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         if (!adapter->login_buf)
 933                 return;
 934
 935         dma_unmap_single(&adapter->vdev->dev, adapter->login_buf_token,
 936                          adapter->login_buf_sz, DMA_TO_DEVICE);
 937         kfree(adapter->login_buf);
 938         adapter->login_buf = NULL;
 939 }
 940
 941 static void release_login_rsp_buffer(struct ibmvnic_adapter *adapter)
 942 {
 943         if (!adapter->login_rsp_buf)
 944                 return;
 945
 946         dma_unmap_single(&adapter->vdev->dev, adapter->login_rsp_buf_token,
 947                          adapter->login_rsp_buf_sz, DMA_FROM_DEVICE);
 948         kfree(adapter->login_rsp_buf);
 949         adapter->login_rsp_buf = NULL;
 950 }
 951
 952 static void release_resources(struct ibmvnic_adapter *adapter)
 953 {
 954         release_vpd_data(adapter);
 955
 956         release_tx_pools(adapter);
 957         release_rx_pools(adapter);
 958
 959         release_napi(adapter);
 960         release_login_buffer(adapter);
 961         release_login_rsp_buffer(adapter);
 962 }
 963
 964 static int set_link_state(struct ibmvnic_adapter *adapter, u8 link_state)
 965 {
 966         struct net_device *netdev = adapter->netdev;
 967         unsigned long timeout = msecs_to_jiffies(20000);
 968         union ibmvnic_crq crq;
 969         bool resend;
 970         int rc;
 971
 972         netdev_dbg(netdev, "setting link state %d\n", link_state);
 973
 974         memset(&crq, 0, sizeof(crq));
 975         crq.logical_link_state.first = IBMVNIC_CRQ_CMD;
 976         crq.logical_link_state.cmd = LOGICAL_LINK_STATE;
 977         crq.logical_link_state.link_state = link_state;
 978
 979         do {
 980                 resend = false;
 981
 982                 reinit_completion(&adapter->init_done);
 983                 rc = ibmvnic_send_crq(adapter, &crq);
 984                 if (rc) {
 985                         netdev_err(netdev, "Failed to set link state\n");
 986                         return rc;
 987                 }
 988
 989                 if (!wait_for_completion_timeout(&adapter->init_done,
 990                                                  timeout)) {
 991                         netdev_err(netdev, "timeout setting link state\n");
 992                         return -1;
 993                 }
 994
 995                 if (adapter->init_done_rc == PARTIALSUCCESS) {
 996                         /* Partuial success, delay and re-send */
 997                         mdelay(1000);
 998                         resend = true;
 999                 } else if (adapter->init_done_rc) {
1000                         netdev_warn(netdev, "Unable to set link state, rc=%d\n",
1001                                     adapter->init_done_rc);
1002                         return adapter->init_done_rc;
1003                 }
1004         } while (resend);
1005
1006         return 0;
1007 }
1008
1009 static int set_real_num_queues(struct net_device *netdev)
1010 {
1011         struct ibmvnic_adapter *adapter = netdev_priv(netdev);
1012         int rc;
1013
1014         netdev_dbg(netdev, "Setting real tx/rx queues (%llx/%llx)\n",
1015                    adapter->req_tx_queues, adapter->req_rx_queues);
1016
1017         rc = netif_set_real_num_tx_queues(netdev, adapter->req_tx_queues);
1018         if (rc) {
1019                 netdev_err(netdev, "failed to set the number of tx queues\n");
1020                 return rc;
1021         }
1022
1023         rc = netif_set_real_num_rx_queues(netdev, adapter->req_rx_queues);
1024         if (rc)
1025                 netdev_err(netdev, "failed to set the number of rx queues\n");
1026
1027         return rc;
1028 }
1029
1030 static int ibmvnic_get_vpd(struct ibmvnic_adapter *adapter)
1031 {
1032         struct device *dev = &adapter->vdev->dev;
1033         union ibmvnic_crq crq;
1034         int len = 0;
1035         int rc;
1036
1037         if (adapter->vpd->buff)
1038                 len = adapter->vpd->len;
1039
1040         mutex_lock(&adapter->fw_lock);
1041         adapter->fw_done_rc = 0;
1042         reinit_completion(&adapter->fw_done);
1043
1044         crq.get_vpd_size.first = IBMVNIC_CRQ_CMD;
1045         crq.get_vpd_size.cmd = GET_VPD_SIZE;
1046         rc = ibmvnic_send_crq(adapter, &crq);
1047         if (rc) {
1048                 mutex_unlock(&adapter->fw_lock);
1049                 return rc;
1050         }
1051
1052         rc = ibmvnic_wait_for_completion(adapter, &adapter->fw_done, 10000);
1053         if (rc) {
1054                 dev_err(dev, "Could not retrieve VPD size, rc = %d\n", rc);
1055                 mutex_unlock(&adapter->fw_lock);
1056                 return rc;
1057         }
1058         mutex_unlock(&adapter->fw_lock);
1059
1060         if (!adapter->vpd->len)
1061                 return -ENODATA;
1062
1063         if (!adapter->vpd->buff)
1064                 adapter->vpd->buff = kzalloc(adapter->vpd->len, GFP_KERNEL);
1065         else if (adapter->vpd->len != len)
1066                 adapter->vpd->buff =
1067                         krealloc(adapter->vpd->buff,
1068                                  adapter->vpd->len, GFP_KERNEL);
1069
1070         if (!adapter->vpd->buff) {
1071                 dev_err(dev, "Could allocate VPD buffer\n");
1072                 return -ENOMEM;
1073         }
1074
1075         adapter->vpd->dma_addr =
1076                 dma_map_single(dev, adapter->vpd->buff, adapter->vpd->len,
1077                                DMA_FROM_DEVICE);
1078         if (dma_mapping_error(dev, adapter->vpd->dma_addr)) {
1079                 dev_err(dev, "Could not map VPD buffer\n");
1080                 kfree(adapter->vpd->buff);
1081                 adapter->vpd->buff = NULL;
1082                 return -ENOMEM;
1083         }
1084
1085         mutex_lock(&adapter->fw_lock);
1086         adapter->fw_done_rc = 0;
1087         reinit_completion(&adapter->fw_done);
1088
1089         crq.get_vpd.first = IBMVNIC_CRQ_CMD;
1090         crq.get_vpd.cmd = GET_VPD;
1091         crq.get_vpd.ioba = cpu_to_be32(adapter->vpd->dma_addr);
1092         crq.get_vpd.len = cpu_to_be32((u32)adapter->vpd->len);
1093         rc = ibmvnic_send_crq(adapter, &crq);
1094         if (rc) {
1095                 kfree(adapter->vpd->buff);
1096                 adapter->vpd->buff = NULL;
1097                 mutex_unlock(&adapter->fw_lock);
1098                 return rc;
1099         }
1100
1101         rc = ibmvnic_wait_for_completion(adapter, &adapter->fw_done, 10000);
1102         if (rc) {
1103                 dev_err(dev, "Unable to retrieve VPD, rc = %d\n", rc);
1104                 kfree(adapter->vpd->buff);
1105                 adapter->vpd->buff = NULL;
1106                 mutex_unlock(&adapter->fw_lock);
1107                 return rc;
1108         }
1109
1110         mutex_unlock(&adapter->fw_lock);
1111         return 0;
1112 }
1113
1114 static int init_resources(struct ibmvnic_adapter *adapter)
1115 {
1116         struct net_device *netdev = adapter->netdev;
1117         int rc;
1118
1119         rc = set_real_num_queues(netdev);
1120         if (rc)
1121                 return rc;
1122
1123         adapter->vpd = kzalloc(sizeof(*adapter->vpd), GFP_KERNEL);
1124         if (!adapter->vpd)
1125                 return -ENOMEM;
1126
1127         /* Vital Product Data (VPD) */
1128         rc = ibmvnic_get_vpd(adapter);
1129         if (rc) {
1130                 netdev_err(netdev, "failed to initialize Vital Product Data (VPD)\n");
1131                 return rc;
1132         }
1133
1134         adapter->map_id = 1;
1135
1136         rc = init_napi(adapter);
1137         if (rc)
1138                 return rc;
1139
1140         send_query_map(adapter);
1141
1142         rc = init_rx_pools(netdev);
1143         if (rc)
1144                 return rc;
1145
1146         rc = init_tx_pools(netdev);
1147         return rc;
1148 }
1149
1150 static int __ibmvnic_open(struct net_device *netdev)
1151 {
1152         struct ibmvnic_adapter *adapter = netdev_priv(netdev);
1153         enum vnic_state prev_state = adapter->state;
1154         int i, rc;
1155
1156         adapter->state = VNIC_OPENING;
1157         replenish_pools(adapter);
1158         ibmvnic_napi_enable(adapter);
1159
1160         /* We're ready to receive frames, enable the sub-crq interrupts and
1161          * set the logical link state to up
1162          */
1163         for (i = 0; i < adapter->req_rx_queues; i++) {
1164                 netdev_dbg(netdev, "Enabling rx_scrq[%d] irq\n", i);
1165                 if (prev_state == VNIC_CLOSED)
1166                         enable_irq(adapter->rx_scrq[i]->irq);
1167                 enable_scrq_irq(adapter, adapter->rx_scrq[i]);
1168         }
1169
1170         for (i = 0; i < adapter->req_tx_queues; i++) {
1171                 netdev_dbg(netdev, "Enabling tx_scrq[%d] irq\n", i);
1172                 if (prev_state == VNIC_CLOSED)
1173                         enable_irq(adapter->tx_scrq[i]->irq);
1174                 enable_scrq_irq(adapter, adapter->tx_scrq[i]);
1175         }
1176
1177         rc = set_link_state(adapter, IBMVNIC_LOGICAL_LNK_UP);
1178         if (rc) {
1179                 ibmvnic_napi_disable(adapter);
1180                 release_resources(adapter);
1181                 return rc;
1182         }
1183
1184         netif_tx_start_all_queues(netdev);
1185
1186         if (prev_state == VNIC_CLOSED) {
1187                 for (i = 0; i < adapter->req_rx_queues; i++)
1188                         napi_schedule(&adapter->napi[i]);
1189         }
1190
1191         adapter->state = VNIC_OPEN;
1192         return rc;
1193 }
1194
1195 static int ibmvnic_open(struct net_device *netdev)
1196 {
1197         struct ibmvnic_adapter *adapter = netdev_priv(netdev);
1198         int rc;
1199
1200         /* If device failover is pending, just set device state and return.
1201          * Device operation will be handled by reset routine.
1202          */
1203         if (adapter->failover_pending) {
1204                 adapter->state = VNIC_OPEN;
1205                 return 0;
1206         }
1207
1208         if (adapter->state != VNIC_CLOSED) {
1209                 rc = ibmvnic_login(netdev);
1210                 if (rc)
1211                         goto out;
1212
1213                 rc = init_resources(adapter);
1214                 if (rc) {
1215                         netdev_err(netdev, "failed to initialize resources\n");
1216                         release_resources(adapter);
1217                         goto out;
1218                 }
1219         }
1220
1221         rc = __ibmvnic_open(netdev);
1222
1223 out:
1224         /*
1225          * If open fails due to a pending failover, set device state and
1226          * return. Device operation will be handled by reset routine.
1227          */
1228         if (rc && adapter->failover_pending) {
1229                 adapter->state = VNIC_OPEN;
1230                 rc = 0;
1231         }
1232         return rc;
1233 }
1234
1235 static void clean_rx_pools(struct ibmvnic_adapter *adapter)
1236 {
1237         struct ibmvnic_rx_pool *rx_pool;
1238         struct ibmvnic_rx_buff *rx_buff;
1239         u64 rx_entries;
1240         int rx_scrqs;
1241         int i, j;
1242
1243         if (!adapter->rx_pool)
1244                 return;
1245
1246         rx_scrqs = adapter->num_active_rx_pools;
1247         rx_entries = adapter->req_rx_add_entries_per_subcrq;
1248
1249         /* Free any remaining skbs in the rx buffer pools */
1250         for (i = 0; i < rx_scrqs; i++) {
1251                 rx_pool = &adapter->rx_pool[i];
1252                 if (!rx_pool || !rx_pool->rx_buff)
1253                         continue;
1254
1255                 netdev_dbg(adapter->netdev, "Cleaning rx_pool[%d]\n", i);
1256                 for (j = 0; j < rx_entries; j++) {
1257                         rx_buff = &rx_pool->rx_buff[j];
1258                         if (rx_buff && rx_buff->skb) {
1259                                 dev_kfree_skb_any(rx_buff->skb);
1260                                 rx_buff->skb = NULL;
1261                         }
1262                 }
1263         }
1264 }
1265
1266 static void clean_one_tx_pool(struct ibmvnic_adapter *adapter,
1267                               struct ibmvnic_tx_pool *tx_pool)
1268 {
1269         struct ibmvnic_tx_buff *tx_buff;
1270         u64 tx_entries;
1271         int i;
1272
1273         if (!tx_pool || !tx_pool->tx_buff)
1274                 return;
1275
1276         tx_entries = tx_pool->num_buffers;
1277
1278         for (i = 0; i < tx_entries; i++) {
1279                 tx_buff = &tx_pool->tx_buff[i];
1280                 if (tx_buff && tx_buff->skb) {
1281                         dev_kfree_skb_any(tx_buff->skb);
1282                         tx_buff->skb = NULL;
1283                 }
1284         }
1285 }
1286
1287 static void clean_tx_pools(struct ibmvnic_adapter *adapter)
1288 {
1289         int tx_scrqs;
1290         int i;
1291
1292         if (!adapter->tx_pool || !adapter->tso_pool)
1293                 return;
1294
1295         tx_scrqs = adapter->num_active_tx_pools;
1296
1297         /* Free any remaining skbs in the tx buffer pools */
1298         for (i = 0; i < tx_scrqs; i++) {
1299                 netdev_dbg(adapter->netdev, "Cleaning tx_pool[%d]\n", i);
1300                 clean_one_tx_pool(adapter, &adapter->tx_pool[i]);
1301                 clean_one_tx_pool(adapter, &adapter->tso_pool[i]);
1302         }
1303 }
1304
1305 static void ibmvnic_disable_irqs(struct ibmvnic_adapter *adapter)
1306 {
1307         struct net_device *netdev = adapter->netdev;
1308         int i;
1309
1310         if (adapter->tx_scrq) {
1311                 for (i = 0; i < adapter->req_tx_queues; i++)
1312                         if (adapter->tx_scrq[i]->irq) {
1313                                 netdev_dbg(netdev,
1314                                            "Disabling tx_scrq[%d] irq\n", i);
1315                                 disable_scrq_irq(adapter, adapter->tx_scrq[i]);
1316                                 disable_irq(adapter->tx_scrq[i]->irq);
1317                         }
1318         }
1319
1320         if (adapter->rx_scrq) {
1321                 for (i = 0; i < adapter->req_rx_queues; i++) {
1322                         if (adapter->rx_scrq[i]->irq) {
1323                                 netdev_dbg(netdev,
1324                                            "Disabling rx_scrq[%d] irq\n", i);
1325                                 disable_scrq_irq(adapter, adapter->rx_scrq[i]);
1326                                 disable_irq(adapter->rx_scrq[i]->irq);
1327                         }
1328                 }
1329         }
1330 }
1331
1332 static void ibmvnic_cleanup(struct net_device *netdev)
1333 {
1334         struct ibmvnic_adapter *adapter = netdev_priv(netdev);
1335
1336         /* ensure that transmissions are stopped if called by do_reset */
1337         if (test_bit(0, &adapter->resetting))
1338                 netif_tx_disable(netdev);
1339         else
1340                 netif_tx_stop_all_queues(netdev);
1341
1342         ibmvnic_napi_disable(adapter);
1343         ibmvnic_disable_irqs(adapter);
1344
1345         clean_rx_pools(adapter);
1346         clean_tx_pools(adapter);
1347 }
1348
1349 static int __ibmvnic_close(struct net_device *netdev)
1350 {
1351         struct ibmvnic_adapter *adapter = netdev_priv(netdev);
1352         int rc = 0;
1353
1354         adapter->state = VNIC_CLOSING;
1355         rc = set_link_state(adapter, IBMVNIC_LOGICAL_LNK_DN);
1356         adapter->state = VNIC_CLOSED;
1357         return rc;
1358 }
1359
1360 static int ibmvnic_close(struct net_device *netdev)
1361 {
1362         struct ibmvnic_adapter *adapter = netdev_priv(netdev);
1363         int rc;
1364
1365         netdev_dbg(netdev, "[S:%d FOP:%d FRR:%d] Closing\n",
1366                    adapter->state, adapter->failover_pending,
1367                    adapter->force_reset_recovery);
1368
1369         /* If device failover is pending, just set device state and return.
1370          * Device operation will be handled by reset routine.
1371          */
1372         if (adapter->failover_pending) {
1373                 adapter->state = VNIC_CLOSED;
1374                 return 0;
1375         }
1376
1377         rc = __ibmvnic_close(netdev);
1378         ibmvnic_cleanup(netdev);
1379
1380         return rc;
1381 }
1382
1383 /**
1384  * build_hdr_data - creates L2/L3/L4 header data buffer
1385  * @hdr_field - bitfield determining needed headers
1386  * @skb - socket buffer
1387  * @hdr_len - array of header lengths
1388  * @tot_len - total length of data
1389  *
1390  * Reads hdr_field to determine which headers are needed by firmware.
