if (rc) {
3783                 dev_err(dev, "Failed to send init request, rc = %d\n", rc);
3784                 return rc;
3785         }
3786
3787         return 0;
3788 }
3789
3790 static int send_version_xchg(struct ibmvnic_adapter *adapter)
3791 {
3792         union ibmvnic_crq crq;
3793
3794         memset(&crq, 0, sizeof(crq));
3795         crq.version_exchange.first = IBMVNIC_CRQ_CMD;
3796         crq.version_exchange.cmd = VERSION_EXCHANGE;
3797         crq.version_exchange.version = cpu_to_be16(ibmvnic_version);
3798
3799         return ibmvnic_send_crq(adapter, &crq);
3800 }
3801
3802 struct vnic_login_client_data {
3803         u8      type;
3804         __be16  len;
3805         char    name[];
3806 } __packed;
3807
3808 static int vnic_client_data_len(struct ibmvnic_adapter *adapter)
3809 {
3810         int len;
3811
3812         /* Calculate the amount of buffer space needed for the
3813          * vnic client data in the login buffer. There are four entries,
3814          * OS name, LPAR name, device name, and a null last entry.
3815          */
3816         len = 4 * sizeof(struct vnic_login_client_data);
3817         len += 6; /* "Linux" plus NULL */
3818         len += strlen(utsname()->nodename) + 1;
3819         len += strlen(adapter->netdev->name) + 1;
3820
3821         return len;
3822 }
3823
3824 static void vnic_add_client_data(struct ibmvnic_adapter *adapter,
3825                                  struct vnic_login_client_data *vlcd)
3826 {
3827         const char *os_name = "Linux";
3828         int len;
3829
3830         /* Type 1 - LPAR OS */
3831         vlcd->type = 1;
3832         len = strlen(os_name) + 1;
3833         vlcd->len = cpu_to_be16(len);
3834         strncpy(vlcd->name, os_name, len);
3835         vlcd = (struct vnic_login_client_data *)(vlcd->name + len);
3836
3837         /* Type 2 - LPAR name */
3838         vlcd->type = 2;
3839         len = strlen(utsname()->nodename) + 1;
3840         vlcd->len = cpu_to_be16(len);
3841         strncpy(vlcd->name, utsname()->nodename, len);
3842         vlcd = (struct vnic_login_client_data *)(vlcd->name + len);
3843
3844         /* Type 3 - device name */
3845         vlcd->type = 3;
3846         len = strlen(adapter->netdev->name) + 1;
3847         vlcd->len = cpu_to_be16(len);
3848         strncpy(vlcd->name, adapter->netdev->name, len);
3849 }
3850
3851 static int send_login(struct ibmvnic_adapter *adapter)
3852 {
3853         struct ibmvnic_login_rsp_buffer *login_rsp_buffer;
3854         struct ibmvnic_login_buffer *login_buffer;
3855         struct device *dev = &adapter->vdev->dev;
3856         struct vnic_login_client_data *vlcd;
3857         dma_addr_t rsp_buffer_token;
3858         dma_addr_t buffer_token;
3859         size_t rsp_buffer_size;
3860         union ibmvnic_crq crq;
3861         int client_data_len;
3862         size_t buffer_size;
3863         __be64 *tx_list_p;
3864         __be64 *rx_list_p;
3865         int rc;
3866         int i;
3867
3868         if (!adapter->tx_scrq || !adapter->rx_scrq) {
3869                 netdev_err(adapter->netdev,
3870                            "RX or TX queues are not allocated, device login failed\n");
3871                 return -1;
3872         }
3873
3874         release_login_buffer(adapter);
3875         release_login_rsp_buffer(adapter);
3876
3877         client_data_len = vnic_client_data_len(adapter);
3878
3879         buffer_size =
3880             sizeof(struct ibmvnic_login_buffer) +
3881             sizeof(u64) * (adapter->req_tx_queues + adapter->req_rx_queues) +
3882             client_data_len;
3883
3884         login_buffer = kzalloc(buffer_size, GFP_ATOMIC);
3885         if (!login_buffer)
3886                 goto buf_alloc_failed;
3887
3888         buffer_token = dma_map_single(dev, login_buffer, buffer_size,
3889                                       DMA_TO_DEVICE);
3890         if (dma_mapping_error(dev, buffer_token)) {
3891                 dev_err(dev, "Couldn't map login buffer\n");
3892                 goto buf_map_failed;
3893         }
3894
3895         rsp_buffer_size = sizeof(struct ibmvnic_login_rsp_buffer) +
3896                           sizeof(u64) * adapter->req_tx_queues +
3897                           sizeof(u64) * adapter->req_rx_queues +
3898                           sizeof(u64) * adapter->req_rx_queues +
3899                           sizeof(u8) * IBMVNIC_TX_DESC_VERSIONS;
3900
3901         login_rsp_buffer = kmalloc(rsp_buffer_size, GFP_ATOMIC);
3902         if (!login_rsp_buffer)
3903                 goto buf_rsp_alloc_failed;
3904
3905         rsp_buffer_token = dma_map_single(dev, login_rsp_buffer,
3906                                           rsp_buffer_size, DMA_FROM_DEVICE);
3907         if (dma_mapping_error(dev, rsp_buffer_token)) {
3908                 dev_err(dev, "Couldn't map login rsp buffer\n");
3909                 goto buf_rsp_map_failed;
3910         }
3911
3912         adapter->login_buf = login_buffer;
3913         adapter->login_buf_token = buffer_token;
3914         adapter->login_buf_sz = buffer_size;
3915         adapter->login_rsp_buf = login_rsp_buffer;
3916         adapter->login_rsp_buf_token = rsp_buffer_token;
3917         adapter->login_rsp_buf_sz = rsp_buffer_size;
3918
3919         login_buffer->len = cpu_to_be32(buffer_size);
3920         login_buffer->version = cpu_to_be32(INITIAL_VERSION_LB);
3921         login_buffer->num_txcomp_subcrqs = cpu_to_be32(adapter->req_tx_queues);
3922         login_buffer->off_txcomp_subcrqs =
3923             cpu_to_be32(sizeof(struct ibmvnic_login_buffer));
3924         login_buffer->num_rxcomp_subcrqs = cpu_to_be32(adapter->req_rx_queues);
3925         login_buffer->off_rxcomp_subcrqs =
3926             cpu_to_be32(sizeof(struct ibmvnic_login_buffer) +
3927                         sizeof(u64) * adapter->req_tx_queues);
3928         login_buffer->login_rsp_ioba = cpu_to_be32(rsp_buffer_token);
3929         login_buffer->login_rsp_len = cpu_to_be32(rsp_buffer_size);
3930
3931         tx_list_p = (__be64 *)((char *)login_buffer +
3932                                       sizeof(struct ibmvnic_login_buffer));
3933         rx_list_p = (__be64 *)((char *)login_buffer +
3934                                       sizeof(struct ibmvnic_login_buffer) +
3935                                       sizeof(u64) * adapter->req_tx_queues);
3936
3937         for (i = 0; i < adapter->req_tx_queues; i++) {
3938                 if (adapter->tx_scrq[i]) {
3939                         tx_list_p[i] = cpu_to_be64(adapter->tx_scrq[i]->
3940                                                    crq_num);
3941                 }
3942         }
3943
3944         for (i = 0; i < adapter->req_rx_queues; i++) {
3945                 if (adapter->rx_scrq[i]) {
3946                         rx_list_p[i] = cpu_to_be64(adapter->rx_scrq[i]->
3947                                                    crq_num);
3948                 }
3949         }
3950
3951         /* Insert vNIC login client data */
3952         vlcd = (struct vnic_login_client_data *)
3953                 ((char *)rx_list_p + (sizeof(u64) * adapter->req_rx_queues));
3954         login_buffer->client_data_offset =
3955                         cpu_to_be32((char *)vlcd - (char *)login_buffer);
3956         login_buffer->client_data_len = cpu_to_be32(client_data_len);
3957
3958         vnic_add_client_data(adapter, vlcd);
3959
3960         netdev_dbg(adapter->netdev, "Login Buffer:\n");
3961         for (i = 0; i < (adapter->login_buf_sz - 1) / 8 + 1; i++) {
3962                 netdev_dbg(adapter->netdev, "%016lx\n",
3963                            ((unsigned long int *)(adapter->login_buf))[i]);
3964         }
3965
3966         memset(&crq, 0, sizeof(crq));
3967         crq.login.first = IBMVNIC_CRQ_CMD;
3968         crq.login.cmd = LOGIN;
3969         crq.login.ioba = cpu_to_be32(buffer_token);
3970         crq.login.len = cpu_to_be32(buffer_size);
3971
3972         adapter->login_pending = true;
3973         rc = ibmvnic_send_crq(adapter, &crq);
3974         if (rc) {
3975                 adapter->login_pending = false;
3976                 netdev_err(adapter->netdev, "Failed to send login, rc=%d\n", rc);
3977                 goto buf_rsp_map_failed;
3978         }
3979
3980         return 0;
3981
3982 buf_rsp_map_failed:
3983         kfree(login_rsp_buffer);
3984         adapter->login_rsp_buf = NULL;
3985 buf_rsp_alloc_failed:
3986         dma_unmap_single(dev, buffer_token, buffer_size, DMA_TO_DEVICE);
3987 buf_map_failed:
3988         kfree(login_buffer);
3989         adapter->login_buf = NULL;
3990 buf_alloc_failed:
3991         return -1;
3992 }
3993
3994 static int send_request_map(struct ibmvnic_adapter *adapter, dma_addr_t addr,
3995                             u32 len, u8 map_id)
3996 {
3997         union ibmvnic_crq crq;
3998
3999         memset(&crq, 0, sizeof(crq));
4000         crq.request_map.first = IBMVNIC_CRQ_CMD;
4001         crq.request_map.cmd = REQUEST_MAP;
4002         crq.request_map.map_id = map_id;
4003         crq.request_map.ioba = cpu_to_be32(addr);
4004         crq.request_map.len = cpu_to_be32(len);
4005         return ibmvnic_send_crq(adapter, &crq);
4006 }
4007
4008 static int send_request_unmap(struct ibmvnic_adapter *adapter, u8 map_id)
