drivers/net/ethernet/ibm/ibmvnic.c

   1 // SPDX-License-Identifier: GPL-2.0-or-later
   2 /**************************************************************************/
   3 /*                                                                        */
   4 /*  IBM System i and System p Virtual NIC Device Driver                   */
   5 /*  Copyright (C) 2014 IBM Corp.                                          */
   6 /*  Santiago Leon (santi_leon@yahoo.com)                                  */
   7 /*  Thomas Falcon (tlfalcon@linux.vnet.ibm.com)                           */
   8 /*  John Allen (jallen@linux.vnet.ibm.com)                                */
   9 /*                                                                        */
  10 /*                                                                        */
  11 /* This module contains the implementation of a virtual ethernet device   */
  12 /* for use with IBM i/p Series LPAR Linux. It utilizes the logical LAN    */
  13 /* option of the RS/6000 Platform Architecture to interface with virtual  */
  14 /* ethernet NICs that are presented to the partition by the hypervisor.   */
  15 /*                                                                         */
  16 /* Messages are passed between the VNIC driver and the VNIC server using  */
  17 /* Command/Response Queues (CRQs) and sub CRQs (sCRQs). CRQs are used to  */
  18 /* issue and receive commands that initiate communication with the server */
  19 /* on driver initialization. Sub CRQs (sCRQs) are similar to CRQs, but    */
  20 /* are used by the driver to notify the server that a packet is           */
  21 /* ready for transmission or that a buffer has been added to receive a    */
  22 /* packet. Subsequently, sCRQs are used by the server to notify the       */
  23 /* driver that a packet transmission has been completed or that a packet  */
  24 /* has been received and placed in a waiting buffer.                      */
  25 /*                                                                        */
  26 /* In lieu of a more conventional "on-the-fly" DMA mapping strategy in    */
  27 /* which skbs are DMA mapped and immediately unmapped when the transmit   */
  28 /* or receive has been completed, the VNIC driver is required to use      */
  29 /* "long term mapping". This entails that large, continuous DMA mapped    */
  30 /* buffers are allocated on driver initialization and these buffers are   */
  31 /* then continuously reused to pass skbs to and from the VNIC server.     */
  32 /*                                                                        */
  33 /**************************************************************************/
  34
  35 #include <linux/module.h>
  36 #include <linux/moduleparam.h>
  37 #include <linux/types.h>
  38 #include <linux/errno.h>
  39 #include <linux/completion.h>
  40 #include <linux/ioport.h>
  41 #include <linux/dma-mapping.h>
  42 #include <linux/kernel.h>
  43 #include <linux/netdevice.h>
  44 #include <linux/etherdevice.h>
  45 #include <linux/skbuff.h>
  46 #include <linux/init.h>
  47 #include <linux/delay.h>
  48 #include <linux/mm.h>
  49 #include <linux/ethtool.h>
  50 #include <linux/proc_fs.h>
  51 #include <linux/if_arp.h>
  52 #include <linux/in.h>
  53 #include <linux/ip.h>
  54 #include <linux/ipv6.h>
  55 #include <linux/irq.h>
  56 #include <linux/kthread.h>
  57 #include <linux/seq_file.h>
  58 #include <linux/interrupt.h>
  59 #include <net/net_namespace.h>
  60 #include <asm/hvcall.h>
  61 #include <linux/atomic.h>
  62 #include <asm/vio.h>
  63 #include <asm/iommu.h>
  64 #include <linux/uaccess.h>
  65 #include <asm/firmware.h>
  66 #include <linux/workqueue.h>
  67 #include <linux/if_vlan.h>
  68 #include <linux/utsname.h>
  69
  70 #include "ibmvnic.h"
  71
  72 static const char ibmvnic_driver_name[] = "ibmvnic";
  73 static const char ibmvnic_driver_string[] = "IBM System i/p Virtual NIC Driver";
  74
  75 MODULE_AUTHOR("Santiago Leon");
  76 MODULE_DESCRIPTION("IBM System i/p Virtual NIC Driver");
  77 MODULE_LICENSE("GPL");
  78 MODULE_VERSION(IBMVNIC_DRIVER_VERSION);
  79
  80 static int ibmvnic_version = IBMVNIC_INITIAL_VERSION;
  81 static int ibmvnic_remove(struct vio_dev *);
  82 static void release_sub_crqs(struct ibmvnic_adapter *, bool);
  83 static int ibmvnic_reset_crq(struct ibmvnic_adapter *);
  84 static int ibmvnic_send_crq_init(struct ibmvnic_adapter *);
  85 static int ibmvnic_reenable_crq_queue(struct ibmvnic_adapter *);
  86 static int ibmvnic_send_crq(struct ibmvnic_adapter *, union ibmvnic_crq *);
  87 static int send_subcrq(struct ibmvnic_adapter *adapter, u64 remote_handle,
  88                        union sub_crq *sub_crq);
  89 static int send_subcrq_indirect(struct ibmvnic_adapter *, u64, u64, u64);
  90 static irqreturn_t ibmvnic_interrupt_rx(int irq, void *instance);
  91 static int enable_scrq_irq(struct ibmvnic_adapter *,
  92                            struct ibmvnic_sub_crq_queue *);
  93 static int disable_scrq_irq(struct ibmvnic_adapter *,
  94                             struct ibmvnic_sub_crq_queue *);
  95 static int pending_scrq(struct ibmvnic_adapter *,
  96                         struct ibmvnic_sub_crq_queue *);
  97 static union sub_crq *ibmvnic_next_scrq(struct ibmvnic_adapter *,
  98                                         struct ibmvnic_sub_crq_queue *);
  99 static int ibmvnic_poll(struct napi_struct *napi, int data);
 100 static void send_query_map(struct ibmvnic_adapter *adapter);
 101 static int send_request_map(struct ibmvnic_adapter *, dma_addr_t, __be32, u8);
 102 static int send_request_unmap(struct ibmvnic_adapter *, u8);
 103 static int send_login(struct ibmvnic_adapter *adapter);
 104 static void send_query_cap(struct ibmvnic_adapter *adapter);
 105 static int init_sub_crqs(struct ibmvnic_adapter *);
 106 static int init_sub_crq_irqs(struct ibmvnic_adapter *adapter);
 107 static int ibmvnic_reset_init(struct ibmvnic_adapter *, bool reset);
 108 static void release_crq_queue(struct ibmvnic_adapter *);
 109 static int __ibmvnic_set_mac(struct net_device *, u8 *);
 110 static int init_crq_queue(struct ibmvnic_adapter *adapter);
 111 static int send_query_phys_parms(struct ibmvnic_adapter *adapter);
 112
 113 struct ibmvnic_stat {
 114         char name[ETH_GSTRING_LEN];
 115         int offset;
 116 };
 117
 118 #define IBMVNIC_STAT_OFF(stat) (offsetof(struct ibmvnic_adapter, stats) + \
 119                              offsetof(struct ibmvnic_statistics, stat))
 120 #define IBMVNIC_GET_STAT(a, off) (*((u64 *)(((unsigned long)(a)) + (off))))
 121
 122 static const struct ibmvnic_stat ibmvnic_stats[] = {
 123         {"rx_packets", IBMVNIC_STAT_OFF(rx_packets)},
 124         {"rx_bytes", IBMVNIC_STAT_OFF(rx_bytes)},
 125         {"tx_packets", IBMVNIC_STAT_OFF(tx_packets)},
 126         {"tx_bytes", IBMVNIC_STAT_OFF(tx_bytes)},
 127         {"ucast_tx_packets", IBMVNIC_STAT_OFF(ucast_tx_packets)},
 128         {"ucast_rx_packets", IBMVNIC_STAT_OFF(ucast_rx_packets)},
 129         {"mcast_tx_packets", IBMVNIC_STAT_OFF(mcast_tx_packets)},
 130         {"mcast_rx_packets", IBMVNIC_STAT_OFF(mcast_rx_packets)},
 131         {"bcast_tx_packets", IBMVNIC_STAT_OFF(bcast_tx_packets)},
 132         {"bcast_rx_packets", IBMVNIC_STAT_OFF(bcast_rx_packets)},
 133         {"align_errors", IBMVNIC_STAT_OFF(align_errors)},
 134         {"fcs_errors", IBMVNIC_STAT_OFF(fcs_errors)},
 135         {"single_collision_frames", IBMVNIC_STAT_OFF(single_collision_frames)},
 136         {"multi_collision_frames", IBMVNIC_STAT_OFF(multi_collision_frames)},
 137         {"sqe_test_errors", IBMVNIC_STAT_OFF(sqe_test_errors)},
 138         {"deferred_tx", IBMVNIC_STAT_OFF(deferred_tx)},
 139         {"late_collisions", IBMVNIC_STAT_OFF(late_collisions)},
 140         {"excess_collisions", IBMVNIC_STAT_OFF(excess_collisions)},
 141         {"internal_mac_tx_errors", IBMVNIC_STAT_OFF(internal_mac_tx_errors)},
 142         {"carrier_sense", IBMVNIC_STAT_OFF(carrier_sense)},
 143         {"too_long_frames", IBMVNIC_STAT_OFF(too_long_frames)},
 144         {"internal_mac_rx_errors", IBMVNIC_STAT_OFF(internal_mac_rx_errors)},
 145 };
 146
 147 static long h_reg_sub_crq(unsigned long unit_address, unsigned long token,
 148                           unsigned long length, unsigned long *number,
 149                           unsigned long *irq)
 150 {
 151         unsigned long retbuf[PLPAR_HCALL_BUFSIZE];
 152         long rc;
 153
 154         rc = plpar_hcall(H_REG_SUB_CRQ, retbuf, unit_address, token, length);
 155         *number = retbuf[0];
 156         *irq = retbuf[1];
 157
 158         return rc;
 159 }
 160
 161 /**
 162  * ibmvnic_wait_for_completion - Check device state and wait for completion
 163  * @adapter: private device data
 164  * @comp_done: completion structure to wait for
 165  * @timeout: time to wait in milliseconds
 166  *
 167  * Wait for a completion signal or until the timeout limit is reached
 168  * while checking that the device is still active.
 169  */
 170 static int ibmvnic_wait_for_completion(struct ibmvnic_adapter *adapter,
 171                                        struct completion *comp_done,
 172                                        unsigned long timeout)
 173 {
 174         struct net_device *netdev;
 175         unsigned long div_timeout;
 176         u8 retry;
 177
 178         netdev = adapter->netdev;
 179         retry = 5;
 180         div_timeout = msecs_to_jiffies(timeout / retry);
 181         while (true) {
 182                 if (!adapter->crq.active) {
 183                         netdev_err(netdev, "Device down!\n");
 184                         return -ENODEV;
 185                 }
 186                 if (!retry--)
 187                         break;
 188                 if (wait_for_completion_timeout(comp_done, div_timeout))
 189                         return 0;
 190         }
 191         netdev_err(netdev, "Operation timed out.\n");
 192         return -ETIMEDOUT;
 193 }
 194
 195 static int alloc_long_term_buff(struct ibmvnic_adapter *adapter,
 196                                 struct ibmvnic_long_term_buff *ltb, int size)
 197 {
 198         struct device *dev = &adapter->vdev->dev;
 199         int rc;
 200
 201         ltb->size = size;
 202         ltb->buff = dma_alloc_coherent(dev, ltb->size, &ltb->addr,
 203                                        GFP_KERNEL);
 204
 205         if (!ltb->buff) {
 206                 dev_err(dev, "Couldn't alloc long term buffer\n");
 207                 return -ENOMEM;
 208         }
 209         ltb->map_id = adapter->map_id;
 210         adapter->map_id++;
 211
 212         mutex_lock(&adapter->fw_lock);
 213         adapter->fw_done_rc = 0;
 214         reinit_completion(&adapter->fw_done);
 215
 216         rc = send_request_map(adapter, ltb->addr, ltb->size, ltb->map_id);
 217         if (rc) {
 218                 dev_err(dev, "send_request_map failed, rc = %d\n", rc);
 219                 goto out;
 220         }
 221
 222         rc = ibmvnic_wait_for_completion(adapter, &adapter->fw_done, 10000);
 223         if (rc) {
 224                 dev_err(dev,
 225                         "Long term map request aborted or timed out,rc = %d\n",
 226                         rc);
 227                 goto out;
 228         }
 229
 230         if (adapter->fw_done_rc) {
 231                 dev_err(dev, "Couldn't map long term buffer,rc = %d\n",
 232                         adapter->fw_done_rc);
 233                 rc = -1;
 234                 goto out;
 235         }
 236         rc = 0;
 237 out:
 238         if (rc) {
 239                 dma_free_coherent(dev, ltb->size, ltb->buff, ltb->addr);
 240                 ltb->buff = NULL;
 241         }
 242         mutex_unlock(&adapter->fw_lock);
 243         return rc;
 244 }
 245
 246 static void free_long_term_buff(struct ibmvnic_adapter *adapter,
 247                                 struct ibmvnic_long_term_buff *ltb)
 248 {
 249         struct device *dev = &adapter->vdev->dev;
 250
 251         if (!ltb->buff)
 252                 return;
 253
 254         /* VIOS automatically unmaps the long term buffer at remote
 255          * end for the following resets:
 256          * FAILOVER, MOBILITY, TIMEOUT.
 257          */
 258         if (adapter->reset_reason != VNIC_RESET_FAILOVER &&
 259             adapter->reset_reason != VNIC_RESET_MOBILITY &&
 260             adapter->reset_reason != VNIC_RESET_TIMEOUT)
 261                 send_request_unmap(adapter, ltb->map_id);
 262         dma_free_coherent(dev, ltb->size, ltb->buff, ltb->addr);
 263         ltb->buff = NULL;
 264         ltb->map_id = 0;
 265 }
 266
 267 static int reset_long_term_buff(struct ibmvnic_adapter *adapter,
 268                                 struct ibmvnic_long_term_buff *ltb)
 269 {
 270         struct device *dev = &adapter->vdev->dev;
 271         int rc;
 272
 273         memset(ltb->buff, 0, ltb->size);
 274
 275         mutex_lock(&adapter->fw_lock);
 276         adapter->fw_done_rc = 0;
 277
 278         reinit_completion(&adapter->fw_done);
 279         rc = send_request_map(adapter, ltb->addr, ltb->size, ltb->map_id);
 280         if (rc) {
 281                 mutex_unlock(&adapter->fw_lock);
 282                 return rc;
 283         }
 284
 285         rc = ibmvnic_wait_for_completion(adapter, &adapter->fw_done, 10000);
 286         if (rc) {
 287                 dev_info(dev,
 288                          "Reset failed, long term map request timed out or aborted\n");
 289                 mutex_unlock(&adapter->fw_lock);
 290                 return rc;
 291         }
 292
 293         if (adapter->fw_done_rc) {
 294                 dev_info(dev,
 295                          "Reset failed, attempting to free and reallocate buffer\n");
 296                 free_long_term_buff(adapter, ltb);
 297                 mutex_unlock(&adapter->fw_lock);
 298                 return alloc_long_term_buff(adapter, ltb, ltb->size);
 299         }
 300         mutex_unlock(&adapter->fw_lock);
 301         return 0;
 302 }
 303
 304 static void deactivate_rx_pools(struct ibmvnic_adapter *adapter)
 305 {
 306         int i;
 307
 308         for (i = 0; i < adapter->num_active_rx_pools; i++)
 309                 adapter->rx_pool[i].active = 0;
 310 }
 311
 312 static void replenish_rx_pool(struct ibmvnic_adapter *adapter,
 313                               struct ibmvnic_rx_pool *pool)
 314 {
 315         int count = pool->size - atomic_read(&pool->available);
 316         u64 handle = adapter->rx_scrq[pool->index]->handle;
 317         struct device *dev = &adapter->vdev->dev;
 318         int buffers_added = 0;
 319         unsigned long lpar_rc;
 320         union sub_crq sub_crq;
 321         struct sk_buff *skb;
 322         unsigned int offset;
 323         dma_addr_t dma_addr;
 324         unsigned char *dst;
 325         int shift = 0;
 326         int index;
 327         int i;
 328
 329         if (!pool->active)
 330                 return;
 331
 332         for (i = 0; i < count; ++i) {
 333                 skb = alloc_skb(pool->buff_size, GFP_ATOMIC);
 334                 if (!skb) {
 335                         dev_err(dev, "Couldn't replenish rx buff\n");
 336                         adapter->replenish_no_mem++;
 337                         break;
 338                 }
 339
 340                 index = pool->free_map[pool->next_free];
 341
 342                 if (pool->rx_buff[index].skb)
 343                         dev_err(dev, "Inconsistent free_map!\n");
 344
 345                 /* Copy the skb to the long term mapped DMA buffer */
 346                 offset = index * pool->buff_size;
 347                 dst = pool->long_term_buff.buff + offset;
 348                 memset(dst, 0, pool->buff_size);
 349                 dma_addr = pool->long_term_buff.addr + offset;
 350                 pool->rx_buff[index].data = dst;
 351
 352                 pool->free_map[pool->next_free] = IBMVNIC_INVALID_MAP;
 353                 pool->rx_buff[index].dma = dma_addr;
 354                 pool->rx_buff[index].skb = skb;
 355                 pool->rx_buff[index].pool_index = pool->index;
 356                 pool->rx_buff[index].size = pool->buff_size;
 357
 358                 memset(&sub_crq, 0, sizeof(sub_crq));
 359                 sub_crq.rx_add.first = IBMVNIC_CRQ_CMD;
 360                 sub_crq.rx_add.correlator =
 361                     cpu_to_be64((u64)&pool->rx_buff[index]);
 362                 sub_crq.rx_add.ioba = cpu_to_be32(dma_addr);
 363                 sub_crq.rx_add.map_id = pool->long_term_buff.map_id;
 364
 365                 /* The length field of the sCRQ is defined to be 24 bits so the
 366                  * buffer size needs to be left shifted by a byte before it is
 367                  * converted to big endian to prevent the last byte from being
 368                  * truncated.
 369                  */
 370 #ifdef __LITTLE_ENDIAN__
 371                 shift = 8;
 372 #endif
 373                 sub_crq.rx_add.len = cpu_to_be32(pool->buff_size << shift);
 374
 375                 lpar_rc = send_subcrq(adapter, handle, &sub_crq);
 376                 if (lpar_rc != H_SUCCESS)
 377                         goto failure;
 378
 379                 buffers_added++;
 380                 adapter->replenish_add_buff_success++;
 381                 pool->next_free = (pool->next_free + 1) % pool->size;
 382         }
 383         atomic_add(buffers_added, &pool->available);
 384         return;
 385
 386 failure:
 387         if (lpar_rc != H_PARAMETER && lpar_rc != H_CLOSED)
 388                 dev_err_ratelimited(dev, "rx: replenish packet buffer failed\n");
 389         pool->free_map[pool->next_free] = index;
 390         pool->rx_buff[index].skb = NULL;
 391
 392         dev_kfree_skb_any(skb);
 393         adapter->replenish_add_buff_failure++;
 394         atomic_add(buffers_added, &pool->available);
 395
 396         if (lpar_rc == H_CLOSED || adapter->failover_pending) {
 397                 /* Disable buffer pool replenishment and report carrier off if
 398                  * queue is closed or pending failover.
 399                  * Firmware guarantees that a signal will be sent to the
 400                  * driver, triggering a reset.
 401                  */
 402                 deactivate_rx_pools(adapter);
 403                 netif_carrier_off(adapter->netdev);
 404         }
 405 }
 406
 407 static void replenish_pools(struct ibmvnic_adapter *adapter)
 408 {
 409         int i;
 410
 411         adapter->replenish_task_cycles++;
 412         for (i = 0; i < adapter->num_active_rx_pools; i++) {
 413                 if (adapter->rx_pool[i].active)
 414                         replenish_rx_pool(adapter, &adapter->rx_pool[i]);
 415         }
 416
 417         netdev_dbg(adapter->netdev, "Replenished %d pools\n", i);
 418 }
 419
 420 static void release_stats_buffers(struct ibmvnic_adapter *adapter)
 421 {
 422         kfree(adapter->tx_stats_buffers);
 423         kfree(adapter->rx_stats_buffers);
 424         adapter->tx_stats_buffers = NULL;
 425         adapter->rx_stats_buffers = NULL;
 426 }
 427
 428 static int init_stats_buffers(struct ibmvnic_adapter *adapter)
 429 {
 430         adapter->tx_stats_buffers =
 431                                 kcalloc(IBMVNIC_MAX_QUEUES,
 432                                         sizeof(struct ibmvnic_tx_queue_stats),
 433                                         GFP_KERNEL);
 434         if (!adapter->tx_stats_buffers)
 435                 return -ENOMEM;
 436
 437         adapter->rx_stats_buffers =
 438                                 kcalloc(IBMVNIC_MAX_QUEUES,
 439                                         sizeof(struct ibmvnic_rx_queue_stats),
 440                                         GFP_KERNEL);
 441         if (!adapter->rx_stats_buffers)
 442                 return -ENOMEM;
 443
 444         return 0;
 445 }
 446
 447 static void release_stats_token(struct ibmvnic_adapter *adapter)
 448 {
 449         struct device *dev = &adapter->vdev->dev;
 450
 451         if (!adapter->stats_token)
 452                 return;
 453
 454         dma_unmap_single(dev, adapter->stats_token,
 455                          sizeof(struct ibmvnic_statistics),
 456                          DMA_FROM_DEVICE);
 457         adapter->stats_token = 0;
 458 }
 459
 460 static int init_stats_token(struct ibmvnic_adapter *adapter)
 461 {
 462         struct device *dev = &adapter->vdev->dev;
 463         dma_addr_t stok;
 464
 465         stok = dma_map_single(dev, &adapter->stats,
 466                               sizeof(struct ibmvnic_statistics),
 467                               DMA_FROM_DEVICE);
 468         if (dma_mapping_error(dev, stok)) {
 469                 dev_err(dev, "Couldn't map stats buffer\n");
 470                 return -1;
 471         }
 472
 473         adapter->stats_token = stok;
 474         netdev_dbg(adapter->netdev, "Stats token initialized (%llx)\n", stok);
 475         return 0;
 476 }
 477
 478 static int reset_rx_pools(struct ibmvnic_adapter *adapter)
 479 {
 480         struct ibmvnic_rx_pool *rx_pool;
 481         u64 buff_size;
 482         int rx_scrqs;
 483         int i, j, rc;
 484
 485         if (!adapter->rx_pool)
 486                 return -1;
 487
 488         buff_size = adapter->cur_rx_buf_sz;
 489         rx_scrqs = adapter->num_active_rx_pools;
 490         for (i = 0; i < rx_scrqs; i++) {
 491                 rx_pool = &adapter->rx_pool[i];
 492
 493                 netdev_dbg(adapter->netdev, "Re-setting rx_pool[%d]\n", i);
 494
 495                 if (rx_pool->buff_size != buff_size) {
 496                         free_long_term_buff(adapter, &rx_pool->long_term_buff);
 497                         rx_pool->buff_size = buff_size;
 498                         rc = alloc_long_term_buff(adapter,
 499                                                   &rx_pool->long_term_buff,
 500                                                   rx_pool->size *
 501                                                   rx_pool->buff_size);
 502                 } else {
 503                         rc = reset_long_term_buff(adapter,
 504                                                   &rx_pool->long_term_buff);
 505                 }
 506
 507                 if (rc)
 508                         return rc;
 509
 510                 for (j = 0; j < rx_pool->size; j++)
 511                         rx_pool->free_map[j] = j;
 512
 513                 memset(rx_pool->rx_buff, 0,
 514                        rx_pool->size * sizeof(struct ibmvnic_rx_buff));
 515
 516                 atomic_set(&rx_pool->available, 0);
 517                 rx_pool->next_alloc = 0;
 518                 rx_pool->next_free = 0;
 519                 rx_pool->active = 1;
 520         }
 521
 522         return 0;
 523 }
 524
 525 static void release_rx_pools(struct ibmvnic_adapter *adapter)
 526 {
 527         struct ibmvnic_rx_pool *rx_pool;
 528         int i, j;
 529
 530         if (!adapter->rx_pool)
 531                 return;
 532
 533         for (i = 0; i < adapter->num_active_rx_pools; i++) {
 534                 rx_pool = &adapter->rx_pool[i];
 535
 536                 netdev_dbg(adapter->netdev, "Releasing rx_pool[%d]\n", i);
 537
 538                 kfree(rx_pool->free_map);
 539                 free_long_term_buff(adapter, &rx_pool->long_term_buff);
 540
 541                 if (!rx_pool->rx_buff)
 542                         continue;
 543
 544                 for (j = 0; j < rx_pool->size; j++) {
 545                         if (rx_pool->rx_buff[j].skb) {
 546                                 dev_kfree_skb_any(rx_pool->rx_buff[j].skb);
 547                                 rx_pool->rx_buff[j].skb = NULL;
 548                         }
 549                 }
 550
 551                 kfree(rx_pool->rx_buff);
 552         }
 553
 554         kfree(adapter->rx_pool);
 555         adapter->rx_pool = NULL;
 556         adapter->num_active_rx_pools = 0;
 557 }
 558
 559 static int init_rx_pools(struct net_device *netdev)
 560 {
 561         struct ibmvnic_adapter *adapter = netdev_priv(netdev);
 562         struct device *dev = &adapter->vdev->dev;
 563         struct ibmvnic_rx_pool *rx_pool;
 564         int rxadd_subcrqs;
 565         u64 buff_size;
 566         int i, j;
 567
 568         rxadd_subcrqs = adapter->num_active_rx_scrqs;
 569         buff_size = adapter->cur_rx_buf_sz;
 570
 571         adapter->rx_pool = kcalloc(rxadd_subcrqs,
 572                                    sizeof(struct ibmvnic_rx_pool),
 573                                    GFP_KERNEL);
 574         if (!adapter->rx_pool) {
 575                 dev_err(dev, "Failed to allocate rx pools\n");
 576                 return -1;
 577         }
 578
 579         adapter->num_active_rx_pools = rxadd_subcrqs;
 580
 581         for (i = 0; i < rxadd_subcrqs; i++) {
 582                 rx_pool = &adapter->rx_pool[i];
 583
 584                 netdev_dbg(adapter->netdev,
 585                            "Initializing rx_pool[%d], %lld buffs, %lld bytes each\n",
 586                            i, adapter->req_rx_add_entries_per_subcrq,
 587                            buff_size);
 588
 589                 rx_pool->size = adapter->req_rx_add_entries_per_subcrq;
 590                 rx_pool->index = i;
 591                 rx_pool->buff_size = buff_size;
 592                 rx_pool->active = 1;
 593
 594                 rx_pool->free_map = kcalloc(rx_pool->size, sizeof(int),
 595                                             GFP_KERNEL);
 596                 if (!rx_pool->free_map) {
 597                         release_rx_pools(adapter);
 598                         return -1;
 599                 }
 600
 601                 rx_pool->rx_buff = kcalloc(rx_pool->size,
 602                                            sizeof(struct ibmvnic_rx_buff),
 603                                            GFP_KERNEL);
 604                 if (!rx_pool->rx_buff) {
 605                         dev_err(dev, "Couldn't alloc rx buffers\n");
 606                         release_rx_pools(adapter);
 607                         return -1;
 608                 }
 609
 610                 if (alloc_long_term_buff(adapter, &rx_pool->long_term_buff,
 611                                          rx_pool->size * rx_pool->buff_size)) {
 612                         release_rx_pools(adapter);
 613                         return -1;
 614                 }
 615
 616                 for (j = 0; j < rx_pool->size; ++j)
 617                         rx_pool->free_map[j] = j;
 618
 619                 atomic_set(&rx_pool->available, 0);
 620                 rx_pool->next_alloc = 0;
 621                 rx_pool->next_free = 0;
 622         }
 623
 624         return 0;
 625 }
 626
 627 static int reset_one_tx_pool(struct ibmvnic_adapter *adapter,
 628                              struct ibmvnic_tx_pool *tx_pool)
 629 {
 630         int rc, i;
 631
 632         rc = reset_long_term_buff(adapter, &tx_pool->long_term_buff);
 633         if (rc)
 634                 return rc;
 635
 636         memset(tx_pool->tx_buff, 0,
 637                tx_pool->num_buffers *
 638                sizeof(struct ibmvnic_tx_buff));
 639
 640         for (i = 0; i < tx_pool->num_buffers; i++)
 641                 tx_pool->free_map[i] = i;
 642
 643         tx_pool->consumer_index = 0;
 644         tx_pool->producer_index = 0;
 645
 646         return 0;
 647 }
 648
 649 static int reset_tx_pools(struct ibmvnic_adapter *adapter)
 650 {
 651         int tx_scrqs;
 652         int i, rc;
 653
 654         if (!adapter->tx_pool)
 655                 return -1;
 656
 657         tx_scrqs = adapter->num_active_tx_pools;
 658         for (i = 0; i < tx_scrqs; i++) {
 659                 rc = reset_one_tx_pool(adapter, &adapter->tso_pool[i]);
 660                 if (rc)
 661                         return rc;
 662                 rc = reset_one_tx_pool(adapter, &adapter->tx_pool[i]);
 663                 if (rc)
 664                         return rc;
 665         }
 666
 667         return 0;
 668 }
 669
 670 static void release_vpd_data(struct ibmvnic_adapter *adapter)
 671 {
 672         if (!adapter->vpd)
 673                 return;
 674
 675         kfree(adapter->vpd->buff);
 676         kfree(adapter->vpd);
 677
 678         adapter->vpd = NULL;
 679 }
 680
 681 static void release_one_tx_pool(struct ibmvnic_adapter *adapter,
 682                                 struct ibmvnic_tx_pool *tx_pool)
 683 {
 684         kfree(tx_pool->tx_buff);
 685         kfree(tx_pool->free_map);
 686         free_long_term_buff(adapter, &tx_pool->long_term_buff);
 687 }
 688
 689 static void release_tx_pools(struct ibmvnic_adapter *adapter)
 690 {
 691         int i;
 692
 693         if (!adapter->tx_pool)
 694                 return;
 695
 696         for (i = 0; i < adapter->num_active_tx_pools; i++) {
 697                 release_one_tx_pool(adapter, &adapter->tx_pool[i]);
 698                 release_one_tx_pool(adapter, &adapter->tso_pool[i]);
 699         }
 700
 701         kfree(adapter->tx_pool);
 702         adapter->tx_pool = NULL;
 703         kfree(adapter->tso_pool);
 704         adapter->tso_pool = NULL;
 705         adapter->num_active_tx_pools = 0;
 706 }
 707
 708 static int init_one_tx_pool(struct net_device *netdev,
 709                             struct ibmvnic_tx_pool *tx_pool,
 710                             int num_entries, int buf_size)