1391  * Builds a buffer containing these headers.  Saves individual header
1392  * lengths and total buffer length to be used to build descriptors.
1393  */
1394 static int build_hdr_data(u8 hdr_field, struct sk_buff *skb,
1395                           int *hdr_len, u8 *hdr_data)
1396 {
1397         int len = 0;
1398         u8 *hdr;
1399
1400         if (skb_vlan_tagged(skb) && !skb_vlan_tag_present(skb))
1401                 hdr_len[0] = sizeof(struct vlan_ethhdr);
1402         else
1403                 hdr_len[0] = sizeof(struct ethhdr);
1404
1405         if (skb->protocol == htons(ETH_P_IP)) {
1406                 hdr_len[1] = ip_hdr(skb)->ihl * 4;
1407                 if (ip_hdr(skb)->protocol == IPPROTO_TCP)
1408                         hdr_len[2] = tcp_hdrlen(skb);
1409                 else if (ip_hdr(skb)->protocol == IPPROTO_UDP)
1410                         hdr_len[2] = sizeof(struct udphdr);
1411         } else if (skb->protocol == htons(ETH_P_IPV6)) {
1412                 hdr_len[1] = sizeof(struct ipv6hdr);
1413                 if (ipv6_hdr(skb)->nexthdr == IPPROTO_TCP)
1414                         hdr_len[2] = tcp_hdrlen(skb);
1415                 else if (ipv6_hdr(skb)->nexthdr == IPPROTO_UDP)
1416                         hdr_len[2] = sizeof(struct udphdr);
1417         } else if (skb->protocol == htons(ETH_P_ARP)) {
1418                 hdr_len[1] = arp_hdr_len(skb->dev);
1419                 hdr_len[2] = 0;
1420         }
1421
1422         memset(hdr_data, 0, 120);
1423         if ((hdr_field >> 6) & 1) {
1424                 hdr = skb_mac_header(skb);
1425                 memcpy(hdr_data, hdr, hdr_len[0]);
1426                 len += hdr_len[0];
1427         }
1428
1429         if ((hdr_field >> 5) & 1) {
1430                 hdr = skb_network_header(skb);
1431                 memcpy(hdr_data + len, hdr, hdr_len[1]);
1432                 len += hdr_len[1];
1433         }
1434
1435         if ((hdr_field >> 4) & 1) {
1436                 hdr = skb_transport_header(skb);
1437                 memcpy(hdr_data + len, hdr, hdr_len[2]);
1438                 len += hdr_len[2];
1439         }
1440         return len;
1441 }
1442
1443 /**
1444  * create_hdr_descs - create header and header extension descriptors
1445  * @hdr_field - bitfield determining needed headers
1446  * @data - buffer containing header data
1447  * @len - length of data buffer
1448  * @hdr_len - array of individual header lengths
1449  * @scrq_arr - descriptor array
1450  *
1451  * Creates header and, if needed, header extension descriptors and
1452  * places them in a descriptor array, scrq_arr
1453  */
1454
1455 static int create_hdr_descs(u8 hdr_field, u8 *hdr_data, int len, int *hdr_len,
1456                             union sub_crq *scrq_arr)
1457 {
1458         union sub_crq hdr_desc;
1459         int tmp_len = len;
1460         int num_descs = 0;
1461         u8 *data, *cur;
1462         int tmp;
1463
1464         while (tmp_len > 0) {
1465                 cur = hdr_data + len - tmp_len;
1466
1467                 memset(&hdr_desc, 0, sizeof(hdr_desc));
1468                 if (cur != hdr_data) {
1469                         data = hdr_desc.hdr_ext.data;
1470                         tmp = tmp_len > 29 ? 29 : tmp_len;
1471                         hdr_desc.hdr_ext.first = IBMVNIC_CRQ_CMD;
1472                         hdr_desc.hdr_ext.type = IBMVNIC_HDR_EXT_DESC;
1473                         hdr_desc.hdr_ext.len = tmp;
1474                 } else {
1475                         data = hdr_desc.hdr.data;
1476                         tmp = tmp_len > 24 ? 24 : tmp_len;
1477                         hdr_desc.hdr.first = IBMVNIC_CRQ_CMD;
1478                         hdr_desc.hdr.type = IBMVNIC_HDR_DESC;
1479                         hdr_desc.hdr.len = tmp;
1480                         hdr_desc.hdr.l2_len = (u8)hdr_len[0];
1481                         hdr_desc.hdr.l3_len = cpu_to_be16((u16)hdr_len[1]);
1482                         hdr_desc.hdr.l4_len = (u8)hdr_len[2];
1483                         hdr_desc.hdr.flag = hdr_field << 1;
1484                 }
1485                 memcpy(data, cur, tmp);
1486                 tmp_len -= tmp;
1487                 *scrq_arr = hdr_desc;
1488                 scrq_arr++;
1489                 num_descs++;
1490         }
1491
1492         return num_descs;
1493 }
1494
1495 /**
1496  * build_hdr_descs_arr - build a header descriptor array
1497  * @skb - socket buffer
1498  * @num_entries - number of descriptors to be sent
1499  * @subcrq - first TX descriptor
1500  * @hdr_field - bit field determining which headers will be sent
1501  *
1502  * This function will build a TX descriptor array with applicable
1503  * L2/L3/L4 packet header descriptors to be sent by send_subcrq_indirect.
1504  */
1505
1506 static void build_hdr_descs_arr(struct ibmvnic_tx_buff *txbuff,
1507                                 int *num_entries, u8 hdr_field)
1508 {
1509         int hdr_len[3] = {0, 0, 0};
1510         int tot_len;
1511         u8 *hdr_data = txbuff->hdr_data;
1512
1513         tot_len = build_hdr_data(hdr_field, txbuff->skb, hdr_len,
1514                                  txbuff->hdr_data);
1515         *num_entries += create_hdr_descs(hdr_field, hdr_data, tot_len, hdr_len,
1516                          txbuff->indir_arr + 1);
1517 }
1518
1519 static int ibmvnic_xmit_workarounds(struct sk_buff *skb,
1520                                     struct net_device *netdev)
1521 {
1522         /* For some backing devices, mishandling of small packets
1523          * can result in a loss of connection or TX stall. Device
1524          * architects recommend that no packet should be smaller
1525          * than the minimum MTU value provided to the driver, so
1526          * pad any packets to that length
1527          */
1528         if (skb->len < netdev->min_mtu)
1529                 return skb_put_padto(skb, netdev->min_mtu);
1530
1531         return 0;
1532 }
1533
1534 static netdev_tx_t ibmvnic_xmit(struct sk_buff *skb, struct net_device *netdev)
1535 {
1536         struct ibmvnic_adapter *adapter = netdev_priv(netdev);
1537         int queue_num = skb_get_queue_mapping(skb);
1538         u8 *hdrs = (u8 *)&adapter->tx_rx_desc_req;
1539         struct device *dev = &adapter->vdev->dev;
1540         struct ibmvnic_tx_buff *tx_buff = NULL;
1541         struct ibmvnic_sub_crq_queue *tx_scrq;
1542         struct ibmvnic_tx_pool *tx_pool;
1543         unsigned int tx_send_failed = 0;
1544         unsigned int tx_map_failed = 0;
1545         unsigned int tx_dropped = 0;
1546         unsigned int tx_packets = 0;
1547         unsigned int tx_bytes = 0;
1548         dma_addr_t data_dma_addr;
1549         struct netdev_queue *txq;
1550         unsigned long lpar_rc;
1551         union sub_crq tx_crq;
1552         unsigned int offset;
1553         int num_entries = 1;
1554         unsigned char *dst;
1555         int index = 0;
1556         u8 proto = 0;
1557         u64 handle;
1558         netdev_tx_t ret = NETDEV_TX_OK;
1559
1560         if (test_bit(0, &adapter->resetting)) {
1561                 dev_kfree_skb_any(skb);
1562
1563                 tx_send_failed++;
1564                 tx_dropped++;
1565                 ret = NETDEV_TX_OK;
1566                 goto out;
1567         }
1568
1569         if (ibmvnic_xmit_workarounds(skb, netdev)) {
1570                 tx_dropped++;
1571                 tx_send_failed++;
1572                 ret = NETDEV_TX_OK;
1573                 goto out;
1574         }
1575         if (skb_is_gso(skb))
1576                 tx_pool = &adapter->tso_pool[queue_num];
1577         else
1578                 tx_pool = &adapter->tx_pool[queue_num];
1579
1580         tx_scrq = adapter->tx_scrq[queue_num];
1581         txq = netdev_get_tx_queue(netdev, skb_get_queue_mapping(skb));
1582         handle = tx_scrq->handle;
1583
1584         index = tx_pool->free_map[tx_pool->consumer_index];
1585
1586         if (index == IBMVNIC_INVALID_MAP) {
1587                 dev_kfree_skb_any(skb);
1588                 tx_send_failed++;
1589                 tx_dropped++;
1590                 ret = NETDEV_TX_OK;
1591                 goto out;
1592         }
1593
1594         tx_pool->free_map[tx_pool->consumer_index] = IBMVNIC_INVALID_MAP;
1595
1596         offset = index * tx_pool->buf_size;
1597         dst = tx_pool->long_term_buff.buff + offset;
1598         memset(dst, 0, tx_pool->buf_size);
1599         data_dma_addr = tx_pool->long_term_buff.addr + offset;
1600
1601         if (skb_shinfo(skb)->nr_frags) {
1602                 int cur, i;
1603
1604                 /* Copy the head */
1605                 skb_copy_from_linear_data(skb, dst, skb_headlen(skb));
1606                 cur = skb_headlen(skb);
1607
1608                 /* Copy the frags */
1609                 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
1610                         const skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
1611
1612                         memcpy(dst + cur,
1613                                page_address(skb_frag_page(frag)) +
1614                                skb_frag_off(frag), skb_frag_size(frag));
1615                         cur += skb_frag_size(frag);
1616                 }
1617         } else {
1618                 skb_copy_from_linear_data(skb, dst, skb->len);
1619         }
1620
1621         /* post changes to long_term_buff *dst before VIOS accessing it */
1622         dma_wmb();
1623
1624         tx_pool->consumer_index =
1625             (tx_pool->consumer_index + 1) % tx_pool->num_buffers;
1626
1627         tx_buff = &tx_pool->tx_buff[index];
1628         tx_buff->skb = skb;
1629         tx_buff->data_dma[0] = data_dma_addr;
1630         tx_buff->data_len[0] = skb->len;
1631         tx_buff->index = index;
1632         tx_buff->pool_index = queue_num;
1633         tx_buff->last_frag = true;
1634
1635         memset(&tx_crq, 0, sizeof(tx_crq));
1636         tx_crq.v1.first = IBMVNIC_CRQ_CMD;
1637         tx_crq.v1.type = IBMVNIC_TX_DESC;
1638         tx_crq.v1.n_crq_elem = 1;
1639         tx_crq.v1.n_sge = 1;
1640         tx_crq.v1.flags1 = IBMVNIC_TX_COMP_NEEDED;
1641
1642         if (skb_is_gso(skb))
1643                 tx_crq.v1.correlator =
1644                         cpu_to_be32(index | IBMVNIC_TSO_POOL_MASK);
1645         else
1646                 tx_crq.v1.correlator = cpu_to_be32(index);
1647         tx_crq.v1.dma_reg = cpu_to_be16(tx_pool->long_term_buff.map_id);
1648         tx_crq.v1.sge_len = cpu_to_be32(skb->len);
1649         tx_crq.v1.ioba = cpu_to_be64(data_dma_addr);
1650
1651         if (adapter->vlan_header_insertion && skb_vlan_tag_present(skb)) {
1652                 tx_crq.v1.flags2 |= IBMVNIC_TX_VLAN_INSERT;
1653                 tx_crq.v1.vlan_id = cpu_to_be16(skb->vlan_tci);
1654         }
1655
1656         if (skb->protocol == htons(ETH_P_IP)) {
1657                 tx_crq.v1.flags1 |= IBMVNIC_TX_PROT_IPV4;
1658                 proto = ip_hdr(skb)->protocol;
1659         } else if (skb->protocol == htons(ETH_P_IPV6)) {
1660                 tx_crq.v1.flags1 |= IBMVNIC_TX_PROT_IPV6;
1661                 proto = ipv6_hdr(skb)->nexthdr;
1662         }
1663
1664         if (proto == IPPROTO_TCP)
1665                 tx_crq.v1.flags1 |= IBMVNIC_TX_PROT_TCP;
1666         else if (proto == IPPROTO_UDP)
1667                 tx_crq.v1.flags1 |= IBMVNIC_TX_PROT_UDP;
1668
1669         if (skb->ip_summed == CHECKSUM_PARTIAL) {
1670                 tx_crq.v1.flags1 |= IBMVNIC_TX_CHKSUM_OFFLOAD;
1671                 hdrs += 2;
1672         }
1673         if (skb_is_gso(skb)) {
1674                 tx_crq.v1.flags1 |= IBMVNIC_TX_LSO;
1675                 tx_crq.v1.mss = cpu_to_be16(skb_shinfo(skb)->gso_size);
1676                 hdrs += 2;
1677         }
1678         /* determine if l2/3/4 headers are sent to firmware */
1679         if ((*hdrs >> 7) & 1) {
1680                 build_hdr_descs_arr(tx_buff, &num_entries, *hdrs);
1681                 tx_crq.v1.n_crq_elem = num_entries;
1682                 tx_buff->num_entries = num_entries;
1683                 tx_buff->indir_arr[0] = tx_crq;
1684                 tx_buff->indir_dma = dma_map_single(dev, tx_buff->indir_arr,
1685                                                     sizeof(tx_buff->indir_arr),
1686                                                     DMA_TO_DEVICE);
1687                 if (dma_mapping_error(dev, tx_buff->indir_dma)) {
1688                         dev_kfree_skb_any(skb);
1689                         tx_buff->skb = NULL;
1690                         if (!firmware_has_feature(FW_FEATURE_CMO))
1691                                 dev_err(dev, "tx: unable to map descriptor array\n");
1692                         tx_map_failed++;
1693                         tx_dropped++;
1694                         ret = NETDEV_TX_OK;
1695                         goto tx_err_out;
1696                 }
1697                 lpar_rc = send_subcrq_indirect(adapter, handle,
1698                                                (u64)tx_buff->indir_dma,
1699                                                (u64)num_entries);
1700                 dma_unmap_single(dev, tx_buff->indir_dma,
1701                                  sizeof(tx_buff->indir_arr), DMA_TO_DEVICE);
1702         } else {
1703                 tx_buff->num_entries = num_entries;
1704                 lpar_rc = send_subcrq(adapter, handle,
1705                                       &tx_crq);
1706         }
1707         if (lpar_rc != H_SUCCESS) {
1708                 if (lpar_rc != H_CLOSED && lpar_rc != H_PARAMETER)
1709                         dev_err_ratelimited(dev, "tx: send failed\n");
1710                 dev_kfree_skb_any(skb);
1711                 tx_buff->skb = NULL;
1712
1713                 if (lpar_rc == H_CLOSED || adapter->failover_pending) {
1714                         /* Disable TX and report carrier off if queue is closed
1715                          * or pending failover.
1716                          * Firmware guarantees that a signal will be sent to the
1717                          * driver, triggering a reset or some other action.