4009 {
4010         union ibmvnic_crq crq;
4011
4012         memset(&crq, 0, sizeof(crq));
4013         crq.request_unmap.first = IBMVNIC_CRQ_CMD;
4014         crq.request_unmap.cmd = REQUEST_UNMAP;
4015         crq.request_unmap.map_id = map_id;
4016         return ibmvnic_send_crq(adapter, &crq);
4017 }
4018
4019 static void send_query_map(struct ibmvnic_adapter *adapter)
4020 {
4021         union ibmvnic_crq crq;
4022
4023         memset(&crq, 0, sizeof(crq));
4024         crq.query_map.first = IBMVNIC_CRQ_CMD;
4025         crq.query_map.cmd = QUERY_MAP;
4026         ibmvnic_send_crq(adapter, &crq);
4027 }
4028
4029 /* Send a series of CRQs requesting various capabilities of the VNIC server */
4030 static void send_query_cap(struct ibmvnic_adapter *adapter)
4031 {
4032         union ibmvnic_crq crq;
4033
4034         atomic_set(&adapter->running_cap_crqs, 0);
4035         memset(&crq, 0, sizeof(crq));
4036         crq.query_capability.first = IBMVNIC_CRQ_CMD;
4037         crq.query_capability.cmd = QUERY_CAPABILITY;
4038
4039         crq.query_capability.capability = cpu_to_be16(MIN_TX_QUEUES);
4040         atomic_inc(&adapter->running_cap_crqs);
4041         ibmvnic_send_crq(adapter, &crq);
4042
4043         crq.query_capability.capability = cpu_to_be16(MIN_RX_QUEUES);
4044         atomic_inc(&adapter->running_cap_crqs);
4045         ibmvnic_send_crq(adapter, &crq);
4046
4047         crq.query_capability.capability = cpu_to_be16(MIN_RX_ADD_QUEUES);
4048         atomic_inc(&adapter->running_cap_crqs);
4049         ibmvnic_send_crq(adapter, &crq);
4050
4051         crq.query_capability.capability = cpu_to_be16(MAX_TX_QUEUES);
4052         atomic_inc(&adapter->running_cap_crqs);
4053         ibmvnic_send_crq(adapter, &crq);
4054
4055         crq.query_capability.capability = cpu_to_be16(MAX_RX_QUEUES);
4056         atomic_inc(&adapter->running_cap_crqs);
4057         ibmvnic_send_crq(adapter, &crq);
4058
4059         crq.query_capability.capability = cpu_to_be16(MAX_RX_ADD_QUEUES);
4060         atomic_inc(&adapter->running_cap_crqs);
4061         ibmvnic_send_crq(adapter, &crq);
4062
4063         crq.query_capability.capability =
4064             cpu_to_be16(MIN_TX_ENTRIES_PER_SUBCRQ);
4065         atomic_inc(&adapter->running_cap_crqs);
4066         ibmvnic_send_crq(adapter, &crq);
4067
4068         crq.query_capability.capability =
4069             cpu_to_be16(MIN_RX_ADD_ENTRIES_PER_SUBCRQ);
4070         atomic_inc(&adapter->running_cap_crqs);
4071         ibmvnic_send_crq(adapter, &crq);
4072
4073         crq.query_capability.capability =
4074             cpu_to_be16(MAX_TX_ENTRIES_PER_SUBCRQ);
4075         atomic_inc(&adapter->running_cap_crqs);
4076         ibmvnic_send_crq(adapter, &crq);
4077
4078         crq.query_capability.capability =
4079             cpu_to_be16(MAX_RX_ADD_ENTRIES_PER_SUBCRQ);
4080         atomic_inc(&adapter->running_cap_crqs);
4081         ibmvnic_send_crq(adapter, &crq);
4082
4083         crq.query_capability.capability = cpu_to_be16(TCP_IP_OFFLOAD);
4084         atomic_inc(&adapter->running_cap_crqs);
4085         ibmvnic_send_crq(adapter, &crq);
4086
4087         crq.query_capability.capability = cpu_to_be16(PROMISC_SUPPORTED);
4088         atomic_inc(&adapter->running_cap_crqs);
4089         ibmvnic_send_crq(adapter, &crq);
4090
4091         crq.query_capability.capability = cpu_to_be16(MIN_MTU);
4092         atomic_inc(&adapter->running_cap_crqs);
4093         ibmvnic_send_crq(adapter, &crq);
4094
4095         crq.query_capability.capability = cpu_to_be16(MAX_MTU);
4096         atomic_inc(&adapter->running_cap_crqs);
4097         ibmvnic_send_crq(adapter, &crq);
4098
4099         crq.query_capability.capability = cpu_to_be16(MAX_MULTICAST_FILTERS);
4100         atomic_inc(&adapter->running_cap_crqs);
4101         ibmvnic_send_crq(adapter, &crq);
4102
4103         crq.query_capability.capability = cpu_to_be16(VLAN_HEADER_INSERTION);
4104         atomic_inc(&adapter->running_cap_crqs);
4105         ibmvnic_send_crq(adapter, &crq);
4106
4107         crq.query_capability.capability = cpu_to_be16(RX_VLAN_HEADER_INSERTION);
4108         atomic_inc(&adapter->running_cap_crqs);
4109         ibmvnic_send_crq(adapter, &crq);
4110
4111         crq.query_capability.capability = cpu_to_be16(MAX_TX_SG_ENTRIES);
4112         atomic_inc(&adapter->running_cap_crqs);
4113         ibmvnic_send_crq(adapter, &crq);
4114
4115         crq.query_capability.capability = cpu_to_be16(RX_SG_SUPPORTED);
4116         atomic_inc(&adapter->running_cap_crqs);
4117         ibmvnic_send_crq(adapter, &crq);
4118
4119         crq.query_capability.capability = cpu_to_be16(OPT_TX_COMP_SUB_QUEUES);
4120         atomic_inc(&adapter->running_cap_crqs);
4121         ibmvnic_send_crq(adapter, &crq);
4122
4123         crq.query_capability.capability = cpu_to_be16(OPT_RX_COMP_QUEUES);
4124         atomic_inc(&adapter->running_cap_crqs);
4125         ibmvnic_send_crq(adapter, &crq);
4126
4127         crq.query_capability.capability =
4128                         cpu_to_be16(OPT_RX_BUFADD_Q_PER_RX_COMP_Q);
4129         atomic_inc(&adapter->running_cap_crqs);
4130         ibmvnic_send_crq(adapter, &crq);
4131
4132         crq.query_capability.capability =
4133                         cpu_to_be16(OPT_TX_ENTRIES_PER_SUBCRQ);
4134         atomic_inc(&adapter->running_cap_crqs);
4135         ibmvnic_send_crq(adapter, &crq);
4136
4137         crq.query_capability.capability =
4138                         cpu_to_be16(OPT_RXBA_ENTRIES_PER_SUBCRQ);
4139         atomic_inc(&adapter->running_cap_crqs);
4140         ibmvnic_send_crq(adapter, &crq);
4141
4142         crq.query_capability.capability = cpu_to_be16(TX_RX_DESC_REQ);
4143         atomic_inc(&adapter->running_cap_crqs);
4144         ibmvnic_send_crq(adapter, &crq);
4145 }
4146
4147 static void send_query_ip_offload(struct ibmvnic_adapter *adapter)
4148 {
4149         int buf_sz = sizeof(struct ibmvnic_query_ip_offload_buffer);
4150         struct device *dev = &adapter->vdev->dev;
4151         union ibmvnic_crq crq;
4152
4153         adapter->ip_offload_tok =
4154                 dma_map_single(dev,
4155                                &adapter->ip_offload_buf,
4156                                buf_sz,
4157                                DMA_FROM_DEVICE);
4158
4159         if (dma_mapping_error(dev, adapter->ip_offload_tok)) {
4160                 if (!firmware_has_feature(FW_FEATURE_CMO))
4161                         dev_err(dev, "Couldn't map offload buffer\n");
4162                 return;
4163         }
4164
4165         memset(&crq, 0, sizeof(crq));
4166         crq.query_ip_offload.first = IBMVNIC_CRQ_CMD;
4167         crq.query_ip_offload.cmd = QUERY_IP_OFFLOAD;
4168         crq.query_ip_offload.len = cpu_to_be32(buf_sz);
4169         crq.query_ip_offload.ioba =
4170             cpu_to_be32(adapter->ip_offload_tok);
4171
4172         ibmvnic_send_crq(adapter, &crq);
4173 }
4174
4175 static void send_control_ip_offload(struct ibmvnic_adapter *adapter)
4176 {
4177         struct ibmvnic_control_ip_offload_buffer *ctrl_buf = &adapter->ip_offload_ctrl;
4178         struct ibmvnic_query_ip_offload_buffer *buf = &adapter->ip_offload_buf;
4179         struct device *dev = &adapter->vdev->dev;
4180         netdev_features_t old_hw_features = 0;
4181         union ibmvnic_crq crq;
4182
4183         adapter->ip_offload_ctrl_tok =
4184                 dma_map_single(dev,
4185                                ctrl_buf,
4186                                sizeof(adapter->ip_offload_ctrl),
4187                                DMA_TO_DEVICE);
4188
4189         if (dma_mapping_error(dev, adapter->ip_offload_ctrl_tok)) {
4190                 dev_err(dev, "Couldn't map ip offload control buffer\n");
4191                 return;
4192         }
4193
4194         ctrl_buf->len = cpu_to_be32(sizeof(adapter->ip_offload_ctrl));
4195         ctrl_buf->version = cpu_to_be32(INITIAL_VERSION_IOB);
4196         ctrl_buf->ipv4_chksum = buf->ipv4_chksum;
4197         ctrl_buf->ipv6_chksum = buf->ipv6_chksum;
4198         ctrl_buf->tcp_ipv4_chksum = buf->tcp_ipv4_chksum;
4199         ctrl_buf->udp_ipv4_chksum = buf->udp_ipv4_chksum;
4200         ctrl_buf->tcp_ipv6_chksum = buf->tcp_ipv6_chksum;
4201         ctrl_buf->udp_ipv6_chksum = buf->udp_ipv6_chksum;
4202         ctrl_buf->large_tx_ipv4 = buf->large_tx_ipv4;
4203         ctrl_buf->large_tx_ipv6 = buf->large_tx_ipv6;
4204
4205         /* large_rx disabled for now, additional features needed */
4206         ctrl_buf->large_rx_ipv4 = 0;
4207         ctrl_buf->large_rx_ipv6 = 0;
4208
4209         if (adapter->state != VNIC_PROBING) {
4210                 old_hw_features = adapter->netdev->hw_features;
4211                 adapter->netdev->hw_features = 0;
4212         }
4213
4214         adapter->netdev->hw_features = NETIF_F_SG | NETIF_F_GSO | NETIF_F_GRO;
4215
4216         if (buf->tcp_ipv4_chksum || buf->udp_ipv4_chksum)
4217                 adapter->netdev->hw_features |= NETIF_F_IP_CSUM;
4218
4219         if (buf->tcp_ipv6_chksum || buf->udp_ipv6_chksum)
4220                 adapter->netdev->hw_features |= NETIF_F_IPV6_CSUM;
4221
4222         if ((adapter->netdev->features &
4223             (NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM)))