 711 {
 712         struct ibmvnic_adapter *adapter = netdev_priv(netdev);
 713         int i;
 714
 715         tx_pool->tx_buff = kcalloc(num_entries,
 716                                    sizeof(struct ibmvnic_tx_buff),
 717                                    GFP_KERNEL);
 718         if (!tx_pool->tx_buff)
 719                 return -1;
 720
 721         if (alloc_long_term_buff(adapter, &tx_pool->long_term_buff,
 722                                  num_entries * buf_size))
 723                 return -1;
 724
 725         tx_pool->free_map = kcalloc(num_entries, sizeof(int), GFP_KERNEL);
 726         if (!tx_pool->free_map)
 727                 return -1;
 728
 729         for (i = 0; i < num_entries; i++)
 730                 tx_pool->free_map[i] = i;
 731
 732         tx_pool->consumer_index = 0;
 733         tx_pool->producer_index = 0;
 734         tx_pool->num_buffers = num_entries;
 735         tx_pool->buf_size = buf_size;
 736
 737         return 0;
 738 }
 739
 740 static int init_tx_pools(struct net_device *netdev)
 741 {
 742         struct ibmvnic_adapter *adapter = netdev_priv(netdev);
 743         int tx_subcrqs;
 744         int i, rc;
 745
 746         tx_subcrqs = adapter->num_active_tx_scrqs;
 747         adapter->tx_pool = kcalloc(tx_subcrqs,
 748                                    sizeof(struct ibmvnic_tx_pool), GFP_KERNEL);
 749         if (!adapter->tx_pool)
 750                 return -1;
 751
 752         adapter->tso_pool = kcalloc(tx_subcrqs,
 753                                     sizeof(struct ibmvnic_tx_pool), GFP_KERNEL);
 754         if (!adapter->tso_pool) {
 755                 kfree(adapter->tx_pool);
 756                 adapter->tx_pool = NULL;
 757                 return -1;
 758         }
 759
 760         adapter->num_active_tx_pools = tx_subcrqs;
 761
 762         for (i = 0; i < tx_subcrqs; i++) {
 763                 rc = init_one_tx_pool(netdev, &adapter->tx_pool[i],
 764                                       adapter->req_tx_entries_per_subcrq,
 765                                       adapter->req_mtu + VLAN_HLEN);
 766                 if (rc) {
 767                         release_tx_pools(adapter);
 768                         return rc;
 769                 }
 770
 771                 rc = init_one_tx_pool(netdev, &adapter->tso_pool[i],
 772                                       IBMVNIC_TSO_BUFS,
 773                                       IBMVNIC_TSO_BUF_SZ);
 774                 if (rc) {
 775                         release_tx_pools(adapter);
 776                         return rc;
 777                 }
 778         }
 779
 780         return 0;
 781 }
 782
 783 static void ibmvnic_napi_enable(struct ibmvnic_adapter *adapter)
 784 {
 785         int i;
 786
 787         if (adapter->napi_enabled)
 788                 return;
 789
 790         for (i = 0; i < adapter->req_rx_queues; i++)
 791                 napi_enable(&adapter->napi[i]);
 792
 793         adapter->napi_enabled = true;
 794 }
 795
 796 static void ibmvnic_napi_disable(struct ibmvnic_adapter *adapter)
 797 {
 798         int i;
 799
 800         if (!adapter->napi_enabled)
 801                 return;
 802
 803         for (i = 0; i < adapter->req_rx_queues; i++) {
 804                 netdev_dbg(adapter->netdev, "Disabling napi[%d]\n", i);
 805                 napi_disable(&adapter->napi[i]);
 806         }
 807
 808         adapter->napi_enabled = false;
 809 }
 810
 811 static int init_napi(struct ibmvnic_adapter *adapter)
 812 {
 813         int i;
 814
 815         adapter->napi = kcalloc(adapter->req_rx_queues,
 816                                 sizeof(struct napi_struct), GFP_KERNEL);
 817         if (!adapter->napi)
 818                 return -ENOMEM;
 819
 820         for (i = 0; i < adapter->req_rx_queues; i++) {
 821                 netdev_dbg(adapter->netdev, "Adding napi[%d]\n", i);
 822                 netif_napi_add(adapter->netdev, &adapter->napi[i],
 823                                ibmvnic_poll, NAPI_POLL_WEIGHT);
 824         }
 825
 826         adapter->num_active_rx_napi = adapter->req_rx_queues;
 827         return 0;
 828 }
 829
 830 static void release_napi(struct ibmvnic_adapter *adapter)
 831 {
 832         int i;
 833
 834         if (!adapter->napi)
 835                 return;
 836
 837         for (i = 0; i < adapter->num_active_rx_napi; i++) {
 838                 netdev_dbg(adapter->netdev, "Releasing napi[%d]\n", i);
 839                 netif_napi_del(&adapter->napi[i]);
 840         }
 841
 842         kfree(adapter->napi);
 843         adapter->napi = NULL;
 844         adapter->num_active_rx_napi = 0;
 845         adapter->napi_enabled = false;
 846 }
 847
 848 static int ibmvnic_login(struct net_device *netdev)
 849 {
 850         struct ibmvnic_adapter *adapter = netdev_priv(netdev);
 851         unsigned long timeout = msecs_to_jiffies(20000);
 852         int retry_count = 0;
 853         int retries = 10;
 854         bool retry;
 855         int rc;
 856
 857         do {
 858                 retry = false;
 859                 if (retry_count > retries) {
 860                         netdev_warn(netdev, "Login attempts exceeded\n");
 861                         return -1;
 862                 }
 863
 864                 adapter->init_done_rc = 0;
 865                 reinit_completion(&adapter->init_done);
 866                 rc = send_login(adapter);
 867                 if (rc)
 868                         return rc;
 869
 870                 if (!wait_for_completion_timeout(&adapter->init_done,
 871                                                  timeout)) {
 872                         netdev_warn(netdev, "Login timed out, retrying...\n");
 873                         retry = true;
 874                         adapter->init_done_rc = 0;
 875                         retry_count++;
 876                         continue;
 877                 }
 878
 879                 if (adapter->init_done_rc == ABORTED) {
 880                         netdev_warn(netdev, "Login aborted, retrying...\n");
 881                         retry = true;
 882                         adapter->init_done_rc = 0;
 883                         retry_count++;
 884                         /* FW or device may be busy, so
 885                          * wait a bit before retrying login
 886                          */
 887                         msleep(500);
 888                 } else if (adapter->init_done_rc == PARTIALSUCCESS) {
 889                         retry_count++;
 890                         release_sub_crqs(adapter, 1);
 891
 892                         retry = true;
 893                         netdev_dbg(netdev,
 894                                    "Received partial success, retrying...\n");
 895                         adapter->init_done_rc = 0;
 896                         reinit_completion(&adapter->init_done);
 897                         send_query_cap(adapter);
 898                         if (!wait_for_completion_timeout(&adapter->init_done,
 899                                                          timeout)) {
 900                                 netdev_warn(netdev,
 901                                             "Capabilities query timed out\n");
 902                                 return -1;
 903                         }
 904
 905                         rc = init_sub_crqs(adapter);
 906                         if (rc) {
 907                                 netdev_warn(netdev,
 908                                             "SCRQ initialization failed\n");
 909                                 return -1;
 910                         }
 911
 912                         rc = init_sub_crq_irqs(adapter);
 913                         if (rc) {
 914                                 netdev_warn(netdev,
 915                                             "SCRQ irq initialization failed\n");
 916                                 return -1;
 917                         }
 918                 } else if (adapter->init_done_rc) {
 919                         netdev_warn(netdev, "Adapter login failed\n");
 920                         return -1;
 921                 }
 922         } while (retry);
 923
 924         __ibmvnic_set_mac(netdev, adapter->mac_addr);
 925
 926         netdev_dbg(netdev, "[S:%d] Login succeeded\n", adapter->state);
 927         return 0;
 928 }
 929
 930 static void release_login_buffer(struct ibmvnic_adapter *adapter)
 931 {
 932         kfree(adapter->login_buf);
 933         adapter->login_buf = NULL;
 934 }
 935
 936 static void release_login_rsp_buffer(struct ibmvnic_adapter *adapter)
 937 {
 938         kfree(adapter->login_rsp_buf);
 939         adapter->login_rsp_buf = NULL;
 940 }
 941
 942 static void release_resources(struct ibmvnic_adapter *adapter)
 943 {
 944         release_vpd_data(adapter);
 945
 946         release_tx_pools(adapter);
 947         release_rx_pools(adapter);
 948
 949         release_napi(adapter);
 950         release_login_buffer(adapter);
 951         release_login_rsp_buffer(adapter);
 952 }
 953
 954 static int set_link_state(struct ibmvnic_adapter *adapter, u8 link_state)
 955 {
 956         struct net_device *netdev = adapter->netdev;
 957         unsigned long timeout = msecs_to_jiffies(20000);
 958         union ibmvnic_crq crq;
 959         bool resend;
 960         int rc;
 961
 962         netdev_dbg(netdev, "setting link state %d\n", link_state);
 963
 964         memset(&crq, 0, sizeof(crq));
 965         crq.logical_link_state.first = IBMVNIC_CRQ_CMD;
 966         crq.logical_link_state.cmd = LOGICAL_LINK_STATE;
 967         crq.logical_link_state.link_state = link_state;
 968
 969         do {
 970                 resend = false;
 971
 972                 reinit_completion(&adapter->init_done);
 973                 rc = ibmvnic_send_crq(adapter, &crq);
 974                 if (rc) {
 975                         netdev_err(netdev, "Failed to set link state\n");
 976                         return rc;
 977                 }
 978
 979                 if (!wait_for_completion_timeout(&adapter->init_done,
 980                                                  timeout)) {
 981                         netdev_err(netdev, "timeout setting link state\n");
 982                         return -1;
 983                 }
 984
 985                 if (adapter->init_done_rc == PARTIALSUCCESS) {
 986                         /* Partuial success, delay and re-send */
 987                         mdelay(1000);
 988                         resend = true;
 989                 } else if (adapter->init_done_rc) {
 990                         netdev_warn(netdev, "Unable to set link state, rc=%d\n",
 991                                     adapter->init_done_rc);
 992                         return adapter->init_done_rc;
 993                 }
 994         } while (resend);
 995
 996         return 0;
 997 }
 998
 999 static int set_real_num_queues(struct net_device *netdev)
1000 {
1001         struct ibmvnic_adapter *adapter = netdev_priv(netdev);
1002         int rc;
1003
1004         netdev_dbg(netdev, "Setting real tx/rx queues (%llx/%llx)\n",
1005                    adapter->req_tx_queues, adapter->req_rx_queues);
1006
1007         rc = netif_set_real_num_tx_queues(netdev, adapter->req_tx_queues);
1008         if (rc) {
1009                 netdev_err(netdev, "failed to set the number of tx queues\n");
1010                 return rc;
1011         }
1012
1013         rc = netif_set_real_num_rx_queues(netdev, adapter->req_rx_queues);
1014         if (rc)
1015                 netdev_err(netdev, "failed to set the number of rx queues\n");
1016
1017         return rc;
1018 }
1019
1020 static int ibmvnic_get_vpd(struct ibmvnic_adapter *adapter)
1021 {
1022         struct device *dev = &adapter->vdev->dev;
1023         union ibmvnic_crq crq;
1024         int len = 0;
1025         int rc;
1026
1027         if (adapter->vpd->buff)
1028                 len = adapter->vpd->len;
1029
1030         mutex_lock(&adapter->fw_lock);
1031         adapter->fw_done_rc = 0;
1032         reinit_completion(&adapter->fw_done);
1033
1034         crq.get_vpd_size.first = IBMVNIC_CRQ_CMD;
1035         crq.get_vpd_size.cmd = GET_VPD_SIZE;
1036         rc = ibmvnic_send_crq(adapter, &crq);
1037         if (rc) {
1038                 mutex_unlock(&adapter->fw_lock);
1039                 return rc;
1040         }
1041
1042         rc = ibmvnic_wait_for_completion(adapter, &adapter->fw_done, 10000);
1043         if (rc) {
1044                 dev_err(dev, "Could not retrieve VPD size, rc = %d\n", rc);
1045                 mutex_unlock(&adapter->fw_lock);
1046                 return rc;
1047         }
1048         mutex_unlock(&adapter->fw_lock);
1049
1050         if (!adapter->vpd->len)
1051                 return -ENODATA;
1052
1053         if (!adapter->vpd->buff)
1054                 adapter->vpd->buff = kzalloc(adapter->vpd->len, GFP_KERNEL);
1055         else if (adapter->vpd->len != len)
1056                 adapter->vpd->buff =
1057                         krealloc(adapter->vpd->buff,
1058                                  adapter->vpd->len, GFP_KERNEL);
1059
1060         if (!adapter->vpd->buff) {
1061                 dev_err(dev, "Could allocate VPD buffer\n");
1062                 return -ENOMEM;
1063         }
1064
1065         adapter->vpd->dma_addr =
1066                 dma_map_single(dev, adapter->vpd->buff, adapter->vpd->len,
1067                                DMA_FROM_DEVICE);
1068         if (dma_mapping_error(dev, adapter->vpd->dma_addr)) {
1069                 dev_err(dev, "Could not map VPD buffer\n");
1070                 kfree(adapter->vpd->buff);
1071                 adapter->vpd->buff = NULL;
1072                 return -ENOMEM;
1073         }
1074
1075         mutex_lock(&adapter->fw_lock);
1076         adapter->fw_done_rc = 0;
1077         reinit_completion(&adapter->fw_done);
1078
1079         crq.get_vpd.first = IBMVNIC_CRQ_CMD;
1080         crq.get_vpd.cmd = GET_VPD;
1081         crq.get_vpd.ioba = cpu_to_be32(adapter->vpd->dma_addr);
1082         crq.get_vpd.len = cpu_to_be32((u32)adapter->vpd->len);
1083         rc = ibmvnic_send_crq(adapter, &crq);
1084         if (rc) {
1085                 kfree(adapter->vpd->buff);
1086                 adapter->vpd->buff = NULL;
1087                 mutex_unlock(&adapter->fw_lock);
1088                 return rc;
1089         }
1090
1091         rc = ibmvnic_wait_for_completion(adapter, &adapter->fw_done, 10000);
1092         if (rc) {
1093                 dev_err(dev, "Unable to retrieve VPD, rc = %d\n", rc);
1094                 kfree(adapter->vpd->buff);
1095                 adapter->vpd->buff = NULL;
1096                 mutex_unlock(&adapter->fw_lock);
1097                 return rc;
1098         }
1099
1100         mutex_unlock(&adapter->fw_lock);
1101         return 0;
1102 }
1103
1104 static int init_resources(struct ibmvnic_adapter *adapter)
1105 {
1106         struct net_device *netdev = adapter->netdev;
1107         int rc;
1108
1109         rc = set_real_num_queues(netdev);
1110         if (rc)
1111                 return rc;
1112
1113         adapter->vpd = kzalloc(sizeof(*adapter->vpd), GFP_KERNEL);
1114         if (!adapter->vpd)
1115                 return -ENOMEM;
1116
1117         /* Vital Product Data (VPD) */
1118         rc = ibmvnic_get_vpd(adapter);
1119         if (rc) {
1120                 netdev_err(netdev, "failed to initialize Vital Product Data (VPD)\n");
1121                 return rc;
1122         }
1123
1124         adapter->map_id = 1;
1125
1126         rc = init_napi(adapter);
1127         if (rc)
1128                 return rc;
1129
1130         send_query_map(adapter);
1131
1132         rc = init_rx_pools(netdev);
1133         if (rc)
1134                 return rc;
1135
1136         rc = init_tx_pools(netdev);
1137         return rc;
1138 }
1139
1140 static int __ibmvnic_open(struct net_device *netdev)
1141 {
1142         struct ibmvnic_adapter *adapter = netdev_priv(netdev);
1143         enum vnic_state prev_state = adapter->state;
1144         int i, rc;
1145
1146         adapter->state = VNIC_OPENING;
1147         replenish_pools(adapter);
1148         ibmvnic_napi_enable(adapter);
1149
1150         /* We're ready to receive frames, enable the sub-crq interrupts and
1151          * set the logical link state to up
1152          */
1153         for (i = 0; i < adapter->req_rx_queues; i++) {
1154                 netdev_dbg(netdev, "Enabling rx_scrq[%d] irq\n", i);
1155                 if (prev_state == VNIC_CLOSED)
1156                         enable_irq(adapter->rx_scrq[i]->irq);
1157                 enable_scrq_irq(adapter, adapter->rx_scrq[i]);
1158         }
1159
1160         for (i = 0; i < adapter->req_tx_queues; i++) {
1161                 netdev_dbg(netdev, "Enabling tx_scrq[%d] irq\n", i);
1162                 if (prev_state == VNIC_CLOSED)
1163                         enable_irq(adapter->tx_scrq[i]->irq);
1164                 enable_scrq_irq(adapter, adapter->tx_scrq[i]);
1165         }
1166
1167         rc = set_link_state(adapter, IBMVNIC_LOGICAL_LNK_UP);
1168         if (rc) {
1169                 ibmvnic_napi_disable(adapter);
1170                 release_resources(adapter);
1171                 return rc;
1172         }
1173
1174         netif_tx_start_all_queues(netdev);
1175
1176         if (prev_state == VNIC_CLOSED) {
1177                 for (i = 0; i < adapter->req_rx_queues; i++)
1178                         napi_schedule(&adapter->napi[i]);
1179         }
1180
1181         adapter->state = VNIC_OPEN;
1182         return rc;
1183 }
1184
1185 static int ibmvnic_open(struct net_device *netdev)
1186 {
1187         struct ibmvnic_adapter *adapter = netdev_priv(netdev);
1188         int rc;
1189
1190         /* If device failover is pending, just set device state and return.
1191          * Device operation will be handled by reset routine.
1192          */
1193         if (adapter->failover_pending) {
1194                 adapter->state = VNIC_OPEN;
1195                 return 0;
1196         }
1197
1198         if (adapter->state != VNIC_CLOSED) {
1199                 rc = ibmvnic_login(netdev);
1200                 if (rc)
1201                         goto out;
1202
1203                 rc = init_resources(adapter);
1204                 if (rc) {
1205                         netdev_err(netdev, "failed to initialize resources\n");
1206                         release_resources(adapter);
1207                         goto out;
1208                 }
1209         }
1210
1211         rc = __ibmvnic_open(netdev);
1212
1213 out:
1214         /*
1215          * If open fails due to a pending failover, set device state and
1216          * return. Device operation will be handled by reset routine.
1217          */
1218         if (rc && adapter->failover_pending) {
1219                 adapter->state = VNIC_OPEN;
1220                 rc = 0;
1221         }
1222         return rc;
1223 }
1224
1225 static void clean_rx_pools(struct ibmvnic_adapter *adapter)
1226 {
1227         struct ibmvnic_rx_pool *rx_pool;
1228         struct ibmvnic_rx_buff *rx_buff;
1229         u64 rx_entries;
1230         int rx_scrqs;
1231         int i, j;
1232
1233         if (!adapter->rx_pool)
1234                 return;
1235
1236         rx_scrqs = adapter->num_active_rx_pools;
1237         rx_entries = adapter->req_rx_add_entries_per_subcrq;
1238
1239         /* Free any remaining skbs in the rx buffer pools */
1240         for (i = 0; i < rx_scrqs; i++) {
1241                 rx_pool = &adapter->rx_pool[i];
1242                 if (!rx_pool || !rx_pool->rx_buff)
1243                         continue;
1244
1245                 netdev_dbg(adapter->netdev, "Cleaning rx_pool[%d]\n", i);
1246                 for (j = 0; j < rx_entries; j++) {
1247                         rx_buff = &rx_pool->rx_buff[j];
1248                         if (rx_buff && rx_buff->skb) {
1249                                 dev_kfree_skb_any(rx_buff->skb);
1250                                 rx_buff->skb = NULL;
1251                         }
1252                 }
1253         }
1254 }
1255
1256 static void clean_one_tx_pool(struct ibmvnic_adapter *adapter,
1257                               struct ibmvnic_tx_pool *tx_pool)
1258 {
1259         struct ibmvnic_tx_buff *tx_buff;
1260         u64 tx_entries;
1261         int i;
1262
1263         if (!tx_pool || !tx_pool->tx_buff)
1264                 return;
1265
1266         tx_entries = tx_pool->num_buffers;
1267
1268         for (i = 0; i < tx_entries; i++) {
1269                 tx_buff = &tx_pool->tx_buff[i];
1270                 if (tx_buff && tx_buff->skb) {
1271                         dev_kfree_skb_any(tx_buff->skb);
1272                         tx_buff->skb = NULL;
1273                 }
1274         }
1275 }
1276
1277 static void clean_tx_pools(struct ibmvnic_adapter *adapter)
1278 {
1279         int tx_scrqs;
1280         int i;
1281
1282         if (!adapter->tx_pool || !adapter->tso_pool)
1283                 return;
1284
1285         tx_scrqs = adapter->num_active_tx_pools;
1286
1287         /* Free any remaining skbs in the tx buffer pools */
1288         for (i = 0; i < tx_scrqs; i++) {
1289                 netdev_dbg(adapter->netdev, "Cleaning tx_pool[%d]\n", i);
1290                 clean_one_tx_pool(adapter, &adapter->tx_pool[i]);
1291                 clean_one_tx_pool(adapter, &adapter->tso_pool[i]);
1292         }
1293 }
1294
1295 static void ibmvnic_disable_irqs(struct ibmvnic_adapter *adapter)
1296 {
1297         struct net_device *netdev = adapter->netdev;
1298         int i;
1299
1300         if (adapter->tx_scrq) {
1301                 for (i = 0; i < adapter->req_tx_queues; i++)
1302                         if (adapter->tx_scrq[i]->irq) {
1303                                 netdev_dbg(netdev,
1304                                            "Disabling tx_scrq[%d] irq\n", i);
1305                                 disable_scrq_irq(adapter, adapter->tx_scrq[i]);
1306                                 disable_irq(adapter->tx_scrq[i]->irq);
1307                         }
1308         }
1309
1310         if (adapter->rx_scrq) {
1311                 for (i = 0; i < adapter->req_rx_queues; i++) {
1312                         if (adapter->rx_scrq[i]->irq) {
1313                                 netdev_dbg(netdev,
1314                                            "Disabling rx_scrq[%d] irq\n", i);
1315                                 disable_scrq_irq(adapter, adapter->rx_scrq[i]);
1316                                 disable_irq(adapter->rx_scrq[i]->irq);
1317                         }
1318                 }
1319         }
1320 }
1321
1322 static void ibmvnic_cleanup(struct net_device *netdev)
1323 {
1324         struct ibmvnic_adapter *adapter = netdev_priv(netdev);
1325
1326         /* ensure that transmissions are stopped if called by do_reset */
1327         if (test_bit(0, &adapter->resetting))
1328                 netif_tx_disable(netdev);
1329         else
1330                 netif_tx_stop_all_queues(netdev);
1331
1332         ibmvnic_napi_disable(adapter);
1333         ibmvnic_disable_irqs(adapter);
1334
1335         clean_rx_pools(adapter);
1336         clean_tx_pools(adapter);
1337 }
1338
1339 static int __ibmvnic_close(struct net_device *netdev)
1340 {
1341         struct ibmvnic_adapter *adapter = netdev_priv(netdev);
1342         int rc = 0;
1343
1344         adapter->state = VNIC_CLOSING;
1345         rc = set_link_state(adapter, IBMVNIC_LOGICAL_LNK_DN);
1346         adapter->state = VNIC_CLOSED;
1347         return rc;
1348 }
1349
1350 static int ibmvnic_close(struct net_device *netdev)
1351 {
1352         struct ibmvnic_adapter *adapter = netdev_priv(netdev);
1353         int rc;
1354
1355         netdev_dbg(netdev, "[S:%d FOP:%d FRR:%d] Closing\n",
1356                    adapter->state, adapter->failover_pending,
1357                    adapter->force_reset_recovery);
1358
1359         /* If device failover is pending, just set device state and return.
1360          * Device operation will be handled by reset routine.
1361          */
1362         if (adapter->failover_pending) {
1363                 adapter->state = VNIC_CLOSED;
1364                 return 0;
1365         }
1366
1367         rc = __ibmvnic_close(netdev);
1368         ibmvnic_cleanup(netdev);
1369
1370         return rc;
1371 }
1372
1373 /**
1374  * build_hdr_data - creates L2/L3/L4 header data buffer
1375  * @hdr_field - bitfield determining needed headers
1376  * @skb - socket buffer
1377  * @hdr_len - array of header lengths
1378  * @tot_len - total length of data
1379  *
1380  * Reads hdr_field to determine which headers are needed by firmware.
1381  * Builds a buffer containing these headers.  Saves individual header
1382  * lengths and total buffer length to be used to build descriptors.
1383  */
1384 static int build_hdr_data(u8 hdr_field, struct sk_buff *skb,
1385                           int *hdr_len, u8 *hdr_data)
1386 {
1387         int len = 0;
1388         u8 *hdr;
1389
1390         if (skb_vlan_tagged(skb) && !skb_vlan_tag_present(skb))
1391                 hdr_len[0] = sizeof(struct vlan_ethhdr);
1392         else
1393                 hdr_len[0] = sizeof(struct ethhdr);
1394
1395         if (skb->protocol == htons(ETH_P_IP)) {
1396                 hdr_len[1] = ip_hdr(skb)->ihl * 4;
1397                 if (ip_hdr(skb)->protocol == IPPROTO_TCP)
1398                         hdr_len[2] = tcp_hdrlen(skb);
1399                 else if (ip_hdr(skb)->protocol == IPPROTO_UDP)
1400                         hdr_len[2] = sizeof(struct udphdr);
1401         } else if (skb->protocol == htons(ETH_P_IPV6)) {
1402                 hdr_len[1] = sizeof(struct ipv6hdr);
1403                 if (ipv6_hdr(skb)->nexthdr == IPPROTO_TCP)
1404                         hdr_len[2] = tcp_hdrlen(skb);
1405                 else if (ipv6_hdr(skb)->nexthdr == IPPROTO_UDP)
1406                         hdr_len[2] = sizeof(struct udphdr);
1407         } else if (skb->protocol == htons(ETH_P_ARP)) {
1408                 hdr_len[1] = arp_hdr_len(skb->dev);
1409                 hdr_len[2] = 0;
1410         }
1411
1412         memset(hdr_data, 0, 120);
1413         if ((hdr_field >> 6) & 1) {
1414                 hdr = skb_mac_header(skb);
1415                 memcpy(hdr_data, hdr, hdr_len[0]);
1416                 len += hdr_len[0];
1417         }
1418
1419         if ((hdr_field >> 5) & 1) {
1420                 hdr = skb_network_header(skb);
1421                 memcpy(hdr_data + len, hdr, hdr_len[1]);
1422                 len += hdr_len[1];
1423         }
1424
1425         if ((hdr_field >> 4) & 1) {
1426                 hdr = skb_transport_header(skb);
1427                 memcpy(hdr_data + len, hdr, hdr_len[2]);
1428                 len += hdr_len[2];
1429         }
1430         return len;
1431 }
1432
1433 /**
1434  * create_hdr_descs - create header and header extension descriptors
1435  * @hdr_field - bitfield determining needed headers
1436  * @data - buffer containing header data
1437  * @len - length of data buffer
1438  * @hdr_len - array of individual header lengths
1439  * @scrq_arr - descriptor array
1440  *
1441  * Creates header and, if needed, header extension descriptors and
1442  * places them in a descriptor array, scrq_arr
1443  */
1444
1445 static int create_hdr_descs(u8 hdr_field, u8 *hdr_data, int len, int *hdr_len,
1446                             union sub_crq *scrq_arr)
1447 {
1448         union sub_crq hdr_desc;
1449         int tmp_len = len;
1450         int num_descs = 0;
1451         u8 *data, *cur;
1452         int tmp;
1453
1454         while (tmp_len > 0) {
1455                 cur = hdr_data + len - tmp_len;
1456
1457                 memset(&hdr_desc, 0, sizeof(hdr_desc));
1458                 if (cur != hdr_data) {
1459                         data = hdr_desc.hdr_ext.data;
1460                         tmp = tmp_len > 29 ? 29 : tmp_len;
1461                         hdr_desc.hdr_ext.first = IBMVNIC_CRQ_CMD;
1462                         hdr_desc.hdr_ext.type = IBMVNIC_HDR_EXT_DESC;
1463                         hdr_desc.hdr_ext.len = tmp;
1464                 } else {
1465                         data = hdr_desc.hdr.data;
1466                         tmp = tmp_len > 24 ? 24 : tmp_len;
1467                         hdr_desc.hdr.first = IBMVNIC_CRQ_CMD;
1468                         hdr_desc.hdr.type = IBMVNIC_HDR_DESC;
1469                         hdr_desc.hdr.len = tmp;
1470                         hdr_desc.hdr.l2_len = (u8)hdr_len[0];
1471                         hdr_desc.hdr.l3_len = cpu_to_be16((u16)hdr_len[1]);
1472                         hdr_desc.hdr.l4_len = (u8)hdr_len[2];
1473                         hdr_desc.hdr.flag = hdr_field << 1;
1474                 }
1475                 memcpy(data, cur, tmp);
1476                 tmp_len -= tmp;
1477                 *scrq_arr = hdr_desc;
1478                 scrq_arr++;
1479                 num_descs++;
1480         }
1481
1482         return num_descs;
1483 }
1484
1485 /**
1486  * build_hdr_descs_arr - build a header descriptor array
1487  * @skb - socket buffer
1488  * @num_entries - number of descriptors to be sent
1489  * @subcrq - first TX descriptor
1490  * @hdr_field - bit field determining which headers will be sent
1491  *
1492  * This function will build a TX descriptor array with applicable
1493  * L2/L3/L4 packet header descriptors to be sent by send_subcrq_indirect.