1718                          */
1719                         netif_tx_stop_all_queues(netdev);
1720                         netif_carrier_off(netdev);
1721                 }
1722
1723                 tx_send_failed++;
1724                 tx_dropped++;
1725                 ret = NETDEV_TX_OK;
1726                 goto tx_err_out;
1727         }
1728
1729         if (atomic_add_return(num_entries, &tx_scrq->used)
1730                                         >= adapter->req_tx_entries_per_subcrq) {
1731                 netdev_dbg(netdev, "Stopping queue %d\n", queue_num);
1732                 netif_stop_subqueue(netdev, queue_num);
1733         }
1734
1735         tx_packets++;
1736         tx_bytes += skb->len;
1737         txq->trans_start = jiffies;
1738         ret = NETDEV_TX_OK;
1739         goto out;
1740
1741 tx_err_out:
1742         /* roll back consumer index and map array*/
1743         if (tx_pool->consumer_index == 0)
1744                 tx_pool->consumer_index =
1745                         tx_pool->num_buffers - 1;
1746         else
1747                 tx_pool->consumer_index--;
1748         tx_pool->free_map[tx_pool->consumer_index] = index;
1749 out:
1750         netdev->stats.tx_dropped += tx_dropped;
1751         netdev->stats.tx_bytes += tx_bytes;
1752         netdev->stats.tx_packets += tx_packets;
1753         adapter->tx_send_failed += tx_send_failed;
1754         adapter->tx_map_failed += tx_map_failed;
1755         adapter->tx_stats_buffers[queue_num].packets += tx_packets;
1756         adapter->tx_stats_buffers[queue_num].bytes += tx_bytes;
1757         adapter->tx_stats_buffers[queue_num].dropped_packets += tx_dropped;
1758
1759         return ret;
1760 }
1761
1762 static void ibmvnic_set_multi(struct net_device *netdev)
1763 {
1764         struct ibmvnic_adapter *adapter = netdev_priv(netdev);
1765         struct netdev_hw_addr *ha;
1766         union ibmvnic_crq crq;
1767
1768         memset(&crq, 0, sizeof(crq));
1769         crq.request_capability.first = IBMVNIC_CRQ_CMD;
1770         crq.request_capability.cmd = REQUEST_CAPABILITY;
1771
1772         if (netdev->flags & IFF_PROMISC) {
1773                 if (!adapter->promisc_supported)
1774                         return;
1775         } else {
1776                 if (netdev->flags & IFF_ALLMULTI) {
1777                         /* Accept all multicast */
1778                         memset(&crq, 0, sizeof(crq));
1779                         crq.multicast_ctrl.first = IBMVNIC_CRQ_CMD;
1780                         crq.multicast_ctrl.cmd = MULTICAST_CTRL;
1781                         crq.multicast_ctrl.flags = IBMVNIC_ENABLE_ALL;
1782                         ibmvnic_send_crq(adapter, &crq);
1783                 } else if (netdev_mc_empty(netdev)) {
1784                         /* Reject all multicast */
1785                         memset(&crq, 0, sizeof(crq));
1786                         crq.multicast_ctrl.first = IBMVNIC_CRQ_CMD;
1787                         crq.multicast_ctrl.cmd = MULTICAST_CTRL;
1788                         crq.multicast_ctrl.flags = IBMVNIC_DISABLE_ALL;
1789                         ibmvnic_send_crq(adapter, &crq);
1790                 } else {
1791                         /* Accept one or more multicast(s) */
1792                         netdev_for_each_mc_addr(ha, netdev) {
1793                                 memset(&crq, 0, sizeof(crq));
1794                                 crq.multicast_ctrl.first = IBMVNIC_CRQ_CMD;
1795                                 crq.multicast_ctrl.cmd = MULTICAST_CTRL;
1796                                 crq.multicast_ctrl.flags = IBMVNIC_ENABLE_MC;
1797                                 ether_addr_copy(&crq.multicast_ctrl.mac_addr[0],
1798                                                 ha->addr);
1799                                 ibmvnic_send_crq(adapter, &crq);
1800                         }
1801                 }
1802         }
1803 }
1804
1805 static int __ibmvnic_set_mac(struct net_device *netdev, u8 *dev_addr)
1806 {
1807         struct ibmvnic_adapter *adapter = netdev_priv(netdev);
1808         union ibmvnic_crq crq;
1809         int rc;
1810
1811         if (!is_valid_ether_addr(dev_addr)) {
1812                 rc = -EADDRNOTAVAIL;
1813                 goto err;
1814         }
1815
1816         memset(&crq, 0, sizeof(crq));
1817         crq.change_mac_addr.first = IBMVNIC_CRQ_CMD;
1818         crq.change_mac_addr.cmd = CHANGE_MAC_ADDR;
1819         ether_addr_copy(&crq.change_mac_addr.mac_addr[0], dev_addr);
1820
1821         mutex_lock(&adapter->fw_lock);
1822         adapter->fw_done_rc = 0;
1823         reinit_completion(&adapter->fw_done);
1824
1825         rc = ibmvnic_send_crq(adapter, &crq);
1826         if (rc) {
1827                 rc = -EIO;
1828                 mutex_unlock(&adapter->fw_lock);
1829                 goto err;
1830         }
1831
1832         rc = ibmvnic_wait_for_completion(adapter, &adapter->fw_done, 10000);
1833         /* netdev->dev_addr is changed in handle_change_mac_rsp function */
1834         if (rc || adapter->fw_done_rc) {
1835                 rc = -EIO;
1836                 mutex_unlock(&adapter->fw_lock);
1837                 goto err;
1838         }
1839         mutex_unlock(&adapter->fw_lock);
1840         return 0;
1841 err:
1842         ether_addr_copy(adapter->mac_addr, netdev->dev_addr);
1843         return rc;
1844 }
1845
1846 static int ibmvnic_set_mac(struct net_device *netdev, void *p)
1847 {
1848         struct ibmvnic_adapter *adapter = netdev_priv(netdev);
1849         struct sockaddr *addr = p;
1850         int rc;
1851
1852         rc = 0;
1853         if (!is_valid_ether_addr(addr->sa_data))
1854                 return -EADDRNOTAVAIL;
1855
1856         ether_addr_copy(adapter->mac_addr, addr->sa_data);
1857         if (adapter->state != VNIC_PROBED)
1858                 rc = __ibmvnic_set_mac(netdev, addr->sa_data);
1859
1860         return rc;
1861 }
1862
1863 /**
1864  * do_change_param_reset returns zero if we are able to keep processing reset
1865  * events, or non-zero if we hit a fatal error and must halt.
1866  */
1867 static int do_change_param_reset(struct ibmvnic_adapter *adapter,
1868                                  struct ibmvnic_rwi *rwi,
1869                                  u32 reset_state)
1870 {
1871         struct net_device *netdev = adapter->netdev;
1872         int i, rc;
1873
1874         netdev_dbg(adapter->netdev, "Change param resetting driver (%d)\n",
1875                    rwi->reset_reason);
1876
1877         netif_carrier_off(netdev);
1878         adapter->reset_reason = rwi->reset_reason;
1879
1880         ibmvnic_cleanup(netdev);
1881
1882         if (reset_state == VNIC_OPEN) {
1883                 rc = __ibmvnic_close(netdev);
1884                 if (rc)
1885                         goto out;
1886         }
1887
1888         release_resources(adapter);
1889         release_sub_crqs(adapter, 1);
1890         release_crq_queue(adapter);
1891
1892         adapter->state = VNIC_PROBED;
1893
1894         rc = init_crq_queue(adapter);
1895
1896         if (rc) {
1897                 netdev_err(adapter->netdev,
1898                            "Couldn't initialize crq. rc=%d\n", rc);
1899                 return rc;
1900         }
1901
1902         rc = ibmvnic_reset_init(adapter, true);
1903         if (rc) {
1904                 rc = IBMVNIC_INIT_FAILED;
1905                 goto out;
1906         }
1907
1908         /* If the adapter was in PROBE state prior to the reset,
1909          * exit here.
1910          */
1911         if (reset_state == VNIC_PROBED)
1912                 goto out;
1913
1914         rc = ibmvnic_login(netdev);
1915         if (rc) {
1916                 goto out;
1917         }
1918
1919         rc = init_resources(adapter);
1920         if (rc)
1921                 goto out;
1922
1923         ibmvnic_disable_irqs(adapter);
1924
1925         adapter->state = VNIC_CLOSED;
1926
1927         if (reset_state == VNIC_CLOSED)
1928                 return 0;
1929
1930         rc = __ibmvnic_open(netdev);
1931         if (rc) {
1932                 rc = IBMVNIC_OPEN_FAILED;
1933                 goto out;
1934         }
1935
1936         /* refresh device's multicast list */
1937         ibmvnic_set_multi(netdev);
1938
1939         /* kick napi */
1940         for (i = 0; i < adapter->req_rx_queues; i++)
1941                 napi_schedule(&adapter->napi[i]);
1942
1943 out:
1944         if (rc)
1945                 adapter->state = reset_state;
1946         return rc;
1947 }
1948
1949 /**
1950  * do_reset returns zero if we are able to keep processing reset events, or
1951  * non-zero if we hit a fatal error and must halt.
1952  */
1953 static int do_reset(struct ibmvnic_adapter *adapter,
1954                     struct ibmvnic_rwi *rwi, u32 reset_state)
1955 {
1956         u64 old_num_rx_queues, old_num_tx_queues;
1957         u64 old_num_rx_slots, old_num_tx_slots;
1958         struct net_device *netdev = adapter->netdev;
1959         int rc;
1960
1961         netdev_dbg(adapter->netdev,
1962                    "[S:%d FOP:%d] Reset reason %d, reset_state %d\n",
1963                    adapter->state, adapter->failover_pending,
1964                    rwi->reset_reason, reset_state);
1965
1966         rtnl_lock();
1967         /*
1968          * Now that we have the rtnl lock, clear any pending failover.
1969          * This will ensure ibmvnic_open() has either completed or will
1970          * block until failover is complete.
1971          */
1972         if (rwi->reset_reason == VNIC_RESET_FAILOVER)
1973                 adapter->failover_pending = false;
1974
1975         netif_carrier_off(netdev);
1976         adapter->reset_reason = rwi->reset_reason;
1977
1978         old_num_rx_queues = adapter->req_rx_queues;
1979         old_num_tx_queues = adapter->req_tx_queues;
1980         old_num_rx_slots = adapter->req_rx_add_entries_per_subcrq;
1981         old_num_tx_slots = adapter->req_tx_entries_per_subcrq;
1982
1983         ibmvnic_cleanup(netdev);
1984
1985         if (reset_state == VNIC_OPEN &&
1986             adapter->reset_reason != VNIC_RESET_MOBILITY &&
1987             adapter->reset_reason != VNIC_RESET_FAILOVER) {
1988                 adapter->state = VNIC_CLOSING;
1989
1990                 /* Release the RTNL lock before link state change and
1991                  * re-acquire after the link state change to allow
1992                  * linkwatch_event to grab the RTNL lock and run during
1993                  * a reset.
1994                  */
1995                 rtnl_unlock();
1996                 rc = set_link_state(adapter, IBMVNIC_LOGICAL_LNK_DN);
1997                 rtnl_lock();
1998                 if (rc)
1999                         goto out;
2000
2001                 if (adapter->state != VNIC_CLOSING) {
2002                         rc = -1;
2003                         goto out;
2004                 }
2005
2006                 adapter->state = VNIC_CLOSED;
2007         }
2008
2009         if (adapter->reset_reason != VNIC_RESET_NON_FATAL) {
2010                 /* remove the closed state so when we call open it appears
2011                  * we are coming from the probed state.
2012                  */
2013                 adapter->state = VNIC_PROBED;
2014
2015                 if (adapter->reset_reason == VNIC_RESET_MOBILITY) {
2016                         rc = ibmvnic_reenable_crq_queue(adapter);
2017                         release_sub_crqs(adapter, 1);
2018                 } else {
2019                         rc = ibmvnic_reset_crq(adapter);
2020                         if (rc == H_CLOSED || rc == H_SUCCESS) {
2021                                 rc = vio_enable_interrupts(adapter->vdev);
2022                                 if (rc)
2023                                         netdev_err(adapter->netdev,
2024                                                    "Reset failed to enable interrupts. rc=%d\n",
2025                                                    rc);
2026                         }
2027                 }
2028
2029                 if (rc) {
2030                         netdev_err(adapter->netdev,
2031                                    "Reset couldn't initialize crq. rc=%d\n", rc);
2032                         goto out;
2033                 }
2034
2035                 rc = ibmvnic_reset_init(adapter, true);
2036                 if (rc) {
2037                         rc = IBMVNIC_INIT_FAILED;
2038                         goto out;
2039                 }
2040
2041                 /* If the adapter was in PROBE state prior to the reset,
2042                  * exit here.
2043                  */
2044                 if (reset_state == VNIC_PROBED) {
2045                         rc = 0;
2046                         goto out;
2047                 }
2048
2049                 rc = ibmvnic_login(netdev);
2050                 if (rc) {
2051                         goto out;
2052                 }
2053
2054                 if (adapter->req_rx_queues != old_num_rx_queues ||
2055                     adapter->req_tx_queues != old_num_tx_queues ||
2056                     adapter->req_rx_add_entries_per_subcrq !=
2057                     old_num_rx_slots ||
2058                     adapter->req_tx_entries_per_subcrq !=
2059                     old_num_tx_slots ||
2060                     !adapter->rx_pool ||
2061                     !adapter->tso_pool ||
2062                     !adapter->tx_pool) {
2063                         release_rx_pools(adapter);
2064                         release_tx_pools(adapter);
2065                         release_napi(adapter);
2066                         release_vpd_data(adapter);
2067
2068                         rc = init_resources(adapter);
2069                         if (rc)
2070                                 goto out;
2071
2072                 } else {
2073                         rc = reset_tx_pools(adapter);
2074                         if (rc) {
2075                                 netdev_dbg(adapter->netdev, "reset tx pools failed (%d)\n",
2076                                            rc);
2077                                 goto out;
2078                         }
2079
2080                         rc = reset_rx_pools(adapter);
2081                         if (rc) {
2082                                 netdev_dbg(adapter->netdev, "reset rx pools failed (%d)\n",
2083                                            rc);
2084                                 goto out;
2085                         }
2086                 }
2087                 ibmvnic_disable_irqs(adapter);
2088         }
2089         adapter->state = VNIC_CLOSED;
2090
2091         if (reset_state == VNIC_CLOSED) {
2092                 rc = 0;
2093                 goto out;
2094         }
2095
2096         rc = __ibmvnic_open(netdev);
2097         if (rc) {
2098                 rc = IBMVNIC_OPEN_FAILED;
2099                 goto out;
2100         }
2101
2102         /* refresh device's multicast list */
2103         ibmvnic_set_multi(netdev);
2104
2105         if (adapter->reset_reason == VNIC_RESET_FAILOVER ||
2106             adapter->reset_reason == VNIC_RESET_MOBILITY) {
2107                 call_netdevice_notifiers(NETDEV_NOTIFY_PEERS, netdev);
2108                 call_netdevice_notifiers(NETDEV_RESEND_IGMP, netdev);
2109         }
2110
2111         rc = 0;
2112
2113 out:
2114         /* restore the adapter state if reset failed */
2115         if (rc)
2116                 adapter->state = reset_state;
2117         rtnl_unlock();
2118
2119         netdev_dbg(adapter->netdev, "[S:%d FOP:%d] Reset done, rc %d\n",
2120                    adapter->state, adapter->failover_pending, rc);
2121         return rc;
2122 }
2123
2124 static int do_hard_reset(struct ibmvnic_adapter *adapter,
2125                          struct ibmvnic_rwi *rwi, u32 reset_state)
2126 {
2127         struct net_device *netdev = adapter->netdev;
2128         int rc;
2129
2130         netdev_dbg(adapter->netdev, "Hard resetting driver (%d)\n",
2131                    rwi->reset_reason);
2132
2133         netif_carrier_off(netdev);
2134         adapter->reset_reason = rwi->reset_reason;
2135
2136         ibmvnic_cleanup(netdev);
2137         release_resources(adapter);
2138         release_sub_crqs(adapter, 0);
2139         release_crq_queue(adapter);
2140
2141         /* remove the closed state so when we call open it appears
2142          * we are coming from the probed state.
2143          */
2144         adapter->state = VNIC_PROBED;
2145
2146         reinit_completion(&adapter->init_done);
2147         rc = init_crq_queue(adapter);
2148         if (rc) {
2149                 netdev_err(adapter->netdev,
2150                            "Couldn't initialize crq. rc=%d\n", rc);
2151                 goto out;
2152         }
2153
2154         rc = ibmvnic_reset_init(adapter, false);
2155         if (rc)
2156                 goto out;
2157
2158         /* If the adapter was in PROBE state prior to the reset,
2159          * exit here.
2160          */
2161         if (reset_state == VNIC_PROBED)
2162                 goto out;
2163
2164         rc = ibmvnic_login(netdev);
2165         if (rc)
2166                 goto out;
2167
2168         rc = init_resources(adapter);
2169         if (rc)
2170                 goto out;
2171
2172         ibmvnic_disable_irqs(adapter);
2173         adapter->state = VNIC_CLOSED;
2174
2175         if (reset_state == VNIC_CLOSED)
2176                 goto out;
2177
2178         rc = __ibmvnic_open(netdev);
2179         if (rc) {
2180                 rc = IBMVNIC_OPEN_FAILED;
2181                 goto out;
2182         }
2183
2184         call_netdevice_notifiers(NETDEV_NOTIFY_PEERS, netdev);
2185         call_netdevice_notifiers(NETDEV_RESEND_IGMP, netdev);
2186 out:
2187         /* restore adapter state if reset failed */
2188         if (rc)
2189                 adapter->state = reset_state;
2190         netdev_dbg(adapter->netdev, "[S:%d FOP:%d] Hard reset done, rc %d\n",
2191                    adapter->state, adapter->failover_pending, rc);
2192         return rc;
2193 }
2194
2195 static struct ibmvnic_rwi *get_next_rwi(struct ibmvnic_adapter *adapter)
2196 {
2197         struct ibmvnic_rwi *rwi;
2198         unsigned long flags;
2199
2200         spin_lock_irqsave(&adapter->rwi_lock, flags);
2201
2202         if (!list_empty(&adapter->rwi_list)) {
2203                 rwi = list_first_entry(&adapter->rwi_list, struct ibmvnic_rwi,
2204                                        list);
2205                 list_del(&rwi->list);
2206         } else {
2207                 rwi = NULL;
2208         }
2209
2210         spin_unlock_irqrestore(&adapter->rwi_lock, flags);
2211         return rwi;
2212 }
2213
2214 static void __ibmvnic_reset(struct work_struct *work)
2215 {
2216         struct ibmvnic_rwi *rwi;
2217         struct ibmvnic_adapter *adapter;
2218         bool saved_state = false;
2219         unsigned long flags;
2220         u32 reset_state;
2221         int rc = 0;
2222
2223         adapter = container_of(work, struct ibmvnic_adapter, ibmvnic_reset);
2224
2225         if (test_and_set_bit_lock(0, &adapter->resetting)) {
2226                 schedule_delayed_work(&adapter->ibmvnic_delayed_reset,
2227                                       IBMVNIC_RESET_DELAY);
2228                 return;
2229         }
2230
2231         rwi = get_next_rwi(adapter);
2232         while (rwi) {
2233                 spin_lock_irqsave(&adapter->state_lock, flags);
2234
2235                 if (adapter->state == VNIC_REMOVING ||
2236                     adapter->state == VNIC_REMOVED) {
2237                         spin_unlock_irqrestore(&adapter->state_lock, flags);
2238                         kfree(rwi);
2239                         rc = EBUSY;
2240                         break;
2241                 }
2242
2243                 if (!saved_state) {
2244                         reset_state = adapter->state;
2245                         saved_state = true;
2246                 }
2247                 spin_unlock_irqrestore(&adapter->state_lock, flags);
2248
2249                 if (rwi->reset_reason == VNIC_RESET_CHANGE_PARAM) {
2250                         /* CHANGE_PARAM requestor holds rtnl_lock */
2251                         rc = do_change_param_reset(adapter, rwi, reset_state);
2252                 } else if (adapter->force_reset_recovery) {
2253                         /*
2254                          * Since we are doing a hard reset now, clear the
2255                          * failover_pending flag so we don't ignore any
2256                          * future MOBILITY or other resets.