4224                 adapter->netdev->hw_features |= NETIF_F_RXCSUM;
4225
4226         if (buf->large_tx_ipv4)
4227                 adapter->netdev->hw_features |= NETIF_F_TSO;
4228         if (buf->large_tx_ipv6)
4229                 adapter->netdev->hw_features |= NETIF_F_TSO6;
4230
4231         if (adapter->state == VNIC_PROBING) {
4232                 adapter->netdev->features |= adapter->netdev->hw_features;
4233         } else if (old_hw_features != adapter->netdev->hw_features) {
4234                 netdev_features_t tmp = 0;
4235
4236                 /* disable features no longer supported */
4237                 adapter->netdev->features &= adapter->netdev->hw_features;
4238                 /* turn on features now supported if previously enabled */
4239                 tmp = (old_hw_features ^ adapter->netdev->hw_features) &
4240                         adapter->netdev->hw_features;
4241                 adapter->netdev->features |=
4242                                 tmp & adapter->netdev->wanted_features;
4243         }
4244
4245         memset(&crq, 0, sizeof(crq));
4246         crq.control_ip_offload.first = IBMVNIC_CRQ_CMD;
4247         crq.control_ip_offload.cmd = CONTROL_IP_OFFLOAD;
4248         crq.control_ip_offload.len =
4249             cpu_to_be32(sizeof(adapter->ip_offload_ctrl));
4250         crq.control_ip_offload.ioba = cpu_to_be32(adapter->ip_offload_ctrl_tok);
4251         ibmvnic_send_crq(adapter, &crq);
4252 }
4253
4254 static void handle_vpd_size_rsp(union ibmvnic_crq *crq,
4255                                 struct ibmvnic_adapter *adapter)
4256 {
4257         struct device *dev = &adapter->vdev->dev;
4258
4259         if (crq->get_vpd_size_rsp.rc.code) {
4260                 dev_err(dev, "Error retrieving VPD size, rc=%x\n",
4261                         crq->get_vpd_size_rsp.rc.code);
4262                 complete(&adapter->fw_done);
4263                 return;
4264         }
4265
4266         adapter->vpd->len = be64_to_cpu(crq->get_vpd_size_rsp.len);
4267         complete(&adapter->fw_done);
4268 }
4269
4270 static void handle_vpd_rsp(union ibmvnic_crq *crq,
4271                            struct ibmvnic_adapter *adapter)
4272 {
4273         struct device *dev = &adapter->vdev->dev;
4274         unsigned char *substr = NULL;
4275         u8 fw_level_len = 0;
4276
4277         memset(adapter->fw_version, 0, 32);
4278
4279         dma_unmap_single(dev, adapter->vpd->dma_addr, adapter->vpd->len,
4280                          DMA_FROM_DEVICE);
4281
4282         if (crq->get_vpd_rsp.rc.code) {
4283                 dev_err(dev, "Error retrieving VPD from device, rc=%x\n",
4284                         crq->get_vpd_rsp.rc.code);
4285                 goto complete;
4286         }
4287
4288         /* get the position of the firmware version info
4289          * located after the ASCII 'RM' substring in the buffer
4290          */
4291         substr = strnstr(adapter->vpd->buff, "RM", adapter->vpd->len);
4292         if (!substr) {
4293                 dev_info(dev, "Warning - No FW level has been provided in the VPD buffer by the VIOS Server\n");
4294                 goto complete;
4295         }
4296
4297         /* get length of firmware level ASCII substring */
4298         if ((substr + 2) < (adapter->vpd->buff + adapter->vpd->len)) {
4299                 fw_level_len = *(substr + 2);
4300         } else {
4301                 dev_info(dev, "Length of FW substr extrapolated VDP buff\n");
4302                 goto complete;
4303         }
4304
4305         /* copy firmware version string from vpd into adapter */
4306         if ((substr + 3 + fw_level_len) <
4307             (adapter->vpd->buff + adapter->vpd->len)) {
4308                 strncpy((char *)adapter->fw_version, substr + 3, fw_level_len);
4309         } else {
4310                 dev_info(dev, "FW substr extrapolated VPD buff\n");
4311         }
4312
4313 complete:
4314         if (adapter->fw_version[0] == '\0')
4315                 strncpy((char *)adapter->fw_version, "N/A", 3 * sizeof(char));
4316         complete(&adapter->fw_done);
4317 }
4318
4319 static void handle_query_ip_offload_rsp(struct ibmvnic_adapter *adapter)
4320 {
4321         struct device *dev = &adapter->vdev->dev;
4322         struct ibmvnic_query_ip_offload_buffer *buf = &adapter->ip_offload_buf;
4323         int i;
4324
4325         dma_unmap_single(dev, adapter->ip_offload_tok,
4326                          sizeof(adapter->ip_offload_buf), DMA_FROM_DEVICE);
4327
4328         netdev_dbg(adapter->netdev, "Query IP Offload Buffer:\n");
4329         for (i = 0; i < (sizeof(adapter->ip_offload_buf) - 1) / 8 + 1; i++)
4330                 netdev_dbg(adapter->netdev, "%016lx\n",
4331                            ((unsigned long int *)(buf))[i]);
4332
4333         netdev_dbg(adapter->netdev, "ipv4_chksum = %d\n", buf->ipv4_chksum);
4334         netdev_dbg(adapter->netdev, "ipv6_chksum = %d\n", buf->ipv6_chksum);
4335         netdev_dbg(adapter->netdev, "tcp_ipv4_chksum = %d\n",
4336                    buf->tcp_ipv4_chksum);
4337         netdev_dbg(adapter->netdev, "tcp_ipv6_chksum = %d\n",
4338                    buf->tcp_ipv6_chksum);
4339         netdev_dbg(adapter->netdev, "udp_ipv4_chksum = %d\n",
4340                    buf->udp_ipv4_chksum);
4341         netdev_dbg(adapter->netdev, "udp_ipv6_chksum = %d\n",
4342                    buf->udp_ipv6_chksum);
4343         netdev_dbg(adapter->netdev, "large_tx_ipv4 = %d\n",
4344                    buf->large_tx_ipv4);
4345         netdev_dbg(adapter->netdev, "large_tx_ipv6 = %d\n",
4346                    buf->large_tx_ipv6);
4347         netdev_dbg(adapter->netdev, "large_rx_ipv4 = %d\n",
4348                    buf->large_rx_ipv4);
4349         netdev_dbg(adapter->netdev, "large_rx_ipv6 = %d\n",
4350                    buf->large_rx_ipv6);
4351         netdev_dbg(adapter->netdev, "max_ipv4_hdr_sz = %d\n",
4352                    buf->max_ipv4_header_size);
4353         netdev_dbg(adapter->netdev, "max_ipv6_hdr_sz = %d\n",
4354                    buf->max_ipv6_header_size);
4355         netdev_dbg(adapter->netdev, "max_tcp_hdr_size = %d\n",
4356                    buf->max_tcp_header_size);
4357         netdev_dbg(adapter->netdev, "max_udp_hdr_size = %d\n",
4358                    buf->max_udp_header_size);
4359         netdev_dbg(adapter->netdev, "max_large_tx_size = %d\n",
4360                    buf->max_large_tx_size);
4361         netdev_dbg(adapter->netdev, "max_large_rx_size = %d\n",
4362                    buf->max_large_rx_size);
4363         netdev_dbg(adapter->netdev, "ipv6_ext_hdr = %d\n",
4364                    buf->ipv6_extension_header);
4365         netdev_dbg(adapter->netdev, "tcp_pseudosum_req = %d\n",
4366                    buf->tcp_pseudosum_req);
4367         netdev_dbg(adapter->netdev, "num_ipv6_ext_hd = %d\n",
4368                    buf->num_ipv6_ext_headers);
4369         netdev_dbg(adapter->netdev, "off_ipv6_ext_hd = %d\n",
4370                    buf->off_ipv6_ext_headers);
4371
4372         send_control_ip_offload(adapter);
4373 }
4374
4375 static const char *ibmvnic_fw_err_cause(u16 cause)
4376 {
4377         switch (cause) {
4378         case ADAPTER_PROBLEM:
4379                 return "adapter problem";
4380         case BUS_PROBLEM:
4381                 return "bus problem";
4382         case FW_PROBLEM:
4383                 return "firmware problem";
4384         case DD_PROBLEM:
4385                 return "device driver problem";
4386         case EEH_RECOVERY:
4387                 return "EEH recovery";
4388         case FW_UPDATED:
4389                 return "firmware updated";
4390         case LOW_MEMORY:
4391                 return "low Memory";
4392         default:
4393                 return "unknown";
4394         }
4395 }
4396
4397 static void handle_error_indication(union ibmvnic_crq *crq,
4398                                     struct ibmvnic_adapter *adapter)
4399 {
4400         struct device *dev = &adapter->vdev->dev;
4401         u16 cause;
4402
4403         cause = be16_to_cpu(crq->error_indication.error_cause);
4404
4405         dev_warn_ratelimited(dev,
4406                              "Firmware reports %serror, cause: %s. Starting recovery...\n",
4407                              crq->error_indication.flags
4408                                 & IBMVNIC_FATAL_ERROR ? "FATAL " : "",
4409                              ibmvnic_fw_err_cause(cause));
4410
4411         if (crq->error_indication.flags & IBMVNIC_FATAL_ERROR)
4412                 ibmvnic_reset(adapter, VNIC_RESET_FATAL);
4413         else
4414                 ibmvnic_reset(adapter, VNIC_RESET_NON_FATAL);
4415 }
4416
4417 static int handle_change_mac_rsp(union ibmvnic_crq *crq,
4418                                  struct ibmvnic_adapter *adapter)
4419 {
4420         struct net_device *netdev = adapter->netdev;
4421         struct device *dev = &adapter->vdev->dev;
4422         long rc;
4423
4424         rc = crq->change_mac_addr_rsp.rc.code;
4425         if (rc) {
4426                 dev_err(dev, "Error %ld in CHANGE_MAC_ADDR_RSP\n", rc);
4427                 goto out;
4428         }
4429         /* crq->change_mac_addr.mac_addr is the requested one
4430          * crq->change_mac_addr_rsp.mac_addr is the returned valid one.