1494  */
1495
1496 static void build_hdr_descs_arr(struct ibmvnic_tx_buff *txbuff,
1497                                 int *num_entries, u8 hdr_field)
1498 {
1499         int hdr_len[3] = {0, 0, 0};
1500         int tot_len;
1501         u8 *hdr_data = txbuff->hdr_data;
1502
1503         tot_len = build_hdr_data(hdr_field, txbuff->skb, hdr_len,
1504                                  txbuff->hdr_data);
1505         *num_entries += create_hdr_descs(hdr_field, hdr_data, tot_len, hdr_len,
1506                          txbuff->indir_arr + 1);
1507 }
1508
1509 static int ibmvnic_xmit_workarounds(struct sk_buff *skb,
1510                                     struct net_device *netdev)
1511 {
1512         /* For some backing devices, mishandling of small packets
1513          * can result in a loss of connection or TX stall. Device
1514          * architects recommend that no packet should be smaller
1515          * than the minimum MTU value provided to the driver, so
1516          * pad any packets to that length
1517          */
1518         if (skb->len < netdev->min_mtu)
1519                 return skb_put_padto(skb, netdev->min_mtu);
1520
1521         return 0;
1522 }
1523
1524 static netdev_tx_t ibmvnic_xmit(struct sk_buff *skb, struct net_device *netdev)
1525 {
1526         struct ibmvnic_adapter *adapter = netdev_priv(netdev);
1527         int queue_num = skb_get_queue_mapping(skb);
1528         u8 *hdrs = (u8 *)&adapter->tx_rx_desc_req;
1529         struct device *dev = &adapter->vdev->dev;
1530         struct ibmvnic_tx_buff *tx_buff = NULL;
1531         struct ibmvnic_sub_crq_queue *tx_scrq;
1532         struct ibmvnic_tx_pool *tx_pool;
1533         unsigned int tx_send_failed = 0;
1534         unsigned int tx_map_failed = 0;
1535         unsigned int tx_dropped = 0;
1536         unsigned int tx_packets = 0;
1537         unsigned int tx_bytes = 0;
1538         dma_addr_t data_dma_addr;
1539         struct netdev_queue *txq;
1540         unsigned long lpar_rc;
1541         union sub_crq tx_crq;
1542         unsigned int offset;
1543         int num_entries = 1;
1544         unsigned char *dst;
1545         int index = 0;
1546         u8 proto = 0;
1547         u64 handle;
1548         netdev_tx_t ret = NETDEV_TX_OK;
1549
1550         if (test_bit(0, &adapter->resetting)) {
1551                 dev_kfree_skb_any(skb);
1552
1553                 tx_send_failed++;
1554                 tx_dropped++;
1555                 ret = NETDEV_TX_OK;
1556                 goto out;
1557         }
1558
1559         if (ibmvnic_xmit_workarounds(skb, netdev)) {
1560                 tx_dropped++;
1561                 tx_send_failed++;
1562                 ret = NETDEV_TX_OK;
1563                 goto out;
1564         }
1565         if (skb_is_gso(skb))
1566                 tx_pool = &adapter->tso_pool[queue_num];
1567         else
1568                 tx_pool = &adapter->tx_pool[queue_num];
1569
1570         tx_scrq = adapter->tx_scrq[queue_num];
1571         txq = netdev_get_tx_queue(netdev, skb_get_queue_mapping(skb));
1572         handle = tx_scrq->handle;
1573
1574         index = tx_pool->free_map[tx_pool->consumer_index];
1575
1576         if (index == IBMVNIC_INVALID_MAP) {
1577                 dev_kfree_skb_any(skb);
1578                 tx_send_failed++;
1579                 tx_dropped++;
1580                 ret = NETDEV_TX_OK;
1581                 goto out;
1582         }
1583
1584         tx_pool->free_map[tx_pool->consumer_index] = IBMVNIC_INVALID_MAP;
1585
1586         offset = index * tx_pool->buf_size;
1587         dst = tx_pool->long_term_buff.buff + offset;
1588         memset(dst, 0, tx_pool->buf_size);
1589         data_dma_addr = tx_pool->long_term_buff.addr + offset;
1590
1591         if (skb_shinfo(skb)->nr_frags) {
1592                 int cur, i;
1593
1594                 /* Copy the head */
1595                 skb_copy_from_linear_data(skb, dst, skb_headlen(skb));
1596                 cur = skb_headlen(skb);
1597
1598                 /* Copy the frags */
1599                 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
1600                         const skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
1601
1602                         memcpy(dst + cur,
1603                                page_address(skb_frag_page(frag)) +
1604                                skb_frag_off(frag), skb_frag_size(frag));
1605                         cur += skb_frag_size(frag);
1606                 }
1607         } else {
1608                 skb_copy_from_linear_data(skb, dst, skb->len);
1609         }
1610
1611         /* post changes to long_term_buff *dst before VIOS accessing it */
1612         dma_wmb();
1613
1614         tx_pool->consumer_index =
1615             (tx_pool->consumer_index + 1) % tx_pool->num_buffers;
1616
1617         tx_buff = &tx_pool->tx_buff[index];
1618         tx_buff->skb = skb;
1619         tx_buff->data_dma[0] = data_dma_addr;
1620         tx_buff->data_len[0] = skb->len;
1621         tx_buff->index = index;
1622         tx_buff->pool_index = queue_num;
1623         tx_buff->last_frag = true;
1624
1625         memset(&tx_crq, 0, sizeof(tx_crq));
1626         tx_crq.v1.first = IBMVNIC_CRQ_CMD;
1627         tx_crq.v1.type = IBMVNIC_TX_DESC;
1628         tx_crq.v1.n_crq_elem = 1;
1629         tx_crq.v1.n_sge = 1;
1630         tx_crq.v1.flags1 = IBMVNIC_TX_COMP_NEEDED;
1631
1632         if (skb_is_gso(skb))
1633                 tx_crq.v1.correlator =
1634                         cpu_to_be32(index | IBMVNIC_TSO_POOL_MASK);
1635         else
1636                 tx_crq.v1.correlator = cpu_to_be32(index);
1637         tx_crq.v1.dma_reg = cpu_to_be16(tx_pool->long_term_buff.map_id);
1638         tx_crq.v1.sge_len = cpu_to_be32(skb->len);
1639         tx_crq.v1.ioba = cpu_to_be64(data_dma_addr);
1640
1641         if (adapter->vlan_header_insertion && skb_vlan_tag_present(skb)) {
1642                 tx_crq.v1.flags2 |= IBMVNIC_TX_VLAN_INSERT;
1643                 tx_crq.v1.vlan_id = cpu_to_be16(skb->vlan_tci);
1644         }
1645
1646         if (skb->protocol == htons(ETH_P_IP)) {
1647                 tx_crq.v1.flags1 |= IBMVNIC_TX_PROT_IPV4;
1648                 proto = ip_hdr(skb)->protocol;
1649         } else if (skb->protocol == htons(ETH_P_IPV6)) {
1650                 tx_crq.v1.flags1 |= IBMVNIC_TX_PROT_IPV6;
1651                 proto = ipv6_hdr(skb)->nexthdr;
1652         }
1653
1654         if (proto == IPPROTO_TCP)
1655                 tx_crq.v1.flags1 |= IBMVNIC_TX_PROT_TCP;
1656         else if (proto == IPPROTO_UDP)
1657                 tx_crq.v1.flags1 |= IBMVNIC_TX_PROT_UDP;
1658
1659         if (skb->ip_summed == CHECKSUM_PARTIAL) {
1660                 tx_crq.v1.flags1 |= IBMVNIC_TX_CHKSUM_OFFLOAD;
1661                 hdrs += 2;
1662         }
1663         if (skb_is_gso(skb)) {
1664                 tx_crq.v1.flags1 |= IBMVNIC_TX_LSO;
1665                 tx_crq.v1.mss = cpu_to_be16(skb_shinfo(skb)->gso_size);
1666                 hdrs += 2;
1667         }
1668         /* determine if l2/3/4 headers are sent to firmware */
1669         if ((*hdrs >> 7) & 1) {
1670                 build_hdr_descs_arr(tx_buff, &num_entries, *hdrs);
1671                 tx_crq.v1.n_crq_elem = num_entries;
1672                 tx_buff->num_entries = num_entries;
1673                 tx_buff->indir_arr[0] = tx_crq;
1674                 tx_buff->indir_dma = dma_map_single(dev, tx_buff->indir_arr,
1675                                                     sizeof(tx_buff->indir_arr),
1676                                                     DMA_TO_DEVICE);
1677                 if (dma_mapping_error(dev, tx_buff->indir_dma)) {
1678                         dev_kfree_skb_any(skb);
1679                         tx_buff->skb = NULL;
1680                         if (!firmware_has_feature(FW_FEATURE_CMO))
1681                                 dev_err(dev, "tx: unable to map descriptor array\n");
1682                         tx_map_failed++;
1683                         tx_dropped++;
1684                         ret = NETDEV_TX_OK;
1685                         goto tx_err_out;
1686                 }
1687                 lpar_rc = send_subcrq_indirect(adapter, handle,
1688                                                (u64)tx_buff->indir_dma,
1689                                                (u64)num_entries);
1690                 dma_unmap_single(dev, tx_buff->indir_dma,
1691                                  sizeof(tx_buff->indir_arr), DMA_TO_DEVICE);
1692         } else {
1693                 tx_buff->num_entries = num_entries;
1694                 lpar_rc = send_subcrq(adapter, handle,
1695                                       &tx_crq);
1696         }
1697         if (lpar_rc != H_SUCCESS) {
1698                 if (lpar_rc != H_CLOSED && lpar_rc != H_PARAMETER)
1699                         dev_err_ratelimited(dev, "tx: send failed\n");
1700                 dev_kfree_skb_any(skb);
1701                 tx_buff->skb = NULL;
1702
1703                 if (lpar_rc == H_CLOSED || adapter->failover_pending) {
1704                         /* Disable TX and report carrier off if queue is closed
1705                          * or pending failover.
1706                          * Firmware guarantees that a signal will be sent to the
1707                          * driver, triggering a reset or some other action.
1708                          */
1709                         netif_tx_stop_all_queues(netdev);
1710                         netif_carrier_off(netdev);
1711                 }
1712
1713                 tx_send_failed++;
1714                 tx_dropped++;
1715                 ret = NETDEV_TX_OK;
1716                 goto tx_err_out;
1717         }
1718
1719         if (atomic_add_return(num_entries, &tx_scrq->used)
1720                                         >= adapter->req_tx_entries_per_subcrq) {
1721                 netdev_dbg(netdev, "Stopping queue %d\n", queue_num);
1722                 netif_stop_subqueue(netdev, queue_num);
1723         }
1724
1725         tx_packets++;
1726         tx_bytes += skb->len;
1727         txq->trans_start = jiffies;
1728         ret = NETDEV_TX_OK;
1729         goto out;
1730
1731 tx_err_out:
1732         /* roll back consumer index and map array*/
1733         if (tx_pool->consumer_index == 0)
1734                 tx_pool->consumer_index =
1735                         tx_pool->num_buffers - 1;
1736         else
1737                 tx_pool->consumer_index--;
1738         tx_pool->free_map[tx_pool->consumer_index] = index;
1739 out:
1740         netdev->stats.tx_dropped += tx_dropped;
1741         netdev->stats.tx_bytes += tx_bytes;
1742         netdev->stats.tx_packets += tx_packets;
1743         adapter->tx_send_failed += tx_send_failed;
1744         adapter->tx_map_failed += tx_map_failed;
1745         adapter->tx_stats_buffers[queue_num].packets += tx_packets;
1746         adapter->tx_stats_buffers[queue_num].bytes += tx_bytes;
1747         adapter->tx_stats_buffers[queue_num].dropped_packets += tx_dropped;
1748
1749         return ret;
1750 }
1751
1752 static void ibmvnic_set_multi(struct net_device *netdev)
1753 {
1754         struct ibmvnic_adapter *adapter = netdev_priv(netdev);
1755         struct netdev_hw_addr *ha;
1756         union ibmvnic_crq crq;
1757
1758         memset(&crq, 0, sizeof(crq));
1759         crq.request_capability.first = IBMVNIC_CRQ_CMD;
1760         crq.request_capability.cmd = REQUEST_CAPABILITY;
1761
1762         if (netdev->flags & IFF_PROMISC) {
1763                 if (!adapter->promisc_supported)
1764                         return;
1765         } else {
1766                 if (netdev->flags & IFF_ALLMULTI) {
1767                         /* Accept all multicast */
1768                         memset(&crq, 0, sizeof(crq));
1769                         crq.multicast_ctrl.first = IBMVNIC_CRQ_CMD;
1770                         crq.multicast_ctrl.cmd = MULTICAST_CTRL;
1771                         crq.multicast_ctrl.flags = IBMVNIC_ENABLE_ALL;
1772                         ibmvnic_send_crq(adapter, &crq);
1773                 } else if (netdev_mc_empty(netdev)) {
1774                         /* Reject all multicast */
1775                         memset(&crq, 0, sizeof(crq));
1776                         crq.multicast_ctrl.first = IBMVNIC_CRQ_CMD;
1777                         crq.multicast_ctrl.cmd = MULTICAST_CTRL;
1778                         crq.multicast_ctrl.flags = IBMVNIC_DISABLE_ALL;
1779                         ibmvnic_send_crq(adapter, &crq);
1780                 } else {
1781                         /* Accept one or more multicast(s) */
1782                         netdev_for_each_mc_addr(ha, netdev) {
1783                                 memset(&crq, 0, sizeof(crq));
1784                                 crq.multicast_ctrl.first = IBMVNIC_CRQ_CMD;
1785                                 crq.multicast_ctrl.cmd = MULTICAST_CTRL;
1786                                 crq.multicast_ctrl.flags = IBMVNIC_ENABLE_MC;
1787                                 ether_addr_copy(&crq.multicast_ctrl.mac_addr[0],
1788                                                 ha->addr);
1789                                 ibmvnic_send_crq(adapter, &crq);
1790                         }
1791                 }
1792         }
1793 }
1794
1795 static int __ibmvnic_set_mac(struct net_device *netdev, u8 *dev_addr)
1796 {
1797         struct ibmvnic_adapter *adapter = netdev_priv(netdev);
1798         union ibmvnic_crq crq;
1799         int rc;
1800
1801         if (!is_valid_ether_addr(dev_addr)) {
1802                 rc = -EADDRNOTAVAIL;
1803                 goto err;
1804         }
1805
1806         memset(&crq, 0, sizeof(crq));
1807         crq.change_mac_addr.first = IBMVNIC_CRQ_CMD;
1808         crq.change_mac_addr.cmd = CHANGE_MAC_ADDR;
1809         ether_addr_copy(&crq.change_mac_addr.mac_addr[0], dev_addr);
1810
1811         mutex_lock(&adapter->fw_lock);
1812         adapter->fw_done_rc = 0;
1813         reinit_completion(&adapter->fw_done);
1814
1815         rc = ibmvnic_send_crq(adapter, &crq);
1816         if (rc) {
1817                 rc = -EIO;
1818                 mutex_unlock(&adapter->fw_lock);
1819                 goto err;
1820         }
1821
1822         rc = ibmvnic_wait_for_completion(adapter, &adapter->fw_done, 10000);
1823         /* netdev->dev_addr is changed in handle_change_mac_rsp function */
1824         if (rc || adapter->fw_done_rc) {
1825                 rc = -EIO;
1826                 mutex_unlock(&adapter->fw_lock);
1827                 goto err;
1828         }
1829         mutex_unlock(&adapter->fw_lock);
1830         return 0;
1831 err:
1832         ether_addr_copy(adapter->mac_addr, netdev->dev_addr);
1833         return rc;
1834 }
1835
1836 static int ibmvnic_set_mac(struct net_device *netdev, void *p)
1837 {
1838         struct ibmvnic_adapter *adapter = netdev_priv(netdev);
1839         struct sockaddr *addr = p;
1840         int rc;
1841
1842         rc = 0;
1843         if (!is_valid_ether_addr(addr->sa_data))
1844                 return -EADDRNOTAVAIL;
1845
1846         ether_addr_copy(adapter->mac_addr, addr->sa_data);
1847         if (adapter->state != VNIC_PROBED)
1848                 rc = __ibmvnic_set_mac(netdev, addr->sa_data);
1849
1850         return rc;
1851 }
1852
1853 /**
1854  * do_change_param_reset returns zero if we are able to keep processing reset
1855  * events, or non-zero if we hit a fatal error and must halt.
1856  */
1857 static int do_change_param_reset(struct ibmvnic_adapter *adapter,
1858                                  struct ibmvnic_rwi *rwi,
1859                                  u32 reset_state)
1860 {
1861         struct net_device *netdev = adapter->netdev;
1862         int i, rc;
1863
1864         netdev_dbg(adapter->netdev, "Change param resetting driver (%d)\n",
1865                    rwi->reset_reason);
1866
1867         netif_carrier_off(netdev);
1868         adapter->reset_reason = rwi->reset_reason;
1869
1870         ibmvnic_cleanup(netdev);
1871
1872         if (reset_state == VNIC_OPEN) {
1873                 rc = __ibmvnic_close(netdev);
1874                 if (rc)
1875                         goto out;
1876         }
1877
1878         release_resources(adapter);
1879         release_sub_crqs(adapter, 1);
1880         release_crq_queue(adapter);
1881
1882         adapter->state = VNIC_PROBED;
1883
1884         rc = init_crq_queue(adapter);
1885
1886         if (rc) {
1887                 netdev_err(adapter->netdev,
1888                            "Couldn't initialize crq. rc=%d\n", rc);
1889                 return rc;
1890         }
1891
1892         rc = ibmvnic_reset_init(adapter, true);
1893         if (rc) {
1894                 rc = IBMVNIC_INIT_FAILED;
1895                 goto out;
1896         }
1897
1898         /* If the adapter was in PROBE state prior to the reset,
1899          * exit here.
1900          */
1901         if (reset_state == VNIC_PROBED)
1902                 goto out;
1903
1904         rc = ibmvnic_login(netdev);
1905         if (rc) {
1906                 goto out;
1907         }
1908
1909         rc = init_resources(adapter);
1910         if (rc)
1911                 goto out;
1912
1913         ibmvnic_disable_irqs(adapter);
1914
1915         adapter->state = VNIC_CLOSED;
1916
1917         if (reset_state == VNIC_CLOSED)
1918                 return 0;
1919
1920         rc = __ibmvnic_open(netdev);
1921         if (rc) {
1922                 rc = IBMVNIC_OPEN_FAILED;
1923                 goto out;
1924         }
1925
1926         /* refresh device's multicast list */
1927         ibmvnic_set_multi(netdev);
1928
1929         /* kick napi */
1930         for (i = 0; i < adapter->req_rx_queues; i++)
1931                 napi_schedule(&adapter->napi[i]);
1932
1933 out:
1934         if (rc)
1935                 adapter->state = reset_state;
1936         return rc;
1937 }
1938
1939 /**
1940  * do_reset returns zero if we are able to keep processing reset events, or
1941  * non-zero if we hit a fatal error and must halt.
1942  */
1943 static int do_reset(struct ibmvnic_adapter *adapter,
1944                     struct ibmvnic_rwi *rwi, u32 reset_state)
1945 {
1946         u64 old_num_rx_queues, old_num_tx_queues;
1947         u64 old_num_rx_slots, old_num_tx_slots;
1948         struct net_device *netdev = adapter->netdev;
1949         int rc;
1950
1951         netdev_dbg(adapter->netdev,
1952                    "[S:%d FOP:%d] Reset reason %d, reset_state %d\n",
1953                    adapter->state, adapter->failover_pending,
1954                    rwi->reset_reason, reset_state);
1955
1956         rtnl_lock();
1957         /*
1958          * Now that we have the rtnl lock, clear any pending failover.
1959          * This will ensure ibmvnic_open() has either completed or will
1960          * block until failover is complete.
1961          */
1962         if (rwi->reset_reason == VNIC_RESET_FAILOVER)
1963                 adapter->failover_pending = false;
1964
1965         netif_carrier_off(netdev);
1966         adapter->reset_reason = rwi->reset_reason;
1967
1968         old_num_rx_queues = adapter->req_rx_queues;
1969         old_num_tx_queues = adapter->req_tx_queues;
1970         old_num_rx_slots = adapter->req_rx_add_entries_per_subcrq;
1971         old_num_tx_slots = adapter->req_tx_entries_per_subcrq;
1972
1973         ibmvnic_cleanup(netdev);
1974
1975         if (reset_state == VNIC_OPEN &&
1976             adapter->reset_reason != VNIC_RESET_MOBILITY &&
1977             adapter->reset_reason != VNIC_RESET_FAILOVER) {
1978                 adapter->state = VNIC_CLOSING;
1979
1980                 /* Release the RTNL lock before link state change and
1981                  * re-acquire after the link state change to allow
1982                  * linkwatch_event to grab the RTNL lock and run during
1983                  * a reset.
1984                  */
1985                 rtnl_unlock();
1986                 rc = set_link_state(adapter, IBMVNIC_LOGICAL_LNK_DN);
1987                 rtnl_lock();
1988                 if (rc)
1989                         goto out;
1990
1991                 if (adapter->state != VNIC_CLOSING) {
1992                         rc = -1;
1993                         goto out;
1994                 }
1995
1996                 adapter->state = VNIC_CLOSED;
1997         }
1998
1999         if (adapter->reset_reason != VNIC_RESET_NON_FATAL) {
2000                 /* remove the closed state so when we call open it appears
2001                  * we are coming from the probed state.
2002                  */
2003                 adapter->state = VNIC_PROBED;
2004
2005                 if (adapter->reset_reason == VNIC_RESET_MOBILITY) {
2006                         rc = ibmvnic_reenable_crq_queue(adapter);
2007                         release_sub_crqs(adapter, 1);
2008                 } else {
2009                         rc = ibmvnic_reset_crq(adapter);
2010                         if (rc == H_CLOSED || rc == H_SUCCESS) {
2011                                 rc = vio_enable_interrupts(adapter->vdev);
2012                                 if (rc)
2013                                         netdev_err(adapter->netdev,
2014                                                    "Reset failed to enable interrupts. rc=%d\n",
2015                                                    rc);
2016                         }
2017                 }
2018
2019                 if (rc) {
2020                         netdev_err(adapter->netdev,
2021                                    "Reset couldn't initialize crq. rc=%d\n", rc);
2022                         goto out;
2023                 }
2024
2025                 rc = ibmvnic_reset_init(adapter, true);
2026                 if (rc) {
2027                         rc = IBMVNIC_INIT_FAILED;
2028                         goto out;
2029                 }
2030
2031                 /* If the adapter was in PROBE state prior to the reset,
2032                  * exit here.
2033                  */
2034                 if (reset_state == VNIC_PROBED) {
2035                         rc = 0;
2036                         goto out;
2037                 }
2038
2039                 rc = ibmvnic_login(netdev);
2040                 if (rc) {
2041                         goto out;
2042                 }
2043
2044                 if (adapter->req_rx_queues != old_num_rx_queues ||
2045                     adapter->req_tx_queues != old_num_tx_queues ||
2046                     adapter->req_rx_add_entries_per_subcrq !=
2047                     old_num_rx_slots ||
2048                     adapter->req_tx_entries_per_subcrq !=
2049                     old_num_tx_slots ||
2050                     !adapter->rx_pool ||
2051                     !adapter->tso_pool ||
2052                     !adapter->tx_pool) {
2053                         release_rx_pools(adapter);
2054                         release_tx_pools(adapter);
2055                         release_napi(adapter);
2056                         release_vpd_data(adapter);
2057
2058                         rc = init_resources(adapter);
2059                         if (rc)
2060                                 goto out;
2061
2062                 } else {
2063                         rc = reset_tx_pools(adapter);
2064                         if (rc) {
2065                                 netdev_dbg(adapter->netdev, "reset tx pools failed (%d)\n",
2066                                            rc);
2067                                 goto out;
2068                         }
2069
2070                         rc = reset_rx_pools(adapter);
2071                         if (rc) {
2072                                 netdev_dbg(adapter->netdev, "reset rx pools failed (%d)\n",
2073                                            rc);
2074                                 goto out;
2075                         }
2076                 }
2077                 ibmvnic_disable_irqs(adapter);
2078         }
2079         adapter->state = VNIC_CLOSED;
2080
2081         if (reset_state == VNIC_CLOSED) {
2082                 rc = 0;
2083                 goto out;
2084         }
2085
2086         rc = __ibmvnic_open(netdev);
2087         if (rc) {
2088                 rc = IBMVNIC_OPEN_FAILED;
2089                 goto out;
2090         }
2091
2092         /* refresh device's multicast list */
2093         ibmvnic_set_multi(netdev);
2094
2095         if (adapter->reset_reason == VNIC_RESET_FAILOVER ||
2096             adapter->reset_reason == VNIC_RESET_MOBILITY) {
2097                 call_netdevice_notifiers(NETDEV_NOTIFY_PEERS, netdev);
2098                 call_netdevice_notifiers(NETDEV_RESEND_IGMP, netdev);
2099         }
2100
2101         rc = 0;
2102
2103 out:
2104         /* restore the adapter state if reset failed */
2105         if (rc)
2106                 adapter->state = reset_state;
2107         rtnl_unlock();
2108
2109         netdev_dbg(adapter->netdev, "[S:%d FOP:%d] Reset done, rc %d\n",
2110                    adapter->state, adapter->failover_pending, rc);
2111         return rc;
2112 }
2113
2114 static int do_hard_reset(struct ibmvnic_adapter *adapter,
2115                          struct ibmvnic_rwi *rwi, u32 reset_state)
2116 {
2117         struct net_device *netdev = adapter->netdev;
2118         int rc;
2119
2120         netdev_dbg(adapter->netdev, "Hard resetting driver (%d)\n",
2121                    rwi->reset_reason);
2122
2123         netif_carrier_off(netdev);
2124         adapter->reset_reason = rwi->reset_reason;
2125
2126         ibmvnic_cleanup(netdev);
2127         release_resources(adapter);
2128         release_sub_crqs(adapter, 0);
2129         release_crq_queue(adapter);
2130
2131         /* remove the closed state so when we call open it appears
2132          * we are coming from the probed state.
2133          */
2134         adapter->state = VNIC_PROBED;
2135
2136         reinit_completion(&adapter->init_done);
2137         rc = init_crq_queue(adapter);
2138         if (rc) {
2139                 netdev_err(adapter->netdev,
2140                            "Couldn't initialize crq. rc=%d\n", rc);
2141                 goto out;
2142         }
2143
2144         rc = ibmvnic_reset_init(adapter, false);
2145         if (rc)
2146                 goto out;
2147
2148         /* If the adapter was in PROBE state prior to the reset,
2149          * exit here.