2257                          */
2258                         adapter->failover_pending = false;
2259
2260                         /* Transport event occurred during previous reset */
2261                         if (adapter->wait_for_reset) {
2262                                 /* Previous was CHANGE_PARAM; caller locked */
2263                                 adapter->force_reset_recovery = false;
2264                                 rc = do_hard_reset(adapter, rwi, reset_state);
2265                         } else {
2266                                 rtnl_lock();
2267                                 adapter->force_reset_recovery = false;
2268                                 rc = do_hard_reset(adapter, rwi, reset_state);
2269                                 rtnl_unlock();
2270                         }
2271                         if (rc) {
2272                                 /* give backing device time to settle down */
2273                                 netdev_dbg(adapter->netdev,
2274                                            "[S:%d] Hard reset failed, waiting 60 secs\n",
2275                                            adapter->state);
2276                                 set_current_state(TASK_UNINTERRUPTIBLE);
2277                                 schedule_timeout(60 * HZ);
2278                         }
2279                 } else {
2280                         rc = do_reset(adapter, rwi, reset_state);
2281                 }
2282                 kfree(rwi);
2283                 adapter->last_reset_time = jiffies;
2284
2285                 if (rc)
2286                         netdev_dbg(adapter->netdev, "Reset failed, rc=%d\n", rc);
2287
2288                 rwi = get_next_rwi(adapter);
2289
2290                 if (rwi && (rwi->reset_reason == VNIC_RESET_FAILOVER ||
2291                             rwi->reset_reason == VNIC_RESET_MOBILITY))
2292                         adapter->force_reset_recovery = true;
2293         }
2294
2295         if (adapter->wait_for_reset) {
2296                 adapter->reset_done_rc = rc;
2297                 complete(&adapter->reset_done);
2298         }
2299
2300         clear_bit_unlock(0, &adapter->resetting);
2301
2302         netdev_dbg(adapter->netdev,
2303                    "[S:%d FRR:%d WFR:%d] Done processing resets\n",
2304                    adapter->state, adapter->force_reset_recovery,
2305                    adapter->wait_for_reset);
2306 }
2307
2308 static void __ibmvnic_delayed_reset(struct work_struct *work)
2309 {
2310         struct ibmvnic_adapter *adapter;
2311
2312         adapter = container_of(work, struct ibmvnic_adapter,
2313                                ibmvnic_delayed_reset.work);
2314         __ibmvnic_reset(&adapter->ibmvnic_reset);
2315 }
2316
2317 static int ibmvnic_reset(struct ibmvnic_adapter *adapter,
2318                          enum ibmvnic_reset_reason reason)
2319 {
2320         struct list_head *entry, *tmp_entry;
2321         struct ibmvnic_rwi *rwi, *tmp;
2322         struct net_device *netdev = adapter->netdev;
2323         unsigned long flags;
2324         int ret;
2325
2326         spin_lock_irqsave(&adapter->rwi_lock, flags);
2327
2328         /*
2329          * If failover is pending don't schedule any other reset.
2330          * Instead let the failover complete. If there is already a
2331          * a failover reset scheduled, we will detect and drop the
2332          * duplicate reset when walking the ->rwi_list below.
2333          */
2334         if (adapter->state == VNIC_REMOVING ||
2335             adapter->state == VNIC_REMOVED ||
2336             (adapter->failover_pending && reason != VNIC_RESET_FAILOVER)) {
2337                 ret = EBUSY;
2338                 netdev_dbg(netdev, "Adapter removing or pending failover, skipping reset\n");
2339                 goto err;
2340         }
2341
2342         if (adapter->state == VNIC_PROBING) {
2343                 netdev_warn(netdev, "Adapter reset during probe\n");
2344                 adapter->init_done_rc = EAGAIN;
2345                 ret = EAGAIN;
2346                 goto err;
2347         }
2348
2349         list_for_each(entry, &adapter->rwi_list) {
2350                 tmp = list_entry(entry, struct ibmvnic_rwi, list);
2351                 if (tmp->reset_reason == reason) {
2352                         netdev_dbg(netdev, "Skipping matching reset, reason=%d\n",
2353                                    reason);
2354                         ret = EBUSY;
2355                         goto err;
2356                 }
2357         }
2358
2359         rwi = kzalloc(sizeof(*rwi), GFP_ATOMIC);
2360         if (!rwi) {
2361                 ret = ENOMEM;
2362                 goto err;
2363         }
2364         /* if we just received a transport event,
2365          * flush reset queue and process this reset
2366          */
2367         if (adapter->force_reset_recovery && !list_empty(&adapter->rwi_list)) {
2368                 list_for_each_safe(entry, tmp_entry, &adapter->rwi_list) {
2369                         list_del(entry);
2370                         kfree(list_entry(entry, struct ibmvnic_rwi, list));
2371                 }
2372         }
2373         rwi->reset_reason = reason;
2374         list_add_tail(&rwi->list, &adapter->rwi_list);
2375         netdev_dbg(adapter->netdev, "Scheduling reset (reason %d)\n", reason);
2376         schedule_work(&adapter->ibmvnic_reset);
2377
2378         ret = 0;
2379 err:
2380         /* ibmvnic_close() below can block, so drop the lock first */
2381         spin_unlock_irqrestore(&adapter->rwi_lock, flags);
2382
2383         if (ret == ENOMEM)
2384                 ibmvnic_close(netdev);
2385
2386         return -ret;
2387 }
2388
2389 static void ibmvnic_tx_timeout(struct net_device *dev, unsigned int txqueue)
2390 {
2391         struct ibmvnic_adapter *adapter = netdev_priv(dev);
2392
2393         if (test_bit(0, &adapter->resetting)) {
2394                 netdev_err(adapter->netdev,
2395                            "Adapter is resetting, skip timeout reset\n");
2396                 return;
2397         }
2398         /* No queuing up reset until at least 5 seconds (default watchdog val)
2399          * after last reset
2400          */
2401         if (time_before(jiffies, (adapter->last_reset_time + dev->watchdog_timeo))) {
2402                 netdev_dbg(dev, "Not yet time to tx timeout.\n");
2403                 return;
2404         }
2405         ibmvnic_reset(adapter, VNIC_RESET_TIMEOUT);
2406 }
2407
2408 static void remove_buff_from_pool(struct ibmvnic_adapter *adapter,
2409                                   struct ibmvnic_rx_buff *rx_buff)
2410 {
2411         struct ibmvnic_rx_pool *pool = &adapter->rx_pool[rx_buff->pool_index];
2412
2413         rx_buff->skb = NULL;
2414
2415         pool->free_map[pool->next_alloc] = (int)(rx_buff - pool->rx_buff);
2416         pool->next_alloc = (pool->next_alloc + 1) % pool->size;
2417
2418         atomic_dec(&pool->available);
2419 }
2420
2421 static int ibmvnic_poll(struct napi_struct *napi, int budget)
2422 {
2423         struct net_device *netdev = napi->dev;
2424         struct ibmvnic_adapter *adapter = netdev_priv(netdev);
2425         int scrq_num = (int)(napi - adapter->napi);
2426         int frames_processed = 0;
2427
2428 restart_poll:
2429         while (frames_processed < budget) {
2430                 struct sk_buff *skb;
2431                 struct ibmvnic_rx_buff *rx_buff;
2432                 union sub_crq *next;
2433                 u32 length;
2434                 u16 offset;
2435                 u8 flags = 0;
2436
2437                 if (unlikely(test_bit(0, &adapter->resetting) &&
2438                              adapter->reset_reason != VNIC_RESET_NON_FATAL)) {
2439                         enable_scrq_irq(adapter, adapter->rx_scrq[scrq_num]);
2440                         napi_complete_done(napi, frames_processed);
2441                         return frames_processed;
2442                 }
2443
2444                 if (!pending_scrq(adapter, adapter->rx_scrq[scrq_num]))
2445                         break;
2446                 /* The queue entry at the current index is peeked at above
2447                  * to determine that there is a valid descriptor awaiting
2448                  * processing. We want to be sure that the current slot
2449                  * holds a valid descriptor before reading its contents.
2450                  */
2451                 dma_rmb();
2452                 next = ibmvnic_next_scrq(adapter, adapter->rx_scrq[scrq_num]);
2453                 rx_buff =
2454                     (struct ibmvnic_rx_buff *)be64_to_cpu(next->
2455                                                           rx_comp.correlator);
2456                 /* do error checking */
2457                 if (next->rx_comp.rc) {
2458                         netdev_dbg(netdev, "rx buffer returned with rc %x\n",
2459                                    be16_to_cpu(next->rx_comp.rc));
2460                         /* free the entry */
2461                         next->rx_comp.first = 0;
2462                         dev_kfree_skb_any(rx_buff->skb);
2463                         remove_buff_from_pool(adapter, rx_buff);
2464                         continue;
2465                 } else if (!rx_buff->skb) {
2466                         /* free the entry */
2467                         next->rx_comp.first = 0;
2468                         remove_buff_from_pool(adapter, rx_buff);
2469                         continue;
2470                 }
2471
2472                 length = be32_to_cpu(next->rx_comp.len);
2473                 offset = be16_to_cpu(next->rx_comp.off_frame_data);
2474                 flags = next->rx_comp.flags;
2475                 skb = rx_buff->skb;
2476                 /* load long_term_buff before copying to skb */
2477                 dma_rmb();
2478                 skb_copy_to_linear_data(skb, rx_buff->data + offset,
2479                                         length);
2480
2481                 /* VLAN Header has been stripped by the system firmware and
2482                  * needs to be inserted by the driver
2483                  */
2484                 if (adapter->rx_vlan_header_insertion &&
2485                     (flags & IBMVNIC_VLAN_STRIPPED))
2486                         __vlan_hwaccel_put_tag(skb, htons(ETH_P_8021Q),
2487                                                ntohs(next->rx_comp.vlan_tci));
2488
2489                 /* free the entry */
2490                 next->rx_comp.first = 0;
2491                 remove_buff_from_pool(adapter, rx_buff);
2492
2493                 skb_put(skb, length);
2494                 skb->protocol = eth_type_trans(skb, netdev);
2495                 skb_record_rx_queue(skb, scrq_num);
2496
2497                 if (flags & IBMVNIC_IP_CHKSUM_GOOD &&
2498                     flags & IBMVNIC_TCP_UDP_CHKSUM_GOOD) {
2499                         skb->ip_summed = CHECKSUM_UNNECESSARY;
2500                 }
2501
2502                 length = skb->len;
2503                 napi_gro_receive(napi, skb); /* send it up */
2504                 netdev->stats.rx_packets++;
2505                 netdev->stats.rx_bytes += length;
2506                 adapter->rx_stats_buffers[scrq_num].packets++;
2507                 adapter->rx_stats_buffers[scrq_num].bytes += length;
2508                 frames_processed++;
2509         }
2510
2511         if (adapter->state != VNIC_CLOSING)
2512                 replenish_rx_pool(adapter, &adapter->rx_pool[scrq_num]);
2513
2514         if (frames_processed < budget) {
2515                 enable_scrq_irq(adapter, adapter->rx_scrq[scrq_num]);
2516                 napi_complete_done(napi, frames_processed);
2517                 if (pending_scrq(adapter, adapter->rx_scrq[scrq_num]) &&
2518                     napi_reschedule(napi)) {
2519                         disable_scrq_irq(adapter, adapter->rx_scrq[scrq_num]);
2520                         goto restart_poll;
2521                 }
2522         }
2523         return frames_processed;
2524 }
2525
2526 static int wait_for_reset(struct ibmvnic_adapter *adapter)
2527 {
2528         int rc, ret;
2529
2530         adapter->fallback.mtu = adapter->req_mtu;
2531         adapter->fallback.rx_queues = adapter->req_rx_queues;
2532         adapter->fallback.tx_queues = adapter->req_tx_queues;
2533         adapter->fallback.rx_entries = adapter->req_rx_add_entries_per_subcrq;
2534         adapter->fallback.tx_entries = adapter->req_tx_entries_per_subcrq;
2535
2536         reinit_completion(&adapter->reset_done);
2537         adapter->wait_for_reset = true;
2538         rc = ibmvnic_reset(adapter, VNIC_RESET_CHANGE_PARAM);
2539
2540         if (rc) {
2541                 ret = rc;
2542                 goto out;
2543         }
2544         rc = ibmvnic_wait_for_completion(adapter, &adapter->reset_done, 60000);
2545         if (rc) {
2546                 ret = -ENODEV;
2547                 goto out;
2548         }
2549
2550         ret = 0;
2551         if (adapter->reset_done_rc) {
2552                 ret = -EIO;
2553                 adapter->desired.mtu = adapter->fallback.mtu;
2554                 adapter->desired.rx_queues = adapter->fallback.rx_queues;
2555                 adapter->desired.tx_queues = adapter->fallback.tx_queues;
2556                 adapter->desired.rx_entries = adapter->fallback.rx_entries;
2557                 adapter->desired.tx_entries = adapter->fallback.tx_entries;
2558
2559                 reinit_completion(&adapter->reset_done);
2560                 adapter->wait_for_reset = true;
2561                 rc = ibmvnic_reset(adapter, VNIC_RESET_CHANGE_PARAM);
2562                 if (rc) {
2563                         ret = rc;
2564                         goto out;
2565                 }
2566                 rc = ibmvnic_wait_for_completion(adapter, &adapter->reset_done,
2567                                                  60000);
2568                 if (rc) {
2569                         ret = -ENODEV;
2570                         goto out;
2571                 }
2572         }
2573 out:
2574         adapter->wait_for_reset = false;
2575
2576         return ret;
2577 }
2578
2579 static int ibmvnic_change_mtu(struct net_device *netdev, int new_mtu)
2580 {
2581         struct ibmvnic_adapter *adapter = netdev_priv(netdev);
2582
2583         adapter->desired.mtu = new_mtu + ETH_HLEN;
2584
2585         return wait_for_reset(adapter);
2586 }
2587
2588 static netdev_features_t ibmvnic_features_check(struct sk_buff *skb,
2589                                                 struct net_device *dev,
2590                                                 netdev_features_t features)
2591 {
2592         /* Some backing hardware adapters can not
2593          * handle packets with a MSS less than 224
2594          * or with only one segment.