4431          */
4432         ether_addr_copy(netdev->dev_addr,
4433                         &crq->change_mac_addr_rsp.mac_addr[0]);
4434         ether_addr_copy(adapter->mac_addr,
4435                         &crq->change_mac_addr_rsp.mac_addr[0]);
4436 out:
4437         complete(&adapter->fw_done);
4438         return rc;
4439 }
4440
4441 static void handle_request_cap_rsp(union ibmvnic_crq *crq,
4442                                    struct ibmvnic_adapter *adapter)
4443 {
4444         struct device *dev = &adapter->vdev->dev;
4445         u64 *req_value;
4446         char *name;
4447
4448         atomic_dec(&adapter->running_cap_crqs);
4449         switch (be16_to_cpu(crq->request_capability_rsp.capability)) {
4450         case REQ_TX_QUEUES:
4451                 req_value = &adapter->req_tx_queues;
4452                 name = "tx";
4453                 break;
4454         case REQ_RX_QUEUES:
4455                 req_value = &adapter->req_rx_queues;
4456                 name = "rx";
4457                 break;
4458         case REQ_RX_ADD_QUEUES:
4459                 req_value = &adapter->req_rx_add_queues;
4460                 name = "rx_add";
4461                 break;
4462         case REQ_TX_ENTRIES_PER_SUBCRQ:
4463                 req_value = &adapter->req_tx_entries_per_subcrq;
4464                 name = "tx_entries_per_subcrq";
4465                 break;
4466         case REQ_RX_ADD_ENTRIES_PER_SUBCRQ:
4467                 req_value = &adapter->req_rx_add_entries_per_subcrq;
4468                 name = "rx_add_entries_per_subcrq";
4469                 break;
4470         case REQ_MTU:
4471                 req_value = &adapter->req_mtu;
4472                 name = "mtu";
4473                 break;
4474         case PROMISC_REQUESTED:
4475                 req_value = &adapter->promisc;
4476                 name = "promisc";
4477                 break;
4478         default:
4479                 dev_err(dev, "Got invalid cap request rsp %d\n",
4480                         crq->request_capability.capability);
4481                 return;
4482         }
4483
4484         switch (crq->request_capability_rsp.rc.code) {
4485         case SUCCESS:
4486                 break;
4487         case PARTIALSUCCESS:
4488                 dev_info(dev, "req=%lld, rsp=%ld in %s queue, retrying.\n",
4489                          *req_value,
4490                          (long int)be64_to_cpu(crq->request_capability_rsp.
4491                                                number), name);
4492
4493                 if (be16_to_cpu(crq->request_capability_rsp.capability) ==
4494                     REQ_MTU) {
4495                         pr_err("mtu of %llu is not supported. Reverting.\n",
4496                                *req_value);
4497                         *req_value = adapter->fallback.mtu;
4498                 } else {
4499                         *req_value =
4500                                 be64_to_cpu(crq->request_capability_rsp.number);
4501                 }
4502
4503                 send_request_cap(adapter, 1);
4504                 return;
4505         default:
4506                 dev_err(dev, "Error %d in request cap rsp\n",
4507                         crq->request_capability_rsp.rc.code);
4508                 return;
4509         }
4510
4511         /* Done receiving requested capabilities, query IP offload support */
4512         if (atomic_read(&adapter->running_cap_crqs) == 0) {
4513                 adapter->wait_capability = false;
4514                 send_query_ip_offload(adapter);
4515         }
4516 }
4517
4518 static int handle_login_rsp(union ibmvnic_crq *login_rsp_crq,
4519                             struct ibmvnic_adapter *adapter)
4520 {
4521         struct device *dev = &adapter->vdev->dev;
4522         struct net_device *netdev = adapter->netdev;
4523         struct ibmvnic_login_rsp_buffer *login_rsp = adapter->login_rsp_buf;
4524         struct ibmvnic_login_buffer *login = adapter->login_buf;
4525         u64 *tx_handle_array;
4526         u64 *rx_handle_array;
4527         int num_tx_pools;
4528         int num_rx_pools;
4529         u64 *size_array;
4530         int i;
4531
4532         /* CHECK: Test/set of login_pending does not need to be atomic
4533          * because only ibmvnic_tasklet tests/clears this.
4534          */
4535         if (!adapter->login_pending) {
4536                 netdev_warn(netdev, "Ignoring unexpected login response\n");
4537                 return 0;
4538         }
4539         adapter->login_pending = false;
4540
4541         dma_unmap_single(dev, adapter->login_buf_token, adapter->login_buf_sz,
4542                          DMA_TO_DEVICE);
4543         dma_unmap_single(dev, adapter->login_rsp_buf_token,
4544                          adapter->login_rsp_buf_sz, DMA_FROM_DEVICE);
4545
4546         /* If the number of queues requested can't be allocated by the
4547          * server, the login response will return with code 1. We will need
4548          * to resend the login buffer with fewer queues requested.
4549          */
4550         if (login_rsp_crq->generic.rc.code) {
4551                 adapter->init_done_rc = login_rsp_crq->generic.rc.code;
4552                 complete(&adapter->init_done);
4553                 return 0;
4554         }
4555
4556         netdev->mtu = adapter->req_mtu - ETH_HLEN;
4557
4558         netdev_dbg(adapter->netdev, "Login Response Buffer:\n");
4559         for (i = 0; i < (adapter->login_rsp_buf_sz - 1) / 8 + 1; i++) {
4560                 netdev_dbg(adapter->netdev, "%016lx\n",
4561                            ((unsigned long int *)(adapter->login_rsp_buf))[i]);
4562         }
4563
4564         /* Sanity checks */
4565         if (login->num_txcomp_subcrqs != login_rsp->num_txsubm_subcrqs ||
4566             (be32_to_cpu(login->num_rxcomp_subcrqs) *
4567              adapter->req_rx_add_queues !=
4568              be32_to_cpu(login_rsp->num_rxadd_subcrqs))) {
4569                 dev_err(dev, "FATAL: Inconsistent login and login rsp\n");
4570                 ibmvnic_reset(adapter, VNIC_RESET_FATAL);
4571                 return -EIO;
4572         }
4573         size_array = (u64 *)((u8 *)(adapter->login_rsp_buf) +
4574                 be32_to_cpu(adapter->login_rsp_buf->off_rxadd_buff_size));
4575         /* variable buffer sizes are not supported, so just read the
4576          * first entry.
4577          */
4578         adapter->cur_rx_buf_sz = be64_to_cpu(size_array[0]);
4579
4580         num_tx_pools = be32_to_cpu(adapter->login_rsp_buf->num_txsubm_subcrqs);
4581         num_rx_pools = be32_to_cpu(adapter->login_rsp_buf->num_rxadd_subcrqs);
4582
4583         tx_handle_array = (u64 *)((u8 *)(adapter->login_rsp_buf) +
4584                                   be32_to_cpu(adapter->login_rsp_buf->off_txsubm_subcrqs));
4585         rx_handle_array = (u64 *)((u8 *)(adapter->login_rsp_buf) +
4586                                   be32_to_cpu(adapter->login_rsp_buf->off_rxadd_subcrqs));
4587
4588         for (i = 0; i < num_tx_pools; i++)
4589                 adapter->tx_scrq[i]->handle = tx_handle_array[i];
4590
4591         for (i = 0; i < num_rx_pools; i++)
4592                 adapter->rx_scrq[i]->handle = rx_handle_array[i];
4593
4594         adapter->num_active_tx_scrqs = num_tx_pools;
4595         adapter->num_active_rx_scrqs = num_rx_pools;
4596         release_login_rsp_buffer(adapter);
4597         release_login_buffer(adapter);
4598         complete(&adapter->init_done);
4599
4600         return 0;
4601 }
4602
4603 static void handle_request_unmap_rsp(union ibmvnic_crq *crq,
4604                                      struct ibmvnic_adapter *adapter)
4605 {
4606         struct device *dev = &adapter->vdev->dev;
4607         long rc;
4608
4609         rc = crq->request_unmap_rsp.rc.code;
4610         if (rc)
4611                 dev_err(dev, "Error %ld in REQUEST_UNMAP_RSP\n", rc);
4612 }
4613
4614 static void handle_query_map_rsp(union ibmvnic_crq *crq,
4615                                  struct ibmvnic_adapter *adapter)
4616 {
4617         struct net_device *netdev = adapter->netdev;
4618         struct device *dev = &adapter->vdev->dev;
4619         long rc;
4620
4621         rc = crq->query_map_rsp.rc.code;
4622         if (rc) {
4623                 dev_err(dev, "Error %ld in QUERY_MAP_RSP\n", rc);
4624                 return;
4625         }
4626         netdev_dbg(netdev, "page_size = %d\ntot_pages = %d\nfree_pages = %d\n",
4627                    crq->query_map_rsp.page_size, crq->query_map_rsp.tot_pages,
4628                    crq->query_map_rsp.free_pages);
4629 }
4630
4631 static void handle_query_cap_rsp(union ibmvnic_crq *crq,
4632                                  struct ibmvnic_adapter *adapter)
4633 {
4634         struct net_device *netdev = adapter->netdev;
4635         struct device *dev = &adapter->vdev->dev;
4636         long rc;
4637
4638         atomic_dec(&adapter->running_cap_crqs);
4639         netdev_dbg(netdev, "Outstanding queries: %d\n",
4640                    atomic_read(&adapter->running_cap_crqs));
4641         rc = crq->query_capability.rc.code;
4642         if (rc) {
4643                 dev_err(dev, "Error %ld in QUERY_CAP_RSP\n", rc);
4644                 goto out;
4645         }
4646
4647         switch (be16_to_cpu(crq->query_capability.capability)) {
4648         case MIN_TX_QUEUES:
4649                 adapter->min_tx_queues =
4650                     be64_to_cpu(crq->query_capability.number);
4651                 netdev_dbg(netdev, "min_tx_queues = %lld\n",
4652                            adapter->min_tx_queues);
4653                 break;
4654         case MIN_RX_QUEUES:
4655                 adapter->min_rx_queues =
4656                     be64_to_cpu(crq->query_capability.number);
4657                 netdev_dbg(netdev, "min_rx_queues = %lld\n",
4658                            adapter->min_rx_queues);
4659                 break;
4660         case MIN_RX_ADD_QUEUES:
4661                 adapter->min_rx_add_queues =
4662                     be64_to_cpu(crq->query_capability.number);
4663                 netdev_dbg(netdev, "min_rx_add_queues = %lld\n",
4664                            adapter->min_rx_add_queues);
4665                 break;
4666         case MAX_TX_QUEUES:
4667                 adapter->max_tx_queues =
4668                     be64_to_cpu(crq->query_capability.number);
4669                 netdev_dbg(netdev, "max_tx_queues = %lld\n",
4670                            adapter->max_tx_queues);
4671                 break;
4672         case MAX_RX_QUEUES:
4673                 adapter->max_rx_queues =
4674                     be64_to_cpu(crq->query_capability.number);
4675                 netdev_dbg(netdev, "max_rx_queues = %lld\n",
4676                            adapter->max_rx_queues);
4677                 break;
4678         case MAX_RX_ADD_QUEUES:
4679                 adapter->max_rx_add_queues =