2150          */
2151         if (reset_state == VNIC_PROBED)
2152                 goto out;
2153
2154         rc = ibmvnic_login(netdev);
2155         if (rc)
2156                 goto out;
2157
2158         rc = init_resources(adapter);
2159         if (rc)
2160                 goto out;
2161
2162         ibmvnic_disable_irqs(adapter);
2163         adapter->state = VNIC_CLOSED;
2164
2165         if (reset_state == VNIC_CLOSED)
2166                 goto out;
2167
2168         rc = __ibmvnic_open(netdev);
2169         if (rc) {
2170                 rc = IBMVNIC_OPEN_FAILED;
2171                 goto out;
2172         }
2173
2174         call_netdevice_notifiers(NETDEV_NOTIFY_PEERS, netdev);
2175         call_netdevice_notifiers(NETDEV_RESEND_IGMP, netdev);
2176 out:
2177         /* restore adapter state if reset failed */
2178         if (rc)
2179                 adapter->state = reset_state;
2180         netdev_dbg(adapter->netdev, "[S:%d FOP:%d] Hard reset done, rc %d\n",
2181                    adapter->state, adapter->failover_pending, rc);
2182         return rc;
2183 }
2184
2185 static struct ibmvnic_rwi *get_next_rwi(struct ibmvnic_adapter *adapter)
2186 {
2187         struct ibmvnic_rwi *rwi;
2188         unsigned long flags;
2189
2190         spin_lock_irqsave(&adapter->rwi_lock, flags);
2191
2192         if (!list_empty(&adapter->rwi_list)) {
2193                 rwi = list_first_entry(&adapter->rwi_list, struct ibmvnic_rwi,
2194                                        list);
2195                 list_del(&rwi->list);
2196         } else {
2197                 rwi = NULL;
2198         }
2199
2200         spin_unlock_irqrestore(&adapter->rwi_lock, flags);
2201         return rwi;
2202 }
2203
2204 static void __ibmvnic_reset(struct work_struct *work)
2205 {
2206         struct ibmvnic_rwi *rwi;
2207         struct ibmvnic_adapter *adapter;
2208         bool saved_state = false;
2209         unsigned long flags;
2210         u32 reset_state;
2211         int rc = 0;
2212
2213         adapter = container_of(work, struct ibmvnic_adapter, ibmvnic_reset);
2214
2215         if (test_and_set_bit_lock(0, &adapter->resetting)) {
2216                 schedule_delayed_work(&adapter->ibmvnic_delayed_reset,
2217                                       IBMVNIC_RESET_DELAY);
2218                 return;
2219         }
2220
2221         rwi = get_next_rwi(adapter);
2222         while (rwi) {
2223                 spin_lock_irqsave(&adapter->state_lock, flags);
2224
2225                 if (adapter->state == VNIC_REMOVING ||
2226                     adapter->state == VNIC_REMOVED) {
2227                         spin_unlock_irqrestore(&adapter->state_lock, flags);
2228                         kfree(rwi);
2229                         rc = EBUSY;
2230                         break;
2231                 }
2232
2233                 if (!saved_state) {
2234                         reset_state = adapter->state;
2235                         saved_state = true;
2236                 }
2237                 spin_unlock_irqrestore(&adapter->state_lock, flags);
2238
2239                 if (rwi->reset_reason == VNIC_RESET_CHANGE_PARAM) {
2240                         /* CHANGE_PARAM requestor holds rtnl_lock */
2241                         rc = do_change_param_reset(adapter, rwi, reset_state);
2242                 } else if (adapter->force_reset_recovery) {
2243                         /*
2244                          * Since we are doing a hard reset now, clear the
2245                          * failover_pending flag so we don't ignore any
2246                          * future MOBILITY or other resets.
2247                          */
2248                         adapter->failover_pending = false;
2249
2250                         /* Transport event occurred during previous reset */
2251                         if (adapter->wait_for_reset) {
2252                                 /* Previous was CHANGE_PARAM; caller locked */
2253                                 adapter->force_reset_recovery = false;
2254                                 rc = do_hard_reset(adapter, rwi, reset_state);
2255                         } else {
2256                                 rtnl_lock();
2257                                 adapter->force_reset_recovery = false;
2258                                 rc = do_hard_reset(adapter, rwi, reset_state);
2259                                 rtnl_unlock();
2260                         }
2261                         if (rc) {
2262                                 /* give backing device time to settle down */
2263                                 netdev_dbg(adapter->netdev,
2264                                            "[S:%d] Hard reset failed, waiting 60 secs\n",
2265                                            adapter->state);
2266                                 set_current_state(TASK_UNINTERRUPTIBLE);
2267                                 schedule_timeout(60 * HZ);
2268                         }
2269                 } else {
2270                         rc = do_reset(adapter, rwi, reset_state);
2271                 }
2272                 kfree(rwi);
2273                 adapter->last_reset_time = jiffies;
2274
2275                 if (rc)
2276                         netdev_dbg(adapter->netdev, "Reset failed, rc=%d\n", rc);
2277
2278                 rwi = get_next_rwi(adapter);
2279
2280                 if (rwi && (rwi->reset_reason == VNIC_RESET_FAILOVER ||
2281                             rwi->reset_reason == VNIC_RESET_MOBILITY))
2282                         adapter->force_reset_recovery = true;
2283         }
2284
2285         if (adapter->wait_for_reset) {
2286                 adapter->reset_done_rc = rc;
2287                 complete(&adapter->reset_done);
2288         }
2289
2290         clear_bit_unlock(0, &adapter->resetting);
2291
2292         netdev_dbg(adapter->netdev,
2293                    "[S:%d FRR:%d WFR:%d] Done processing resets\n",
2294                    adapter->state, adapter->force_reset_recovery,
2295                    adapter->wait_for_reset);
2296 }
2297
2298 static void __ibmvnic_delayed_reset(struct work_struct *work)
2299 {
2300         struct ibmvnic_adapter *adapter;
2301
2302         adapter = container_of(work, struct ibmvnic_adapter,
2303                                ibmvnic_delayed_reset.work);
2304         __ibmvnic_reset(&adapter->ibmvnic_reset);
2305 }
2306
2307 static int ibmvnic_reset(struct ibmvnic_adapter *adapter,
2308                          enum ibmvnic_reset_reason reason)
2309 {
2310         struct list_head *entry, *tmp_entry;
2311         struct ibmvnic_rwi *rwi, *tmp;
2312         struct net_device *netdev = adapter->netdev;
2313         unsigned long flags;
2314         int ret;
2315
2316         spin_lock_irqsave(&adapter->rwi_lock, flags);
2317
2318         /*
2319          * If failover is pending don't schedule any other reset.
2320          * Instead let the failover complete. If there is already a
2321          * a failover reset scheduled, we will detect and drop the
2322          * duplicate reset when walking the ->rwi_list below.
2323          */
2324         if (adapter->state == VNIC_REMOVING ||
2325             adapter->state == VNIC_REMOVED ||
2326             (adapter->failover_pending && reason != VNIC_RESET_FAILOVER)) {
2327                 ret = EBUSY;
2328                 netdev_dbg(netdev, "Adapter removing or pending failover, skipping reset\n");
2329                 goto err;
2330         }
2331
2332         if (adapter->state == VNIC_PROBING) {
2333                 netdev_warn(netdev, "Adapter reset during probe\n");
2334                 adapter->init_done_rc = EAGAIN;
2335                 ret = EAGAIN;
2336                 goto err;
2337         }
2338
2339         list_for_each(entry, &adapter->rwi_list) {
2340                 tmp = list_entry(entry, struct ibmvnic_rwi, list);
2341                 if (tmp->reset_reason == reason) {
2342                         netdev_dbg(netdev, "Skipping matching reset, reason=%d\n",
2343                                    reason);
2344                         ret = EBUSY;
2345                         goto err;
2346                 }
2347         }
2348
2349         rwi = kzalloc(sizeof(*rwi), GFP_ATOMIC);
2350         if (!rwi) {
2351                 ret = ENOMEM;
2352                 goto err;
2353         }
2354         /* if we just received a transport event,
2355          * flush reset queue and process this reset
2356          */
2357         if (adapter->force_reset_recovery && !list_empty(&adapter->rwi_list)) {
2358                 list_for_each_safe(entry, tmp_entry, &adapter->rwi_list) {
2359                         list_del(entry);
2360                         kfree(list_entry(entry, struct ibmvnic_rwi, list));
2361                 }
2362         }
2363         rwi->reset_reason = reason;
2364         list_add_tail(&rwi->list, &adapter->rwi_list);
2365         netdev_dbg(adapter->netdev, "Scheduling reset (reason %d)\n", reason);
2366         schedule_work(&adapter->ibmvnic_reset);
2367
2368         ret = 0;
2369 err:
2370         /* ibmvnic_close() below can block, so drop the lock first */
2371         spin_unlock_irqrestore(&adapter->rwi_lock, flags);
2372
2373         if (ret == ENOMEM)
2374                 ibmvnic_close(netdev);
2375
2376         return -ret;
2377 }
2378
2379 static void ibmvnic_tx_timeout(struct net_device *dev, unsigned int txqueue)
2380 {
2381         struct ibmvnic_adapter *adapter = netdev_priv(dev);
2382
2383         if (test_bit(0, &adapter->resetting)) {
2384                 netdev_err(adapter->netdev,
2385                            "Adapter is resetting, skip timeout reset\n");
2386                 return;
2387         }
2388         /* No queuing up reset until at least 5 seconds (default watchdog val)
2389          * after last reset
2390          */
2391         if (time_before(jiffies, (adapter->last_reset_time + dev->watchdog_timeo))) {
2392                 netdev_dbg(dev, "Not yet time to tx timeout.\n");
2393                 return;
2394         }
2395         ibmvnic_reset(adapter, VNIC_RESET_TIMEOUT);
2396 }
2397
2398 static void remove_buff_from_pool(struct ibmvnic_adapter *adapter,
2399                                   struct ibmvnic_rx_buff *rx_buff)
2400 {
2401         struct ibmvnic_rx_pool *pool = &adapter->rx_pool[rx_buff->pool_index];
2402
2403         rx_buff->skb = NULL;
2404
2405         pool->free_map[pool->next_alloc] = (int)(rx_buff - pool->rx_buff);
2406         pool->next_alloc = (pool->next_alloc + 1) % pool->size;
2407
2408         atomic_dec(&pool->available);
2409 }
2410
2411 static int ibmvnic_poll(struct napi_struct *napi, int budget)
2412 {
2413         struct net_device *netdev = napi->dev;
2414         struct ibmvnic_adapter *adapter = netdev_priv(netdev);
2415         int scrq_num = (int)(napi - adapter->napi);
2416         int frames_processed = 0;
2417
2418 restart_poll:
2419         while (frames_processed < budget) {
2420                 struct sk_buff *skb;
2421                 struct ibmvnic_rx_buff *rx_buff;
2422                 union sub_crq *next;
2423                 u32 length;
2424                 u16 offset;
2425                 u8 flags = 0;
2426
2427                 if (unlikely(test_bit(0, &adapter->resetting) &&
2428                              adapter->reset_reason != VNIC_RESET_NON_FATAL)) {
2429                         enable_scrq_irq(adapter, adapter->rx_scrq[scrq_num]);
2430                         napi_complete_done(napi, frames_processed);
2431                         return frames_processed;
2432                 }
2433
2434                 if (!pending_scrq(adapter, adapter->rx_scrq[scrq_num]))
2435                         break;
2436                 /* The queue entry at the current index is peeked at above
2437                  * to determine that there is a valid descriptor awaiting
2438                  * processing. We want to be sure that the current slot
2439                  * holds a valid descriptor before reading its contents.
2440                  */
2441                 dma_rmb();
2442                 next = ibmvnic_next_scrq(adapter, adapter->rx_scrq[scrq_num]);
2443                 rx_buff =
2444                     (struct ibmvnic_rx_buff *)be64_to_cpu(next->
2445                                                           rx_comp.correlator);
2446                 /* do error checking */
2447                 if (next->rx_comp.rc) {
2448                         netdev_dbg(netdev, "rx buffer returned with rc %x\n",
2449                                    be16_to_cpu(next->rx_comp.rc));
2450                         /* free the entry */
2451                         next->rx_comp.first = 0;
2452                         dev_kfree_skb_any(rx_buff->skb);
2453                         remove_buff_from_pool(adapter, rx_buff);
2454                         continue;
2455                 } else if (!rx_buff->skb) {
2456                         /* free the entry */
2457                         next->rx_comp.first = 0;
2458                         remove_buff_from_pool(adapter, rx_buff);
2459                         continue;
2460                 }
2461
2462                 length = be32_to_cpu(next->rx_comp.len);
2463                 offset = be16_to_cpu(next->rx_comp.off_frame_data);
2464                 flags = next->rx_comp.flags;
2465                 skb = rx_buff->skb;
2466                 /* load long_term_buff before copying to skb */
2467                 dma_rmb();
2468                 skb_copy_to_linear_data(skb, rx_buff->data + offset,
2469                                         length);
2470
2471                 /* VLAN Header has been stripped by the system firmware and
2472                  * needs to be inserted by the driver
2473                  */
2474                 if (adapter->rx_vlan_header_insertion &&
2475                     (flags & IBMVNIC_VLAN_STRIPPED))
2476                         __vlan_hwaccel_put_tag(skb, htons(ETH_P_8021Q),
2477                                                ntohs(next->rx_comp.vlan_tci));
2478
2479                 /* free the entry */
2480                 next->rx_comp.first = 0;
2481                 remove_buff_from_pool(adapter, rx_buff);
2482
2483                 skb_put(skb, length);
2484                 skb->protocol = eth_type_trans(skb, netdev);
2485                 skb_record_rx_queue(skb, scrq_num);
2486
2487                 if (flags & IBMVNIC_IP_CHKSUM_GOOD &&
2488                     flags & IBMVNIC_TCP_UDP_CHKSUM_GOOD) {
2489                         skb->ip_summed = CHECKSUM_UNNECESSARY;
2490                 }
2491
2492                 length = skb->len;
2493                 napi_gro_receive(napi, skb); /* send it up */
2494                 netdev->stats.rx_packets++;
2495                 netdev->stats.rx_bytes += length;
2496                 adapter->rx_stats_buffers[scrq_num].packets++;
2497                 adapter->rx_stats_buffers[scrq_num].bytes += length;
2498                 frames_processed++;
2499         }
2500
2501         if (adapter->state != VNIC_CLOSING)
2502                 replenish_rx_pool(adapter, &adapter->rx_pool[scrq_num]);
2503
2504         if (frames_processed < budget) {
2505                 enable_scrq_irq(adapter, adapter->rx_scrq[scrq_num]);
2506                 napi_complete_done(napi, frames_processed);
2507                 if (pending_scrq(adapter, adapter->rx_scrq[scrq_num]) &&
2508                     napi_reschedule(napi)) {
2509                         disable_scrq_irq(adapter, adapter->rx_scrq[scrq_num]);
2510                         goto restart_poll;
2511                 }
2512         }
2513         return frames_processed;
2514 }
2515
2516 static int wait_for_reset(struct ibmvnic_adapter *adapter)
2517 {
2518         int rc, ret;
2519
2520         adapter->fallback.mtu = adapter->req_mtu;
2521         adapter->fallback.rx_queues = adapter->req_rx_queues;
2522         adapter->fallback.tx_queues = adapter->req_tx_queues;
2523         adapter->fallback.rx_entries = adapter->req_rx_add_entries_per_subcrq;
2524         adapter->fallback.tx_entries = adapter->req_tx_entries_per_subcrq;
2525
2526         reinit_completion(&adapter->reset_done);
2527         adapter->wait_for_reset = true;
2528         rc = ibmvnic_reset(adapter, VNIC_RESET_CHANGE_PARAM);
2529
2530         if (rc) {
2531                 ret = rc;
2532                 goto out;
2533         }
2534         rc = ibmvnic_wait_for_completion(adapter, &adapter->reset_done, 60000);
2535         if (rc) {
2536                 ret = -ENODEV;
2537                 goto out;
2538         }
2539
2540         ret = 0;
2541         if (adapter->reset_done_rc) {
2542                 ret = -EIO;
2543                 adapter->desired.mtu = adapter->fallback.mtu;
2544                 adapter->desired.rx_queues = adapter->fallback.rx_queues;
2545                 adapter->desired.tx_queues = adapter->fallback.tx_queues;
2546                 adapter->desired.rx_entries = adapter->fallback.rx_entries;
2547                 adapter->desired.tx_entries = adapter->fallback.tx_entries;
2548
2549                 reinit_completion(&adapter->reset_done);
2550                 adapter->wait_for_reset = true;
2551                 rc = ibmvnic_reset(adapter, VNIC_RESET_CHANGE_PARAM);
2552                 if (rc) {
2553                         ret = rc;
2554                         goto out;
2555                 }
2556                 rc = ibmvnic_wait_for_completion(adapter, &adapter->reset_done,
2557                                                  60000);
2558                 if (rc) {
2559                         ret = -ENODEV;
2560                         goto out;
2561                 }
2562         }
2563 out:
2564         adapter->wait_for_reset = false;
2565
2566         return ret;
2567 }
2568
2569 static int ibmvnic_change_mtu(struct net_device *netdev, int new_mtu)
2570 {
2571         struct ibmvnic_adapter *adapter = netdev_priv(netdev);
2572
2573         adapter->desired.mtu = new_mtu + ETH_HLEN;
2574
2575         return wait_for_reset(adapter);
2576 }
2577
2578 static netdev_features_t ibmvnic_features_check(struct sk_buff *skb,
2579                                                 struct net_device *dev,
2580                                                 netdev_features_t features)
2581 {
2582         /* Some backing hardware adapters can not
2583          * handle packets with a MSS less than 224
2584          * or with only one segment.
2585          */
2586         if (skb_is_gso(skb)) {
2587                 if (skb_shinfo(skb)->gso_size < 224 ||
2588                     skb_shinfo(skb)->gso_segs == 1)
2589                         features &= ~NETIF_F_GSO_MASK;
2590         }
2591
2592         return features;
2593 }
2594
2595 static const struct net_device_ops ibmvnic_netdev_ops = {
2596         .ndo_open               = ibmvnic_open,
2597         .ndo_stop               = ibmvnic_close,
2598         .ndo_start_xmit         = ibmvnic_xmit,
2599         .ndo_set_rx_mode        = ibmvnic_set_multi,
2600         .ndo_set_mac_address    = ibmvnic_set_mac,
2601         .ndo_validate_addr      = eth_validate_addr,
2602         .ndo_tx_timeout         = ibmvnic_tx_timeout,
2603         .ndo_change_mtu         = ibmvnic_change_mtu,
2604         .ndo_features_check     = ibmvnic_features_check,
2605 };
2606
2607 /* ethtool functions */
2608
2609 static int ibmvnic_get_link_ksettings(struct net_device *netdev,
2610                                       struct ethtool_link_ksettings *cmd)
2611 {
2612         struct ibmvnic_adapter *adapter = netdev_priv(netdev);
2613         int rc;
2614
2615         rc = send_query_phys_parms(adapter);
2616         if (rc) {
2617                 adapter->speed = SPEED_UNKNOWN;
2618                 adapter->duplex = DUPLEX_UNKNOWN;
2619         }
2620         cmd->base.speed = adapter->speed;
2621         cmd->base.duplex = adapter->duplex;
2622         cmd->base.port = PORT_FIBRE;
2623         cmd->base.phy_address = 0;
2624         cmd->base.autoneg = AUTONEG_ENABLE;
2625
2626         return 0;
2627 }
2628
2629 static void ibmvnic_get_drvinfo(struct net_device *netdev,
2630                                 struct ethtool_drvinfo *info)
2631 {
2632         struct ibmvnic_adapter *adapter = netdev_priv(netdev);
2633
2634         strlcpy(info->driver, ibmvnic_driver_name, sizeof(info->driver));
2635         strlcpy(info->version, IBMVNIC_DRIVER_VERSION, sizeof(info->version));
2636         strlcpy(info->fw_version, adapter->fw_version,
2637                 sizeof(info->fw_version));
2638 }
2639
2640 static u32 ibmvnic_get_msglevel(struct net_device *netdev)
2641 {
2642         struct ibmvnic_adapter *adapter = netdev_priv(netdev);
2643
2644         return adapter->msg_enable;
2645 }
2646
2647 static void ibmvnic_set_msglevel(struct net_device *netdev, u32 data)
2648 {
2649         struct ibmvnic_adapter *adapter = netdev_priv(netdev);
2650
2651         adapter->msg_enable = data;
2652 }
2653
2654 static u32 ibmvnic_get_link(struct net_device *netdev)
2655 {
2656         struct ibmvnic_adapter *adapter = netdev_priv(netdev);
2657
2658         /* Don't need to send a query because we request a logical link up at
2659          * init and then we wait for link state indications
2660          */
2661         return adapter->logical_link_state;
2662 }
2663
2664 static void ibmvnic_get_ringparam(struct net_device *netdev,
2665                                   struct ethtool_ringparam *ring)
2666 {
2667         struct ibmvnic_adapter *adapter = netdev_priv(netdev);
2668
2669         ring->rx_max_pending = adapter->max_rx_add_entries_per_subcrq;
2670         ring->tx_max_pending = adapter->max_tx_entries_per_subcrq;
2671         ring->rx_mini_max_pending = 0;
2672         ring->rx_jumbo_max_pending = 0;
2673         ring->rx_pending = adapter->req_rx_add_entries_per_subcrq;
2674         ring->tx_pending = adapter->req_tx_entries_per_subcrq;
2675         ring->rx_mini_pending = 0;
2676         ring->rx_jumbo_pending = 0;
2677 }
2678
2679 static int ibmvnic_set_ringparam(struct net_device *netdev,
2680                                  struct ethtool_ringparam *ring)
2681 {
2682         struct ibmvnic_adapter *adapter = netdev_priv(netdev);
2683
2684         if (ring->rx_pending > adapter->max_rx_add_entries_per_subcrq  ||
2685             ring->tx_pending > adapter->max_tx_entries_per_subcrq) {
2686                 netdev_err(netdev, "Invalid request.\n");
2687                 netdev_err(netdev, "Max tx buffers = %llu\n",
2688                            adapter->max_rx_add_entries_per_subcrq);
2689                 netdev_err(netdev, "Max rx buffers = %llu\n",
2690                            adapter->max_tx_entries_per_subcrq);
2691                 return -EINVAL;
2692         }
2693
2694         adapter->desired.rx_entries = ring->rx_pending;
2695         adapter->desired.tx_entries = ring->tx_pending;
2696
2697         return wait_for_reset(adapter);
2698 }
2699
2700 static void ibmvnic_get_channels(struct net_device *netdev,
2701                                  struct ethtool_channels *channels)
2702 {
2703         struct ibmvnic_adapter *adapter = netdev_priv(netdev);
2704
2705         channels->max_rx = adapter->max_rx_queues;
2706         channels->max_tx = adapter->max_tx_queues;
2707         channels->max_other = 0;
2708         channels->max_combined = 0;
2709         channels->rx_count = adapter->req_rx_queues;
2710         channels->tx_count = adapter->req_tx_queues;
2711         channels->other_count = 0;
2712         channels->combined_count = 0;
2713 }
2714
2715 static int ibmvnic_set_channels(struct net_device *netdev,
2716                                 struct ethtool_channels *channels)
2717 {
2718         struct ibmvnic_adapter *adapter = netdev_priv(netdev);
2719
2720         adapter->desired.rx_queues = channels->rx_count;
2721         adapter->desired.tx_queues = channels->tx_count;
2722
2723         return wait_for_reset(adapter);
2724 }
2725
2726 static void ibmvnic_get_strings(struct net_device *dev, u32 stringset, u8 *data)
2727 {
2728         struct ibmvnic_adapter *adapter = netdev_priv(dev);
2729         int i;
2730
2731         if (stringset != ETH_SS_STATS)
2732                 return;
2733
2734         for (i = 0; i < ARRAY_SIZE(ibmvnic_stats); i++, data += ETH_GSTRING_LEN)
2735                 memcpy(data, ibmvnic_stats[i].name, ETH_GSTRING_LEN);
2736
2737         for (i = 0; i < adapter->req_tx_queues; i++) {
2738                 snprintf(data, ETH_GSTRING_LEN, "tx%d_packets", i);
2739                 data += ETH_GSTRING_LEN;
2740
2741                 snprintf(data, ETH_GSTRING_LEN, "tx%d_bytes", i);
2742                 data += ETH_GSTRING_LEN;
2743
2744                 snprintf(data, ETH_GSTRING_LEN, "tx%d_dropped_packets", i);
2745                 data += ETH_GSTRING_LEN;
2746         }
2747
2748         for (i = 0; i < adapter->req_rx_queues; i++) {
2749                 snprintf(data, ETH_GSTRING_LEN, "rx%d_packets", i);
2750                 data += ETH_GSTRING_LEN;
2751
2752                 snprintf(data, ETH_GSTRING_LEN, "rx%d_bytes", i);
2753                 data += ETH_GSTRING_LEN;
2754
2755                 snprintf(data, ETH_GSTRING_LEN, "rx%d_interrupts", i);
2756                 data += ETH_GSTRING_LEN;
2757         }
2758 }
2759
2760 static int ibmvnic_get_sset_count(struct net_device *dev, int sset)
2761 {
2762         struct ibmvnic_adapter *adapter = netdev_priv(dev);
2763
2764         switch (sset) {
2765         case ETH_SS_STATS:
2766                 return ARRAY_SIZE(ibmvnic_stats) +
2767                        adapter->req_tx_queues * NUM_TX_STATS +
2768                        adapter->req_rx_queues * NUM_RX_STATS;
2769         default:
2770                 return -EOPNOTSUPP;
2771         }
2772 }
2773
2774 static void ibmvnic_get_ethtool_stats(struct net_device *dev,
2775                                       struct ethtool_stats *stats, u64 *data)
2776 {
2777         struct ibmvnic_adapter *adapter = netdev_priv(dev);
2778         union ibmvnic_crq crq;
2779         int i, j;
2780         int rc;
2781
2782         memset(&crq, 0, sizeof(crq));
2783         crq.request_statistics.first = IBMVNIC_CRQ_CMD;
2784         crq.request_statistics.cmd = REQUEST_STATISTICS;
2785         crq.request_statistics.ioba = cpu_to_be32(adapter->stats_token);
2786         crq.request_statistics.len =
2787             cpu_to_be32(sizeof(struct ibmvnic_statistics));
2788
2789         /* Wait for data to be written */
2790         reinit_completion(&adapter->stats_done);
2791         rc = ibmvnic_send_crq(adapter, &crq);
2792         if (rc)
2793                 return;
2794         rc = ibmvnic_wait_for_completion(adapter, &adapter->stats_done, 10000);
2795         if (rc)
2796                 return;
2797
2798         for (i = 0; i < ARRAY_SIZE(ibmvnic_stats); i++)
2799                 data[i] = be64_to_cpu(IBMVNIC_GET_STAT
2800                                       (adapter, ibmvnic_stats[i].offset));
2801
2802         for (j = 0; j < adapter->req_tx_queues; j++) {
2803                 data[i] = adapter->tx_stats_buffers[j].packets;
2804                 i++;
2805                 data[i] = adapter->tx_stats_buffers[j].bytes;
2806                 i++;
2807                 data[i] = adapter->tx_stats_buffers[j].dropped_packets;
2808                 i++;
2809         }
2810
2811         for (j = 0; j < adapter->req_rx_queues; j++) {
2812                 data[i] = adapter->rx_stats_buffers[j].packets;
2813                 i++;
2814                 data[i] = adapter->rx_stats_buffers[j].bytes;
2815                 i++;
2816                 data[i] = adapter->rx_stats_buffers[j].interrupts;
2817                 i++;
2818         }
2819 }
2820
2821 static const struct ethtool_ops ibmvnic_ethtool_ops = {
2822         .get_drvinfo            = ibmvnic_get_drvinfo,
2823         .get_msglevel           = ibmvnic_get_msglevel,
2824         .set_msglevel           = ibmvnic_set_msglevel,
2825         .get_link               = ibmvnic_get_link,
2826         .get_ringparam          = ibmvnic_get_ringparam,
2827         .set_ringparam          = ibmvnic_set_ringparam,
2828         .get_channels           = ibmvnic_get_channels,
2829         .set_channels           = ibmvnic_set_channels,
2830         .get_strings            = ibmvnic_get_strings,
2831         .get_sset_count         = ibmvnic_get_sset_count,
2832         .get_ethtool_stats      = ibmvnic_get_ethtool_stats,
2833         .get_link_ksettings     = ibmvnic_get_link_ksettings,
2834 };
2835
2836 /* Routines for managing CRQs/sCRQs  */
2837
2838 static int reset_one_sub_crq_queue(struct ibmvnic_adapter *adapter,
2839                                    struct ibmvnic_sub_crq_queue *scrq)
2840 {
2841         int rc;
2842
2843         if (!scrq) {
2844                 netdev_dbg(adapter->netdev, "Invalid scrq reset.\n");
2845                 return -EINVAL;
2846         }
2847
2848         if (scrq->irq) {
2849                 free_irq(scrq->irq, scrq);
2850                 irq_dispose_mapping(scrq->irq);
2851                 scrq->irq = 0;
2852         }
2853         if (scrq->msgs) {
2854                 memset(scrq->msgs, 0, 4 * PAGE_SIZE);
2855                 atomic_set(&scrq->used, 0);
2856                 scrq->cur = 0;
2857         } else {
2858                 netdev_dbg(adapter->netdev, "Invalid scrq reset\n");
2859                 return -EINVAL;
2860         }
2861
2862         rc = h_reg_sub_crq(adapter->vdev->unit_address, scrq->msg_token,
2863                            4 * PAGE_SIZE, &scrq->crq_num, &scrq->hw_irq);
2864         return rc;
2865 }
2866
2867 static int reset_sub_crq_queues(struct ibmvnic_adapter *adapter)
2868 {
2869         int i, rc;
2870
2871         if (!adapter->tx_scrq || !adapter->rx_scrq)
2872                 return -EINVAL;
2873
2874         for (i = 0; i < adapter->req_tx_queues; i++) {
2875                 netdev_dbg(adapter->netdev, "Re-setting tx_scrq[%d]\n", i);
2876                 rc = reset_one_sub_crq_queue(adapter, adapter->tx_scrq[i]);
2877                 if (rc)
2878                         return rc;
2879         }
2880
2881         for (i = 0; i < adapter->req_rx_queues; i++) {
2882                 netdev_dbg(adapter->netdev, "Re-setting rx_scrq[%d]\n", i);
2883                 rc = reset_one_sub_crq_queue(adapter, adapter->rx_scrq[i]);
2884                 if (rc)
2885                         return rc;
2886         }
2887
2888         return rc;
2889 }
2890
2891 static void release_sub_crq_queue(struct ibmvnic_adapter *adapter,
2892                                   struct ibmvnic_sub_crq_queue *scrq,
2893                                   bool do_h_free)
2894 {
2895         struct device *dev = &adapter->vdev->dev;
2896         long rc;
2897
2898         netdev_dbg(adapter->netdev, "Releasing sub-CRQ\n");
2899
2900         if (do_h_free) {
2901                 /* Close the sub-crqs */
2902                 do {
2903                         rc = plpar_hcall_norets(H_FREE_SUB_CRQ,
2904                                                 adapter->vdev->unit_address,
2905                                                 scrq->crq_num);
2906                 } while (rc == H_BUSY || H_IS_LONG_BUSY(rc));
2907
2908                 if (rc) {
2909                         netdev_err(adapter->netdev,
2910                                    "Failed to release sub-CRQ %16lx, rc = %ld\n",
2911                                    scrq->crq_num, rc);
2912                 }
2913         }
2914
2915         dma_unmap_single(dev, scrq->msg_token, 4 * PAGE_SIZE,
2916                          DMA_BIDIRECTIONAL);
2917         free_pages((unsigned long)scrq->msgs, 2);
2918         kfree(scrq);
2919 }
2920
2921 static struct ibmvnic_sub_crq_queue *init_sub_crq_queue(struct ibmvnic_adapter
2922                                                         *adapter)
2923 {
2924         struct device *dev = &adapter->vdev->dev;
2925         struct ibmvnic_sub_crq_queue *scrq;
2926         int rc;
2927
2928         scrq = kzalloc(sizeof(*scrq), GFP_KERNEL);
2929         if (!scrq)
2930                 return NULL;
2931
2932         scrq->msgs =
2933                 (union sub_crq *)__get_free_pages(GFP_KERNEL | __GFP_ZERO, 2);
2934         if (!scrq->msgs) {
2935                 dev_warn(dev, "Couldn't allocate crq queue messages page\n");
2936                 goto zero_page_failed;
2937         }
2938
2939         scrq->msg_token = dma_map_single(dev, scrq->msgs, 4 * PAGE_SIZE,
2940                                          DMA_BIDIRECTIONAL);
2941         if (dma_mapping_error(dev, scrq->msg_token)) {
2942                 dev_warn(dev, "Couldn't map crq queue messages page\n");
2943                 goto map_failed;
2944         }
2945
2946         rc = h_reg_sub_crq(adapter->vdev->unit_address, scrq->msg_token,
2947                            4 * PAGE_SIZE, &scrq->crq_num, &scrq->hw_irq);
2948
2949         if (rc == H_RESOURCE)
2950                 rc = ibmvnic_reset_crq(adapter);
2951
2952         if (rc == H_CLOSED) {
2953                 dev_warn(dev, "Partner adapter not ready, waiting.\n");
2954         } else if (rc) {
2955                 dev_warn(dev, "Error %d registering sub-crq\n", rc);
2956                 goto reg_failed;
2957         }
2958
2959         scrq->adapter = adapter;
2960         scrq->size = 4 * PAGE_SIZE / sizeof(*scrq->msgs);
2961         spin_lock_init(&scrq->lock);
2962
2963         netdev_dbg(adapter->netdev,
2964                    "sub-crq initialized, num %lx, hw_irq=%lx, irq=%x\n",
2965                    scrq->crq_num, scrq->hw_irq, scrq->irq);
2966
2967         return scrq;
2968
2969 reg_failed:
2970         dma_unmap_single(dev, scrq->msg_token, 4 * PAGE_SIZE,
2971                          DMA_BIDIRECTIONAL);
2972 map_failed:
2973         free_pages((unsigned long)scrq->msgs, 2);
2974 zero_page_failed:
2975         kfree(scrq);
2976
2977         return NULL;
2978 }
2979
2980 static void release_sub_crqs(struct ibmvnic_adapter *adapter, bool do_h_free)
2981 {
2982         int i;
2983
2984         if (adapter->tx_scrq) {
2985                 for (i = 0; i < adapter->num_active_tx_scrqs; i++) {
2986                         if (!adapter->tx_scrq[i])
2987                                 continue;
2988
2989                         netdev_dbg(adapter->netdev, "Releasing tx_scrq[%d]\n",
2990                                    i);
2991                         if (adapter->tx_scrq[i]->irq) {
2992                                 free_irq(adapter->tx_scrq[i]->irq,
2993                                          adapter->tx_scrq[i]);
2994                                 irq_dispose_mapping(adapter->tx_scrq[i]->irq);
2995                                 adapter->tx_scrq[i]->irq = 0;
2996                         }
2997
2998                         release_sub_crq_queue(adapter, adapter->tx_scrq[i],
2999                                               do_h_free);
3000                 }
3001
3002                 kfree(adapter->tx_scrq);
3003                 adapter->tx_scrq = NULL;
3004                 adapter->num_active_tx_scrqs = 0;
3005         }
3006
3007         if (adapter->rx_scrq) {
3008                 for (i = 0; i < adapter->num_active_rx_scrqs; i++) {
3009                         if (!adapter->rx_scrq[i])
3010                                 continue;
3011
3012                         netdev_dbg(adapter->netdev, "Releasing rx_scrq[%d]\n",
3013                                    i);
3014                         if (adapter->rx_scrq[i]->irq) {
3015                                 free_irq(adapter->rx_scrq[i]->irq,
3016                                          adapter->rx_scrq[i]);
3017                                 irq_dispose_mapping(adapter->rx_scrq[i]->irq);
3018                                 adapter->rx_scrq[i]->irq = 0;
3019                         }
3020
3021                         release_sub_crq_queue(adapter, adapter->rx_scrq[i],
3022                                               do_h_free);
3023                 }
3024
3025                 kfree(adapter->rx_scrq);
3026                 adapter->rx_scrq = NULL;
3027                 adapter->num_active_rx_scrqs = 0;
3028         }
3029 }
3030
3031 static int disable_scrq_irq(struct ibmvnic_adapter *adapter,
3032                             struct ibmvnic_sub_crq_queue *scrq)
3033 {
3034         struct device *dev = &adapter->vdev->dev;
3035         unsigned long rc;
3036
3037         rc = plpar_hcall_norets(H_VIOCTL, adapter->vdev->unit_address,
3038                                 H_DISABLE_VIO_INTERRUPT, scrq->hw_irq, 0, 0);
3039         if (rc)
3040                 dev_err(dev, "Couldn't disable scrq irq 0x%lx. rc=%ld\n",
3041                         scrq->hw_irq, rc);
3042         return rc;
3043 }
3044
3045 static int enable_scrq_irq(struct ibmvnic_adapter *adapter,
3046                            struct ibmvnic_sub_crq_queue *scrq)
3047 {
3048         struct device *dev = &adapter->vdev->dev;
3049         unsigned long rc;
3050
3051         if (scrq->hw_irq > 0x100000000ULL) {
3052                 dev_err(dev, "bad hw_irq = %lx\n", scrq->hw_irq);
3053                 return 1;
3054         }
3055
3056         if (test_bit(0, &adapter->resetting) &&
3057             adapter->reset_reason == VNIC_RESET_MOBILITY) {
3058                 u64 val = (0xff000000) | scrq->hw_irq;
3059
3060                 rc = plpar_hcall_norets(H_EOI, val);
3061                 /* H_EOI would fail with rc = H_FUNCTION when running
3062                  * in XIVE mode which is expected, but not an error.