2595          */
2596         if (skb_is_gso(skb)) {
2597                 if (skb_shinfo(skb)->gso_size < 224 ||
2598                     skb_shinfo(skb)->gso_segs == 1)
2599                         features &= ~NETIF_F_GSO_MASK;
2600         }
2601
2602         return features;
2603 }
2604
2605 static const struct net_device_ops ibmvnic_netdev_ops = {
2606         .ndo_open               = ibmvnic_open,
2607         .ndo_stop               = ibmvnic_close,
2608         .ndo_start_xmit         = ibmvnic_xmit,
2609         .ndo_set_rx_mode        = ibmvnic_set_multi,
2610         .ndo_set_mac_address    = ibmvnic_set_mac,
2611         .ndo_validate_addr      = eth_validate_addr,
2612         .ndo_tx_timeout         = ibmvnic_tx_timeout,
2613         .ndo_change_mtu         = ibmvnic_change_mtu,
2614         .ndo_features_check     = ibmvnic_features_check,
2615 };
2616
2617 /* ethtool functions */
2618
2619 static int ibmvnic_get_link_ksettings(struct net_device *netdev,
2620                                       struct ethtool_link_ksettings *cmd)
2621 {
2622         struct ibmvnic_adapter *adapter = netdev_priv(netdev);
2623         int rc;
2624
2625         rc = send_query_phys_parms(adapter);
2626         if (rc) {
2627                 adapter->speed = SPEED_UNKNOWN;
2628                 adapter->duplex = DUPLEX_UNKNOWN;
2629         }
2630         cmd->base.speed = adapter->speed;
2631         cmd->base.duplex = adapter->duplex;
2632         cmd->base.port = PORT_FIBRE;
2633         cmd->base.phy_address = 0;
2634         cmd->base.autoneg = AUTONEG_ENABLE;
2635
2636         return 0;
2637 }
2638
2639 static void ibmvnic_get_drvinfo(struct net_device *netdev,
2640                                 struct ethtool_drvinfo *info)
2641 {
2642         struct ibmvnic_adapter *adapter = netdev_priv(netdev);
2643
2644         strlcpy(info->driver, ibmvnic_driver_name, sizeof(info->driver));
2645         strlcpy(info->version, IBMVNIC_DRIVER_VERSION, sizeof(info->version));
2646         strlcpy(info->fw_version, adapter->fw_version,
2647                 sizeof(info->fw_version));
2648 }
2649
2650 static u32 ibmvnic_get_msglevel(struct net_device *netdev)
2651 {
2652         struct ibmvnic_adapter *adapter = netdev_priv(netdev);
2653
2654         return adapter->msg_enable;
2655 }
2656
2657 static void ibmvnic_set_msglevel(struct net_device *netdev, u32 data)
2658 {
2659         struct ibmvnic_adapter *adapter = netdev_priv(netdev);
2660
2661         adapter->msg_enable = data;
2662 }
2663
2664 static u32 ibmvnic_get_link(struct net_device *netdev)
2665 {
2666         struct ibmvnic_adapter *adapter = netdev_priv(netdev);
2667
2668         /* Don't need to send a query because we request a logical link up at
2669          * init and then we wait for link state indications
2670          */
2671         return adapter->logical_link_state;
2672 }
2673
2674 static void ibmvnic_get_ringparam(struct net_device *netdev,
2675                                   struct ethtool_ringparam *ring)
2676 {
2677         struct ibmvnic_adapter *adapter = netdev_priv(netdev);
2678
2679         ring->rx_max_pending = adapter->max_rx_add_entries_per_subcrq;
2680         ring->tx_max_pending = adapter->max_tx_entries_per_subcrq;
2681         ring->rx_mini_max_pending = 0;
2682         ring->rx_jumbo_max_pending = 0;
2683         ring->rx_pending = adapter->req_rx_add_entries_per_subcrq;
2684         ring->tx_pending = adapter->req_tx_entries_per_subcrq;
2685         ring->rx_mini_pending = 0;
2686         ring->rx_jumbo_pending = 0;
2687 }
2688
2689 static int ibmvnic_set_ringparam(struct net_device *netdev,
2690                                  struct ethtool_ringparam *ring)
2691 {
2692         struct ibmvnic_adapter *adapter = netdev_priv(netdev);
2693
2694         if (ring->rx_pending > adapter->max_rx_add_entries_per_subcrq  ||
2695             ring->tx_pending > adapter->max_tx_entries_per_subcrq) {
2696                 netdev_err(netdev, "Invalid request.\n");
2697                 netdev_err(netdev, "Max tx buffers = %llu\n",
2698                            adapter->max_rx_add_entries_per_subcrq);
2699                 netdev_err(netdev, "Max rx buffers = %llu\n",
2700                            adapter->max_tx_entries_per_subcrq);
2701                 return -EINVAL;
2702         }
2703
2704         adapter->desired.rx_entries = ring->rx_pending;
2705         adapter->desired.tx_entries = ring->tx_pending;
2706
2707         return wait_for_reset(adapter);
2708 }
2709
2710 static void ibmvnic_get_channels(struct net_device *netdev,
2711                                  struct ethtool_channels *channels)
2712 {
2713         struct ibmvnic_adapter *adapter = netdev_priv(netdev);
2714
2715         channels->max_rx = adapter->max_rx_queues;
2716         channels->max_tx = adapter->max_tx_queues;
2717         channels->max_other = 0;
2718         channels->max_combined = 0;
2719         channels->rx_count = adapter->req_rx_queues;
2720         channels->tx_count = adapter->req_tx_queues;
2721         channels->other_count = 0;
2722         channels->combined_count = 0;
2723 }
2724
2725 static int ibmvnic_set_channels(struct net_device *netdev,
2726                                 struct ethtool_channels *channels)
2727 {
2728         struct ibmvnic_adapter *adapter = netdev_priv(netdev);
2729
2730         adapter->desired.rx_queues = channels->rx_count;
2731         adapter->desired.tx_queues = channels->tx_count;
2732
2733         return wait_for_reset(adapter);
2734 }
2735
2736 static void ibmvnic_get_strings(struct net_device *dev, u32 stringset, u8 *data)
2737 {
2738         struct ibmvnic_adapter *adapter = netdev_priv(dev);
2739         int i;
2740
2741         if (stringset != ETH_SS_STATS)
2742                 return;
2743
2744         for (i = 0; i < ARRAY_SIZE(ibmvnic_stats); i++, data += ETH_GSTRING_LEN)
2745                 memcpy(data, ibmvnic_stats[i].name, ETH_GSTRING_LEN);
2746
2747         for (i = 0; i < adapter->req_tx_queues; i++) {
2748                 snprintf(data, ETH_GSTRING_LEN, "tx%d_packets", i);
2749                 data += ETH_GSTRING_LEN;
2750
2751                 snprintf(data, ETH_GSTRING_LEN, "tx%d_bytes", i);
2752                 data += ETH_GSTRING_LEN;
2753
2754                 snprintf(data, ETH_GSTRING_LEN, "tx%d_dropped_packets", i);
2755                 data += ETH_GSTRING_LEN;
2756         }
2757
2758         for (i = 0; i < adapter->req_rx_queues; i++) {
2759                 snprintf(data, ETH_GSTRING_LEN, "rx%d_packets", i);
2760                 data += ETH_GSTRING_LEN;
2761
2762                 snprintf(data, ETH_GSTRING_LEN, "rx%d_bytes", i);
2763                 data += ETH_GSTRING_LEN;
2764
2765                 snprintf(data, ETH_GSTRING_LEN, "rx%d_interrupts", i);
2766                 data += ETH_GSTRING_LEN;
2767         }
2768 }
2769
2770 static int ibmvnic_get_sset_count(struct net_device *dev, int sset)
2771 {
2772         struct ibmvnic_adapter *adapter = netdev_priv(dev);
2773
2774         switch (sset) {
2775         case ETH_SS_STATS:
2776                 return ARRAY_SIZE(ibmvnic_stats) +
2777                        adapter->req_tx_queues * NUM_TX_STATS +
2778                        adapter->req_rx_queues * NUM_RX_STATS;
2779         default:
2780                 return -EOPNOTSUPP;
2781         }
2782 }
2783
2784 static void ibmvnic_get_ethtool_stats(struct net_device *dev,
2785                                       struct ethtool_stats *stats, u64 *data)
2786 {
2787         struct ibmvnic_adapter *adapter = netdev_priv(dev);
2788         union ibmvnic_crq crq;
2789         int i, j;
2790         int rc;
2791
2792         memset(&crq, 0, sizeof(crq));
2793         crq.request_statistics.first = IBMVNIC_CRQ_CMD;
2794         crq.request_statistics.cmd = REQUEST_STATISTICS;
2795         crq.request_statistics.ioba = cpu_to_be32(adapter->stats_token);
2796         crq.request_statistics.len =
2797             cpu_to_be32(sizeof(struct ibmvnic_statistics));
2798
2799         /* Wait for data to be written */
2800         reinit_completion(&adapter->stats_done);
2801         rc = ibmvnic_send_crq(adapter, &crq);
2802         if (rc)
2803                 return;
2804         rc = ibmvnic_wait_for_completion(adapter, &adapter->stats_done, 10000);
2805         if (rc)
2806                 return;
2807
2808         for (i = 0; i < ARRAY_SIZE(ibmvnic_stats); i++)
2809                 data[i] = be64_to_cpu(IBMVNIC_GET_STAT
2810                                       (adapter, ibmvnic_stats[i].offset));
2811
2812         for (j = 0; j < adapter->req_tx_queues; j++) {
2813                 data[i] = adapter->tx_stats_buffers[j].packets;
2814                 i++;
2815                 data[i] = adapter->tx_stats_buffers[j].bytes;
2816                 i++;
2817                 data[i] = adapter->tx_stats_buffers[j].dropped_packets;
2818                 i++;
2819         }
2820
2821         for (j = 0; j < adapter->req_rx_queues; j++) {
2822                 data[i] = adapter->rx_stats_buffers[j].packets;
2823                 i++;
2824                 data[i] = adapter->rx_stats_buffers[j].bytes;
2825                 i++;
2826                 data[i] = adapter->rx_stats_buffers[j].interrupts;
2827                 i++;
2828         }
2829 }
2830
2831 static const struct ethtool_ops ibmvnic_ethtool_ops = {
2832         .get_drvinfo            = ibmvnic_get_drvinfo,
2833         .get_msglevel           = ibmvnic_get_msglevel,
2834         .set_msglevel           = ibmvnic_set_msglevel,
2835         .get_link               = ibmvnic_get_link,
2836         .get_ringparam          = ibmvnic_get_ringparam,
2837         .set_ringparam          = ibmvnic_set_ringparam,
2838         .get_channels           = ibmvnic_get_channels,
2839         .set_channels           = ibmvnic_set_channels,
2840         .get_strings            = ibmvnic_get_strings,
2841         .get_sset_count         = ibmvnic_get_sset_count,
2842         .get_ethtool_stats      = ibmvnic_get_ethtool_stats,
2843         .get_link_ksettings     = ibmvnic_get_link_ksettings,
2844 };
2845
2846 /* Routines for managing CRQs/sCRQs  */
2847
2848 static int reset_one_sub_crq_queue(struct ibmvnic_adapter *adapter,
2849                                    struct ibmvnic_sub_crq_queue *scrq)
2850 {
2851         int rc;
2852
2853         if (!scrq) {
2854                 netdev_dbg(adapter->netdev, "Invalid scrq reset.\n");
2855                 return -EINVAL;
2856         }
2857
2858         if (scrq->irq) {
2859                 free_irq(scrq->irq, scrq);
2860                 irq_dispose_mapping(scrq->irq);
2861                 scrq->irq = 0;
2862         }
2863         if (scrq->msgs) {
2864                 memset(scrq->msgs, 0, 4 * PAGE_SIZE);
2865                 atomic_set(&scrq->used, 0);
2866                 scrq->cur = 0;
2867         } else {
2868                 netdev_dbg(adapter->netdev, "Invalid scrq reset\n");
2869                 return -EINVAL;
2870         }
2871
2872         rc = h_reg_sub_crq(adapter->vdev->unit_address, scrq->msg_token,
2873                            4 * PAGE_SIZE, &scrq->crq_num, &scrq->hw_irq);
2874         return rc;
2875 }
2876
2877 static int reset_sub_crq_queues(struct ibmvnic_adapter *adapter)
2878 {
2879         int i, rc;
2880
2881         if (!adapter->tx_scrq || !adapter->rx_scrq)
2882                 return -EINVAL;
2883
2884         for (i = 0; i < adapter->req_tx_queues; i++) {
2885                 netdev_dbg(adapter->netdev, "Re-setting tx_scrq[%d]\n", i);
2886                 rc = reset_one_sub_crq_queue(adapter, adapter->tx_scrq[i]);
2887                 if (rc)
2888                         return rc;
2889         }
2890
2891         for (i = 0; i < adapter->req_rx_queues; i++) {
2892                 netdev_dbg(adapter->netdev, "Re-setting rx_scrq[%d]\n", i);
2893                 rc = reset_one_sub_crq_queue(adapter, adapter->rx_scrq[i]);
2894                 if (rc)
2895                         return rc;
2896         }
2897
2898         return rc;
2899 }
2900
2901 static void release_sub_crq_queue(struct ibmvnic_adapter *adapter,
2902                                   struct ibmvnic_sub_crq_queue *scrq,
2903                                   bool do_h_free)
2904 {
2905         struct device *dev = &adapter->vdev->dev;
2906         long rc;
2907
2908         netdev_dbg(adapter->netdev, "Releasing sub-CRQ\n");
2909
2910         if (do_h_free) {
2911                 /* Close the sub-crqs */
2912                 do {
2913                         rc = plpar_hcall_norets(H_FREE_SUB_CRQ,
2914                                                 adapter->vdev->unit_address,
2915                                                 scrq->crq_num);
2916                 } while (rc == H_BUSY || H_IS_LONG_BUSY(rc));
2917
2918                 if (rc) {
2919                         netdev_err(adapter->netdev,
2920                                    "Failed to release sub-CRQ %16lx, rc = %ld\n",
2921                                    scrq->crq_num, rc);
2922                 }
2923         }
2924
2925         dma_unmap_single(dev, scrq->msg_token, 4 * PAGE_SIZE,
2926                          DMA_BIDIRECTIONAL);
2927         free_pages((unsigned long)scrq->msgs, 2);
2928         kfree(scrq);
2929 }
2930
2931 static struct ibmvnic_sub_crq_queue *init_sub_crq_queue(struct ibmvnic_adapter
2932                                                         *adapter)
2933 {
2934         struct device *dev = &adapter->vdev->dev;
2935         struct ibmvnic_sub_crq_queue *scrq;
2936         int rc;
2937
2938         scrq = kzalloc(sizeof(*scrq), GFP_KERNEL);
2939         if (!scrq)
2940                 return NULL;
2941
2942         scrq->msgs =
2943                 (union sub_crq *)__get_free_pages(GFP_KERNEL | __GFP_ZERO, 2);
2944         if (!scrq->msgs) {
2945                 dev_warn(dev, "Couldn't allocate crq queue messages page\n");
2946                 goto zero_page_failed;
2947         }
2948
2949         scrq->msg_token = dma_map_single(dev, scrq->msgs, 4 * PAGE_SIZE,
2950                                          DMA_BIDIRECTIONAL);
2951         if (dma_mapping_error(dev, scrq->msg_token)) {
2952                 dev_warn(dev, "Couldn't map crq queue messages page\n");
2953                 goto map_failed;
2954         }
2955
2956         rc = h_reg_sub_crq(adapter->vdev->unit_address, scrq->msg_token,
2957                            4 * PAGE_SIZE, &scrq->crq_num, &scrq->hw_irq);
2958
2959         if (rc == H_RESOURCE)
2960                 rc = ibmvnic_reset_crq(adapter);
2961
2962         if (rc == H_CLOSED) {
2963                 dev_warn(dev, "Partner adapter not ready, waiting.\n");
2964         } else if (rc) {
2965                 dev_warn(dev, "Error %d registering sub-crq\n", rc);
2966                 goto reg_failed;
2967         }
2968
2969         scrq->adapter = adapter;
2970         scrq->size = 4 * PAGE_SIZE / sizeof(*scrq->msgs);
2971         spin_lock_init(&scrq->lock);
2972
2973         netdev_dbg(adapter->netdev,
2974                    "sub-crq initialized, num %lx, hw_irq=%lx, irq=%x\n",
2975                    scrq->crq_num, scrq->hw_irq, scrq->irq);
2976
2977         return scrq;
2978
2979 reg_failed:
2980         dma_unmap_single(dev, scrq->msg_token, 4 * PAGE_SIZE,
2981                          DMA_BIDIRECTIONAL);
2982 map_failed:
2983         free_pages((unsigned long)scrq->msgs, 2);
2984 zero_page_failed:
2985         kfree(scrq);
2986
2987         return NULL;
2988 }
2989
2990 static void release_sub_crqs(struct ibmvnic_adapter *adapter, bool do_h_free)
2991 {
2992         int i;
2993
2994         if (adapter->tx_scrq) {
2995                 for (i = 0; i < adapter->num_active_tx_scrqs; i++) {
2996                         if (!adapter->tx_scrq[i])
2997                                 continue;
2998
2999                         netdev_dbg(adapter->netdev, "Releasing tx_scrq[%d]\n",
3000                                    i);
3001                         if (adapter->tx_scrq[i]->irq) {
3002                                 free_irq(adapter->tx_scrq[i]->irq,
3003                                          adapter->tx_scrq[i]);
3004                                 irq_dispose_mapping(adapter->tx_scrq[i]->irq);
3005                                 adapter->tx_scrq[i]->irq = 0;
3006                         }
3007
3008                         release_sub_crq_queue(adapter, adapter->tx_scrq[i],
3009                                               do_h_free);
3010                 }
3011
3012                 kfree(adapter->tx_scrq);
3013                 adapter->tx_scrq = NULL;
3014                 adapter->num_active_tx_scrqs = 0;
3015         }
3016
3017         if (adapter->rx_scrq) {
3018                 for (i = 0; i < adapter->num_active_rx_scrqs; i++) {
3019                         if (!adapter->rx_scrq[i])
3020                                 continue;
3021
3022                         netdev_dbg(adapter->netdev, "Releasing rx_scrq[%d]\n",
3023                                    i);
3024                         if (adapter->rx_scrq[i]->irq) {
3025                                 free_irq(adapter->rx_scrq[i]->irq,
3026                                          adapter->rx_scrq[i]);
3027                                 irq_dispose_mapping(adapter->rx_scrq[i]->irq);
3028                                 adapter->rx_scrq[i]->irq = 0;
3029                         }
3030
3031                         release_sub_crq_queue(adapter, adapter->rx_scrq[i],
3032                                               do_h_free);
3033                 }
3034
3035                 kfree(adapter->rx_scrq);
3036                 adapter->rx_scrq = NULL;
3037                 adapter->num_active_rx_scrqs = 0;
3038         }
3039 }
3040
3041 static int disable_scrq_irq(struct ibmvnic_adapter *adapter,
3042                             struct ibmvnic_sub_crq_queue *scrq)
3043 {
3044         struct device *dev = &adapter->vdev->dev;
3045         unsigned long rc;
3046
3047         rc = plpar_hcall_norets(H_VIOCTL, adapter->vdev->unit_address,
3048                                 H_DISABLE_VIO_INTERRUPT, scrq->hw_irq, 0, 0);
3049         if (rc)
3050                 dev_err(dev, "Couldn't disable scrq irq 0x%lx. rc=%ld\n",
3051                         scrq->hw_irq, rc);
3052         return rc;
3053 }
3054
3055 static int enable_scrq_irq(struct ibmvnic_adapter *adapter,
3056                            struct ibmvnic_sub_crq_queue *scrq)
3057 {
3058         struct device *dev = &adapter->vdev->dev;
3059         unsigned long rc;
3060
3061         if (scrq->hw_irq > 0x100000000ULL) {
3062                 dev_err(dev, "bad hw_irq = %lx\n", scrq->hw_irq);
3063                 return 1;
3064         }
3065
3066         if (test_bit(0, &adapter->resetting) &&
3067             adapter->reset_reason == VNIC_RESET_MOBILITY) {
3068                 u64 val = (0xff000000) | scrq->hw_irq;
3069
3070                 rc = plpar_hcall_norets(H_EOI, val);
3071                 /* H_EOI would fail with rc = H_FUNCTION when running
3072                  * in XIVE mode which is expected, but not an error.