4680                     be64_to_cpu(crq->query_capability.number);
4681                 netdev_dbg(netdev, "max_rx_add_queues = %lld\n",
4682                            adapter->max_rx_add_queues);
4683                 break;
4684         case MIN_TX_ENTRIES_PER_SUBCRQ:
4685                 adapter->min_tx_entries_per_subcrq =
4686                     be64_to_cpu(crq->query_capability.number);
4687                 netdev_dbg(netdev, "min_tx_entries_per_subcrq = %lld\n",
4688                            adapter->min_tx_entries_per_subcrq);
4689                 break;
4690         case MIN_RX_ADD_ENTRIES_PER_SUBCRQ:
4691                 adapter->min_rx_add_entries_per_subcrq =
4692                     be64_to_cpu(crq->query_capability.number);
4693                 netdev_dbg(netdev, "min_rx_add_entrs_per_subcrq = %lld\n",
4694                            adapter->min_rx_add_entries_per_subcrq);
4695                 break;
4696         case MAX_TX_ENTRIES_PER_SUBCRQ:
4697                 adapter->max_tx_entries_per_subcrq =
4698                     be64_to_cpu(crq->query_capability.number);
4699                 netdev_dbg(netdev, "max_tx_entries_per_subcrq = %lld\n",
4700                            adapter->max_tx_entries_per_subcrq);
4701                 break;
4702         case MAX_RX_ADD_ENTRIES_PER_SUBCRQ:
4703                 adapter->max_rx_add_entries_per_subcrq =
4704                     be64_to_cpu(crq->query_capability.number);
4705                 netdev_dbg(netdev, "max_rx_add_entrs_per_subcrq = %lld\n",
4706                            adapter->max_rx_add_entries_per_subcrq);
4707                 break;
4708         case TCP_IP_OFFLOAD:
4709                 adapter->tcp_ip_offload =
4710                     be64_to_cpu(crq->query_capability.number);
4711                 netdev_dbg(netdev, "tcp_ip_offload = %lld\n",
4712                            adapter->tcp_ip_offload);
4713                 break;
4714         case PROMISC_SUPPORTED:
4715                 adapter->promisc_supported =
4716                     be64_to_cpu(crq->query_capability.number);
4717                 netdev_dbg(netdev, "promisc_supported = %lld\n",
4718                            adapter->promisc_supported);
4719                 break;
4720         case MIN_MTU:
4721                 adapter->min_mtu = be64_to_cpu(crq->query_capability.number);
4722                 netdev->min_mtu = adapter->min_mtu - ETH_HLEN;
4723                 netdev_dbg(netdev, "min_mtu = %lld\n", adapter->min_mtu);
4724                 break;
4725         case MAX_MTU:
4726                 adapter->max_mtu = be64_to_cpu(crq->query_capability.number);
4727                 netdev->max_mtu = adapter->max_mtu - ETH_HLEN;
4728                 netdev_dbg(netdev, "max_mtu = %lld\n", adapter->max_mtu);
4729                 break;
4730         case MAX_MULTICAST_FILTERS:
4731                 adapter->max_multicast_filters =
4732                     be64_to_cpu(crq->query_capability.number);
4733                 netdev_dbg(netdev, "max_multicast_filters = %lld\n",
4734                            adapter->max_multicast_filters);
4735                 break;
4736         case VLAN_HEADER_INSERTION:
4737                 adapter->vlan_header_insertion =
4738                     be64_to_cpu(crq->query_capability.number);
4739                 if (adapter->vlan_header_insertion)
4740                         netdev->features |= NETIF_F_HW_VLAN_STAG_TX;
4741                 netdev_dbg(netdev, "vlan_header_insertion = %lld\n",
4742                            adapter->vlan_header_insertion);
4743                 break;
4744         case RX_VLAN_HEADER_INSERTION:
4745                 adapter->rx_vlan_header_insertion =
4746                     be64_to_cpu(crq->query_capability.number);
4747                 netdev_dbg(netdev, "rx_vlan_header_insertion = %lld\n",
4748                            adapter->rx_vlan_header_insertion);
4749                 break;
4750         case MAX_TX_SG_ENTRIES:
4751                 adapter->max_tx_sg_entries =
4752                     be64_to_cpu(crq->query_capability.number);
4753                 netdev_dbg(netdev, "max_tx_sg_entries = %lld\n",
4754                            adapter->max_tx_sg_entries);
4755                 break;
4756         case RX_SG_SUPPORTED:
4757                 adapter->rx_sg_supported =
4758                     be64_to_cpu(crq->query_capability.number);
4759                 netdev_dbg(netdev, "rx_sg_supported = %lld\n",
4760                            adapter->rx_sg_supported);
4761                 break;
4762         case OPT_TX_COMP_SUB_QUEUES:
4763                 adapter->opt_tx_comp_sub_queues =
4764                     be64_to_cpu(crq->query_capability.number);
4765                 netdev_dbg(netdev, "opt_tx_comp_sub_queues = %lld\n",
4766                            adapter->opt_tx_comp_sub_queues);
4767                 break;
4768         case OPT_RX_COMP_QUEUES:
4769                 adapter->opt_rx_comp_queues =
4770                     be64_to_cpu(crq->query_capability.number);
4771                 netdev_dbg(netdev, "opt_rx_comp_queues = %lld\n",
4772                            adapter->opt_rx_comp_queues);
4773                 break;
4774         case OPT_RX_BUFADD_Q_PER_RX_COMP_Q:
4775                 adapter->opt_rx_bufadd_q_per_rx_comp_q =
4776                     be64_to_cpu(crq->query_capability.number);
4777                 netdev_dbg(netdev, "opt_rx_bufadd_q_per_rx_comp_q = %lld\n",
4778                            adapter->opt_rx_bufadd_q_per_rx_comp_q);
4779                 break;
4780         case OPT_TX_ENTRIES_PER_SUBCRQ:
4781                 adapter->opt_tx_entries_per_subcrq =
4782                     be64_to_cpu(crq->query_capability.number);
4783                 netdev_dbg(netdev, "opt_tx_entries_per_subcrq = %lld\n",
4784                            adapter->opt_tx_entries_per_subcrq);
4785                 break;
4786         case OPT_RXBA_ENTRIES_PER_SUBCRQ:
4787                 adapter->opt_rxba_entries_per_subcrq =
4788                     be64_to_cpu(crq->query_capability.number);
4789                 netdev_dbg(netdev, "opt_rxba_entries_per_subcrq = %lld\n",
4790                            adapter->opt_rxba_entries_per_subcrq);
4791                 break;
4792         case TX_RX_DESC_REQ:
4793                 adapter->tx_rx_desc_req = crq->query_capability.number;
4794                 netdev_dbg(netdev, "tx_rx_desc_req = %llx\n",
4795                            adapter->tx_rx_desc_req);
4796                 break;
4797
4798         default:
4799                 netdev_err(netdev, "Got invalid cap rsp %d\n",
4800                            crq->query_capability.capability);
4801         }
4802
4803 out:
4804         if (atomic_read(&adapter->running_cap_crqs) == 0) {
4805                 adapter->wait_capability = false;
4806                 send_request_cap(adapter, 0);
4807         }
4808 }
4809
4810 static int send_query_phys_parms(struct ibmvnic_adapter *adapter)
4811 {
4812         union ibmvnic_crq crq;
4813         int rc;
4814
4815         memset(&crq, 0, sizeof(crq));
4816         crq.query_phys_parms.first = IBMVNIC_CRQ_CMD;
4817         crq.query_phys_parms.cmd = QUERY_PHYS_PARMS;
4818
4819         mutex_lock(&adapter->fw_lock);
4820         adapter->fw_done_rc = 0;
4821         reinit_completion(&adapter->fw_done);
4822
4823         rc = ibmvnic_send_crq(adapter, &crq);
4824         if (rc) {
4825                 mutex_unlock(&adapter->fw_lock);
4826                 return rc;
4827         }
4828
4829         rc = ibmvnic_wait_for_completion(adapter, &adapter->fw_done, 10000);
4830         if (rc) {
4831                 mutex_unlock(&adapter->fw_lock);
4832                 return rc;
4833         }
4834
4835         mutex_unlock(&adapter->fw_lock);
4836         return adapter->fw_done_rc ? -EIO : 0;
4837 }
4838
4839 static int handle_query_phys_parms_rsp(union ibmvnic_crq *crq,
4840                                        struct ibmvnic_adapter *adapter)
4841 {
4842         struct net_device *netdev = adapter->netdev;
4843         int rc;
4844         __be32 rspeed = cpu_to_be32(crq->query_phys_parms_rsp.speed);
4845
4846         rc = crq->query_phys_parms_rsp.rc.code;
4847         if (rc) {
4848                 netdev_err(netdev, "Error %d in QUERY_PHYS_PARMS\n", rc);
4849                 return rc;
4850         }
4851         switch (rspeed) {
4852         case IBMVNIC_10MBPS:
4853                 adapter->speed = SPEED_10;
4854                 break;
4855         case IBMVNIC_100MBPS:
4856                 adapter->speed = SPEED_100;
4857                 break;
4858         case IBMVNIC_1GBPS:
4859                 adapter->speed = SPEED_1000;
4860                 break;
4861         case IBMVNIC_10GBPS:
4862                 adapter->speed = SPEED_10000;
4863                 break;
4864         case IBMVNIC_25GBPS:
4865                 adapter->speed = SPEED_25000;
4866                 break;
4867         case IBMVNIC_40GBPS:
4868                 adapter->speed = SPEED_40000;
4869                 break;
4870         case IBMVNIC_50GBPS:
4871                 adapter->speed = SPEED_50000;
4872                 break;
4873         case IBMVNIC_100GBPS:
4874                 adapter->speed = SPEED_100000;
4875                 break;
4876         case IBMVNIC_200GBPS:
4877                 adapter->speed = SPEED_200000;
4878                 break;
4879         default:
4880                 if (netif_carrier_ok(netdev))
4881                         netdev_warn(netdev, "Unknown speed 0x%08x\n", rspeed);
4882                 adapter->speed = SPEED_UNKNOWN;
4883         }
4884         if (crq->query_phys_parms_rsp.flags1 & IBMVNIC_FULL_DUPLEX)
4885                 adapter->duplex = DUPLEX_FULL;
4886         else if (crq->query_phys_parms_rsp.flags1 & IBMVNIC_HALF_DUPLEX)
4887                 adapter->duplex = DUPLEX_HALF;
4888         else
4889                 adapter->duplex = DUPLEX_UNKNOWN;
4890
4891         return rc;
4892 }
4893
4894 static void ibmvnic_handle_crq(union ibmvnic_crq *crq,
4895                                struct ibmvnic_adapter *adapter)
4896 {
4897         struct ibmvnic_generic_crq *gen_crq = &crq->generic;
4898         struct net_device *netdev = adapter->netdev;
4899         struct device *dev = &adapter->vdev->dev;
4900         u64 *u64_crq = (u64 *)crq;
4901         long rc;
4902
4903         netdev_dbg(netdev, "Handling CRQ: %016lx %016lx\n",
4904                    (unsigned long int)cpu_to_be64(u64_crq[0]),
4905                    (unsigned long int)cpu_to_be64(u64_crq[1]));
4906         switch (gen_crq->first) {
4907         case IBMVNIC_CRQ_INIT_RSP:
4908                 switch (gen_crq->cmd) {
4909                 case IBMVNIC_CRQ_INIT:
4910                         dev_info(dev, "Partner initialized\n");
4911                         adapter->from_passive_init = true;
4912                         /* Discard any stale login responses from prev reset.