3063                  */
3064                 if (rc && rc != H_FUNCTION)
3065                         dev_err(dev, "H_EOI FAILED irq 0x%llx. rc=%ld\n",
3066                                 val, rc);
3067         }
3068
3069         rc = plpar_hcall_norets(H_VIOCTL, adapter->vdev->unit_address,
3070                                 H_ENABLE_VIO_INTERRUPT, scrq->hw_irq, 0, 0);
3071         if (rc)
3072                 dev_err(dev, "Couldn't enable scrq irq 0x%lx. rc=%ld\n",
3073                         scrq->hw_irq, rc);
3074         return rc;
3075 }
3076
3077 static int ibmvnic_complete_tx(struct ibmvnic_adapter *adapter,
3078                                struct ibmvnic_sub_crq_queue *scrq)
3079 {
3080         struct device *dev = &adapter->vdev->dev;
3081         struct ibmvnic_tx_pool *tx_pool;
3082         struct ibmvnic_tx_buff *txbuff;
3083         union sub_crq *next;
3084         int index;
3085         int i, j;
3086
3087 restart_loop:
3088         while (pending_scrq(adapter, scrq)) {
3089                 unsigned int pool = scrq->pool_index;
3090                 int num_entries = 0;
3091
3092                 /* The queue entry at the current index is peeked at above
3093                  * to determine that there is a valid descriptor awaiting
3094                  * processing. We want to be sure that the current slot
3095                  * holds a valid descriptor before reading its contents.
3096                  */
3097                 dma_rmb();
3098
3099                 next = ibmvnic_next_scrq(adapter, scrq);
3100                 for (i = 0; i < next->tx_comp.num_comps; i++) {
3101                         if (next->tx_comp.rcs[i])
3102                                 dev_err(dev, "tx error %x\n",
3103                                         next->tx_comp.rcs[i]);
3104                         index = be32_to_cpu(next->tx_comp.correlators[i]);
3105                         if (index & IBMVNIC_TSO_POOL_MASK) {
3106                                 tx_pool = &adapter->tso_pool[pool];
3107                                 index &= ~IBMVNIC_TSO_POOL_MASK;
3108                         } else {
3109                                 tx_pool = &adapter->tx_pool[pool];
3110                         }
3111
3112                         txbuff = &tx_pool->tx_buff[index];
3113
3114                         for (j = 0; j < IBMVNIC_MAX_FRAGS_PER_CRQ; j++) {
3115                                 if (!txbuff->data_dma[j])
3116                                         continue;
3117
3118                                 txbuff->data_dma[j] = 0;
3119                         }
3120
3121                         if (txbuff->last_frag) {
3122                                 dev_kfree_skb_any(txbuff->skb);
3123                                 txbuff->skb = NULL;
3124                         }
3125
3126                         num_entries += txbuff->num_entries;
3127
3128                         tx_pool->free_map[tx_pool->producer_index] = index;
3129                         tx_pool->producer_index =
3130                                 (tx_pool->producer_index + 1) %
3131                                         tx_pool->num_buffers;
3132                 }
3133                 /* remove tx_comp scrq*/
3134                 next->tx_comp.first = 0;
3135
3136                 if (atomic_sub_return(num_entries, &scrq->used) <=
3137                     (adapter->req_tx_entries_per_subcrq / 2) &&
3138                     __netif_subqueue_stopped(adapter->netdev,
3139                                              scrq->pool_index)) {
3140                         netif_wake_subqueue(adapter->netdev, scrq->pool_index);
3141                         netdev_dbg(adapter->netdev, "Started queue %d\n",
3142                                    scrq->pool_index);
3143                 }
3144         }
3145
3146         enable_scrq_irq(adapter, scrq);
3147
3148         if (pending_scrq(adapter, scrq)) {
3149                 disable_scrq_irq(adapter, scrq);
3150                 goto restart_loop;
3151         }
3152
3153         return 0;
3154 }
3155
3156 static irqreturn_t ibmvnic_interrupt_tx(int irq, void *instance)
3157 {
3158         struct ibmvnic_sub_crq_queue *scrq = instance;
3159         struct ibmvnic_adapter *adapter = scrq->adapter;
3160
3161         disable_scrq_irq(adapter, scrq);
3162         ibmvnic_complete_tx(adapter, scrq);
3163
3164         return IRQ_HANDLED;
3165 }
3166
3167 static irqreturn_t ibmvnic_interrupt_rx(int irq, void *instance)
3168 {
3169         struct ibmvnic_sub_crq_queue *scrq = instance;
3170         struct ibmvnic_adapter *adapter = scrq->adapter;
3171
3172         /* When booting a kdump kernel we can hit pending interrupts
3173          * prior to completing driver initialization.
3174          */
3175         if (unlikely(adapter->state != VNIC_OPEN))
3176                 return IRQ_NONE;
3177
3178         adapter->rx_stats_buffers[scrq->scrq_num].interrupts++;
3179
3180         if (napi_schedule_prep(&adapter->napi[scrq->scrq_num])) {
3181                 disable_scrq_irq(adapter, scrq);
3182                 __napi_schedule(&adapter->napi[scrq->scrq_num]);
3183         }
3184
3185         return IRQ_HANDLED;
3186 }
3187
3188 static int init_sub_crq_irqs(struct ibmvnic_adapter *adapter)
3189 {
3190         struct device *dev = &adapter->vdev->dev;
3191         struct ibmvnic_sub_crq_queue *scrq;
3192         int i = 0, j = 0;
3193         int rc = 0;
3194
3195         for (i = 0; i < adapter->req_tx_queues; i++) {
3196                 netdev_dbg(adapter->netdev, "Initializing tx_scrq[%d] irq\n",
3197                            i);
3198                 scrq = adapter->tx_scrq[i];
3199                 scrq->irq = irq_create_mapping(NULL, scrq->hw_irq);
3200
3201                 if (!scrq->irq) {
3202                         rc = -EINVAL;
3203                         dev_err(dev, "Error mapping irq\n");
3204                         goto req_tx_irq_failed;
3205                 }
3206
3207                 snprintf(scrq->name, sizeof(scrq->name), "ibmvnic-%x-tx%d",
3208                          adapter->vdev->unit_address, i);
3209                 rc = request_irq(scrq->irq, ibmvnic_interrupt_tx,
3210                                  0, scrq->name, scrq);
3211
3212                 if (rc) {
3213                         dev_err(dev, "Couldn't register tx irq 0x%x. rc=%d\n",
3214                                 scrq->irq, rc);
3215                         irq_dispose_mapping(scrq->irq);
3216                         goto req_tx_irq_failed;
3217                 }
3218         }
3219
3220         for (i = 0; i < adapter->req_rx_queues; i++) {
3221                 netdev_dbg(adapter->netdev, "Initializing rx_scrq[%d] irq\n",
3222                            i);
3223                 scrq = adapter->rx_scrq[i];
3224                 scrq->irq = irq_create_mapping(NULL, scrq->hw_irq);
3225                 if (!scrq->irq) {
3226                         rc = -EINVAL;
3227                         dev_err(dev, "Error mapping irq\n");
3228                         goto req_rx_irq_failed;
3229                 }
3230                 snprintf(scrq->name, sizeof(scrq->name), "ibmvnic-%x-rx%d",
3231                          adapter->vdev->unit_address, i);
3232                 rc = request_irq(scrq->irq, ibmvnic_interrupt_rx,
3233                                  0, scrq->name, scrq);
3234                 if (rc) {
3235                         dev_err(dev, "Couldn't register rx irq 0x%x. rc=%d\n",
3236                                 scrq->irq, rc);
3237                         irq_dispose_mapping(scrq->irq);
3238                         goto req_rx_irq_failed;
3239                 }
3240         }
3241         return rc;
3242
3243 req_rx_irq_failed:
3244         for (j = 0; j < i; j++) {
3245                 free_irq(adapter->rx_scrq[j]->irq, adapter->rx_scrq[j]);
3246                 irq_dispose_mapping(adapter->rx_scrq[j]->irq);
3247         }
3248         i = adapter->req_tx_queues;
3249 req_tx_irq_failed:
3250         for (j = 0; j < i; j++) {
3251                 free_irq(adapter->tx_scrq[j]->irq, adapter->tx_scrq[j]);
3252                 irq_dispose_mapping(adapter->tx_scrq[j]->irq);
3253         }
3254         release_sub_crqs(adapter, 1);
3255         return rc;
3256 }
3257
3258 static int init_sub_crqs(struct ibmvnic_adapter *adapter)
3259 {
3260         struct device *dev = &adapter->vdev->dev;
3261         struct ibmvnic_sub_crq_queue **allqueues;
3262         int registered_queues = 0;
3263         int total_queues;
3264         int more = 0;
3265         int i;
3266
3267         total_queues = adapter->req_tx_queues + adapter->req_rx_queues;
3268
3269         allqueues = kcalloc(total_queues, sizeof(*allqueues), GFP_KERNEL);
3270         if (!allqueues)
3271                 return -1;
3272
3273         for (i = 0; i < total_queues; i++) {
3274                 allqueues[i] = init_sub_crq_queue(adapter);
3275                 if (!allqueues[i]) {
3276                         dev_warn(dev, "Couldn't allocate all sub-crqs\n");
3277                         break;
3278                 }
3279                 registered_queues++;
3280         }
3281
3282         /* Make sure we were able to register the minimum number of queues */
3283         if (registered_queues <
3284             adapter->min_tx_queues + adapter->min_rx_queues) {
3285                 dev_err(dev, "Fatal: Couldn't init  min number of sub-crqs\n");
3286                 goto tx_failed;
3287         }
3288
3289         /* Distribute the failed allocated queues*/
3290         for (i = 0; i < total_queues - registered_queues + more ; i++) {
3291                 netdev_dbg(adapter->netdev, "Reducing number of queues\n");
3292                 switch (i % 3) {
3293                 case 0:
3294                         if (adapter->req_rx_queues > adapter->min_rx_queues)
3295                                 adapter->req_rx_queues--;
3296                         else
3297                                 more++;
3298                         break;
3299                 case 1:
3300                         if (adapter->req_tx_queues > adapter->min_tx_queues)
3301                                 adapter->req_tx_queues--;
3302                         else
3303                                 more++;
3304                         break;
3305                 }
3306         }
3307
3308         adapter->tx_scrq = kcalloc(adapter->req_tx_queues,
3309                                    sizeof(*adapter->tx_scrq), GFP_KERNEL);
3310         if (!adapter->tx_scrq)
3311                 goto tx_failed;
3312
3313         for (i = 0; i < adapter->req_tx_queues; i++) {
3314                 adapter->tx_scrq[i] = allqueues[i];
3315                 adapter->tx_scrq[i]->pool_index = i;
3316                 adapter->num_active_tx_scrqs++;
3317         }
3318
3319         adapter->rx_scrq = kcalloc(adapter->req_rx_queues,
3320                                    sizeof(*adapter->rx_scrq), GFP_KERNEL);
3321         if (!adapter->rx_scrq)
3322                 goto rx_failed;
3323
3324         for (i = 0; i < adapter->req_rx_queues; i++) {
3325                 adapter->rx_scrq[i] = allqueues[i + adapter->req_tx_queues];
3326                 adapter->rx_scrq[i]->scrq_num = i;
3327                 adapter->num_active_rx_scrqs++;
3328         }
3329
3330         kfree(allqueues);
3331         return 0;
3332
3333 rx_failed:
3334         kfree(adapter->tx_scrq);
3335         adapter->tx_scrq = NULL;
3336 tx_failed:
3337         for (i = 0; i < registered_queues; i++)
3338                 release_sub_crq_queue(adapter, allqueues[i], 1);
3339         kfree(allqueues);
3340         return -1;
3341 }
3342
3343 static void send_request_cap(struct ibmvnic_adapter *adapter, int retry)
3344 {
3345         struct device *dev = &adapter->vdev->dev;
3346         union ibmvnic_crq crq;
3347         int max_entries;
3348         int cap_reqs;
3349
3350         /* We send out 6 or 7 REQUEST_CAPABILITY CRQs below (depending on
3351          * the PROMISC flag). Initialize this count upfront. When the tasklet
3352          * receives a response to all of these, it will send the next protocol
3353          * message (QUERY_IP_OFFLOAD).
3354          */
3355         if (!(adapter->netdev->flags & IFF_PROMISC) ||
3356             adapter->promisc_supported)
3357                 cap_reqs = 7;
3358         else
3359                 cap_reqs = 6;
3360
3361         if (!retry) {
3362                 /* Sub-CRQ entries are 32 byte long */
3363                 int entries_page = 4 * PAGE_SIZE / (sizeof(u64) * 4);
3364
3365                 atomic_set(&adapter->running_cap_crqs, cap_reqs);
3366
3367                 if (adapter->min_tx_entries_per_subcrq > entries_page ||
3368                     adapter->min_rx_add_entries_per_subcrq > entries_page) {
3369                         dev_err(dev, "Fatal, invalid entries per sub-crq\n");
3370                         return;
3371                 }
3372
3373                 if (adapter->desired.mtu)
3374                         adapter->req_mtu = adapter->desired.mtu;
3375                 else
3376                         adapter->req_mtu = adapter->netdev->mtu + ETH_HLEN;
3377
3378                 if (!adapter->desired.tx_entries)
3379                         adapter->desired.tx_entries =
3380                                         adapter->max_tx_entries_per_subcrq;
3381                 if (!adapter->desired.rx_entries)
3382                         adapter->desired.rx_entries =
3383                                         adapter->max_rx_add_entries_per_subcrq;
3384
3385                 max_entries = IBMVNIC_MAX_LTB_SIZE /
3386                               (adapter->req_mtu + IBMVNIC_BUFFER_HLEN);
3387
3388                 if ((adapter->req_mtu + IBMVNIC_BUFFER_HLEN) *
3389                         adapter->desired.tx_entries > IBMVNIC_MAX_LTB_SIZE) {
3390                         adapter->desired.tx_entries = max_entries;
3391                 }
3392
3393                 if ((adapter->req_mtu + IBMVNIC_BUFFER_HLEN) *
3394                         adapter->desired.rx_entries > IBMVNIC_MAX_LTB_SIZE) {
3395                         adapter->desired.rx_entries = max_entries;
3396                 }
3397
3398                 if (adapter->desired.tx_entries)
3399                         adapter->req_tx_entries_per_subcrq =
3400                                         adapter->desired.tx_entries;
3401                 else
3402                         adapter->req_tx_entries_per_subcrq =
3403                                         adapter->max_tx_entries_per_subcrq;
3404
3405                 if (adapter->desired.rx_entries)
3406                         adapter->req_rx_add_entries_per_subcrq =
3407                                         adapter->desired.rx_entries;
3408                 else
3409                         adapter->req_rx_add_entries_per_subcrq =
3410                                         adapter->max_rx_add_entries_per_subcrq;
3411
3412                 if (adapter->desired.tx_queues)
3413                         adapter->req_tx_queues =
3414                                         adapter->desired.tx_queues;
3415                 else
3416                         adapter->req_tx_queues =
3417                                         adapter->opt_tx_comp_sub_queues;
3418
3419                 if (adapter->desired.rx_queues)
3420                         adapter->req_rx_queues =
3421                                         adapter->desired.rx_queues;
3422                 else
3423                         adapter->req_rx_queues =
3424                                         adapter->opt_rx_comp_queues;
3425
3426                 adapter->req_rx_add_queues = adapter->max_rx_add_queues;
3427         } else {
3428                 atomic_add(cap_reqs, &adapter->running_cap_crqs);
3429         }
3430         memset(&crq, 0, sizeof(crq));
3431         crq.request_capability.first = IBMVNIC_CRQ_CMD;
3432         crq.request_capability.cmd = REQUEST_CAPABILITY;
3433
3434         crq.request_capability.capability = cpu_to_be16(REQ_TX_QUEUES);
3435         crq.request_capability.number = cpu_to_be64(adapter->req_tx_queues);
3436         cap_reqs--;
3437         ibmvnic_send_crq(adapter, &crq);
3438
3439         crq.request_capability.capability = cpu_to_be16(REQ_RX_QUEUES);
3440         crq.request_capability.number = cpu_to_be64(adapter->req_rx_queues);
3441         cap_reqs--;
3442         ibmvnic_send_crq(adapter, &crq);
3443
3444         crq.request_capability.capability = cpu_to_be16(REQ_RX_ADD_QUEUES);
3445         crq.request_capability.number = cpu_to_be64(adapter->req_rx_add_queues);
3446         cap_reqs--;
3447         ibmvnic_send_crq(adapter, &crq);
3448
3449         crq.request_capability.capability =
3450             cpu_to_be16(REQ_TX_ENTRIES_PER_SUBCRQ);
3451         crq.request_capability.number =
3452             cpu_to_be64(adapter->req_tx_entries_per_subcrq);
3453         cap_reqs--;
3454         ibmvnic_send_crq(adapter, &crq);
3455
3456         crq.request_capability.capability =
3457             cpu_to_be16(REQ_RX_ADD_ENTRIES_PER_SUBCRQ);
3458         crq.request_capability.number =
3459             cpu_to_be64(adapter->req_rx_add_entries_per_subcrq);
3460         cap_reqs--;
3461         ibmvnic_send_crq(adapter, &crq);
3462
3463         crq.request_capability.capability = cpu_to_be16(REQ_MTU);
3464         crq.request_capability.number = cpu_to_be64(adapter->req_mtu);
3465         cap_reqs--;
3466         ibmvnic_send_crq(adapter, &crq);
3467
3468         if (adapter->netdev->flags & IFF_PROMISC) {
3469                 if (adapter->promisc_supported) {
3470                         crq.request_capability.capability =
3471                             cpu_to_be16(PROMISC_REQUESTED);
3472                         crq.request_capability.number = cpu_to_be64(1);
3473                         cap_reqs--;
3474                         ibmvnic_send_crq(adapter, &crq);
3475                 }
3476         } else {
3477                 crq.request_capability.capability =
3478                     cpu_to_be16(PROMISC_REQUESTED);
3479                 crq.request_capability.number = cpu_to_be64(0);
3480                 cap_reqs--;
3481                 ibmvnic_send_crq(adapter, &crq);
3482         }
3483
3484         /* Keep at end to catch any discrepancy between expected and actual
3485          * CRQs sent.