3073                  */
3074                 if (rc && rc != H_FUNCTION)
3075                         dev_err(dev, "H_EOI FAILED irq 0x%llx. rc=%ld\n",
3076                                 val, rc);
3077         }
3078
3079         rc = plpar_hcall_norets(H_VIOCTL, adapter->vdev->unit_address,
3080                                 H_ENABLE_VIO_INTERRUPT, scrq->hw_irq, 0, 0);
3081         if (rc)
3082                 dev_err(dev, "Couldn't enable scrq irq 0x%lx. rc=%ld\n",
3083                         scrq->hw_irq, rc);
3084         return rc;
3085 }
3086
3087 static int ibmvnic_complete_tx(struct ibmvnic_adapter *adapter,
3088                                struct ibmvnic_sub_crq_queue *scrq)
3089 {
3090         struct device *dev = &adapter->vdev->dev;
3091         struct ibmvnic_tx_pool *tx_pool;
3092         struct ibmvnic_tx_buff *txbuff;
3093         union sub_crq *next;
3094         int index;
3095         int i, j;
3096
3097 restart_loop:
3098         while (pending_scrq(adapter, scrq)) {
3099                 unsigned int pool = scrq->pool_index;
3100                 int num_entries = 0;
3101
3102                 /* The queue entry at the current index is peeked at above
3103                  * to determine that there is a valid descriptor awaiting
3104                  * processing. We want to be sure that the current slot
3105                  * holds a valid descriptor before reading its contents.
3106                  */
3107                 dma_rmb();
3108
3109                 next = ibmvnic_next_scrq(adapter, scrq);
3110                 for (i = 0; i < next->tx_comp.num_comps; i++) {
3111                         if (next->tx_comp.rcs[i])
3112                                 dev_err(dev, "tx error %x\n",
3113                                         next->tx_comp.rcs[i]);
3114                         index = be32_to_cpu(next->tx_comp.correlators[i]);
3115                         if (index & IBMVNIC_TSO_POOL_MASK) {
3116                                 tx_pool = &adapter->tso_pool[pool];
3117                                 index &= ~IBMVNIC_TSO_POOL_MASK;
3118                         } else {
3119                                 tx_pool = &adapter->tx_pool[pool];
3120                         }
3121
3122                         txbuff = &tx_pool->tx_buff[index];
3123
3124                         for (j = 0; j < IBMVNIC_MAX_FRAGS_PER_CRQ; j++) {
3125                                 if (!txbuff->data_dma[j])
3126                                         continue;
3127
3128                                 txbuff->data_dma[j] = 0;
3129                         }
3130
3131                         if (txbuff->last_frag) {
3132                                 dev_kfree_skb_any(txbuff->skb);
3133                                 txbuff->skb = NULL;
3134                         }
3135
3136                         num_entries += txbuff->num_entries;
3137
3138                         tx_pool->free_map[tx_pool->producer_index] = index;
3139                         tx_pool->producer_index =
3140                                 (tx_pool->producer_index + 1) %
3141                                         tx_pool->num_buffers;
3142                 }
3143                 /* remove tx_comp scrq*/
3144                 next->tx_comp.first = 0;
3145
3146                 if (atomic_sub_return(num_entries, &scrq->used) <=
3147                     (adapter->req_tx_entries_per_subcrq / 2) &&
3148                     __netif_subqueue_stopped(adapter->netdev,
3149                                              scrq->pool_index)) {
3150                         netif_wake_subqueue(adapter->netdev, scrq->pool_index);
3151                         netdev_dbg(adapter->netdev, "Started queue %d\n",
3152                                    scrq->pool_index);
3153                 }
3154         }
3155
3156         enable_scrq_irq(adapter, scrq);
3157
3158         if (pending_scrq(adapter, scrq)) {
3159                 disable_scrq_irq(adapter, scrq);
3160                 goto restart_loop;
3161         }
3162
3163         return 0;
3164 }
3165
3166 static irqreturn_t ibmvnic_interrupt_tx(int irq, void *instance)
3167 {
3168         struct ibmvnic_sub_crq_queue *scrq = instance;
3169         struct ibmvnic_adapter *adapter = scrq->adapter;
3170
3171         disable_scrq_irq(adapter, scrq);
3172         ibmvnic_complete_tx(adapter, scrq);
3173
3174         return IRQ_HANDLED;
3175 }
3176
3177 static irqreturn_t ibmvnic_interrupt_rx(int irq, void *instance)
3178 {
3179         struct ibmvnic_sub_crq_queue *scrq = instance;
3180         struct ibmvnic_adapter *adapter = scrq->adapter;
3181
3182         /* When booting a kdump kernel we can hit pending interrupts
3183          * prior to completing driver initialization.
3184          */
3185         if (unlikely(adapter->state != VNIC_OPEN))
3186                 return IRQ_NONE;
3187
3188         adapter->rx_stats_buffers[scrq->scrq_num].interrupts++;
3189
3190         if (napi_schedule_prep(&adapter->napi[scrq->scrq_num])) {
3191                 disable_scrq_irq(adapter, scrq);
3192                 __napi_schedule(&adapter->napi[scrq->scrq_num]);
3193         }
3194
3195         return IRQ_HANDLED;
3196 }
3197
3198 static int init_sub_crq_irqs(struct ibmvnic_adapter *adapter)
3199 {
3200         struct device *dev = &adapter->vdev->dev;
3201         struct ibmvnic_sub_crq_queue *scrq;
3202         int i = 0, j = 0;
3203         int rc = 0;
3204
3205         for (i = 0; i < adapter->req_tx_queues; i++) {
3206                 netdev_dbg(adapter->netdev, "Initializing tx_scrq[%d] irq\n",
3207                            i);
3208                 scrq = adapter->tx_scrq[i];
3209                 scrq->irq = irq_create_mapping(NULL, scrq->hw_irq);
3210
3211                 if (!scrq->irq) {
3212                         rc = -EINVAL;
3213                         dev_err(dev, "Error mapping irq\n");
3214                         goto req_tx_irq_failed;
3215                 }
3216
3217                 snprintf(scrq->name, sizeof(scrq->name), "ibmvnic-%x-tx%d",
3218                          adapter->vdev->unit_address, i);
3219                 rc = request_irq(scrq->irq, ibmvnic_interrupt_tx,
3220                                  0, scrq->name, scrq);
3221
3222                 if (rc) {
3223                         dev_err(dev, "Couldn't register tx irq 0x%x. rc=%d\n",
3224                                 scrq->irq, rc);
3225                         irq_dispose_mapping(scrq->irq);
3226                         goto req_tx_irq_failed;
3227                 }
3228         }
3229
3230         for (i = 0; i < adapter->req_rx_queues; i++) {
3231                 netdev_dbg(adapter->netdev, "Initializing rx_scrq[%d] irq\n",
3232                            i);
3233                 scrq = adapter->rx_scrq[i];
3234                 scrq->irq = irq_create_mapping(NULL, scrq->hw_irq);
3235                 if (!scrq->irq) {
3236                         rc = -EINVAL;
3237                         dev_err(dev, "Error mapping irq\n");
3238                         goto req_rx_irq_failed;
3239                 }
3240                 snprintf(scrq->name, sizeof(scrq->name), "ibmvnic-%x-rx%d",
3241                          adapter->vdev->unit_address, i);
3242                 rc = request_irq(scrq->irq, ibmvnic_interrupt_rx,
3243                                  0, scrq->name, scrq);
3244                 if (rc) {
3245                         dev_err(dev, "Couldn't register rx irq 0x%x. rc=%d\n",
3246                                 scrq->irq, rc);
3247                         irq_dispose_mapping(scrq->irq);
3248                         goto req_rx_irq_failed;
3249                 }
3250         }
3251         return rc;
3252
3253 req_rx_irq_failed:
3254         for (j = 0; j < i; j++) {
3255                 free_irq(adapter->rx_scrq[j]->irq, adapter->rx_scrq[j]);
3256                 irq_dispose_mapping(adapter->rx_scrq[j]->irq);
3257         }
3258         i = adapter->req_tx_queues;
3259 req_tx_irq_failed:
3260         for (j = 0; j < i; j++) {
3261                 free_irq(adapter->tx_scrq[j]->irq, adapter->tx_scrq[j]);
3262                 irq_dispose_mapping(adapter->tx_scrq[j]->irq);
3263         }
3264         release_sub_crqs(adapter, 1);
3265         return rc;
3266 }
3267
3268 static int init_sub_crqs(struct ibmvnic_adapter *adapter)
3269 {
3270         struct device *dev = &adapter->vdev->dev;
3271         struct ibmvnic_sub_crq_queue **allqueues;
3272         int registered_queues = 0;
3273         int total_queues;
3274         int more = 0;
3275         int i;
3276
3277         total_queues = adapter->req_tx_queues + adapter->req_rx_queues;
3278
3279         allqueues = kcalloc(total_queues, sizeof(*allqueues), GFP_KERNEL);
3280         if (!allqueues)
3281                 return -1;
3282
3283         for (i = 0; i < total_queues; i++) {
3284                 allqueues[i] = init_sub_crq_queue(adapter);
3285                 if (!allqueues[i]) {
3286                         dev_warn(dev, "Couldn't allocate all sub-crqs\n");
3287                         break;
3288                 }
3289                 registered_queues++;
3290         }
3291
3292         /* Make sure we were able to register the minimum number of queues */
3293         if (registered_queues <
3294             adapter->min_tx_queues + adapter->min_rx_queues) {
3295                 dev_err(dev, "Fatal: Couldn't init  min number of sub-crqs\n");
3296                 goto tx_failed;
3297         }
3298
3299         /* Distribute the failed allocated queues*/
3300         for (i = 0; i < total_queues - registered_queues + more ; i++) {
3301                 netdev_dbg(adapter->netdev, "Reducing number of queues\n");
3302                 switch (i % 3) {
3303                 case 0:
3304                         if (adapter->req_rx_queues > adapter->min_rx_queues)
3305                                 adapter->req_rx_queues--;
3306                         else
3307                                 more++;
3308                         break;
3309                 case 1:
3310                         if (adapter->req_tx_queues > adapter->min_tx_queues)
3311                                 adapter->req_tx_queues--;
3312                         else
3313                                 more++;
3314                         break;
3315                 }
3316         }
3317
3318         adapter->tx_scrq = kcalloc(adapter->req_tx_queues,
3319                                    sizeof(*adapter->tx_scrq), GFP_KERNEL);
3320         if (!adapter->tx_scrq)
3321                 goto tx_failed;
3322
3323         for (i = 0; i < adapter->req_tx_queues; i++) {
3324                 adapter->tx_scrq[i] = allqueues[i];
3325                 adapter->tx_scrq[i]->pool_index = i;
3326                 adapter->num_active_tx_scrqs++;
3327         }
3328
3329         adapter->rx_scrq = kcalloc(adapter->req_rx_queues,
3330                                    sizeof(*adapter->rx_scrq), GFP_KERNEL);
3331         if (!adapter->rx_scrq)
3332                 goto rx_failed;
3333
3334         for (i = 0; i < adapter->req_rx_queues; i++) {
3335                 adapter->rx_scrq[i] = allqueues[i + adapter->req_tx_queues];
3336                 adapter->rx_scrq[i]->scrq_num = i;
3337                 adapter->num_active_rx_scrqs++;
3338         }
3339
3340         kfree(allqueues);
3341         return 0;
3342
3343 rx_failed:
3344         kfree(adapter->tx_scrq);
3345         adapter->tx_scrq = NULL;
3346 tx_failed:
3347         for (i = 0; i < registered_queues; i++)
3348                 release_sub_crq_queue(adapter, allqueues[i], 1);
3349         kfree(allqueues);
3350         return -1;
3351 }
3352
3353 static void send_request_cap(struct ibmvnic_adapter *adapter, int retry)
3354 {
3355         struct device *dev = &adapter->vdev->dev;
3356         union ibmvnic_crq crq;
3357         int max_entries;
3358         int cap_reqs;
3359
3360         /* We send out 6 or 7 REQUEST_CAPABILITY CRQs below (depending on
3361          * the PROMISC flag). Initialize this count upfront. When the tasklet
3362          * receives a response to all of these, it will send the next protocol
3363          * message (QUERY_IP_OFFLOAD).
3364          */
3365         if (!(adapter->netdev->flags & IFF_PROMISC) ||
3366             adapter->promisc_supported)
3367                 cap_reqs = 7;
3368         else
3369                 cap_reqs = 6;
3370
3371         if (!retry) {
3372                 /* Sub-CRQ entries are 32 byte long */
3373                 int entries_page = 4 * PAGE_SIZE / (sizeof(u64) * 4);
3374
3375                 atomic_set(&adapter->running_cap_crqs, cap_reqs);
3376
3377                 if (adapter->min_tx_entries_per_subcrq > entries_page ||
3378                     adapter->min_rx_add_entries_per_subcrq > entries_page) {
3379                         dev_err(dev, "Fatal, invalid entries per sub-crq\n");
3380                         return;
3381                 }
3382
3383                 if (adapter->desired.mtu)
3384                         adapter->req_mtu = adapter->desired.mtu;
3385                 else
3386                         adapter->req_mtu = adapter->netdev->mtu + ETH_HLEN;
3387
3388                 if (!adapter->desired.tx_entries)
3389                         adapter->desired.tx_entries =
3390                                         adapter->max_tx_entries_per_subcrq;
3391                 if (!adapter->desired.rx_entries)
3392                         adapter->desired.rx_entries =
3393                                         adapter->max_rx_add_entries_per_subcrq;
3394
3395                 max_entries = IBMVNIC_MAX_LTB_SIZE /
3396                               (adapter->req_mtu + IBMVNIC_BUFFER_HLEN);
3397
3398                 if ((adapter->req_mtu + IBMVNIC_BUFFER_HLEN) *
3399                         adapter->desired.tx_entries > IBMVNIC_MAX_LTB_SIZE) {
3400                         adapter->desired.tx_entries = max_entries;
3401                 }
3402
3403                 if ((adapter->req_mtu + IBMVNIC_BUFFER_HLEN) *
3404                         adapter->desired.rx_entries > IBMVNIC_MAX_LTB_SIZE) {
3405                         adapter->desired.rx_entries = max_entries;
3406                 }
3407
3408                 if (adapter->desired.tx_entries)
3409                         adapter->req_tx_entries_per_subcrq =
3410                                         adapter->desired.tx_entries;
3411                 else
3412                         adapter->req_tx_entries_per_subcrq =
3413                                         adapter->max_tx_entries_per_subcrq;
3414
3415                 if (adapter->desired.rx_entries)
3416                         adapter->req_rx_add_entries_per_subcrq =
3417                                         adapter->desired.rx_entries;
3418                 else
3419                         adapter->req_rx_add_entries_per_subcrq =
3420                                         adapter->max_rx_add_entries_per_subcrq;
3421
3422                 if (adapter->desired.tx_queues)
3423                         adapter->req_tx_queues =
3424                                         adapter->desired.tx_queues;
3425                 else
3426                         adapter->req_tx_queues =
3427                                         adapter->opt_tx_comp_sub_queues;
3428
3429                 if (adapter->desired.rx_queues)
3430                         adapter->req_rx_queues =
3431                                         adapter->desired.rx_queues;
3432                 else
3433                         adapter->req_rx_queues =
3434                                         adapter->opt_rx_comp_queues;
3435
3436                 adapter->req_rx_add_queues = adapter->max_rx_add_queues;
3437         } else {
3438                 atomic_add(cap_reqs, &adapter->running_cap_crqs);
3439         }
3440         memset(&crq, 0, sizeof(crq));
3441         crq.request_capability.first = IBMVNIC_CRQ_CMD;
3442         crq.request_capability.cmd = REQUEST_CAPABILITY;
3443
3444         crq.request_capability.capability = cpu_to_be16(REQ_TX_QUEUES);
3445         crq.request_capability.number = cpu_to_be64(adapter->req_tx_queues);
3446         cap_reqs--;
3447         ibmvnic_send_crq(adapter, &crq);
3448
3449         crq.request_capability.capability = cpu_to_be16(REQ_RX_QUEUES);
3450         crq.request_capability.number = cpu_to_be64(adapter->req_rx_queues);
3451         cap_reqs--;
3452         ibmvnic_send_crq(adapter, &crq);
3453
3454         crq.request_capability.capability = cpu_to_be16(REQ_RX_ADD_QUEUES);
3455         crq.request_capability.number = cpu_to_be64(adapter->req_rx_add_queues);
3456         cap_reqs--;
3457         ibmvnic_send_crq(adapter, &crq);
3458
3459         crq.request_capability.capability =
3460             cpu_to_be16(REQ_TX_ENTRIES_PER_SUBCRQ);
3461         crq.request_capability.number =
3462             cpu_to_be64(adapter->req_tx_entries_per_subcrq);
3463         cap_reqs--;
3464         ibmvnic_send_crq(adapter, &crq);
3465
3466         crq.request_capability.capability =
3467             cpu_to_be16(REQ_RX_ADD_ENTRIES_PER_SUBCRQ);
3468         crq.request_capability.number =
3469             cpu_to_be64(adapter->req_rx_add_entries_per_subcrq);
3470         cap_reqs--;
3471         ibmvnic_send_crq(adapter, &crq);
3472
3473         crq.request_capability.capability = cpu_to_be16(REQ_MTU);
3474         crq.request_capability.number = cpu_to_be64(adapter->req_mtu);
3475         cap_reqs--;
3476         ibmvnic_send_crq(adapter, &crq);
3477
3478         if (adapter->netdev->flags & IFF_PROMISC) {
3479                 if (adapter->promisc_supported) {
3480                         crq.request_capability.capability =
3481                             cpu_to_be16(PROMISC_REQUESTED);
3482                         crq.request_capability.number = cpu_to_be64(1);
3483                         cap_reqs--;
3484                         ibmvnic_send_crq(adapter, &crq);
3485                 }
3486         } else {
3487                 crq.request_capability.capability =
3488                     cpu_to_be16(PROMISC_REQUESTED);
3489                 crq.request_capability.number = cpu_to_be64(0);
3490                 cap_reqs--;
3491                 ibmvnic_send_crq(adapter, &crq);
3492         }
3493
3494         /* Keep at end to catch any discrepancy between expected and actual
3495          * CRQs sent.