4913                          * CHECK: should we clear even on INIT_COMPLETE?
4914                          */
4915                         adapter->login_pending = false;
4916
4917                         if (!completion_done(&adapter->init_done)) {
4918                                 complete(&adapter->init_done);
4919                                 adapter->init_done_rc = -EIO;
4920                         }
4921                         ibmvnic_reset(adapter, VNIC_RESET_FAILOVER);
4922                         break;
4923                 case IBMVNIC_CRQ_INIT_COMPLETE:
4924                         dev_info(dev, "Partner initialization complete\n");
4925                         adapter->crq.active = true;
4926                         send_version_xchg(adapter);
4927                         break;
4928                 default:
4929                         dev_err(dev, "Unknown crq cmd: %d\n", gen_crq->cmd);
4930                 }
4931                 return;
4932         case IBMVNIC_CRQ_XPORT_EVENT:
4933                 netif_carrier_off(netdev);
4934                 adapter->crq.active = false;
4935                 /* terminate any thread waiting for a response
4936                  * from the device
4937                  */
4938                 if (!completion_done(&adapter->fw_done)) {
4939                         adapter->fw_done_rc = -EIO;
4940                         complete(&adapter->fw_done);
4941                 }
4942                 if (!completion_done(&adapter->stats_done))
4943                         complete(&adapter->stats_done);
4944                 if (test_bit(0, &adapter->resetting))
4945                         adapter->force_reset_recovery = true;
4946                 if (gen_crq->cmd == IBMVNIC_PARTITION_MIGRATED) {
4947                         dev_info(dev, "Migrated, re-enabling adapter\n");
4948                         ibmvnic_reset(adapter, VNIC_RESET_MOBILITY);
4949                 } else if (gen_crq->cmd == IBMVNIC_DEVICE_FAILOVER) {
4950                         dev_info(dev, "Backing device failover detected\n");
4951                         adapter->failover_pending = true;
4952                 } else {
4953                         /* The adapter lost the connection */
4954                         dev_err(dev, "Virtual Adapter failed (rc=%d)\n",
4955                                 gen_crq->cmd);
4956                         ibmvnic_reset(adapter, VNIC_RESET_FATAL);
4957                 }
4958                 return;
4959         case IBMVNIC_CRQ_CMD_RSP:
4960                 break;
4961         default:
4962                 dev_err(dev, "Got an invalid msg type 0x%02x\n",
4963                         gen_crq->first);
4964                 return;
4965         }
4966
4967         switch (gen_crq->cmd) {
4968         case VERSION_EXCHANGE_RSP:
4969                 rc = crq->version_exchange_rsp.rc.code;
4970                 if (rc) {
4971                         dev_err(dev, "Error %ld in VERSION_EXCHG_RSP\n", rc);
4972                         break;
4973                 }
4974                 ibmvnic_version =
4975                             be16_to_cpu(crq->version_exchange_rsp.version);
4976                 dev_info(dev, "Partner protocol version is %d\n",
4977                          ibmvnic_version);
4978                 send_query_cap(adapter);
4979                 break;
4980         case QUERY_CAPABILITY_RSP:
4981                 handle_query_cap_rsp(crq, adapter);
4982                 break;
4983         case QUERY_MAP_RSP:
4984                 handle_query_map_rsp(crq, adapter);
4985                 break;
4986         case REQUEST_MAP_RSP:
4987                 adapter->fw_done_rc = crq->request_map_rsp.rc.code;
4988                 complete(&adapter->fw_done);
4989                 break;
4990         case REQUEST_UNMAP_RSP:
4991                 handle_request_unmap_rsp(crq, adapter);
4992                 break;
4993         case REQUEST_CAPABILITY_RSP:
4994                 handle_request_cap_rsp(crq, adapter);
4995                 break;
4996         case LOGIN_RSP:
4997                 netdev_dbg(netdev, "Got Login Response\n");
4998                 handle_login_rsp(crq, adapter);
4999                 break;
5000         case LOGICAL_LINK_STATE_RSP:
5001                 netdev_dbg(netdev,
5002                            "Got Logical Link State Response, state: %d rc: %d\n",
5003                            crq->logical_link_state_rsp.link_state,
5004                            crq->logical_link_state_rsp.rc.code);
5005                 adapter->logical_link_state =
5006                     crq->logical_link_state_rsp.link_state;
5007                 adapter->init_done_rc = crq->logical_link_state_rsp.rc.code;
5008                 complete(&adapter->init_done);
5009                 break;
5010         case LINK_STATE_INDICATION:
5011                 netdev_dbg(netdev, "Got Logical Link State Indication\n");
5012                 adapter->phys_link_state =
5013                     crq->link_state_indication.phys_link_state;
5014                 adapter->logical_link_state =
5015                     crq->link_state_indication.logical_link_state;
5016                 if (adapter->phys_link_state && adapter->logical_link_state)
5017                         netif_carrier_on(netdev);
5018                 else
5019                         netif_carrier_off(netdev);
5020                 break;
5021         case CHANGE_MAC_ADDR_RSP:
5022                 netdev_dbg(netdev, "Got MAC address change Response\n");
5023                 adapter->fw_done_rc = handle_change_mac_rsp(crq, adapter);
5024                 break;
5025         case ERROR_INDICATION:
5026                 netdev_dbg(netdev, "Got Error Indication\n");
5027                 handle_error_indication(crq, adapter);
5028                 break;
5029         case REQUEST_STATISTICS_RSP:
5030                 netdev_dbg(netdev, "Got Statistics Response\n");
5031                 complete(&adapter->stats_done);
5032                 break;
5033         case QUERY_IP_OFFLOAD_RSP:
5034                 netdev_dbg(netdev, "Got Query IP offload Response\n");
5035                 handle_query_ip_offload_rsp(adapter);
5036                 break;
5037         case MULTICAST_CTRL_RSP:
5038                 netdev_dbg(netdev, "Got multicast control Response\n");
5039                 break;
5040         case CONTROL_IP_OFFLOAD_RSP:
5041                 netdev_dbg(netdev, "Got Control IP offload Response\n");
5042                 dma_unmap_single(dev, adapter->ip_offload_ctrl_tok,
5043                                  sizeof(adapter->ip_offload_ctrl),
5044                                  DMA_TO_DEVICE);
5045                 complete(&adapter->init_done);
5046                 break;
5047         case COLLECT_FW_TRACE_RSP:
5048                 netdev_dbg(netdev, "Got Collect firmware trace Response\n");
5049                 complete(&adapter->fw_done);
5050                 break;
5051         case GET_VPD_SIZE_RSP:
5052                 handle_vpd_size_rsp(crq, adapter);
5053                 break;
5054         case GET_VPD_RSP:
5055                 handle_vpd_rsp(crq, adapter);
5056                 break;
5057         case QUERY_PHYS_PARMS_RSP:
5058                 adapter->fw_done_rc = handle_query_phys_parms_rsp(crq, adapter);
5059                 complete(&adapter->fw_done);
5060                 break;
5061         default:
5062                 netdev_err(netdev, "Got an invalid cmd type 0x%02x\n",
5063                            gen_crq->cmd);
5064         }
5065 }
5066
5067 static irqreturn_t ibmvnic_interrupt(int irq, void *instance)
5068 {
5069         struct ibmvnic_adapter *adapter = instance;
5070
5071         tasklet_schedule(&adapter->tasklet);
5072         return IRQ_HANDLED;
5073 }
5074
5075 static void ibmvnic_tasklet(struct tasklet_struct *t)
5076 {
5077         struct ibmvnic_adapter *adapter = from_tasklet(adapter, t, tasklet);
5078         struct ibmvnic_crq_queue *queue = &adapter->crq;
5079         union ibmvnic_crq *crq;
5080         unsigned long flags;
5081         bool done = false;
5082
5083         spin_lock_irqsave(&queue->lock, flags);
5084         while (!done) {
5085                 /* Pull all the valid messages off the CRQ */
5086                 while ((crq = ibmvnic_next_crq(adapter)) != NULL) {
5087                         ibmvnic_handle_crq(crq, adapter);
5088                         crq->generic.first = 0;
5089                 }
5090
5091                 /* remain in tasklet until all
5092                  * capabilities responses are received
5093                  */
5094                 if (!adapter->wait_capability)
5095                         done = true;
5096         }
5097         /* if capabilities CRQ's were sent in this tasklet, the following