3486          */
3487         WARN_ON(cap_reqs != 0);
3488 }
3489
3490 static int pending_scrq(struct ibmvnic_adapter *adapter,
3491                         struct ibmvnic_sub_crq_queue *scrq)
3492 {
3493         union sub_crq *entry = &scrq->msgs[scrq->cur];
3494
3495         if (entry->generic.first & IBMVNIC_CRQ_CMD_RSP)
3496                 return 1;
3497         else
3498                 return 0;
3499 }
3500
3501 static union sub_crq *ibmvnic_next_scrq(struct ibmvnic_adapter *adapter,
3502                                         struct ibmvnic_sub_crq_queue *scrq)
3503 {
3504         union sub_crq *entry;
3505         unsigned long flags;
3506
3507         spin_lock_irqsave(&scrq->lock, flags);
3508         entry = &scrq->msgs[scrq->cur];
3509         if (entry->generic.first & IBMVNIC_CRQ_CMD_RSP) {
3510                 if (++scrq->cur == scrq->size)
3511                         scrq->cur = 0;
3512         } else {
3513                 entry = NULL;
3514         }
3515         spin_unlock_irqrestore(&scrq->lock, flags);
3516
3517         /* Ensure that the entire buffer descriptor has been
3518          * loaded before reading its contents
3519          */
3520         dma_rmb();
3521
3522         return entry;
3523 }
3524
3525 static union ibmvnic_crq *ibmvnic_next_crq(struct ibmvnic_adapter *adapter)
3526 {
3527         struct ibmvnic_crq_queue *queue = &adapter->crq;
3528         union ibmvnic_crq *crq;
3529
3530         crq = &queue->msgs[queue->cur];
3531         if (crq->generic.first & IBMVNIC_CRQ_CMD_RSP) {
3532                 if (++queue->cur == queue->size)
3533                         queue->cur = 0;
3534         } else {
3535                 crq = NULL;
3536         }
3537
3538         return crq;
3539 }
3540
3541 static void print_subcrq_error(struct device *dev, int rc, const char *func)
3542 {
3543         switch (rc) {
3544         case H_PARAMETER:
3545                 dev_warn_ratelimited(dev,
3546                                      "%s failed: Send request is malformed or adapter failover pending. (rc=%d)\n",
3547                                      func, rc);
3548                 break;
3549         case H_CLOSED:
3550                 dev_warn_ratelimited(dev,
3551                                      "%s failed: Backing queue closed. Adapter is down or failover pending. (rc=%d)\n",
3552                                      func, rc);
3553                 break;
3554         default:
3555                 dev_err_ratelimited(dev, "%s failed: (rc=%d)\n", func, rc);
3556                 break;
3557         }
3558 }
3559
3560 static int send_subcrq(struct ibmvnic_adapter *adapter, u64 remote_handle,
3561                        union sub_crq *sub_crq)
3562 {
3563         unsigned int ua = adapter->vdev->unit_address;
3564         struct device *dev = &adapter->vdev->dev;
3565         u64 *u64_crq = (u64 *)sub_crq;
3566         int rc;
3567
3568         netdev_dbg(adapter->netdev,
3569                    "Sending sCRQ %016lx: %016lx %016lx %016lx %016lx\n",
3570                    (unsigned long int)cpu_to_be64(remote_handle),
3571                    (unsigned long int)cpu_to_be64(u64_crq[0]),
3572                    (unsigned long int)cpu_to_be64(u64_crq[1]),
3573                    (unsigned long int)cpu_to_be64(u64_crq[2]),
3574                    (unsigned long int)cpu_to_be64(u64_crq[3]));
3575
3576         /* Make sure the hypervisor sees the complete request */
3577         mb();
3578
3579         rc = plpar_hcall_norets(H_SEND_SUB_CRQ, ua,
3580                                 cpu_to_be64(remote_handle),
3581                                 cpu_to_be64(u64_crq[0]),
3582                                 cpu_to_be64(u64_crq[1]),
3583                                 cpu_to_be64(u64_crq[2]),
3584                                 cpu_to_be64(u64_crq[3]));
3585
3586         if (rc)
3587                 print_subcrq_error(dev, rc, __func__);
3588
3589         return rc;
3590 }
3591
3592 static int send_subcrq_indirect(struct ibmvnic_adapter *adapter,
3593                                 u64 remote_handle, u64 ioba, u64 num_entries)
3594 {
3595         unsigned int ua = adapter->vdev->unit_address;
3596         struct device *dev = &adapter->vdev->dev;
3597         int rc;
3598
3599         /* Make sure the hypervisor sees the complete request */
3600         mb();
3601         rc = plpar_hcall_norets(H_SEND_SUB_CRQ_INDIRECT, ua,
3602                                 cpu_to_be64(remote_handle),
3603                                 ioba, num_entries);
3604
3605         if (rc)
3606                 print_subcrq_error(dev, rc, __func__);
3607
3608         return rc;
3609 }
3610
3611 static int ibmvnic_send_crq(struct ibmvnic_adapter *adapter,
3612                             union ibmvnic_crq *crq)
3613 {
3614         unsigned int ua = adapter->vdev->unit_address;
3615         struct device *dev = &adapter->vdev->dev;
3616         u64 *u64_crq = (u64 *)crq;
3617         int rc;
3618
3619         netdev_dbg(adapter->netdev, "Sending CRQ: %016lx %016lx\n",
3620                    (unsigned long int)cpu_to_be64(u64_crq[0]),
3621                    (unsigned long int)cpu_to_be64(u64_crq[1]));
3622
3623         if (!adapter->crq.active &&
3624             crq->generic.first != IBMVNIC_CRQ_INIT_CMD) {
3625                 dev_warn(dev, "Invalid request detected while CRQ is inactive, possible device state change during reset\n");
3626                 return -EINVAL;
3627         }
3628
3629         /* Make sure the hypervisor sees the complete request */
3630         mb();
3631
3632         rc = plpar_hcall_norets(H_SEND_CRQ, ua,
3633                                 cpu_to_be64(u64_crq[0]),
3634                                 cpu_to_be64(u64_crq[1]));
3635
3636         if (rc) {
3637                 if (rc == H_CLOSED) {
3638                         dev_warn(dev, "CRQ Queue closed\n");
3639                         /* do not reset, report the fail, wait for passive init from server */
3640                 }
3641
3642                 dev_warn(dev, "Send error (rc=%d)\n", rc);
3643         }
3644
3645         return rc;
3646 }
3647
3648 static int ibmvnic_send_crq_init(struct ibmvnic_adapter *adapter)
3649 {
3650         struct device *dev = &adapter->vdev->dev;
3651         union ibmvnic_crq crq;
3652         int retries = 100;
3653         int rc;
3654
3655         memset(&crq, 0, sizeof(crq));
3656         crq.generic.first = IBMVNIC_CRQ_INIT_CMD;
3657         crq.generic.cmd = IBMVNIC_CRQ_INIT;
3658         netdev_dbg(adapter->netdev, "Sending CRQ init\n");
3659
3660         do {
3661                 rc = ibmvnic_send_crq(adapter, &crq);
3662                 if (rc != H_CLOSED)
3663                         break;
3664                 retries--;
3665                 msleep(50);
3666
3667         } while (retries > 0);
3668
3669         if (rc) {
3670                 dev_err(dev, "Failed to send init request, rc = %d\n", rc);
3671                 return rc;
3672         }
3673
3674         return 0;
3675 }
3676
3677 static int send_version_xchg(struct ibmvnic_adapter *adapter)
3678 {
3679         union ibmvnic_crq crq;
3680
3681         memset(&crq, 0, sizeof(crq));
3682         crq.version_exchange.first = IBMVNIC_CRQ_CMD;
3683         crq.version_exchange.cmd = VERSION_EXCHANGE;
3684         crq.version_exchange.version = cpu_to_be16(ibmvnic_version);
3685
3686         return ibmvnic_send_crq(adapter, &crq);
3687 }
3688
3689 struct vnic_login_client_data {
3690         u8      type;
3691         __be16  len;
3692         char    name[];
3693 } __packed;
3694
3695 static int vnic_client_data_len(struct ibmvnic_adapter *adapter)
3696 {
3697         int len;
3698
3699         /* Calculate the amount of buffer space needed for the
3700          * vnic client data in the login buffer. There are four entries,
3701          * OS name, LPAR name, device name, and a null last entry.
3702          */
3703         len = 4 * sizeof(struct vnic_login_client_data);
3704         len += 6; /* "Linux" plus NULL */
3705         len += strlen(utsname()->nodename) + 1;
3706         len += strlen(adapter->netdev->name) + 1;
3707
3708         return len;
3709 }
3710
3711 static void vnic_add_client_data(struct ibmvnic_adapter *adapter,
3712                                  struct vnic_login_client_data *vlcd)
3713 {
3714         const char *os_name = "Linux";
3715         int len;
3716
3717         /* Type 1 - LPAR OS */
3718         vlcd->type = 1;
3719         len = strlen(os_name) + 1;
3720         vlcd->len = cpu_to_be16(len);
3721         strncpy(vlcd->name, os_name, len);
3722         vlcd = (struct vnic_login_client_data *)(vlcd->name + len);
3723
3724         /* Type 2 - LPAR name */
3725         vlcd->type = 2;
3726         len = strlen(utsname()->nodename) + 1;
3727         vlcd->len = cpu_to_be16(len);
3728         strncpy(vlcd->name, utsname()->nodename, len);
3729         vlcd = (struct vnic_login_client_data *)(vlcd->name + len);
3730
3731         /* Type 3 - device name */
3732         vlcd->type = 3;
3733         len = strlen(adapter->netdev->name) + 1;
3734         vlcd->len = cpu_to_be16(len);
3735         strncpy(vlcd->name, adapter->netdev->name, len);
3736 }
3737
3738 static int send_login(struct ibmvnic_adapter *adapter)
3739 {
3740         struct ibmvnic_login_rsp_buffer *login_rsp_buffer;
3741         struct ibmvnic_login_buffer *login_buffer;
3742         struct device *dev = &adapter->vdev->dev;
3743         struct vnic_login_client_data *vlcd;
3744         dma_addr_t rsp_buffer_token;
3745         dma_addr_t buffer_token;
3746         size_t rsp_buffer_size;
3747         union ibmvnic_crq crq;
3748         int client_data_len;
3749         size_t buffer_size;
3750         __be64 *tx_list_p;
3751         __be64 *rx_list_p;
3752         int rc;
3753         int i;
3754
3755         if (!adapter->tx_scrq || !adapter->rx_scrq) {
3756                 netdev_err(adapter->netdev,
3757                            "RX or TX queues are not allocated, device login failed\n");
3758                 return -1;
3759         }
3760
3761         release_login_buffer(adapter);
3762         release_login_rsp_buffer(adapter);
3763
3764         client_data_len = vnic_client_data_len(adapter);
3765
3766         buffer_size =
3767             sizeof(struct ibmvnic_login_buffer) +
3768             sizeof(u64) * (adapter->req_tx_queues + adapter->req_rx_queues) +
3769             client_data_len;
3770
3771         login_buffer = kzalloc(buffer_size, GFP_ATOMIC);
3772         if (!login_buffer)
3773                 goto buf_alloc_failed;
3774
3775         buffer_token = dma_map_single(dev, login_buffer, buffer_size,
3776                                       DMA_TO_DEVICE);
3777         if (dma_mapping_error(dev, buffer_token)) {
3778                 dev_err(dev, "Couldn't map login buffer\n");
3779                 goto buf_map_failed;
3780         }
3781
3782         rsp_buffer_size = sizeof(struct ibmvnic_login_rsp_buffer) +
3783                           sizeof(u64) * adapter->req_tx_queues +
3784                           sizeof(u64) * adapter->req_rx_queues +
3785                           sizeof(u64) * adapter->req_rx_queues +
3786                           sizeof(u8) * IBMVNIC_TX_DESC_VERSIONS;
3787
3788         login_rsp_buffer = kmalloc(rsp_buffer_size, GFP_ATOMIC);
3789         if (!login_rsp_buffer)
3790                 goto buf_rsp_alloc_failed;
3791
3792         rsp_buffer_token = dma_map_single(dev, login_rsp_buffer,
3793                                           rsp_buffer_size, DMA_FROM_DEVICE);
3794         if (dma_mapping_error(dev, rsp_buffer_token)) {
3795                 dev_err(dev, "Couldn't map login rsp buffer\n");
3796                 goto buf_rsp_map_failed;
3797         }
3798
3799         adapter->login_buf = login_buffer;
3800         adapter->login_buf_token = buffer_token;
3801         adapter->login_buf_sz = buffer_size;
3802         adapter->login_rsp_buf = login_rsp_buffer;
3803         adapter->login_rsp_buf_token = rsp_buffer_token;
3804         adapter->login_rsp_buf_sz = rsp_buffer_size;
3805
3806         login_buffer->len = cpu_to_be32(buffer_size);
3807         login_buffer->version = cpu_to_be32(INITIAL_VERSION_LB);
3808         login_buffer->num_txcomp_subcrqs = cpu_to_be32(adapter->req_tx_queues);
3809         login_buffer->off_txcomp_subcrqs =
3810             cpu_to_be32(sizeof(struct ibmvnic_login_buffer));
3811         login_buffer->num_rxcomp_subcrqs = cpu_to_be32(adapter->req_rx_queues);
3812         login_buffer->off_rxcomp_subcrqs =
3813             cpu_to_be32(sizeof(struct ibmvnic_login_buffer) +
3814                         sizeof(u64) * adapter->req_tx_queues);
3815         login_buffer->login_rsp_ioba = cpu_to_be32(rsp_buffer_token);
3816         login_buffer->login_rsp_len = cpu_to_be32(rsp_buffer_size);
3817
3818         tx_list_p = (__be64 *)((char *)login_buffer +
3819                                       sizeof(struct ibmvnic_login_buffer));
3820         rx_list_p = (__be64 *)((char *)login_buffer +
3821                                       sizeof(struct ibmvnic_login_buffer) +
3822                                       sizeof(u64) * adapter->req_tx_queues);
3823
3824         for (i = 0; i < adapter->req_tx_queues; i++) {
3825                 if (adapter->tx_scrq[i]) {
3826                         tx_list_p[i] = cpu_to_be64(adapter->tx_scrq[i]->
3827                                                    crq_num);
3828                 }
3829         }
3830
3831         for (i = 0; i < adapter->req_rx_queues; i++) {
3832                 if (adapter->rx_scrq[i]) {
3833                         rx_list_p[i] = cpu_to_be64(adapter->rx_scrq[i]->
3834                                                    crq_num);
3835                 }
3836         }
3837
3838         /* Insert vNIC login client data */
3839         vlcd = (struct vnic_login_client_data *)
3840                 ((char *)rx_list_p + (sizeof(u64) * adapter->req_rx_queues));
3841         login_buffer->client_data_offset =
3842                         cpu_to_be32((char *)vlcd - (char *)login_buffer);
3843         login_buffer->client_data_len = cpu_to_be32(client_data_len);
3844
3845         vnic_add_client_data(adapter, vlcd);
3846
3847         netdev_dbg(adapter->netdev, "Login Buffer:\n");
3848         for (i = 0; i < (adapter->login_buf_sz - 1) / 8 + 1; i++) {
3849                 netdev_dbg(adapter->netdev, "%016lx\n",
3850                            ((unsigned long int *)(adapter->login_buf))[i]);
3851         }
3852
3853         memset(&crq, 0, sizeof(crq));
3854         crq.login.first = IBMVNIC_CRQ_CMD;
3855         crq.login.cmd = LOGIN;
3856         crq.login.ioba = cpu_to_be32(buffer_token);
3857         crq.login.len = cpu_to_be32(buffer_size);
3858
3859         adapter->login_pending = true;
3860         rc = ibmvnic_send_crq(adapter, &crq);
3861         if (rc) {
3862                 adapter->login_pending = false;
3863                 netdev_err(adapter->netdev, "Failed to send login, rc=%d\n", rc);
3864                 goto buf_rsp_map_failed;
3865         }
3866
3867         return 0;
3868
3869 buf_rsp_map_failed:
3870         kfree(login_rsp_buffer);
3871         adapter->login_rsp_buf = NULL;
3872 buf_rsp_alloc_failed:
3873         dma_unmap_single(dev, buffer_token, buffer_size, DMA_TO_DEVICE);
3874 buf_map_failed:
3875         kfree(login_buffer);
3876         adapter->login_buf = NULL;
3877 buf_alloc_failed:
3878         return -1;
3879 }
3880
3881 static int send_request_map(struct ibmvnic_adapter *adapter, dma_addr_t addr,
3882                             u32 len, u8 map_id)
3883 {
3884         union ibmvnic_crq crq;
3885
3886         memset(&crq, 0, sizeof(crq));
3887         crq.request_map.first = IBMVNIC_CRQ_CMD;
3888         crq.request_map.cmd = REQUEST_MAP;
3889         crq.request_map.map_id = map_id;
3890         crq.request_map.ioba = cpu_to_be32(addr);
3891         crq.request_map.len = cpu_to_be32(len);
3892         return ibmvnic_send_crq(adapter, &crq);
3893 }
3894
3895 static int send_request_unmap(struct ibmvnic_adapter *adapter, u8 map_id)
3896 {
3897         union ibmvnic_crq crq;
3898
3899         memset(&crq, 0, sizeof(crq));
3900         crq.request_unmap.first = IBMVNIC_CRQ_CMD;
3901         crq.request_unmap.cmd = REQUEST_UNMAP;
3902         crq.request_unmap.map_id = map_id;
3903         return ibmvnic_send_crq(adapter, &crq);
3904 }
3905
3906 static void send_query_map(struct ibmvnic_adapter *adapter)
3907 {
3908         union ibmvnic_crq crq;
3909
3910         memset(&crq, 0, sizeof(crq));
3911         crq.query_map.first = IBMVNIC_CRQ_CMD;
3912         crq.query_map.cmd = QUERY_MAP;
3913         ibmvnic_send_crq(adapter, &crq);
3914 }
3915
3916 /* Send a series of CRQs requesting various capabilities of the VNIC server */
3917 static void send_query_cap(struct ibmvnic_adapter *adapter)
3918 {
3919         union ibmvnic_crq crq;
3920         int cap_reqs;
3921
3922         /* We send out 25 QUERY_CAPABILITY CRQs below.  Initialize this count
3923          * upfront. When the tasklet receives a response to all of these, it
3924          * can send out the next protocol messaage (REQUEST_CAPABILITY).
3925          */
3926         cap_reqs = 25;
3927
3928         atomic_set(&adapter->running_cap_crqs, cap_reqs);
3929
3930         memset(&crq, 0, sizeof(crq));
3931         crq.query_capability.first = IBMVNIC_CRQ_CMD;
3932         crq.query_capability.cmd = QUERY_CAPABILITY;
3933
3934         crq.query_capability.capability = cpu_to_be16(MIN_TX_QUEUES);
3935         ibmvnic_send_crq(adapter, &crq);
3936         cap_reqs--;
3937
3938         crq.query_capability.capability = cpu_to_be16(MIN_RX_QUEUES);
3939         ibmvnic_send_crq(adapter, &crq);
3940         cap_reqs--;
3941
3942         crq.query_capability.capability = cpu_to_be16(MIN_RX_ADD_QUEUES);
3943         ibmvnic_send_crq(adapter, &crq);
3944         cap_reqs--;
3945
3946         crq.query_capability.capability = cpu_to_be16(MAX_TX_QUEUES);
3947         ibmvnic_send_crq(adapter, &crq);
3948         cap_reqs--;
3949
3950         crq.query_capability.capability = cpu_to_be16(MAX_RX_QUEUES);
3951         ibmvnic_send_crq(adapter, &crq);
3952         cap_reqs--;
3953
3954         crq.query_capability.capability = cpu_to_be16(MAX_RX_ADD_QUEUES);
3955         ibmvnic_send_crq(adapter, &crq);
3956         cap_reqs--;
3957
3958         crq.query_capability.capability =
3959             cpu_to_be16(MIN_TX_ENTRIES_PER_SUBCRQ);
3960         ibmvnic_send_crq(adapter, &crq);
3961         cap_reqs--;
3962
3963         crq.query_capability.capability =
3964             cpu_to_be16(MIN_RX_ADD_ENTRIES_PER_SUBCRQ);
3965         ibmvnic_send_crq(adapter, &crq);
3966         cap_reqs--;
3967
3968         crq.query_capability.capability =
3969             cpu_to_be16(MAX_TX_ENTRIES_PER_SUBCRQ);
3970         ibmvnic_send_crq(adapter, &crq);
3971         cap_reqs--;
3972
3973         crq.query_capability.capability =
3974             cpu_to_be16(MAX_RX_ADD_ENTRIES_PER_SUBCRQ);
3975         ibmvnic_send_crq(adapter, &crq);
3976         cap_reqs--;
3977
3978         crq.query_capability.capability = cpu_to_be16(TCP_IP_OFFLOAD);
3979         ibmvnic_send_crq(adapter, &crq);
3980         cap_reqs--;
3981
3982         crq.query_capability.capability = cpu_to_be16(PROMISC_SUPPORTED);
3983         ibmvnic_send_crq(adapter, &crq);
3984         cap_reqs--;
3985
3986         crq.query_capability.capability = cpu_to_be16(MIN_MTU);
3987         ibmvnic_send_crq(adapter, &crq);
3988         cap_reqs--;
3989
3990         crq.query_capability.capability = cpu_to_be16(MAX_MTU);
3991         ibmvnic_send_crq(adapter, &crq);
3992         cap_reqs--;
3993
3994         crq.query_capability.capability = cpu_to_be16(MAX_MULTICAST_FILTERS);
3995         ibmvnic_send_crq(adapter, &crq);
3996         cap_reqs--;
3997
3998         crq.query_capability.capability = cpu_to_be16(VLAN_HEADER_INSERTION);
3999         ibmvnic_send_crq(adapter, &crq);
4000         cap_reqs--;
4001
4002         crq.query_capability.capability = cpu_to_be16(RX_VLAN_HEADER_INSERTION);
4003         ibmvnic_send_crq(adapter, &crq);
4004         cap_reqs--;
4005
4006         crq.query_capability.capability = cpu_to_be16(MAX_TX_SG_ENTRIES);
4007         ibmvnic_send_crq(adapter, &crq);
4008         cap_reqs--;
4009
4010         crq.query_capability.capability = cpu_to_be16(RX_SG_SUPPORTED);
4011         ibmvnic_send_crq(adapter, &crq);
4012         cap_reqs--;
4013
4014         crq.query_capability.capability = cpu_to_be16(OPT_TX_COMP_SUB_QUEUES);
4015         ibmvnic_send_crq(adapter, &crq);
4016         cap_reqs--;
4017
4018         crq.query_capability.capability = cpu_to_be16(OPT_RX_COMP_QUEUES);
4019         ibmvnic_send_crq(adapter, &crq);
4020         cap_reqs--;
4021
4022         crq.query_capability.capability =
4023                         cpu_to_be16(OPT_RX_BUFADD_Q_PER_RX_COMP_Q);
4024         ibmvnic_send_crq(adapter, &crq);
4025         cap_reqs--;
4026
4027         crq.query_capability.capability =
4028                         cpu_to_be16(OPT_TX_ENTRIES_PER_SUBCRQ);
4029         ibmvnic_send_crq(adapter, &crq);
4030         cap_reqs--;
4031
4032         crq.query_capability.capability =
4033                         cpu_to_be16(OPT_RXBA_ENTRIES_PER_SUBCRQ);
4034         ibmvnic_send_crq(adapter, &crq);
4035         cap_reqs--;
4036
4037         crq.query_capability.capability = cpu_to_be16(TX_RX_DESC_REQ);
4038
4039         ibmvnic_send_crq(adapter, &crq);
4040         cap_reqs--;
4041
4042         /* Keep at end to catch any discrepancy between expected and actual
4043          * CRQs sent.
4044          */
4045         WARN_ON(cap_reqs != 0);
4046 }
4047
4048 static void send_query_ip_offload(struct ibmvnic_adapter *adapter)
4049 {
4050         int buf_sz = sizeof(struct ibmvnic_query_ip_offload_buffer);
4051         struct device *dev = &adapter->vdev->dev;
4052         union ibmvnic_crq crq;
4053
4054         adapter->ip_offload_tok =
4055                 dma_map_single(dev,
4056                                &adapter->ip_offload_buf,
4057                                buf_sz,
4058                                DMA_FROM_DEVICE);
4059
4060         if (dma_mapping_error(dev, adapter->ip_offload_tok)) {
4061                 if (!firmware_has_feature(FW_FEATURE_CMO))
4062                         dev_err(dev, "Couldn't map offload buffer\n");
4063                 return;
4064         }
4065
4066         memset(&crq, 0, sizeof(crq));
4067         crq.query_ip_offload.first = IBMVNIC_CRQ_CMD;
4068         crq.query_ip_offload.cmd = QUERY_IP_OFFLOAD;
4069         crq.query_ip_offload.len = cpu_to_be32(buf_sz);
4070         crq.query_ip_offload.ioba =
4071             cpu_to_be32(adapter->ip_offload_tok);
4072
4073         ibmvnic_send_crq(adapter, &crq);
4074 }
4075
4076 static void send_control_ip_offload(struct ibmvnic_adapter *adapter)
4077 {
4078         struct ibmvnic_control_ip_offload_buffer *ctrl_buf = &adapter->ip_offload_ctrl;
4079         struct ibmvnic_query_ip_offload_buffer *buf = &adapter->ip_offload_buf;
4080         struct device *dev = &adapter->vdev->dev;
4081         netdev_features_t old_hw_features = 0;
4082         union ibmvnic_crq crq;
4083
4084         adapter->ip_offload_ctrl_tok =
4085                 dma_map_single(dev,
4086                                ctrl_buf,
4087                                sizeof(adapter->ip_offload_ctrl),
4088                                DMA_TO_DEVICE);
4089
4090         if (dma_mapping_error(dev, adapter->ip_offload_ctrl_tok)) {
4091                 dev_err(dev, "Couldn't map ip offload control buffer\n");
4092                 return;
4093         }
4094
4095         ctrl_buf->len = cpu_to_be32(sizeof(adapter->ip_offload_ctrl));
4096         ctrl_buf->version = cpu_to_be32(INITIAL_VERSION_IOB);
4097         ctrl_buf->ipv4_chksum = buf->ipv4_chksum;
4098         ctrl_buf->ipv6_chksum = buf->ipv6_chksum;
4099         ctrl_buf->tcp_ipv4_chksum = buf->tcp_ipv4_chksum;
4100         ctrl_buf->udp_ipv4_chksum = buf->udp_ipv4_chksum;
4101         ctrl_buf->tcp_ipv6_chksum = buf->tcp_ipv6_chksum;
4102         ctrl_buf->udp_ipv6_chksum = buf->udp_ipv6_chksum;
4103         ctrl_buf->large_tx_ipv4 = buf->large_tx_ipv4;
4104         ctrl_buf->large_tx_ipv6 = buf->large_tx_ipv6;
4105
4106         /* large_rx disabled for now, additional features needed */
4107         ctrl_buf->large_rx_ipv4 = 0;
4108         ctrl_buf->large_rx_ipv6 = 0;
4109
4110         if (adapter->state != VNIC_PROBING) {
4111                 old_hw_features = adapter->netdev->hw_features;
4112                 adapter->netdev->hw_features = 0;
4113         }
4114
4115         adapter->netdev->hw_features = NETIF_F_SG | NETIF_F_GSO | NETIF_F_GRO;
4116
4117         if (buf->tcp_ipv4_chksum || buf->udp_ipv4_chksum)
4118                 adapter->netdev->hw_features |= NETIF_F_IP_CSUM;
4119
4120         if (buf->tcp_ipv6_chksum || buf->udp_ipv6_chksum)
4121                 adapter->netdev->hw_features |= NETIF_F_IPV6_CSUM;
4122
4123         if ((adapter->netdev->features &
4124             (NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM)))
4125                 adapter->netdev->hw_features |= NETIF_F_RXCSUM;
4126
4127         if (buf->large_tx_ipv4)
4128                 adapter->netdev->hw_features |= NETIF_F_TSO;
4129         if (buf->large_tx_ipv6)
4130                 adapter->netdev->hw_features |= NETIF_F_TSO6;
4131
4132         if (adapter->state == VNIC_PROBING) {
4133                 adapter->netdev->features |= adapter->netdev->hw_features;
4134         } else if (old_hw_features != adapter->netdev->hw_features) {
4135                 netdev_features_t tmp = 0;
4136
4137                 /* disable features no longer supported */
4138                 adapter->netdev->features &= adapter->netdev->hw_features;
4139                 /* turn on features now supported if previously enabled */
4140                 tmp = (old_hw_features ^ adapter->netdev->hw_features) &
4141                         adapter->netdev->hw_features;
4142                 adapter->netdev->features |=
4143                                 tmp & adapter->netdev->wanted_features;
4144         }
4145
4146         memset(&crq, 0, sizeof(crq));
4147         crq.control_ip_offload.first = IBMVNIC_CRQ_CMD;
4148         crq.control_ip_offload.cmd = CONTROL_IP_OFFLOAD;
4149         crq.control_ip_offload.len =
4150             cpu_to_be32(sizeof(adapter->ip_offload_ctrl));
4151         crq.control_ip_offload.ioba = cpu_to_be32(adapter->ip_offload_ctrl_tok);
4152         ibmvnic_send_crq(adapter, &crq);
4153 }
4154
4155 static void handle_vpd_size_rsp(union ibmvnic_crq *crq,
4156                                 struct ibmvnic_adapter *adapter)
4157 {
4158         struct device *dev = &adapter->vdev->dev;
4159
4160         if (crq->get_vpd_size_rsp.rc.code) {
4161                 dev_err(dev, "Error retrieving VPD size, rc=%x\n",
4162                         crq->get_vpd_size_rsp.rc.code);
4163                 complete(&adapter->fw_done);
4164                 return;
4165         }
4166
4167         adapter->vpd->len = be64_to_cpu(crq->get_vpd_size_rsp.len);
4168         complete(&adapter->fw_done);
4169 }
4170
4171 static void handle_vpd_rsp(union ibmvnic_crq *crq,
4172                            struct ibmvnic_adapter *adapter)
4173 {
4174         struct device *dev = &adapter->vdev->dev;
4175         unsigned char *substr = NULL;
4176         u8 fw_level_len = 0;
4177
4178         memset(adapter->fw_version, 0, 32);
4179
4180         dma_unmap_single(dev, adapter->vpd->dma_addr, adapter->vpd->len,
4181                          DMA_FROM_DEVICE);
4182
4183         if (crq->get_vpd_rsp.rc.code) {
4184                 dev_err(dev, "Error retrieving VPD from device, rc=%x\n",
4185                         crq->get_vpd_rsp.rc.code);
4186                 goto complete;
4187         }
4188
4189         /* get the position of the firmware version info
4190          * located after the ASCII 'RM' substring in the buffer
4191          */
4192         substr = strnstr(adapter->vpd->buff, "RM", adapter->vpd->len);
4193         if (!substr) {
4194                 dev_info(dev, "Warning - No FW level has been provided in the VPD buffer by the VIOS Server\n");
4195                 goto complete;
4196         }
4197
4198         /* get length of firmware level ASCII substring */
4199         if ((substr + 2) < (adapter->vpd->buff + adapter->vpd->len)) {
4200                 fw_level_len = *(substr + 2);
4201         } else {
4202                 dev_info(dev, "Length of FW substr extrapolated VDP buff\n");
4203                 goto complete;
4204         }
4205
4206         /* copy firmware version string from vpd into adapter */
4207         if ((substr + 3 + fw_level_len) <
4208             (adapter->vpd->buff + adapter->vpd->len)) {
4209                 strncpy((char *)adapter->fw_version, substr + 3, fw_level_len);
4210         } else {
4211                 dev_info(dev, "FW substr extrapolated VPD buff\n");
4212         }
4213
4214 complete:
4215         if (adapter->fw_version[0] == '\0')
4216                 strncpy((char *)adapter->fw_version, "N/A", 3 * sizeof(char));
4217         complete(&adapter->fw_done);
4218 }
4219
4220 static void handle_query_ip_offload_rsp(struct ibmvnic_adapter *adapter)
4221 {
4222         struct device *dev = &adapter->vdev->dev;
4223         struct ibmvnic_query_ip_offload_buffer *buf = &adapter->ip_offload_buf;
4224         int i;
4225
4226         dma_unmap_single(dev, adapter->ip_offload_tok,
4227                          sizeof(adapter->ip_offload_buf), DMA_FROM_DEVICE);
4228
4229         netdev_dbg(adapter->netdev, "Query IP Offload Buffer:\n");
4230         for (i = 0; i < (sizeof(adapter->ip_offload_buf) - 1) / 8 + 1; i++)
4231                 netdev_dbg(adapter->netdev, "%016lx\n",
4232                            ((unsigned long int *)(buf))[i]);
4233
4234         netdev_dbg(adapter->netdev, "ipv4_chksum = %d\n", buf->ipv4_chksum);
4235         netdev_dbg(adapter->netdev, "ipv6_chksum = %d\n", buf->ipv6_chksum);
4236         netdev_dbg(adapter->netdev, "tcp_ipv4_chksum = %d\n",
4237                    buf->tcp_ipv4_chksum);
4238         netdev_dbg(adapter->netdev, "tcp_ipv6_chksum = %d\n",
4239                    buf->tcp_ipv6_chksum);
4240         netdev_dbg(adapter->netdev, "udp_ipv4_chksum = %d\n",
4241                    buf->udp_ipv4_chksum);
4242         netdev_dbg(adapter->netdev, "udp_ipv6_chksum = %d\n",
4243                    buf->udp_ipv6_chksum);
4244         netdev_dbg(adapter->netdev, "large_tx_ipv4 = %d\n",
4245                    buf->large_tx_ipv4);
4246         netdev_dbg(adapter->netdev, "large_tx_ipv6 = %d\n",
4247                    buf->large_tx_ipv6);
4248         netdev_dbg(adapter->netdev, "large_rx_ipv4 = %d\n",
4249                    buf->large_rx_ipv4);
4250         netdev_dbg(adapter->netdev, "large_rx_ipv6 = %d\n",
4251                    buf->large_rx_ipv6);
4252         netdev_dbg(adapter->netdev, "max_ipv4_hdr_sz = %d\n",
4253                    buf->max_ipv4_header_size);
4254         netdev_dbg(adapter->netdev, "max_ipv6_hdr_sz = %d\n",
4255                    buf->max_ipv6_header_size);
4256         netdev_dbg(adapter->netdev, "max_tcp_hdr_size = %d\n",
4257                    buf->max_tcp_header_size);
4258         netdev_dbg(adapter->netdev, "max_udp_hdr_size = %d\n",
4259                    buf->max_udp_header_size);
4260         netdev_dbg(adapter->netdev, "max_large_tx_size = %d\n",
4261                    buf->max_large_tx_size);
4262         netdev_dbg(adapter->netdev, "max_large_rx_size = %d\n",
4263                    buf->max_large_rx_size);
4264         netdev_dbg(adapter->netdev, "ipv6_ext_hdr = %d\n",
4265                    buf->ipv6_extension_header);
4266         netdev_dbg(adapter->netdev, "tcp_pseudosum_req = %d\n",
4267                    buf->tcp_pseudosum_req);
4268         netdev_dbg(adapter->netdev, "num_ipv6_ext_hd = %d\n",
4269                    buf->num_ipv6_ext_headers);
4270         netdev_dbg(adapter->netdev, "off_ipv6_ext_hd = %d\n",
4271                    buf->off_ipv6_ext_headers);
4272
4273         send_control_ip_offload(adapter);
4274 }
4275
4276 static const char *ibmvnic_fw_err_cause(u16 cause)
4277 {
4278         switch (cause) {
4279         case ADAPTER_PROBLEM:
4280                 return "adapter problem";
4281         case BUS_PROBLEM:
4282                 return "bus problem";
4283         case FW_PROBLEM:
4284                 return "firmware problem";
4285         case DD_PROBLEM:
4286                 return "device driver problem";
4287         case EEH_RECOVERY:
4288                 return "EEH recovery";
4289         case FW_UPDATED:
4290                 return "firmware updated";
4291         case LOW_MEMORY:
4292                 return "low Memory";
4293         default:
4294                 return "unknown";
4295         }
4296 }
4297
4298 static void handle_error_indication(union ibmvnic_crq *crq,
4299                                     struct ibmvnic_adapter *adapter)
4300 {
4301         struct device *dev = &adapter->vdev->dev;
4302         u16 cause;
4303
4304         cause = be16_to_cpu(crq->error_indication.error_cause);
4305
4306         dev_warn_ratelimited(dev,
4307                              "Firmware reports %serror, cause: %s. Starting recovery...\n",
4308                              crq->error_indication.flags
4309                                 & IBMVNIC_FATAL_ERROR ? "FATAL " : "",
4310                              ibmvnic_fw_err_cause(cause));
4311
4312         if (crq->error_indication.flags & IBMVNIC_FATAL_ERROR)
4313                 ibmvnic_reset(adapter, VNIC_RESET_FATAL);
4314         else
4315                 ibmvnic_reset(adapter, VNIC_RESET_NON_FATAL);
4316 }
4317
4318 static int handle_change_mac_rsp(union ibmvnic_crq *crq,
4319                                  struct ibmvnic_adapter *adapter)
4320 {
4321         struct net_device *netdev = adapter->netdev;
4322         struct device *dev = &adapter->vdev->dev;
4323         long rc;
4324
4325         rc = crq->change_mac_addr_rsp.rc.code;
4326         if (rc) {
4327                 dev_err(dev, "Error %ld in CHANGE_MAC_ADDR_RSP\n", rc);
4328                 goto out;
4329         }
4330         /* crq->change_mac_addr.mac_addr is the requested one
4331          * crq->change_mac_addr_rsp.mac_addr is the returned valid one.