3496          */
3497         WARN_ON(cap_reqs != 0);
3498 }
3499
3500 static int pending_scrq(struct ibmvnic_adapter *adapter,
3501                         struct ibmvnic_sub_crq_queue *scrq)
3502 {
3503         union sub_crq *entry = &scrq->msgs[scrq->cur];
3504
3505         if (entry->generic.first & IBMVNIC_CRQ_CMD_RSP)
3506                 return 1;
3507         else
3508                 return 0;
3509 }
3510
3511 static union sub_crq *ibmvnic_next_scrq(struct ibmvnic_adapter *adapter,
3512                                         struct ibmvnic_sub_crq_queue *scrq)
3513 {
3514         union sub_crq *entry;
3515         unsigned long flags;
3516
3517         spin_lock_irqsave(&scrq->lock, flags);
3518         entry = &scrq->msgs[scrq->cur];
3519         if (entry->generic.first & IBMVNIC_CRQ_CMD_RSP) {
3520                 if (++scrq->cur == scrq->size)
3521                         scrq->cur = 0;
3522         } else {
3523                 entry = NULL;
3524         }
3525         spin_unlock_irqrestore(&scrq->lock, flags);
3526
3527         /* Ensure that the entire buffer descriptor has been
3528          * loaded before reading its contents
3529          */
3530         dma_rmb();
3531
3532         return entry;
3533 }
3534
3535 static union ibmvnic_crq *ibmvnic_next_crq(struct ibmvnic_adapter *adapter)
3536 {
3537         struct ibmvnic_crq_queue *queue = &adapter->crq;
3538         union ibmvnic_crq *crq;
3539
3540         crq = &queue->msgs[queue->cur];
3541         if (crq->generic.first & IBMVNIC_CRQ_CMD_RSP) {
3542                 if (++queue->cur == queue->size)
3543                         queue->cur = 0;
3544         } else {
3545                 crq = NULL;
3546         }
3547
3548         return crq;
3549 }
3550
3551 static void print_subcrq_error(struct device *dev, int rc, const char *func)
3552 {
3553         switch (rc) {
3554         case H_PARAMETER:
3555                 dev_warn_ratelimited(dev,
3556                                      "%s failed: Send request is malformed or adapter failover pending. (rc=%d)\n",
3557                                      func, rc);
3558                 break;
3559         case H_CLOSED:
3560                 dev_warn_ratelimited(dev,
3561                                      "%s failed: Backing queue closed. Adapter is down or failover pending. (rc=%d)\n",
3562                                      func, rc);
3563                 break;
3564         default:
3565                 dev_err_ratelimited(dev, "%s failed: (rc=%d)\n", func, rc);
3566                 break;
3567         }
3568 }
3569
3570 static int send_subcrq(struct ibmvnic_adapter *adapter, u64 remote_handle,
3571                        union sub_crq *sub_crq)
3572 {
3573         unsigned int ua = adapter->vdev->unit_address;
3574         struct device *dev = &adapter->vdev->dev;
3575         u64 *u64_crq = (u64 *)sub_crq;
3576         int rc;
3577
3578         netdev_dbg(adapter->netdev,
3579                    "Sending sCRQ %016lx: %016lx %016lx %016lx %016lx\n",
3580                    (unsigned long int)cpu_to_be64(remote_handle),
3581                    (unsigned long int)cpu_to_be64(u64_crq[0]),
3582                    (unsigned long int)cpu_to_be64(u64_crq[1]),
3583                    (unsigned long int)cpu_to_be64(u64_crq[2]),
3584                    (unsigned long int)cpu_to_be64(u64_crq[3]));
3585
3586         /* Make sure the hypervisor sees the complete request */
3587         mb();
3588
3589         rc = plpar_hcall_norets(H_SEND_SUB_CRQ, ua,
3590                                 cpu_to_be64(remote_handle),
3591                                 cpu_to_be64(u64_crq[0]),
3592                                 cpu_to_be64(u64_crq[1]),
3593                                 cpu_to_be64(u64_crq[2]),
3594                                 cpu_to_be64(u64_crq[3]));
3595
3596         if (rc)
3597                 print_subcrq_error(dev, rc, __func__);
3598
3599         return rc;
3600 }
3601
3602 static int send_subcrq_indirect(struct ibmvnic_adapter *adapter,
3603                                 u64 remote_handle, u64 ioba, u64 num_entries)
3604 {
3605         unsigned int ua = adapter->vdev->unit_address;
3606         struct device *dev = &adapter->vdev->dev;
3607         int rc;
3608
3609         /* Make sure the hypervisor sees the complete request */
3610         mb();
3611         rc = plpar_hcall_norets(H_SEND_SUB_CRQ_INDIRECT, ua,
3612                                 cpu_to_be64(remote_handle),
3613                                 ioba, num_entries);
3614
3615         if (rc)
3616                 print_subcrq_error(dev, rc, __func__);
3617
3618         return rc;
3619 }
3620
3621 static int ibmvnic_send_crq(struct ibmvnic_adapter *adapter,
3622                             union ibmvnic_crq *crq)
3623 {
3624         unsigned int ua = adapter->vdev->unit_address;
3625         struct device *dev = &adapter->vdev->dev;
3626         u64 *u64_crq = (u64 *)crq;
3627         int rc;
3628
3629         netdev_dbg(adapter->netdev, "Sending CRQ: %016lx %016lx\n",
3630                    (unsigned long int)cpu_to_be64(u64_crq[0]),
3631                    (unsigned long int)cpu_to_be64(u64_crq[1]));
3632
3633         if (!adapter->crq.active &&
3634             crq->generic.first != IBMVNIC_CRQ_INIT_CMD) {
3635                 dev_warn(dev, "Invalid request detected while CRQ is inactive, possible device state change during reset\n");
3636                 return -EINVAL;
3637         }
3638
3639         /* Make sure the hypervisor sees the complete request */
3640         mb();
3641
3642         rc = plpar_hcall_norets(H_SEND_CRQ, ua,
3643                                 cpu_to_be64(u64_crq[0]),
3644                                 cpu_to_be64(u64_crq[1]));
3645
3646         if (rc) {
3647                 if (rc == H_CLOSED) {
3648                         dev_warn(dev, "CRQ Queue closed\n");
3649                         /* do not reset, report the fail, wait for passive init from server */
3650                 }
3651
3652                 dev_warn(dev, "Send error (rc=%d)\n", rc);
3653         }
3654
3655         return rc;
3656 }
3657
3658 static int ibmvnic_send_crq_init(struct ibmvnic_adapter *adapter)
3659 {
3660         struct device *dev = &adapter->vdev->dev;
3661         union ibmvnic_crq crq;
3662         int retries = 100;
3663         int rc;
3664
3665         memset(&crq, 0, sizeof(crq));
3666         crq.generic.first = IBMVNIC_CRQ_INIT_CMD;
3667         crq.generic.cmd = IBMVNIC_CRQ_INIT;
3668         netdev_dbg(adapter->netdev, "Sending CRQ init\n");
3669
3670         do {
3671                 rc = ibmvnic_send_crq(adapter, &crq);
3672                 if (rc != H_CLOSED)
3673                         break;
3674                 retries--;
3675                 msleep(50);
3676
3677         } while (retries > 0);
3678
3679         if (rc) {
3680                 dev_err(dev, "Failed to send init request, rc = %d\n", rc);
3681                 return rc;
3682         }
3683
3684         return 0;
3685 }
3686
3687 static int send_version_xchg(struct ibmvnic_adapter *adapter)
3688 {
3689         union ibmvnic_crq crq;
3690
3691         memset(&crq, 0, sizeof(crq));
3692         crq.version_exchange.first = IBMVNIC_CRQ_CMD;
3693         crq.version_exchange.cmd = VERSION_EXCHANGE;
3694         crq.version_exchange.version = cpu_to_be16(ibmvnic_version);
3695
3696         return ibmvnic_send_crq(adapter, &crq);
3697 }
3698
3699 struct vnic_login_client_data {
3700         u8      type;
3701         __be16  len;
3702         char    name[];
3703 } __packed;
3704
3705 static int vnic_client_data_len(struct ibmvnic_adapter *adapter)
3706 {
3707         int len;
3708
3709         /* Calculate the amount of buffer space needed for the
3710          * vnic client data in the login buffer. There are four entries,
3711          * OS name, LPAR name, device name, and a null last entry.
3712          */
3713         len = 4 * sizeof(struct vnic_login_client_data);
3714         len += 6; /* "Linux" plus NULL */
3715         len += strlen(utsname()->nodename) + 1;
3716         len += strlen(adapter->netdev->name) + 1;
3717
3718         return len;
3719 }
3720
3721 static void vnic_add_client_data(struct ibmvnic_adapter *adapter,
3722                                  struct vnic_login_client_data *vlcd)
3723 {
3724         const char *os_name = "Linux";
3725         int len;
3726
3727         /* Type 1 - LPAR OS */
3728         vlcd->type = 1;
3729         len = strlen(os_name) + 1;
3730         vlcd->len = cpu_to_be16(len);
3731         strncpy(vlcd->name, os_name, len);
3732         vlcd = (struct vnic_login_client_data *)(vlcd->name + len);
3733
3734         /* Type 2 - LPAR name */
3735         vlcd->type = 2;
3736         len = strlen(utsname()->nodename) + 1;
3737         vlcd->len = cpu_to_be16(len);
3738         strncpy(vlcd->name, utsname()->nodename, len);
3739         vlcd = (struct vnic_login_client_data *)(vlcd->name + len);
3740
3741         /* Type 3 - device name */
3742         vlcd->type = 3;
3743         len = strlen(adapter->netdev->name) + 1;
3744         vlcd->len = cpu_to_be16(len);
3745         strncpy(vlcd->name, adapter->netdev->name, len);
3746 }
3747
3748 static int send_login(struct ibmvnic_adapter *adapter)
3749 {
3750         struct ibmvnic_login_rsp_buffer *login_rsp_buffer;
3751         struct ibmvnic_login_buffer *login_buffer;
3752         struct device *dev = &adapter->vdev->dev;
3753         struct vnic_login_client_data *vlcd;
3754         dma_addr_t rsp_buffer_token;
3755         dma_addr_t buffer_token;
3756         size_t rsp_buffer_size;
3757         union ibmvnic_crq crq;
3758         int client_data_len;
3759         size_t buffer_size;
3760         __be64 *tx_list_p;
3761         __be64 *rx_list_p;
3762         int rc;
3763         int i;
3764
3765         if (!adapter->tx_scrq || !adapter->rx_scrq) {
3766                 netdev_err(adapter->netdev,
3767                            "RX or TX queues are not allocated, device login failed\n");
3768                 return -1;
3769         }
3770
3771         release_login_buffer(adapter);
3772         release_login_rsp_buffer(adapter);
3773
3774         client_data_len = vnic_client_data_len(adapter);
3775
3776         buffer_size =
3777             sizeof(struct ibmvnic_login_buffer) +
3778             sizeof(u64) * (adapter->req_tx_queues + adapter->req_rx_queues) +
3779             client_data_len;
3780
3781         login_buffer = kzalloc(buffer_size, GFP_ATOMIC);
3782         if (!login_buffer)
3783                 goto buf_alloc_failed;
3784
3785         buffer_token = dma_map_single(dev, login_buffer, buffer_size,
3786                                       DMA_TO_DEVICE);
3787         if (dma_mapping_error(dev, buffer_token)) {
3788                 dev_err(dev, "Couldn't map login buffer\n");
3789                 goto buf_map_failed;
3790         }
3791
3792         rsp_buffer_size = sizeof(struct ibmvnic_login_rsp_buffer) +
3793                           sizeof(u64) * adapter->req_tx_queues +
3794                           sizeof(u64) * adapter->req_rx_queues +
3795                           sizeof(u64) * adapter->req_rx_queues +
3796                           sizeof(u8) * IBMVNIC_TX_DESC_VERSIONS;
3797
3798         login_rsp_buffer = kmalloc(rsp_buffer_size, GFP_ATOMIC);
3799         if (!login_rsp_buffer)
3800                 goto buf_rsp_alloc_failed;
3801
3802         rsp_buffer_token = dma_map_single(dev, login_rsp_buffer,
3803                                           rsp_buffer_size, DMA_FROM_DEVICE);
3804         if (dma_mapping_error(dev, rsp_buffer_token)) {
3805                 dev_err(dev, "Couldn't map login rsp buffer\n");
3806                 goto buf_rsp_map_failed;
3807         }
3808
3809         adapter->login_buf = login_buffer;
3810         adapter->login_buf_token = buffer_token;
3811         adapter->login_buf_sz = buffer_size;
3812         adapter->login_rsp_buf = login_rsp_buffer;
3813         adapter->login_rsp_buf_token = rsp_buffer_token;
3814         adapter->login_rsp_buf_sz = rsp_buffer_size;
3815
3816         login_buffer->len = cpu_to_be32(buffer_size);
3817         login_buffer->version = cpu_to_be32(INITIAL_VERSION_LB);
3818         login_buffer->num_txcomp_subcrqs = cpu_to_be32(adapter->req_tx_queues);
3819         login_buffer->off_txcomp_subcrqs =
3820             cpu_to_be32(sizeof(struct ibmvnic_login_buffer));
3821         login_buffer->num_rxcomp_subcrqs = cpu_to_be32(adapter->req_rx_queues);
3822         login_buffer->off_rxcomp_subcrqs =
3823             cpu_to_be32(sizeof(struct ibmvnic_login_buffer) +
3824                         sizeof(u64) * adapter->req_tx_queues);
3825         login_buffer->login_rsp_ioba = cpu_to_be32(rsp_buffer_token);
3826         login_buffer->login_rsp_len = cpu_to_be32(rsp_buffer_size);
3827
3828         tx_list_p = (__be64 *)((char *)login_buffer +
3829                                       sizeof(struct ibmvnic_login_buffer));
3830         rx_list_p = (__be64 *)((char *)login_buffer +
3831                                       sizeof(struct ibmvnic_login_buffer) +
3832                                       sizeof(u64) * adapter->req_tx_queues);
3833
3834         for (i = 0; i < adapter->req_tx_queues; i++) {
3835                 if (adapter->tx_scrq[i]) {
3836                         tx_list_p[i] = cpu_to_be64(adapter->tx_scrq[i]->
3837                                                    crq_num);
3838                 }
3839         }
3840
3841         for (i = 0; i < adapter->req_rx_queues; i++) {
3842                 if (adapter->rx_scrq[i]) {
3843                         rx_list_p[i] = cpu_to_be64(adapter->rx_scrq[i]->
3844                                                    crq_num);
3845                 }
3846         }
3847
3848         /* Insert vNIC login client data */
3849         vlcd = (struct vnic_login_client_data *)
3850                 ((char *)rx_list_p + (sizeof(u64) * adapter->req_rx_queues));
3851         login_buffer->client_data_offset =
3852                         cpu_to_be32((char *)vlcd - (char *)login_buffer);
3853         login_buffer->client_data_len = cpu_to_be32(client_data_len);
3854
3855         vnic_add_client_data(adapter, vlcd);
3856
3857         netdev_dbg(adapter->netdev, "Login Buffer:\n");
3858         for (i = 0; i < (adapter->login_buf_sz - 1) / 8 + 1; i++) {
3859                 netdev_dbg(adapter->netdev, "%016lx\n",
3860                            ((unsigned long int *)(adapter->login_buf))[i]);
3861         }
3862
3863         memset(&crq, 0, sizeof(crq));
3864         crq.login.first = IBMVNIC_CRQ_CMD;
3865         crq.login.cmd = LOGIN;
3866         crq.login.ioba = cpu_to_be32(buffer_token);
3867         crq.login.len = cpu_to_be32(buffer_size);
3868
3869         adapter->login_pending = true;
3870         rc = ibmvnic_send_crq(adapter, &crq);
3871         if (rc) {
3872                 adapter->login_pending = false;
3873                 netdev_err(adapter->netdev, "Failed to send login, rc=%d\n", rc);
3874                 goto buf_send_failed;
3875         }
3876
3877         return 0;
3878
3879 buf_send_failed:
3880         dma_unmap_single(dev, rsp_buffer_token, rsp_buffer_size,
3881                          DMA_FROM_DEVICE);
3882 buf_rsp_map_failed:
3883         kfree(login_rsp_buffer);
3884         adapter->login_rsp_buf = NULL;
3885 buf_rsp_alloc_failed:
3886         dma_unmap_single(dev, buffer_token, buffer_size, DMA_TO_DEVICE);
3887 buf_map_failed:
3888         kfree(login_buffer);
3889         adapter->login_buf = NULL;
3890 buf_alloc_failed:
3891         return -1;
3892 }
3893
3894 static int send_request_map(struct ibmvnic_adapter *adapter, dma_addr_t addr,
3895                             u32 len, u8 map_id)
3896 {
3897         union ibmvnic_crq crq;
3898
3899         memset(&crq, 0, sizeof(crq));
3900         crq.request_map.first = IBMVNIC_CRQ_CMD;
3901         crq.request_map.cmd = REQUEST_MAP;
3902         crq.request_map.map_id = map_id;
3903         crq.request_map.ioba = cpu_to_be32(addr);
3904         crq.request_map.len = cpu_to_be32(len);
3905         return ibmvnic_send_crq(adapter, &crq);
3906 }
3907
3908 static int send_request_unmap(struct ibmvnic_adapter *adapter, u8 map_id)
3909 {
3910         union ibmvnic_crq crq;
3911
3912         memset(&crq, 0, sizeof(crq));
3913         crq.request_unmap.first = IBMVNIC_CRQ_CMD;
3914         crq.request_unmap.cmd = REQUEST_UNMAP;
3915         crq.request_unmap.map_id = map_id;
3916         return ibmvnic_send_crq(adapter, &crq);
3917 }
3918
3919 static void send_query_map(struct ibmvnic_adapter *adapter)
3920 {
3921         union ibmvnic_crq crq;
3922
3923         memset(&crq, 0, sizeof(crq));
3924         crq.query_map.first = IBMVNIC_CRQ_CMD;
3925         crq.query_map.cmd = QUERY_MAP;
3926         ibmvnic_send_crq(adapter, &crq);
3927 }
3928
3929 /* Send a series of CRQs requesting various capabilities of the VNIC server */
3930 static void send_query_cap(struct ibmvnic_adapter *adapter)
3931 {
3932         union ibmvnic_crq crq;
3933         int cap_reqs;
3934
3935         /* We send out 25 QUERY_CAPABILITY CRQs below.  Initialize this count
3936          * upfront. When the tasklet receives a response to all of these, it
3937          * can send out the next protocol messaage (REQUEST_CAPABILITY).