5098          * tasklet must wait until all responses are received
5099          */
5100         if (atomic_read(&adapter->running_cap_crqs) != 0)
5101                 adapter->wait_capability = true;
5102         spin_unlock_irqrestore(&queue->lock, flags);
5103 }
5104
5105 static int ibmvnic_reenable_crq_queue(struct ibmvnic_adapter *adapter)
5106 {
5107         struct vio_dev *vdev = adapter->vdev;
5108         int rc;
5109
5110         do {
5111                 rc = plpar_hcall_norets(H_ENABLE_CRQ, vdev->unit_address);
5112         } while (rc == H_IN_PROGRESS || rc == H_BUSY || H_IS_LONG_BUSY(rc));
5113
5114         if (rc)
5115                 dev_err(&vdev->dev, "Error enabling adapter (rc=%d)\n", rc);
5116
5117         return rc;
5118 }
5119
5120 static int ibmvnic_reset_crq(struct ibmvnic_adapter *adapter)
5121 {
5122         struct ibmvnic_crq_queue *crq = &adapter->crq;
5123         struct device *dev = &adapter->vdev->dev;
5124         struct vio_dev *vdev = adapter->vdev;
5125         int rc;
5126
5127         /* Close the CRQ */
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         /* Clean out the queue */
5133         if (!crq->msgs)
5134                 return -EINVAL;
5135
5136         memset(crq->msgs, 0, PAGE_SIZE);
5137         crq->cur = 0;
5138         crq->active = false;
5139
5140         /* And re-open it again */
5141         rc = plpar_hcall_norets(H_REG_CRQ, vdev->unit_address,
5142                                 crq->msg_token, PAGE_SIZE);
5143
5144         if (rc == H_CLOSED)
5145                 /* Adapter is good, but other end is not ready */
5146                 dev_warn(dev, "Partner adapter not ready\n");
5147         else if (rc != 0)
5148                 dev_warn(dev, "Couldn't register crq (rc=%d)\n", rc);
5149
5150         return rc;
5151 }
5152
5153 static void release_crq_queue(struct ibmvnic_adapter *adapter)
5154 {
5155         struct ibmvnic_crq_queue *crq = &adapter->crq;
5156         struct vio_dev *vdev = adapter->vdev;
5157         long rc;
5158
5159         if (!crq->msgs)
5160                 return;
5161
5162         netdev_dbg(adapter->netdev, "Releasing CRQ\n");
5163         free_irq(vdev->irq, adapter);
5164         tasklet_kill(&adapter->tasklet);
5165         do {
5166                 rc = plpar_hcall_norets(H_FREE_CRQ, vdev->unit_address);
5167         } while (rc == H_BUSY || H_IS_LONG_BUSY(rc));
5168
5169         dma_unmap_single(&vdev->dev, crq->msg_token, PAGE_SIZE,
5170                          DMA_BIDIRECTIONAL);
5171         free_page((unsigned long)crq->msgs);
5172         crq->msgs = NULL;
5173         crq->active = false;
5174 }
5175
5176 static int init_crq_queue(struct ibmvnic_adapter *adapter)
5177 {
5178         struct ibmvnic_crq_queue *crq = &adapter->crq;
5179         struct device *dev = &adapter->vdev->dev;
5180         struct vio_dev *vdev = adapter->vdev;
5181         int rc, retrc = -ENOMEM;
5182
5183         if (crq->msgs)
5184                 return 0;
5185
5186         crq->msgs = (union ibmvnic_crq *)get_zeroed_page(GFP_KERNEL);
5187         /* Should we allocate more than one page? */
5188
5189         if (!crq->msgs)
5190                 return -ENOMEM;
5191
5192         crq->size = PAGE_SIZE / sizeof(*crq->msgs);
5193         crq->msg_token = dma_map_single(dev, crq->msgs, PAGE_SIZE,
5194                                         DMA_BIDIRECTIONAL);
5195         if (dma_mapping_error(dev, crq->msg_token))
5196                 goto map_failed;
5197
5198         rc = plpar_hcall_norets(H_REG_CRQ, vdev->unit_address,
5199                                 crq->msg_token, PAGE_SIZE);
5200
5201         if (rc == H_RESOURCE)
5202                 /* maybe kexecing and resource is busy. try a reset */
5203                 rc = ibmvnic_reset_crq(adapter);
5204         retrc = rc;
5205
5206         if (rc == H_CLOSED) {
5207                 dev_warn(dev, "Partner adapter not ready\n");
5208         } else if (rc) {
5209                 dev_warn(dev, "Error %d opening adapter\n", rc);
5210                 goto reg_crq_failed;
5211         }
5212
5213         retrc = 0;
5214
5215         tasklet_setup(&adapter->tasklet, (void *)ibmvnic_tasklet);
5216
5217         netdev_dbg(adapter->netdev, "registering irq 0x%x\n", vdev->irq);
5218         snprintf(crq->name, sizeof(crq->name), "ibmvnic-%x",
5219                  adapter->vdev->unit_address);
5220         rc = request_irq(vdev->irq, ibmvnic_interrupt, 0, crq->name, adapter);
5221         if (rc) {
5222                 dev_err(dev, "Couldn't register irq 0x%x. rc=%d\n",
5223                         vdev->irq, rc);
5224                 goto req_irq_failed;
5225         }
5226
5227         rc = vio_enable_interrupts(vdev);
5228         if (rc) {
5229                 dev_err(dev, "Error %d enabling interrupts\n", rc);
5230                 goto req_irq_failed;
5231         }
5232
5233         crq->cur = 0;
5234         spin_lock_init(&crq->lock);
5235
5236         return retrc;
5237
5238 req_irq_failed:
5239         tasklet_kill(&adapter->tasklet);
5240         do {
5241                 rc = plpar_hcall_norets(H_FREE_CRQ, vdev->unit_address);
5242         } while (rc == H_BUSY || H_IS_LONG_BUSY(rc));
5243 reg_crq_failed:
5244         dma_unmap_single(dev, crq->msg_token, PAGE_SIZE, DMA_BIDIRECTIONAL);
5245 map_failed:
5246         free_page((unsigned long)crq->msgs);
5247         crq->msgs = NULL;
5248         return retrc;
5249 }
5250
5251 static int ibmvnic_reset_init(struct ibmvnic_adapter *adapter, bool reset)
5252 {
5253         struct device *dev = &adapter->vdev->dev;
5254         unsigned long timeout = msecs_to_jiffies(20000);
5255         u64 old_num_rx_queues, old_num_tx_queues;
5256         int rc;
5257
5258         adapter->from_passive_init = false;
5259
5260         if (reset) {
5261                 old_num_rx_queues = adapter->req_rx_queues;
5262                 old_num_tx_queues = adapter->req_tx_queues;
5263                 reinit_completion(&adapter->init_done);
5264         }
5265
5266         adapter->init_done_rc = 0;
5267         rc = ibmvnic_send_crq_init(adapter);
5268         if (rc) {
5269                 dev_err(dev, "Send crq init failed with error %d\n", rc);
5270                 return rc;
5271         }
5272
5273         if (!wait_for_completion_timeout(&adapter->init_done, timeout)) {
5274                 dev_err(dev, "Initialization sequence timed out\n");
5275                 return -1;
5276         }
5277
5278         if (adapter->init_done_rc) {
5279                 release_crq_queue(adapter);
5280                 return adapter->init_done_rc;
5281         }
5282
5283         if (adapter->from_passive_init) {
5284                 adapter->state = VNIC_OPEN;
5285                 adapter->from_passive_init = false;
5286                 return -1;
5287         }
5288
5289         if (reset &&
5290             test_bit(0, &adapter->resetting) && !adapter->wait_for_reset &&
5291             adapter->reset_reason != VNIC_RESET_MOBILITY) {
5292                 if (adapter->req_rx_queues != old_num_rx_queues ||
5293                     adapter->req_tx_queues != old_num_tx_queues) {
5294                         release_sub_crqs(adapter, 0);
5295                         rc = init_sub_crqs(adapter);
5296                 } else {
5297                         rc = reset_sub_crq_queues(adapter);
5298                 }
5299         } else {
5300                 rc = init_sub_crqs(adapter);
5301         }
5302
5303         if (rc) {
5304                 dev_err(dev, "Initialization of sub crqs failed\n");
5305                 release_crq_queue(adapter);
5306                 return rc;
5307         }
5308
5309         rc = init_sub_crq_irqs(adapter);
5310         if (rc) {
5311                 dev_err(dev, "Failed to initialize sub crq irqs\n");
5312                 release_crq_queue(adapter);
5313         }
5314
5315         return rc;
5316 }
5317
5318 static struct device_attribute dev_attr_failover;
5319
5320 static int ibmvnic_probe(struct vio_dev *dev, const struct vio_device_id *id)
5321 {
5322         struct ibmvnic_adapter *adapter;
5323         struct net_device *netdev;
5324         unsigned char *mac_addr_p;
5325         int rc;
5326
5327         dev_dbg(&dev->dev, "entering ibmvnic_probe for UA 0x%x\n",
5328                 dev->unit_address);
5329
5330         mac_addr_p = (unsigned char *)vio_get_attribute(dev,
5331                                                         VETH_MAC_ADDR, NULL);
5332         if (!mac_addr_p) {
5333                 dev_err(&dev->dev,
5334                         "(%s:%3.3d) ERROR: Can't find MAC_ADDR attribute\n",
5335                         __FILE__, __LINE__);
5336                 return 0;
5337         }
5338
5339         netdev = alloc_etherdev_mq(sizeof(struct ibmvnic_adapter),