4332          */
4333         ether_addr_copy(netdev->dev_addr,
4334                         &crq->change_mac_addr_rsp.mac_addr[0]);
4335         ether_addr_copy(adapter->mac_addr,
4336                         &crq->change_mac_addr_rsp.mac_addr[0]);
4337 out:
4338         complete(&adapter->fw_done);
4339         return rc;
4340 }
4341
4342 static void handle_request_cap_rsp(union ibmvnic_crq *crq,
4343                                    struct ibmvnic_adapter *adapter)
4344 {
4345         struct device *dev = &adapter->vdev->dev;
4346         u64 *req_value;
4347         char *name;
4348
4349         atomic_dec(&adapter->running_cap_crqs);
4350         netdev_dbg(adapter->netdev, "Outstanding request-caps: %d\n",
4351                    atomic_read(&adapter->running_cap_crqs));
4352         switch (be16_to_cpu(crq->request_capability_rsp.capability)) {
4353         case REQ_TX_QUEUES:
4354                 req_value = &adapter->req_tx_queues;
4355                 name = "tx";
4356                 break;
4357         case REQ_RX_QUEUES:
4358                 req_value = &adapter->req_rx_queues;
4359                 name = "rx";
4360                 break;
4361         case REQ_RX_ADD_QUEUES:
4362                 req_value = &adapter->req_rx_add_queues;
4363                 name = "rx_add";
4364                 break;
4365         case REQ_TX_ENTRIES_PER_SUBCRQ:
4366                 req_value = &adapter->req_tx_entries_per_subcrq;
4367                 name = "tx_entries_per_subcrq";
4368                 break;
4369         case REQ_RX_ADD_ENTRIES_PER_SUBCRQ:
4370                 req_value = &adapter->req_rx_add_entries_per_subcrq;
4371                 name = "rx_add_entries_per_subcrq";
4372                 break;
4373         case REQ_MTU:
4374                 req_value = &adapter->req_mtu;
4375                 name = "mtu";
4376                 break;
4377         case PROMISC_REQUESTED:
4378                 req_value = &adapter->promisc;
4379                 name = "promisc";
4380                 break;
4381         default:
4382                 dev_err(dev, "Got invalid cap request rsp %d\n",
4383                         crq->request_capability.capability);
4384                 return;
4385         }
4386
4387         switch (crq->request_capability_rsp.rc.code) {
4388         case SUCCESS:
4389                 break;
4390         case PARTIALSUCCESS:
4391                 dev_info(dev, "req=%lld, rsp=%ld in %s queue, retrying.\n",
4392                          *req_value,
4393                          (long int)be64_to_cpu(crq->request_capability_rsp.
4394                                                number), name);
4395
4396                 if (be16_to_cpu(crq->request_capability_rsp.capability) ==
4397                     REQ_MTU) {
4398                         pr_err("mtu of %llu is not supported. Reverting.\n",
4399                                *req_value);
4400                         *req_value = adapter->fallback.mtu;
4401                 } else {
4402                         *req_value =
4403                                 be64_to_cpu(crq->request_capability_rsp.number);
4404                 }
4405
4406                 send_request_cap(adapter, 1);
4407                 return;
4408         default:
4409                 dev_err(dev, "Error %d in request cap rsp\n",
4410                         crq->request_capability_rsp.rc.code);
4411                 return;
4412         }
4413
4414         /* Done receiving requested capabilities, query IP offload support */
4415         if (atomic_read(&adapter->running_cap_crqs) == 0) {
4416                 adapter->wait_capability = false;
4417                 send_query_ip_offload(adapter);
4418         }
4419 }
4420
4421 static int handle_login_rsp(union ibmvnic_crq *login_rsp_crq,
4422                             struct ibmvnic_adapter *adapter)
4423 {
4424         struct device *dev = &adapter->vdev->dev;
4425         struct net_device *netdev = adapter->netdev;
4426         struct ibmvnic_login_rsp_buffer *login_rsp = adapter->login_rsp_buf;
4427         struct ibmvnic_login_buffer *login = adapter->login_buf;
4428         u64 *tx_handle_array;
4429         u64 *rx_handle_array;
4430         int num_tx_pools;
4431         int num_rx_pools;
4432         u64 *size_array;
4433         int i;
4434
4435         /* CHECK: Test/set of login_pending does not need to be atomic
4436          * because only ibmvnic_tasklet tests/clears this.
4437          */
4438         if (!adapter->login_pending) {
4439                 netdev_warn(netdev, "Ignoring unexpected login response\n");
4440                 return 0;
4441         }
4442         adapter->login_pending = false;
4443
4444         dma_unmap_single(dev, adapter->login_buf_token, adapter->login_buf_sz,
4445                          DMA_TO_DEVICE);
4446         dma_unmap_single(dev, adapter->login_rsp_buf_token,
4447                          adapter->login_rsp_buf_sz, DMA_FROM_DEVICE);
4448
4449         /* If the number of queues requested can't be allocated by the
4450          * server, the login response will return with code 1. We will need
4451          * to resend the login buffer with fewer queues requested.
4452          */
4453         if (login_rsp_crq->generic.rc.code) {
4454                 adapter->init_done_rc = login_rsp_crq->generic.rc.code;
4455                 complete(&adapter->init_done);
4456                 return 0;
4457         }
4458
4459         if (adapter->failover_pending) {
4460                 adapter->init_done_rc = -EAGAIN;
4461                 netdev_dbg(netdev, "Failover pending, ignoring login response\n");
4462                 complete(&adapter->init_done);
4463                 /* login response buffer will be released on reset */
4464                 return 0;
4465         }
4466
4467         netdev->mtu = adapter->req_mtu - ETH_HLEN;
4468
4469         netdev_dbg(adapter->netdev, "Login Response Buffer:\n");
4470         for (i = 0; i < (adapter->login_rsp_buf_sz - 1) / 8 + 1; i++) {
4471                 netdev_dbg(adapter->netdev, "%016lx\n",
4472                            ((unsigned long int *)(adapter->login_rsp_buf))[i]);
4473         }
4474
4475         /* Sanity checks */
4476         if (login->num_txcomp_subcrqs != login_rsp->num_txsubm_subcrqs ||
4477             (be32_to_cpu(login->num_rxcomp_subcrqs) *
4478              adapter->req_rx_add_queues !=
4479              be32_to_cpu(login_rsp->num_rxadd_subcrqs))) {
4480                 dev_err(dev, "FATAL: Inconsistent login and login rsp\n");
4481                 ibmvnic_reset(adapter, VNIC_RESET_FATAL);
4482                 return -EIO;
4483         }
4484         size_array = (u64 *)((u8 *)(adapter->login_rsp_buf) +
4485                 be32_to_cpu(adapter->login_rsp_buf->off_rxadd_buff_size));
4486         /* variable buffer sizes are not supported, so just read the
4487          * first entry.
4488          */
4489         adapter->cur_rx_buf_sz = be64_to_cpu(size_array[0]);
4490
4491         num_tx_pools = be32_to_cpu(adapter->login_rsp_buf->num_txsubm_subcrqs);
4492         num_rx_pools = be32_to_cpu(adapter->login_rsp_buf->num_rxadd_subcrqs);
4493
4494         tx_handle_array = (u64 *)((u8 *)(adapter->login_rsp_buf) +
4495                                   be32_to_cpu(adapter->login_rsp_buf->off_txsubm_subcrqs));
4496         rx_handle_array = (u64 *)((u8 *)(adapter->login_rsp_buf) +
4497                                   be32_to_cpu(adapter->login_rsp_buf->off_rxadd_subcrqs));
4498
4499         for (i = 0; i < num_tx_pools; i++)
4500                 adapter->tx_scrq[i]->handle = tx_handle_array[i];
4501
4502         for (i = 0; i < num_rx_pools; i++)
4503                 adapter->rx_scrq[i]->handle = rx_handle_array[i];
4504
4505         adapter->num_active_tx_scrqs = num_tx_pools;
4506         adapter->num_active_rx_scrqs = num_rx_pools;
4507         release_login_rsp_buffer(adapter);
4508         release_login_buffer(adapter);
4509         complete(&adapter->init_done);
4510
4511         return 0;
4512 }
4513
4514 static void handle_request_unmap_rsp(union ibmvnic_crq *crq,
4515                                      struct ibmvnic_adapter *adapter)
4516 {
4517         struct device *dev = &adapter->vdev->dev;
4518         long rc;
4519
4520         rc = crq->request_unmap_rsp.rc.code;
4521         if (rc)
4522                 dev_err(dev, "Error %ld in REQUEST_UNMAP_RSP\n", rc);
4523 }
4524
4525 static void handle_query_map_rsp(union ibmvnic_crq *crq,
4526                                  struct ibmvnic_adapter *adapter)
4527 {
4528         struct net_device *netdev = adapter->netdev;
4529         struct device *dev = &adapter->vdev->dev;
4530         long rc;
4531
4532         rc = crq->query_map_rsp.rc.code;
4533         if (rc) {
4534                 dev_err(dev, "Error %ld in QUERY_MAP_RSP\n", rc);
4535                 return;
4536         }
4537         netdev_dbg(netdev, "page_size = %d\ntot_pages = %d\nfree_pages = %d\n",
4538                    crq->query_map_rsp.page_size, crq->query_map_rsp.tot_pages,
4539                    crq->query_map_rsp.free_pages);
4540 }
4541
4542 static void handle_query_cap_rsp(union ibmvnic_crq *crq,
4543                                  struct ibmvnic_adapter *adapter)
4544 {
4545         struct net_device *netdev = adapter->netdev;
4546         struct device *dev = &adapter->vdev->dev;
4547         long rc;
4548
4549         atomic_dec(&adapter->running_cap_crqs);
4550         netdev_dbg(netdev, "Outstanding queries: %d\n",
4551                    atomic_read(&adapter->running_cap_crqs));
4552         rc = crq->query_capability.rc.code;
4553         if (rc) {
4554                 dev_err(dev, "Error %ld in QUERY_CAP_RSP\n", rc);
4555                 goto out;
4556         }
4557
4558         switch (be16_to_cpu(crq->query_capability.capability)) {
4559         case MIN_TX_QUEUES:
4560                 adapter->min_tx_queues =
4561                     be64_to_cpu(crq->query_capability.number);
4562                 netdev_dbg(netdev, "min_tx_queues = %lld\n",
4563                            adapter->min_tx_queues);
4564                 break;
4565         case MIN_RX_QUEUES:
4566                 adapter->min_rx_queues =
4567                     be64_to_cpu(crq->query_capability.number);
4568                 netdev_dbg(netdev, "min_rx_queues = %lld\n",
4569                            adapter->min_rx_queues);
4570                 break;
4571         case MIN_RX_ADD_QUEUES:
4572                 adapter->min_rx_add_queues =
4573                     be64_to_cpu(crq->query_capability.number);
4574                 netdev_dbg(netdev, "min_rx_add_queues = %lld\n",
4575                            adapter->min_rx_add_queues);
4576                 break;
4577         case MAX_TX_QUEUES:
4578                 adapter->max_tx_queues =
4579                     be64_to_cpu(crq->query_capability.number);
4580                 netdev_dbg(netdev, "max_tx_queues = %lld\n",
4581                            adapter->max_tx_queues);
4582                 break;
4583         case MAX_RX_QUEUES:
4584                 adapter->max_rx_queues =
4585                     be64_to_cpu(crq->query_capability.number);
4586                 netdev_dbg(netdev, "max_rx_queues = %lld\n",
4587                            adapter->max_rx_queues);
4588                 break;
4589         case MAX_RX_ADD_QUEUES:
4590                 adapter->max_rx_add_queues =
4591                     be64_to_cpu(crq->query_capability.number);
4592                 netdev_dbg(netdev, "max_rx_add_queues = %lld\n",
4593                            adapter->max_rx_add_queues);
4594                 break;
4595         case MIN_TX_ENTRIES_PER_SUBCRQ:
4596                 adapter->min_tx_entries_per_subcrq =
4597                     be64_to_cpu(crq->query_capability.number);
4598                 netdev_dbg(netdev, "min_tx_entries_per_subcrq = %lld\n",
4599                            adapter->min_tx_entries_per_subcrq);
4600                 break;
4601         case MIN_RX_ADD_ENTRIES_PER_SUBCRQ:
4602                 adapter->min_rx_add_entries_per_subcrq =
4603                     be64_to_cpu(crq->query_capability.number);
4604                 netdev_dbg(netdev, "min_rx_add_entrs_per_subcrq = %lld\n",
4605                            adapter->min_rx_add_entries_per_subcrq);
4606                 break;
4607         case MAX_TX_ENTRIES_PER_SUBCRQ:
4608                 adapter->max_tx_entries_per_subcrq =
4609                     be64_to_cpu(crq->query_capability.number);
4610                 netdev_dbg(netdev, "max_tx_entries_per_subcrq = %lld\n",
4611                            adapter->max_tx_entries_per_subcrq);
4612                 break;
4613         case MAX_RX_ADD_ENTRIES_PER_SUBCRQ:
4614                 adapter->max_rx_add_entries_per_subcrq =
4615                     be64_to_cpu(crq->query_capability.number);
4616                 netdev_dbg(netdev, "max_rx_add_entrs_per_subcrq = %lld\n",
4617                            adapter->max_rx_add_entries_per_subcrq);
4618                 break;
4619         case TCP_IP_OFFLOAD:
4620                 adapter->tcp_ip_offload =
4621                     be64_to_cpu(crq->query_capability.number);
4622                 netdev_dbg(netdev, "tcp_ip_offload = %lld\n",
4623                            adapter->tcp_ip_offload);
4624                 break;
4625         case PROMISC_SUPPORTED:
4626                 adapter->promisc_supported =
4627                     be64_to_cpu(crq->query_capability.number);
4628                 netdev_dbg(netdev, "promisc_supported = %lld\n",
4629                            adapter->promisc_supported);
4630                 break;
4631         case MIN_MTU:
4632                 adapter->min_mtu = be64_to_cpu(crq->query_capability.number);
4633                 netdev->min_mtu = adapter->min_mtu - ETH_HLEN;
4634                 netdev_dbg(netdev, "min_mtu = %lld\n", adapter->min_mtu);
4635                 break;
4636         case MAX_MTU:
4637                 adapter->max_mtu = be64_to_cpu(crq->query_capability.number);
4638                 netdev->max_mtu = adapter->max_mtu - ETH_HLEN;
4639                 netdev_dbg(netdev, "max_mtu = %lld\n", adapter->max_mtu);
4640                 break;
4641         case MAX_MULTICAST_FILTERS:
4642                 adapter->max_multicast_filters =
4643                     be64_to_cpu(crq->query_capability.number);
4644                 netdev_dbg(netdev, "max_multicast_filters = %lld\n",
4645                            adapter->max_multicast_filters);
4646                 break;
4647         case VLAN_HEADER_INSERTION:
4648                 adapter->vlan_header_insertion =
4649                     be64_to_cpu(crq->query_capability.number);
4650                 if (adapter->vlan_header_insertion)
4651                         netdev->features |= NETIF_F_HW_VLAN_STAG_TX;
4652                 netdev_dbg(netdev, "vlan_header_insertion = %lld\n",
4653                            adapter->vlan_header_insertion);
4654                 break;
4655         case RX_VLAN_HEADER_INSERTION:
4656                 adapter->rx_vlan_header_insertion =
4657                     be64_to_cpu(crq->query_capability.number);
4658                 netdev_dbg(netdev, "rx_vlan_header_insertion = %lld\n",
4659                            adapter->rx_vlan_header_insertion);
4660                 break;
4661         case MAX_TX_SG_ENTRIES:
4662                 adapter->max_tx_sg_entries =
4663                     be64_to_cpu(crq->query_capability.number);
4664                 netdev_dbg(netdev, "max_tx_sg_entries = %lld\n",
4665                            adapter->max_tx_sg_entries);
4666                 break;
4667         case RX_SG_SUPPORTED:
4668                 adapter->rx_sg_supported =
4669                     be64_to_cpu(crq->query_capability.number);
4670                 netdev_dbg(netdev, "rx_sg_supported = %lld\n",
4671                            adapter->rx_sg_supported);
4672                 break;
4673         case OPT_TX_COMP_SUB_QUEUES:
4674                 adapter->opt_tx_comp_sub_queues =
4675                     be64_to_cpu(crq->query_capability.number);
4676                 netdev_dbg(netdev, "opt_tx_comp_sub_queues = %lld\n",
4677                            adapter->opt_tx_comp_sub_queues);
4678                 break;
4679         case OPT_RX_COMP_QUEUES:
4680                 adapter->opt_rx_comp_queues =
4681                     be64_to_cpu(crq->query_capability.number);
4682                 netdev_dbg(netdev, "opt_rx_comp_queues = %lld\n",
4683                            adapter->opt_rx_comp_queues);
4684                 break;
4685         case OPT_RX_BUFADD_Q_PER_RX_COMP_Q:
4686                 adapter->opt_rx_bufadd_q_per_rx_comp_q =
4687                     be64_to_cpu(crq->query_capability.number);
4688                 netdev_dbg(netdev, "opt_rx_bufadd_q_per_rx_comp_q = %lld\n",
4689                            adapter->opt_rx_bufadd_q_per_rx_comp_q);
4690                 break;
4691         case OPT_TX_ENTRIES_PER_SUBCRQ:
4692                 adapter->opt_tx_entries_per_subcrq =
4693                     be64_to_cpu(crq->query_capability.number);
4694                 netdev_dbg(netdev, "opt_tx_entries_per_subcrq = %lld\n",
4695                            adapter->opt_tx_entries_per_subcrq);
4696                 break;
4697         case OPT_RXBA_ENTRIES_PER_SUBCRQ:
4698                 adapter->opt_rxba_entries_per_subcrq =
4699                     be64_to_cpu(crq->query_capability.number);
4700                 netdev_dbg(netdev, "opt_rxba_entries_per_subcrq = %lld\n",
4701                            adapter->opt_rxba_entries_per_subcrq);
4702                 break;
4703         case TX_RX_DESC_REQ:
4704                 adapter->tx_rx_desc_req = crq->query_capability.number;
4705                 netdev_dbg(netdev, "tx_rx_desc_req = %llx\n",
4706                            adapter->tx_rx_desc_req);
4707                 break;
4708
4709         default:
4710                 netdev_err(netdev, "Got invalid cap rsp %d\n",
4711                            crq->query_capability.capability);
4712         }
4713
4714 out:
4715         if (atomic_read(&adapter->running_cap_crqs) == 0) {
4716                 adapter->wait_capability = false;
4717                 send_request_cap(adapter, 0);
4718         }
4719 }
4720
4721 static int send_query_phys_parms(struct ibmvnic_adapter *adapter)
4722 {
4723         union ibmvnic_crq crq;
4724         int rc;
4725
4726         memset(&crq, 0, sizeof(crq));
4727         crq.query_phys_parms.first = IBMVNIC_CRQ_CMD;
4728         crq.query_phys_parms.cmd = QUERY_PHYS_PARMS;
4729
4730         mutex_lock(&adapter->fw_lock);
4731         adapter->fw_done_rc = 0;
4732         reinit_completion(&adapter->fw_done);
4733
4734         rc = ibmvnic_send_crq(adapter, &crq);
4735         if (rc) {
4736                 mutex_unlock(&adapter->fw_lock);
4737                 return rc;
4738         }
4739
4740         rc = ibmvnic_wait_for_completion(adapter, &adapter->fw_done, 10000);
4741         if (rc) {
4742                 mutex_unlock(&adapter->fw_lock);
4743                 return rc;
4744         }
4745
4746         mutex_unlock(&adapter->fw_lock);
4747         return adapter->fw_done_rc ? -EIO : 0;
4748 }
4749
4750 static int handle_query_phys_parms_rsp(union ibmvnic_crq *crq,
4751                                        struct ibmvnic_adapter *adapter)
4752 {
4753         struct net_device *netdev = adapter->netdev;
4754         int rc;
4755         __be32 rspeed = cpu_to_be32(crq->query_phys_parms_rsp.speed);
4756
4757         rc = crq->query_phys_parms_rsp.rc.code;
4758         if (rc) {
4759                 netdev_err(netdev, "Error %d in QUERY_PHYS_PARMS\n", rc);
4760                 return rc;
4761         }
4762         switch (rspeed) {
4763         case IBMVNIC_10MBPS:
4764                 adapter->speed = SPEED_10;
4765                 break;
4766         case IBMVNIC_100MBPS:
4767                 adapter->speed = SPEED_100;
4768                 break;
4769         case IBMVNIC_1GBPS:
4770                 adapter->speed = SPEED_1000;
4771                 break;
4772         case IBMVNIC_10GBPS:
4773                 adapter->speed = SPEED_10000;
4774                 break;
4775         case IBMVNIC_25GBPS:
4776                 adapter->speed = SPEED_25000;
4777                 break;
4778         case IBMVNIC_40GBPS:
4779                 adapter->speed = SPEED_40000;
4780                 break;
4781         case IBMVNIC_50GBPS:
4782                 adapter->speed = SPEED_50000;
4783                 break;
4784         case IBMVNIC_100GBPS:
4785                 adapter->speed = SPEED_100000;
4786                 break;
4787         case IBMVNIC_200GBPS:
4788                 adapter->speed = SPEED_200000;
4789                 break;
4790         default:
4791                 if (netif_carrier_ok(netdev))
4792                         netdev_warn(netdev, "Unknown speed 0x%08x\n", rspeed);
4793                 adapter->speed = SPEED_UNKNOWN;
4794         }
4795         if (crq->query_phys_parms_rsp.flags1 & IBMVNIC_FULL_DUPLEX)
4796                 adapter->duplex = DUPLEX_FULL;
4797         else if (crq->query_phys_parms_rsp.flags1 & IBMVNIC_HALF_DUPLEX)
4798                 adapter->duplex = DUPLEX_HALF;
4799         else
4800                 adapter->duplex = DUPLEX_UNKNOWN;
4801
4802         return rc;
4803 }
4804
4805 static void ibmvnic_handle_crq(union ibmvnic_crq *crq,
4806                                struct ibmvnic_adapter *adapter)
4807 {
4808         struct ibmvnic_generic_crq *gen_crq = &crq->generic;
4809         struct net_device *netdev = adapter->netdev;
4810         struct device *dev = &adapter->vdev->dev;
4811         u64 *u64_crq = (u64 *)crq;
4812         long rc;
4813
4814         netdev_dbg(netdev, "Handling CRQ: %016lx %016lx\n",
4815                    (unsigned long int)cpu_to_be64(u64_crq[0]),
4816                    (unsigned long int)cpu_to_be64(u64_crq[1]));
4817         switch (gen_crq->first) {
4818         case IBMVNIC_CRQ_INIT_RSP:
4819                 switch (gen_crq->cmd) {
4820                 case IBMVNIC_CRQ_INIT:
4821                         dev_info(dev, "Partner initialized\n");
4822                         adapter->from_passive_init = true;
4823                         /* Discard any stale login responses from prev reset.
4824                          * CHECK: should we clear even on INIT_COMPLETE?