5340                                    IBMVNIC_MAX_QUEUES);
5341         if (!netdev)
5342                 return -ENOMEM;
5343
5344         adapter = netdev_priv(netdev);
5345         adapter->state = VNIC_PROBING;
5346         dev_set_drvdata(&dev->dev, netdev);
5347         adapter->vdev = dev;
5348         adapter->netdev = netdev;
5349         adapter->login_pending = false;
5350
5351         ether_addr_copy(adapter->mac_addr, mac_addr_p);
5352         ether_addr_copy(netdev->dev_addr, adapter->mac_addr);
5353         netdev->irq = dev->irq;
5354         netdev->netdev_ops = &ibmvnic_netdev_ops;
5355         netdev->ethtool_ops = &ibmvnic_ethtool_ops;
5356         SET_NETDEV_DEV(netdev, &dev->dev);
5357
5358         spin_lock_init(&adapter->stats_lock);
5359
5360         INIT_WORK(&adapter->ibmvnic_reset, __ibmvnic_reset);
5361         INIT_DELAYED_WORK(&adapter->ibmvnic_delayed_reset,
5362                           __ibmvnic_delayed_reset);
5363         INIT_LIST_HEAD(&adapter->rwi_list);
5364         spin_lock_init(&adapter->rwi_lock);
5365         spin_lock_init(&adapter->state_lock);
5366         mutex_init(&adapter->fw_lock);
5367         init_completion(&adapter->init_done);
5368         init_completion(&adapter->fw_done);
5369         init_completion(&adapter->reset_done);
5370         init_completion(&adapter->stats_done);
5371         clear_bit(0, &adapter->resetting);
5372
5373         do {
5374                 rc = init_crq_queue(adapter);
5375                 if (rc) {
5376                         dev_err(&dev->dev, "Couldn't initialize crq. rc=%d\n",
5377                                 rc);
5378                         goto ibmvnic_init_fail;
5379                 }
5380
5381                 rc = ibmvnic_reset_init(adapter, false);
5382                 if (rc && rc != EAGAIN)
5383                         goto ibmvnic_init_fail;
5384         } while (rc == EAGAIN);
5385
5386         rc = init_stats_buffers(adapter);
5387         if (rc)
5388                 goto ibmvnic_init_fail;
5389
5390         rc = init_stats_token(adapter);
5391         if (rc)
5392                 goto ibmvnic_stats_fail;
5393
5394         netdev->mtu = adapter->req_mtu - ETH_HLEN;
5395         netdev->min_mtu = adapter->min_mtu - ETH_HLEN;
5396         netdev->max_mtu = adapter->max_mtu - ETH_HLEN;
5397
5398         rc = device_create_file(&dev->dev, &dev_attr_failover);
5399         if (rc)
5400                 goto ibmvnic_dev_file_err;
5401
5402         netif_carrier_off(netdev);
5403         rc = register_netdev(netdev);
5404         if (rc) {
5405                 dev_err(&dev->dev, "failed to register netdev rc=%d\n", rc);
5406                 goto ibmvnic_register_fail;
5407         }
5408         dev_info(&dev->dev, "ibmvnic registered\n");
5409
5410         adapter->state = VNIC_PROBED;
5411
5412         adapter->wait_for_reset = false;
5413         adapter->last_reset_time = jiffies;
5414         return 0;
5415
5416 ibmvnic_register_fail:
5417         device_remove_file(&dev->dev, &dev_attr_failover);
5418
5419 ibmvnic_dev_file_err:
5420         release_stats_token(adapter);
5421
5422 ibmvnic_stats_fail:
5423         release_stats_buffers(adapter);
5424
5425 ibmvnic_init_fail:
5426         release_sub_crqs(adapter, 1);
5427         release_crq_queue(adapter);
5428         mutex_destroy(&adapter->fw_lock);
5429         free_netdev(netdev);
5430
5431         return rc;
5432 }
5433
5434 static int ibmvnic_remove(struct vio_dev *dev)
5435 {
5436         struct net_device *netdev = dev_get_drvdata(&dev->dev);
5437         struct ibmvnic_adapter *adapter = netdev_priv(netdev);
5438         unsigned long flags;
5439
5440         spin_lock_irqsave(&adapter->state_lock, flags);
5441         if (test_bit(0, &adapter->resetting)) {
5442                 spin_unlock_irqrestore(&adapter->state_lock, flags);
5443                 return -EBUSY;
5444         }
5445
5446         adapter->state = VNIC_REMOVING;
5447         spin_unlock_irqrestore(&adapter->state_lock, flags);
5448
5449         flush_work(&adapter->ibmvnic_reset);
5450         flush_delayed_work(&adapter->ibmvnic_delayed_reset);
5451
5452         rtnl_lock();
5453         unregister_netdevice(netdev);
5454
5455         release_resources(adapter);
5456         release_sub_crqs(adapter, 1);
5457         release_crq_queue(adapter);
5458
5459         release_stats_token(adapter);
5460         release_stats_buffers(adapter);
5461
5462         adapter->state = VNIC_REMOVED;
5463
5464         rtnl_unlock();
5465         mutex_destroy(&adapter->fw_lock);
5466         device_remove_file(&dev->dev, &dev_attr_failover);
5467         free_netdev(netdev);
5468         dev_set_drvdata(&dev->dev, NULL);
5469
5470         return 0;
5471 }
5472
5473 static ssize_t failover_store(struct device *dev, struct device_attribute *attr,
5474                               const char *buf, size_t count)
5475 {
5476         struct net_device *netdev = dev_get_drvdata(dev);
5477         struct ibmvnic_adapter *adapter = netdev_priv(netdev);
5478         unsigned long retbuf[PLPAR_HCALL_BUFSIZE];
5479         __be64 session_token;
5480         long rc;
5481
5482         if (!sysfs_streq(buf, "1"))
5483                 return -EINVAL;
5484
5485         rc = plpar_hcall(H_VIOCTL, retbuf, adapter->vdev->unit_address,
5486                          H_GET_SESSION_TOKEN, 0, 0, 0);
5487         if (rc) {
5488                 netdev_err(netdev, "Couldn't retrieve session token, rc %ld\n",
5489                            rc);
5490                 return -EINVAL;
5491         }
5492
5493         session_token = (__be64)retbuf[0];
5494         netdev_dbg(netdev, "Initiating client failover, session id %llx\n",
5495                    be64_to_cpu(session_token));
5496         rc = plpar_hcall_norets(H_VIOCTL, adapter->vdev->unit_address,
5497                                 H_SESSION_ERR_DETECTED, session_token, 0, 0);
5498         if (rc) {
5499                 netdev_err(netdev, "Client initiated failover failed, rc %ld\n",
5500                            rc);
5501                 return -EINVAL;
5502         }
5503
5504         return count;
5505 }
5506
5507 static DEVICE_ATTR_WO(failover);
5508
5509 static unsigned long ibmvnic_get_desired_dma(struct vio_dev *vdev)
5510 {
5511         struct net_device *netdev = dev_get_drvdata(&vdev->dev);
5512         struct ibmvnic_adapter *adapter;
5513         struct iommu_table *tbl;
5514         unsigned long ret = 0;
5515         int i;
5516
5517         tbl = get_iommu_table_base(&vdev->dev);
5518
5519         /* netdev inits at probe time along with the structures we need below*/
5520         if (!netdev)
5521                 return IOMMU_PAGE_ALIGN(IBMVNIC_IO_ENTITLEMENT_DEFAULT, tbl);
5522
5523         adapter = netdev_priv(netdev);
5524
5525         ret += PAGE_SIZE; /* the crq message queue */
5526         ret += IOMMU_PAGE_ALIGN(sizeof(struct ibmvnic_statistics), tbl);
5527
5528         for (i = 0; i < adapter->req_tx_queues + adapter->req_rx_queues; i++)
5529                 ret += 4 * PAGE_SIZE; /* the scrq message queue */
5530
5531         for (i = 0; i < adapter->num_active_rx_pools; i++)
5532                 ret += adapter->rx_pool[i].size *
5533                     IOMMU_PAGE_ALIGN(adapter->rx_pool[i].buff_size, tbl);
5534
5535         return ret;
5536 }
5537
5538 static int ibmvnic_resume(struct device *dev)
5539 {
5540         struct net_device *netdev = dev_get_drvdata(dev);
5541         struct ibmvnic_adapter *adapter = netdev_priv(netdev);
5542
5543         if (adapter->state != VNIC_OPEN)
5544                 return 0;
5545
5546         tasklet_schedule(&adapter->tasklet);
5547
5548         return 0;
5549 }
5550
5551 static const struct vio_device_id ibmvnic_device_table[] = {
5552         {"network", "IBM,vnic"},
5553         {"", "" }
5554 };
5555 MODULE_DEVICE_TABLE(vio, ibmvnic_device_table);
5556
5557 static const struct dev_pm_ops ibmvnic_pm_ops = {
5558         .resume = ibmvnic_resume
5559 };
5560
5561 static struct vio_driver ibmvnic_driver = {
5562         .id_table       = ibmvnic_device_table,
5563         .probe          = ibmvnic_probe,
5564         .remove         = ibmvnic_remove,
5565         .get_desired_dma = ibmvnic_get_desired_dma,
5566         .name           = ibmvnic_driver_name,
5567         .pm             = &ibmvnic_pm_ops,
5568 };
5569
5570 /* module functions */
5571 static int __init ibmvnic_module_init(void)
5572 {
5573         pr_info("%s: %s %s\n", ibmvnic_driver_name, ibmvnic_driver_string,
5574                 IBMVNIC_DRIVER_VERSION);
5575
5576         return vio_register_driver(&ibmvnic_driver);
5577 }
5578
5579 static void __exit ibmvnic_module_exit(void)
5580 {
5581         vio_unregister_driver(&ibmvnic_driver);
5582 }
5583
5584 module_init(ibmvnic_module_init);
5585 module_exit(ibmvnic_module_exit);