4825                          */
4826                         adapter->login_pending = false;
4827
4828                         if (!completion_done(&adapter->init_done)) {
4829                                 complete(&adapter->init_done);
4830                                 adapter->init_done_rc = -EIO;
4831                         }
4832                         rc = ibmvnic_reset(adapter, VNIC_RESET_FAILOVER);
4833                         if (rc && rc != -EBUSY) {
4834                                 /* We were unable to schedule the failover
4835                                  * reset either because the adapter was still
4836                                  * probing (eg: during kexec) or we could not
4837                                  * allocate memory. Clear the failover_pending
4838                                  * flag since no one else will. We ignore
4839                                  * EBUSY because it means either FAILOVER reset
4840                                  * is already scheduled or the adapter is
4841                                  * being removed.
4842                                  */
4843                                 netdev_err(netdev,
4844                                            "Error %ld scheduling failover reset\n",
4845                                            rc);
4846                                 adapter->failover_pending = false;
4847                         }
4848                         break;
4849                 case IBMVNIC_CRQ_INIT_COMPLETE:
4850                         dev_info(dev, "Partner initialization complete\n");
4851                         adapter->crq.active = true;
4852                         send_version_xchg(adapter);
4853                         break;
4854                 default:
4855                         dev_err(dev, "Unknown crq cmd: %d\n", gen_crq->cmd);
4856                 }
4857                 return;
4858         case IBMVNIC_CRQ_XPORT_EVENT:
4859                 netif_carrier_off(netdev);
4860                 adapter->crq.active = false;
4861                 /* terminate any thread waiting for a response
4862                  * from the device
4863                  */
4864                 if (!completion_done(&adapter->fw_done)) {
4865                         adapter->fw_done_rc = -EIO;
4866                         complete(&adapter->fw_done);
4867                 }
4868
4869                 /* if we got here during crq-init, retry crq-init */
4870                 if (!completion_done(&adapter->init_done)) {
4871                         adapter->init_done_rc = -EAGAIN;
4872                         complete(&adapter->init_done);
4873                 }
4874
4875                 if (!completion_done(&adapter->stats_done))
4876                         complete(&adapter->stats_done);
4877                 if (test_bit(0, &adapter->resetting))
4878                         adapter->force_reset_recovery = true;
4879                 if (gen_crq->cmd == IBMVNIC_PARTITION_MIGRATED) {
4880                         dev_info(dev, "Migrated, re-enabling adapter\n");
4881                         ibmvnic_reset(adapter, VNIC_RESET_MOBILITY);
4882                 } else if (gen_crq->cmd == IBMVNIC_DEVICE_FAILOVER) {
4883                         dev_info(dev, "Backing device failover detected\n");
4884                         adapter->failover_pending = true;
4885                 } else {
4886                         /* The adapter lost the connection */
4887                         dev_err(dev, "Virtual Adapter failed (rc=%d)\n",
4888                                 gen_crq->cmd);
4889                         ibmvnic_reset(adapter, VNIC_RESET_FATAL);
4890                 }
4891                 return;
4892         case IBMVNIC_CRQ_CMD_RSP:
4893                 break;
4894         default:
4895                 dev_err(dev, "Got an invalid msg type 0x%02x\n",
4896                         gen_crq->first);
4897                 return;
4898         }
4899
4900         switch (gen_crq->cmd) {
4901         case VERSION_EXCHANGE_RSP:
4902                 rc = crq->version_exchange_rsp.rc.code;
4903                 if (rc) {
4904                         dev_err(dev, "Error %ld in VERSION_EXCHG_RSP\n", rc);
4905                         break;
4906                 }
4907                 ibmvnic_version =
4908                             be16_to_cpu(crq->version_exchange_rsp.version);
4909                 dev_info(dev, "Partner protocol version is %d\n",
4910                          ibmvnic_version);
4911                 send_query_cap(adapter);
4912                 break;
4913         case QUERY_CAPABILITY_RSP:
4914                 handle_query_cap_rsp(crq, adapter);
4915                 break;
4916         case QUERY_MAP_RSP:
4917                 handle_query_map_rsp(crq, adapter);
4918                 break;
4919         case REQUEST_MAP_RSP:
4920                 adapter->fw_done_rc = crq->request_map_rsp.rc.code;
4921                 complete(&adapter->fw_done);
4922                 break;
4923         case REQUEST_UNMAP_RSP:
4924                 handle_request_unmap_rsp(crq, adapter);
4925                 break;
4926         case REQUEST_CAPABILITY_RSP:
4927                 handle_request_cap_rsp(crq, adapter);
4928                 break;
4929         case LOGIN_RSP:
4930                 netdev_dbg(netdev, "Got Login Response\n");
4931                 handle_login_rsp(crq, adapter);
4932                 break;
4933         case LOGICAL_LINK_STATE_RSP:
4934                 netdev_dbg(netdev,
4935                            "Got Logical Link State Response, state: %d rc: %d\n",
4936                            crq->logical_link_state_rsp.link_state,
4937                            crq->logical_link_state_rsp.rc.code);
4938                 adapter->logical_link_state =
4939                     crq->logical_link_state_rsp.link_state;
4940                 adapter->init_done_rc = crq->logical_link_state_rsp.rc.code;
4941                 complete(&adapter->init_done);
4942                 break;
4943         case LINK_STATE_INDICATION:
4944                 netdev_dbg(netdev, "Got Logical Link State Indication\n");
4945                 adapter->phys_link_state =
4946                     crq->link_state_indication.phys_link_state;
4947                 adapter->logical_link_state =
4948                     crq->link_state_indication.logical_link_state;
4949                 if (adapter->phys_link_state && adapter->logical_link_state)
4950                         netif_carrier_on(netdev);
4951                 else
4952                         netif_carrier_off(netdev);
4953                 break;
4954         case CHANGE_MAC_ADDR_RSP:
4955                 netdev_dbg(netdev, "Got MAC address change Response\n");
4956                 adapter->fw_done_rc = handle_change_mac_rsp(crq, adapter);
4957                 break;
4958         case ERROR_INDICATION:
4959                 netdev_dbg(netdev, "Got Error Indication\n");
4960                 handle_error_indication(crq, adapter);
4961                 break;
4962         case REQUEST_STATISTICS_RSP:
4963                 netdev_dbg(netdev, "Got Statistics Response\n");
4964                 complete(&adapter->stats_done);
4965                 break;
4966         case QUERY_IP_OFFLOAD_RSP:
4967                 netdev_dbg(netdev, "Got Query IP offload Response\n");
4968                 handle_query_ip_offload_rsp(adapter);
4969                 break;
4970         case MULTICAST_CTRL_RSP:
4971                 netdev_dbg(netdev, "Got multicast control Response\n");
4972                 break;
4973         case CONTROL_IP_OFFLOAD_RSP:
4974                 netdev_dbg(netdev, "Got Control IP offload Response\n");
4975                 dma_unmap_single(dev, adapter->ip_offload_ctrl_tok,
4976                                  sizeof(adapter->ip_offload_ctrl),
4977                                  DMA_TO_DEVICE);
4978                 complete(&adapter->init_done);
4979                 break;
4980         case COLLECT_FW_TRACE_RSP:
4981                 netdev_dbg(netdev, "Got Collect firmware trace Response\n");
4982                 complete(&adapter->fw_done);
4983                 break;
4984         case GET_VPD_SIZE_RSP:
4985                 handle_vpd_size_rsp(crq, adapter);
4986                 break;
4987         case GET_VPD_RSP:
4988                 handle_vpd_rsp(crq, adapter);
4989                 break;
4990         case QUERY_PHYS_PARMS_RSP:
4991                 adapter->fw_done_rc = handle_query_phys_parms_rsp(crq, adapter);
4992                 complete(&adapter->fw_done);
4993                 break;
4994         default:
4995                 netdev_err(netdev, "Got an invalid cmd type 0x%02x\n",
4996                            gen_crq->cmd);
4997         }
4998 }
4999
5000 static irqreturn_t ibmvnic_interrupt(int irq, void *instance)
5001 {
5002         struct ibmvnic_adapter *adapter = instance;
5003
5004         tasklet_schedule(&adapter->tasklet);
5005         return IRQ_HANDLED;
5006 }
5007
5008 static void ibmvnic_tasklet(struct tasklet_struct *t)
5009 {
5010         struct ibmvnic_adapter *adapter = from_tasklet(adapter, t, tasklet);
5011         struct ibmvnic_crq_queue *queue = &adapter->crq;
5012         union ibmvnic_crq *crq;
5013         unsigned long flags;
5014         bool done = false;
5015
5016         spin_lock_irqsave(&queue->lock, flags);
5017         while (!done) {
5018                 /* Pull all the valid messages off the CRQ */
5019                 while ((crq = ibmvnic_next_crq(adapter)) != NULL) {
5020                         /* This barrier makes sure ibmvnic_next_crq()'s
5021                          * crq->generic.first & IBMVNIC_CRQ_CMD_RSP is loaded
5022                          * before ibmvnic_handle_crq()'s
5023                          * switch(gen_crq->first) and switch(gen_crq->cmd).
5024                          */
5025                         dma_rmb();
5026                         ibmvnic_handle_crq(crq, adapter);
5027                         crq->generic.first = 0;
5028                 }
5029         }
5030         /* if capabilities CRQ's were sent in this tasklet, the following
5031          * tasklet must wait until all responses are received
5032          */
5033         if (atomic_read(&adapter->running_cap_crqs) != 0)
5034                 adapter->wait_capability = true;
5035         spin_unlock_irqrestore(&queue->lock, flags);
5036 }
5037
5038 static int ibmvnic_reenable_crq_queue(struct ibmvnic_adapter *adapter)
5039 {
5040         struct vio_dev *vdev = adapter->vdev;
5041         int rc;
5042
5043         do {
5044                 rc = plpar_hcall_norets(H_ENABLE_CRQ, vdev->unit_address);
5045         } while (rc == H_IN_PROGRESS || rc == H_BUSY || H_IS_LONG_BUSY(rc));
5046
5047         if (rc)
5048                 dev_err(&vdev->dev, "Error enabling adapter (rc=%d)\n", rc);
5049
5050         return rc;
5051 }
5052
5053 static int ibmvnic_reset_crq(struct ibmvnic_adapter *adapter)
5054 {
5055         struct ibmvnic_crq_queue *crq = &adapter->crq;
5056         struct device *dev = &adapter->vdev->dev;
5057         struct vio_dev *vdev = adapter->vdev;
5058         int rc;
5059
5060         /* Close the CRQ */
5061         do {
5062                 rc = plpar_hcall_norets(H_FREE_CRQ, vdev->unit_address);
5063         } while (rc == H_BUSY || H_IS_LONG_BUSY(rc));
5064
5065         /* Clean out the queue */
5066         if (!crq->msgs)
5067                 return -EINVAL;
5068
5069         memset(crq->msgs, 0, PAGE_SIZE);
5070         crq->cur = 0;
5071         crq->active = false;
5072
5073         /* And re-open it again */
5074         rc = plpar_hcall_norets(H_REG_CRQ, vdev->unit_address,
5075                                 crq->msg_token, PAGE_SIZE);
5076
5077         if (rc == H_CLOSED)
5078                 /* Adapter is good, but other end is not ready */
5079                 dev_warn(dev, "Partner adapter not ready\n");
5080         else if (rc != 0)
5081                 dev_warn(dev, "Couldn't register crq (rc=%d)\n", rc);
5082
5083         return rc;
5084 }
5085
5086 static void release_crq_queue(struct ibmvnic_adapter *adapter)
5087 {
5088         struct ibmvnic_crq_queue *crq = &adapter->crq;
5089         struct vio_dev *vdev = adapter->vdev;
5090         long rc;
5091
5092         if (!crq->msgs)
5093                 return;
5094
5095         netdev_dbg(adapter->netdev, "Releasing CRQ\n");
5096         free_irq(vdev->irq, adapter);
5097         tasklet_kill(&adapter->tasklet);
5098         do {
5099                 rc = plpar_hcall_norets(H_FREE_CRQ, vdev->unit_address);
5100         } while (rc == H_BUSY || H_IS_LONG_BUSY(rc));
5101
5102         dma_unmap_single(&vdev->dev, crq->msg_token, PAGE_SIZE,
5103                          DMA_BIDIRECTIONAL);
5104         free_page((unsigned long)crq->msgs);
5105         crq->msgs = NULL;
5106         crq->active = false;
5107 }
5108
5109 static int init_crq_queue(struct ibmvnic_adapter *adapter)
5110 {
5111         struct ibmvnic_crq_queue *crq = &adapter->crq;
5112         struct device *dev = &adapter->vdev->dev;
5113         struct vio_dev *vdev = adapter->vdev;
5114         int rc, retrc = -ENOMEM;
5115
5116         if (crq->msgs)
5117                 return 0;
5118
5119         crq->msgs = (union ibmvnic_crq *)get_zeroed_page(GFP_KERNEL);
5120         /* Should we allocate more than one page? */
5121
5122         if (!crq->msgs)
5123                 return -ENOMEM;
5124
5125         crq->size = PAGE_SIZE / sizeof(*crq->msgs);
5126         crq->msg_token = dma_map_single(dev, crq->msgs, PAGE_SIZE,
5127                                         DMA_BIDIRECTIONAL);
5128         if (dma_mapping_error(dev, crq->msg_token))
5129                 goto map_failed;
5130
5131         rc = plpar_hcall_norets(H_REG_CRQ, vdev->unit_address,
5132                                 crq->msg_token, PAGE_SIZE);
5133
5134         if (rc == H_RESOURCE)
5135                 /* maybe kexecing and resource is busy. try a reset */
5136                 rc = ibmvnic_reset_crq(adapter);
5137         retrc = rc;
5138
5139         if (rc == H_CLOSED) {
5140                 dev_warn(dev, "Partner adapter not ready\n");
5141         } else if (rc) {
5142                 dev_warn(dev, "Error %d opening adapter\n", rc);
5143                 goto reg_crq_failed;
5144         }
5145
5146         retrc = 0;
5147
5148         tasklet_setup(&adapter->tasklet, (void *)ibmvnic_tasklet);
5149
5150         netdev_dbg(adapter->netdev, "registering irq 0x%x\n", vdev->irq);
5151         snprintf(crq->name, sizeof(crq->name), "ibmvnic-%x",
5152                  adapter->vdev->unit_address);
5153         rc = request_irq(vdev->irq, ibmvnic_interrupt, 0, crq->name, adapter);
5154         if (rc) {
5155                 dev_err(dev, "Couldn't register irq 0x%x. rc=%d\n",
5156                         vdev->irq, rc);
5157                 goto req_irq_failed;
5158         }
5159
5160         rc = vio_enable_interrupts(vdev);
5161         if (rc) {
5162                 dev_err(dev, "Error %d enabling interrupts\n", rc);
5163                 goto req_irq_failed;
5164         }
5165
5166         crq->cur = 0;
5167         spin_lock_init(&crq->lock);
5168
5169         /* process any CRQs that were queued before we enabled interrupts */
5170         tasklet_schedule(&adapter->tasklet);
5171
5172         return retrc;
5173
5174 req_irq_failed:
5175         tasklet_kill(&adapter->tasklet);
5176         do {
5177                 rc = plpar_hcall_norets(H_FREE_CRQ, vdev->unit_address);
5178         } while (rc == H_BUSY || H_IS_LONG_BUSY(rc));
5179 reg_crq_failed:
5180         dma_unmap_single(dev, crq->msg_token, PAGE_SIZE, DMA_BIDIRECTIONAL);
5181 map_failed:
5182         free_page((unsigned long)crq->msgs);
5183         crq->msgs = NULL;
5184         return retrc;
5185 }
5186
5187 static int ibmvnic_reset_init(struct ibmvnic_adapter *adapter, bool reset)
5188 {
5189         struct device *dev = &adapter->vdev->dev;
5190         unsigned long timeout = msecs_to_jiffies(20000);
5191         u64 old_num_rx_queues = adapter->req_rx_queues;
5192         u64 old_num_tx_queues = adapter->req_tx_queues;
5193         int rc;
5194
5195         adapter->from_passive_init = false;
5196
5197         if (reset)
5198                 reinit_completion(&adapter->init_done);
5199
5200         adapter->init_done_rc = 0;
5201         rc = ibmvnic_send_crq_init(adapter);
5202         if (rc) {
5203                 dev_err(dev, "Send crq init failed with error %d\n", rc);
5204                 return rc;
5205         }
5206
5207         if (!wait_for_completion_timeout(&adapter->init_done, timeout)) {
5208                 dev_err(dev, "Initialization sequence timed out\n");
5209                 return -1;
5210         }
5211
5212         if (adapter->init_done_rc) {
5213                 release_crq_queue(adapter);
5214                 return adapter->init_done_rc;
5215         }
5216
5217         if (adapter->from_passive_init) {
5218                 adapter->state = VNIC_OPEN;
5219                 adapter->from_passive_init = false;
5220                 return -1;
5221         }
5222
5223         if (reset &&
5224             test_bit(0, &adapter->resetting) && !adapter->wait_for_reset &&
5225             adapter->reset_reason != VNIC_RESET_MOBILITY) {
5226                 if (adapter->req_rx_queues != old_num_rx_queues ||
5227                     adapter->req_tx_queues != old_num_tx_queues) {
5228                         release_sub_crqs(adapter, 0);
5229                         rc = init_sub_crqs(adapter);
5230                 } else {
5231                         /* no need to reinitialize completely, but we do
5232                          * need to clean up transmits that were in flight
5233                          * when we processed the reset.  Failure to do so
5234                          * will confound the upper layer, usually TCP, by
5235                          * creating the illusion of transmits that are
5236                          * awaiting completion.
5237                          */
5238                         clean_tx_pools(adapter);
5239
5240                         rc = reset_sub_crq_queues(adapter);
5241                 }
5242         } else {
5243                 rc = init_sub_crqs(adapter);
5244         }
5245
5246         if (rc) {
5247                 dev_err(dev, "Initialization of sub crqs failed\n");
5248                 release_crq_queue(adapter);
5249                 return rc;
5250         }
5251
5252         rc = init_sub_crq_irqs(adapter);
5253         if (rc) {
5254                 dev_err(dev, "Failed to initialize sub crq irqs\n");
5255                 release_crq_queue(adapter);
5256         }
5257
5258         return rc;
5259 }
5260
5261 static struct device_attribute dev_attr_failover;
5262
5263 static int ibmvnic_probe(struct vio_dev *dev, const struct vio_device_id *id)
5264 {
5265         struct ibmvnic_adapter *adapter;
5266         struct net_device *netdev;
5267         unsigned char *mac_addr_p;
5268         int rc;
5269
5270         dev_dbg(&dev->dev, "entering ibmvnic_probe for UA 0x%x\n",
5271                 dev->unit_address);
5272
5273         mac_addr_p = (unsigned char *)vio_get_attribute(dev,
5274                                                         VETH_MAC_ADDR, NULL);
5275         if (!mac_addr_p) {
5276                 dev_err(&dev->dev,
5277                         "(%s:%3.3d) ERROR: Can't find MAC_ADDR attribute\n",
5278                         __FILE__, __LINE__);
5279                 return 0;
5280         }
5281
5282         netdev = alloc_etherdev_mq(sizeof(struct ibmvnic_adapter),
5283                                    IBMVNIC_MAX_QUEUES);
5284         if (!netdev)
5285                 return -ENOMEM;
5286
5287         adapter = netdev_priv(netdev);
5288         adapter->state = VNIC_PROBING;
5289         dev_set_drvdata(&dev->dev, netdev);
5290         adapter->vdev = dev;
5291         adapter->netdev = netdev;
5292         adapter->login_pending = false;
5293
5294         ether_addr_copy(adapter->mac_addr, mac_addr_p);
5295         ether_addr_copy(netdev->dev_addr, adapter->mac_addr);
5296         netdev->irq = dev->irq;
5297         netdev->netdev_ops = &ibmvnic_netdev_ops;
5298         netdev->ethtool_ops = &ibmvnic_ethtool_ops;
5299         SET_NETDEV_DEV(netdev, &dev->dev);
5300
5301         spin_lock_init(&adapter->stats_lock);
5302
5303         INIT_WORK(&adapter->ibmvnic_reset, __ibmvnic_reset);
5304         INIT_DELAYED_WORK(&adapter->ibmvnic_delayed_reset,
5305                           __ibmvnic_delayed_reset);
5306         INIT_LIST_HEAD(&adapter->rwi_list);
5307         spin_lock_init(&adapter->rwi_lock);
5308         spin_lock_init(&adapter->state_lock);
5309         mutex_init(&adapter->fw_lock);
5310         init_completion(&adapter->init_done);
5311         init_completion(&adapter->fw_done);
5312         init_completion(&adapter->reset_done);
5313         init_completion(&adapter->stats_done);
5314         clear_bit(0, &adapter->resetting);
5315
5316         do {
5317                 rc = init_crq_queue(adapter);
5318                 if (rc) {
5319                         dev_err(&dev->dev, "Couldn't initialize crq. rc=%d\n",
5320                                 rc);
5321                         goto ibmvnic_init_fail;
5322                 }
5323
5324                 rc = ibmvnic_reset_init(adapter, false);
5325                 if (rc && rc != EAGAIN)
5326                         goto ibmvnic_init_fail;
5327         } while (rc == EAGAIN);
5328
5329         rc = init_stats_buffers(adapter);
5330         if (rc)
5331                 goto ibmvnic_init_fail;
5332
5333         rc = init_stats_token(adapter);
5334         if (rc)
5335                 goto ibmvnic_stats_fail;
5336
5337         netdev->mtu = adapter->req_mtu - ETH_HLEN;
5338         netdev->min_mtu = adapter->min_mtu - ETH_HLEN;
5339         netdev->max_mtu = adapter->max_mtu - ETH_HLEN;
5340
5341         rc = device_create_file(&dev->dev, &dev_attr_failover);
5342         if (rc)
5343                 goto ibmvnic_dev_file_err;
5344
5345         netif_carrier_off(netdev);
5346
5347         adapter->state = VNIC_PROBED;
5348
5349         adapter->wait_for_reset = false;
5350         adapter->last_reset_time = jiffies;
5351
5352         rc = register_netdev(netdev);
5353         if (rc) {
5354                 dev_err(&dev->dev, "failed to register netdev rc=%d\n", rc);
5355                 goto ibmvnic_register_fail;
5356         }
5357         dev_info(&dev->dev, "ibmvnic registered\n");
5358
5359         return 0;
5360
5361 ibmvnic_register_fail:
5362         device_remove_file(&dev->dev, &dev_attr_failover);
5363
5364 ibmvnic_dev_file_err:
5365         release_stats_token(adapter);
5366
5367 ibmvnic_stats_fail:
5368         release_stats_buffers(adapter);
5369
5370 ibmvnic_init_fail:
5371         release_sub_crqs(adapter, 1);
5372         release_crq_queue(adapter);
5373         mutex_destroy(&adapter->fw_lock);
5374         free_netdev(netdev);
5375
5376         return rc;
5377 }
5378
5379 static int ibmvnic_remove(struct vio_dev *dev)
5380 {
5381         struct net_device *netdev = dev_get_drvdata(&dev->dev);
5382         struct ibmvnic_adapter *adapter = netdev_priv(netdev);
5383         unsigned long flags;
5384
5385         spin_lock_irqsave(&adapter->state_lock, flags);
5386
5387         /* If ibmvnic_reset() is scheduling a reset, wait for it to
5388          * finish. Then, set the state to REMOVING to prevent it from
5389          * scheduling any more work and to have reset functions ignore
5390          * any resets that have already been scheduled. Drop the lock
5391          * after setting state, so __ibmvnic_reset() which is called
5392          * from the flush_work() below, can make progress.
5393          */
5394         spin_lock(&adapter->rwi_lock);
5395         adapter->state = VNIC_REMOVING;
5396         spin_unlock(&adapter->rwi_lock);
5397
5398         spin_unlock_irqrestore(&adapter->state_lock, flags);
5399
5400         flush_work(&adapter->ibmvnic_reset);
5401         flush_delayed_work(&adapter->ibmvnic_delayed_reset);
5402
5403         rtnl_lock();
5404         unregister_netdevice(netdev);
5405
5406         release_resources(adapter);
5407         release_sub_crqs(adapter, 1);
5408         release_crq_queue(adapter);
5409
5410         release_stats_token(adapter);
5411         release_stats_buffers(adapter);
5412
5413         adapter->state = VNIC_REMOVED;
5414
5415         rtnl_unlock();
5416         mutex_destroy(&adapter->fw_lock);
5417         device_remove_file(&dev->dev, &dev_attr_failover);
5418         free_netdev(netdev);
5419         dev_set_drvdata(&dev->dev, NULL);
5420
5421         return 0;
5422 }
5423
5424 static ssize_t failover_store(struct device *dev, struct device_attribute *attr,
5425                               const char *buf, size_t count)
5426 {
5427         struct net_device *netdev = dev_get_drvdata(dev);
5428         struct ibmvnic_adapter *adapter = netdev_priv(netdev);
5429         unsigned long retbuf[PLPAR_HCALL_BUFSIZE];
5430         __be64 session_token;
5431         long rc;
5432
5433         if (!sysfs_streq(buf, "1"))
5434                 return -EINVAL;
5435
5436         rc = plpar_hcall(H_VIOCTL, retbuf, adapter->vdev->unit_address,
5437                          H_GET_SESSION_TOKEN, 0, 0, 0);
5438         if (rc) {
5439                 netdev_err(netdev, "Couldn't retrieve session token, rc %ld\n",
5440                            rc);
5441                 return -EINVAL;
5442         }
5443
5444         session_token = (__be64)retbuf[0];
5445         netdev_dbg(netdev, "Initiating client failover, session id %llx\n",
5446                    be64_to_cpu(session_token));
5447         rc = plpar_hcall_norets(H_VIOCTL, adapter->vdev->unit_address,
5448                                 H_SESSION_ERR_DETECTED, session_token, 0, 0);
5449         if (rc) {
5450                 netdev_err(netdev, "Client initiated failover failed, rc %ld\n",
5451                            rc);
5452                 return -EINVAL;
5453         }
5454
5455         return count;
5456 }
5457
5458 static DEVICE_ATTR_WO(failover);
5459
5460 static unsigned long ibmvnic_get_desired_dma(struct vio_dev *vdev)
5461 {
5462         struct net_device *netdev = dev_get_drvdata(&vdev->dev);
5463         struct ibmvnic_adapter *adapter;
5464         struct iommu_table *tbl;
5465         unsigned long ret = 0;
5466         int i;
5467
5468         tbl = get_iommu_table_base(&vdev->dev);
5469
5470         /* netdev inits at probe time along with the structures we need below*/
5471         if (!netdev)
5472                 return IOMMU_PAGE_ALIGN(IBMVNIC_IO_ENTITLEMENT_DEFAULT, tbl);
5473
5474         adapter = netdev_priv(netdev);
5475
5476         ret += PAGE_SIZE; /* the crq message queue */
5477         ret += IOMMU_PAGE_ALIGN(sizeof(struct ibmvnic_statistics), tbl);
5478
5479         for (i = 0; i < adapter->req_tx_queues + adapter->req_rx_queues; i++)
5480                 ret += 4 * PAGE_SIZE; /* the scrq message queue */
5481
5482         for (i = 0; i < adapter->num_active_rx_pools; i++)
5483                 ret += adapter->rx_pool[i].size *
5484                     IOMMU_PAGE_ALIGN(adapter->rx_pool[i].buff_size, tbl);
5485
5486         return ret;
5487 }
5488
5489 static int ibmvnic_resume(struct device *dev)
5490 {
5491         struct net_device *netdev = dev_get_drvdata(dev);
5492         struct ibmvnic_adapter *adapter = netdev_priv(netdev);
5493
5494         if (adapter->state != VNIC_OPEN)
5495                 return 0;
5496
5497         tasklet_schedule(&adapter->tasklet);
5498
5499         return 0;
5500 }
5501
5502 static const struct vio_device_id ibmvnic_device_table[] = {
5503         {"network", "IBM,vnic"},
5504         {"", "" }
5505 };
5506 MODULE_DEVICE_TABLE(vio, ibmvnic_device_table);
5507
5508 static const struct dev_pm_ops ibmvnic_pm_ops = {
5509         .resume = ibmvnic_resume
5510 };
5511
5512 static struct vio_driver ibmvnic_driver = {
5513         .id_table       = ibmvnic_device_table,
5514         .probe          = ibmvnic_probe,
5515         .remove         = ibmvnic_remove,
5516         .get_desired_dma = ibmvnic_get_desired_dma,
5517         .name           = ibmvnic_driver_name,
5518         .pm             = &ibmvnic_pm_ops,
5519 };
5520
5521 /* module functions */
5522 static int __init ibmvnic_module_init(void)
5523 {
5524         pr_info("%s: %s %s\n", ibmvnic_driver_name, ibmvnic_driver_string,
5525                 IBMVNIC_DRIVER_VERSION);
5526
5527         return vio_register_driver(&ibmvnic_driver);
5528 }
5529
5530 static void __exit ibmvnic_module_exit(void)
5531 {
5532         vio_unregister_driver(&ibmvnic_driver);
5533 }
5534
5535 module_init(ibmvnic_module_init);
5536 module_exit(ibmvnic_module_exit);