3938          */
3939         cap_reqs = 25;
3940
3941         atomic_set(&adapter->running_cap_crqs, cap_reqs);
3942
3943         memset(&crq, 0, sizeof(crq));
3944         crq.query_capability.first = IBMVNIC_CRQ_CMD;
3945         crq.query_capability.cmd = QUERY_CAPABILITY;
3946
3947         crq.query_capability.capability = cpu_to_be16(MIN_TX_QUEUES);
3948         ibmvnic_send_crq(adapter, &crq);
3949         cap_reqs--;
3950
3951         crq.query_capability.capability = cpu_to_be16(MIN_RX_QUEUES);
3952         ibmvnic_send_crq(adapter, &crq);
3953         cap_reqs--;
3954
3955         crq.query_capability.capability = cpu_to_be16(MIN_RX_ADD_QUEUES);
3956         ibmvnic_send_crq(adapter, &crq);
3957         cap_reqs--;
3958
3959         crq.query_capability.capability = cpu_to_be16(MAX_TX_QUEUES);
3960         ibmvnic_send_crq(adapter, &crq);
3961         cap_reqs--;
3962
3963         crq.query_capability.capability = cpu_to_be16(MAX_RX_QUEUES);
3964         ibmvnic_send_crq(adapter, &crq);
3965         cap_reqs--;
3966
3967         crq.query_capability.capability = cpu_to_be16(MAX_RX_ADD_QUEUES);
3968         ibmvnic_send_crq(adapter, &crq);
3969         cap_reqs--;
3970
3971         crq.query_capability.capability =
3972             cpu_to_be16(MIN_TX_ENTRIES_PER_SUBCRQ);
3973         ibmvnic_send_crq(adapter, &crq);
3974         cap_reqs--;
3975
3976         crq.query_capability.capability =
3977             cpu_to_be16(MIN_RX_ADD_ENTRIES_PER_SUBCRQ);
3978         ibmvnic_send_crq(adapter, &crq);
3979         cap_reqs--;
3980
3981         crq.query_capability.capability =
3982             cpu_to_be16(MAX_TX_ENTRIES_PER_SUBCRQ);
3983         ibmvnic_send_crq(adapter, &crq);
3984         cap_reqs--;
3985
3986         crq.query_capability.capability =
3987             cpu_to_be16(MAX_RX_ADD_ENTRIES_PER_SUBCRQ);
3988         ibmvnic_send_crq(adapter, &crq);
3989         cap_reqs--;
3990
3991         crq.query_capability.capability = cpu_to_be16(TCP_IP_OFFLOAD);
3992         ibmvnic_send_crq(adapter, &crq);
3993         cap_reqs--;
3994
3995         crq.query_capability.capability = cpu_to_be16(PROMISC_SUPPORTED);
3996         ibmvnic_send_crq(adapter, &crq);
3997         cap_reqs--;
3998
3999         crq.query_capability.capability = cpu_to_be16(MIN_MTU);
4000         ibmvnic_send_crq(adapter, &crq);
4001         cap_reqs--;
4002
4003         crq.query_capability.capability = cpu_to_be16(MAX_MTU);
4004         ibmvnic_send_crq(adapter, &crq);
4005         cap_reqs--;
4006
4007         crq.query_capability.capability = cpu_to_be16(MAX_MULTICAST_FILTERS);
4008         ibmvnic_send_crq(adapter, &crq);
4009         cap_reqs--;
4010
4011         crq.query_capability.capability = cpu_to_be16(VLAN_HEADER_INSERTION);
4012         ibmvnic_send_crq(adapter, &crq);
4013         cap_reqs--;
4014
4015         crq.query_capability.capability = cpu_to_be16(RX_VLAN_HEADER_INSERTION);
4016         ibmvnic_send_crq(adapter, &crq);
4017         cap_reqs--;
4018
4019         crq.query_capability.capability = cpu_to_be16(MAX_TX_SG_ENTRIES);
4020         ibmvnic_send_crq(adapter, &crq);
4021         cap_reqs--;
4022
4023         crq.query_capability.capability = cpu_to_be16(RX_SG_SUPPORTED);
4024         ibmvnic_send_crq(adapter, &crq);
4025         cap_reqs--;
4026
4027         crq.query_capability.capability = cpu_to_be16(OPT_TX_COMP_SUB_QUEUES);
4028         ibmvnic_send_crq(adapter, &crq);
4029         cap_reqs--;
4030
4031         crq.query_capability.capability = cpu_to_be16(OPT_RX_COMP_QUEUES);
4032         ibmvnic_send_crq(adapter, &crq);
4033         cap_reqs--;
4034
4035         crq.query_capability.capability =
4036                         cpu_to_be16(OPT_RX_BUFADD_Q_PER_RX_COMP_Q);
4037         ibmvnic_send_crq(adapter, &crq);
4038         cap_reqs--;
4039
4040         crq.query_capability.capability =
4041                         cpu_to_be16(OPT_TX_ENTRIES_PER_SUBCRQ);
4042         ibmvnic_send_crq(adapter, &crq);
4043         cap_reqs--;
4044
4045         crq.query_capability.capability =
4046                         cpu_to_be16(OPT_RXBA_ENTRIES_PER_SUBCRQ);
4047         ibmvnic_send_crq(adapter, &crq);
4048         cap_reqs--;
4049
4050         crq.query_capability.capability = cpu_to_be16(TX_RX_DESC_REQ);
4051
4052         ibmvnic_send_crq(adapter, &crq);
4053         cap_reqs--;
4054
4055         /* Keep at end to catch any discrepancy between expected and actual
4056          * CRQs sent.
4057          */
4058         WARN_ON(cap_reqs != 0);
4059 }
4060
4061 static void send_query_ip_offload(struct ibmvnic_adapter *adapter)
4062 {
4063         int buf_sz = sizeof(struct ibmvnic_query_ip_offload_buffer);
4064         struct device *dev = &adapter->vdev->dev;
4065         union ibmvnic_crq crq;
4066
4067         adapter->ip_offload_tok =
4068                 dma_map_single(dev,
4069                                &adapter->ip_offload_buf,
4070                                buf_sz,
4071                                DMA_FROM_DEVICE);
4072
4073         if (dma_mapping_error(dev, adapter->ip_offload_tok)) {
4074                 if (!firmware_has_feature(FW_FEATURE_CMO))
4075                         dev_err(dev, "Couldn't map offload buffer\n");
4076                 return;
4077         }
4078
4079         memset(&crq, 0, sizeof(crq));
4080         crq.query_ip_offload.first = IBMVNIC_CRQ_CMD;
4081         crq.query_ip_offload.cmd = QUERY_IP_OFFLOAD;
4082         crq.query_ip_offload.len = cpu_to_be32(buf_sz);
4083         crq.query_ip_offload.ioba =
4084             cpu_to_be32(adapter->ip_offload_tok);
4085
4086         ibmvnic_send_crq(adapter, &crq);
4087 }
4088
4089 static void send_control_ip_offload(struct ibmvnic_adapter *adapter)
4090 {
4091         struct ibmvnic_control_ip_offload_buffer *ctrl_buf = &adapter->ip_offload_ctrl;
4092         struct ibmvnic_query_ip_offload_buffer *buf = &adapter->ip_offload_buf;
4093         struct device *dev = &adapter->vdev->dev;
4094         netdev_features_t old_hw_features = 0;
4095         union ibmvnic_crq crq;
4096
4097         adapter->ip_offload_ctrl_tok =
4098                 dma_map_single(dev,
4099                                ctrl_buf,
4100                                sizeof(adapter->ip_offload_ctrl),
4101                                DMA_TO_DEVICE);
4102
4103         if (dma_mapping_error(dev, adapter->ip_offload_ctrl_tok)) {
4104                 dev_err(dev, "Couldn't map ip offload control buffer\n");
4105                 return;
4106         }
4107
4108         ctrl_buf->len = cpu_to_be32(sizeof(adapter->ip_offload_ctrl));
4109         ctrl_buf->version = cpu_to_be32(INITIAL_VERSION_IOB);
4110         ctrl_buf->ipv4_chksum = buf->ipv4_chksum;
4111         ctrl_buf->ipv6_chksum = buf->ipv6_chksum;
4112         ctrl_buf->tcp_ipv4_chksum = buf->tcp_ipv4_chksum;
4113         ctrl_buf->udp_ipv4_chksum = buf->udp_ipv4_chksum;
4114         ctrl_buf->tcp_ipv6_chksum = buf->tcp_ipv6_chksum;
4115         ctrl_buf->udp_ipv6_chksum = buf->udp_ipv6_chksum;
4116         ctrl_buf->large_tx_ipv4 = buf->large_tx_ipv4;
4117         ctrl_buf->large_tx_ipv6 = buf->large_tx_ipv6;
4118
4119         /* large_rx disabled for now, additional features needed */
4120         ctrl_buf->large_rx_ipv4 = 0;
4121         ctrl_buf->large_rx_ipv6 = 0;
4122
4123         if (adapter->state != VNIC_PROBING) {
4124                 old_hw_features = adapter->netdev->hw_features;
4125                 adapter->netdev->hw_features = 0;
4126         }
4127
4128         adapter->netdev->hw_features = NETIF_F_SG | NETIF_F_GSO | NETIF_F_GRO;
4129
4130         if (buf->tcp_ipv4_chksum || buf->udp_ipv4_chksum)
4131                 adapter->netdev->hw_features |= NETIF_F_IP_CSUM;
4132
4133         if (buf->tcp_ipv6_chksum || buf->udp_ipv6_chksum)
4134                 adapter->netdev->hw_features |= NETIF_F_IPV6_CSUM;
4135
4136         if ((adapter->netdev->features &
4137             (NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM)))
4138                 adapter->netdev->hw_features |= NETIF_F_RXCSUM;
4139
4140         if (buf->large_tx_ipv4)
4141                 adapter->netdev->hw_features |= NETIF_F_TSO;
4142         if (buf->large_tx_ipv6)
4143                 adapter->netdev->hw_features |= NETIF_F_TSO6;
4144
4145         if (adapter->state == VNIC_PROBING) {
4146                 adapter->netdev->features |= adapter->netdev->hw_features;
4147         } else if (old_hw_features != adapter->netdev->hw_features) {
4148                 netdev_features_t tmp = 0;
4149
4150                 /* disable features no longer supported */
4151                 adapter->netdev->features &= adapter->netdev->hw_features;
4152                 /* turn on features now supported if previously enabled */
4153                 tmp = (old_hw_features ^ adapter->netdev->hw_features) &
4154                         adapter->netdev->hw_features;
4155                 adapter->netdev->features |=
4156                                 tmp & adapter->netdev->wanted_features;
4157         }
4158
4159         memset(&crq, 0, sizeof(crq));
4160         crq.control_ip_offload.first = IBMVNIC_CRQ_CMD;
4161         crq.control_ip_offload.cmd = CONTROL_IP_OFFLOAD;
4162         crq.control_ip_offload.len =
4163             cpu_to_be32(sizeof(adapter->ip_offload_ctrl));
4164         crq.control_ip_offload.ioba = cpu_to_be32(adapter->ip_offload_ctrl_tok);
4165         ibmvnic_send_crq(adapter, &crq);
4166 }
4167
4168 static void handle_vpd_size_rsp(union ibmvnic_crq *crq,
4169                                 struct ibmvnic_adapter *adapter)
4170 {
4171         struct device *dev = &adapter->vdev->dev;
4172
4173         if (crq->get_vpd_size_rsp.rc.code) {
4174                 dev_err(dev, "Error retrieving VPD size, rc=%x\n",
4175                         crq->get_vpd_size_rsp.rc.code);
4176                 complete(&adapter->fw_done);
4177                 return;
4178         }
4179
4180         adapter->vpd->len = be64_to_cpu(crq->get_vpd_size_rsp.len);
4181         complete(&adapter->fw_done);
4182 }
4183
4184 static void handle_vpd_rsp(union ibmvnic_crq *crq,
4185                            struct ibmvnic_adapter *adapter)
4186 {
4187         struct device *dev = &adapter->vdev->dev;
4188         unsigned char *substr = NULL;
4189         u8 fw_level_len = 0;
4190
4191         memset(adapter->fw_version, 0, 32);
4192
4193         dma_unmap_single(dev, adapter->vpd->dma_addr, adapter->vpd->len,
4194                          DMA_FROM_DEVICE);
4195
4196         if (crq->get_vpd_rsp.rc.code) {
4197                 dev_err(dev, "Error retrieving VPD from device, rc=%x\n",
4198                         crq->get_vpd_rsp.rc.code);
4199                 goto complete;
4200         }
4201
4202         /* get the position of the firmware version info
4203          * located after the ASCII 'RM' substring in the buffer
4204          */
4205         substr = strnstr(adapter->vpd->buff, "RM", adapter->vpd->len);
4206         if (!substr) {
4207                 dev_info(dev, "Warning - No FW level has been provided in the VPD buffer by the VIOS Server\n");
4208                 goto complete;
4209         }
4210
4211         /* get length of firmware level ASCII substring */
4212         if ((substr + 2) < (adapter->vpd->buff + adapter->vpd->len)) {
4213                 fw_level_len = *(substr + 2);
4214         } else {
4215                 dev_info(dev, "Length of FW substr extrapolated VDP buff\n");
4216                 goto complete;
4217         }
4218
4219         /* copy firmware version string from vpd into adapter */
4220         if ((substr + 3 + fw_level_len) <
4221             (adapter->vpd->buff + adapter->vpd->len)) {
4222                 strncpy((char *)adapter->fw_version, substr + 3, fw_level_len);
4223         } else {
4224                 dev_info(dev, "FW substr extrapolated VPD buff\n");
4225         }
4226
4227 complete:
4228         if (adapter->fw_version[0] == '\0')
4229                 strncpy((char *)adapter->fw_version, "N/A", 3 * sizeof(char));
4230         complete(&adapter->fw_done);
4231 }
4232
4233 static void handle_query_ip_offload_rsp(struct ibmvnic_adapter *adapter)
4234 {
4235         struct device *dev = &adapter->vdev->dev;
4236         struct ibmvnic_query_ip_offload_buffer *buf = &adapter->ip_offload_buf;
4237         int i;
4238
4239         dma_unmap_single(dev, adapter->ip_offload_tok,
4240                          sizeof(adapter->ip_offload_buf), DMA_FROM_DEVICE);
4241
4242         netdev_dbg(adapter->netdev, "Query IP Offload Buffer:\n");
4243         for (i = 0; i < (sizeof(adapter->ip_offload_buf) - 1) / 8 + 1; i++)
4244                 netdev_dbg(adapter->netdev, "%016lx\n",
4245                            ((unsigned long int *)(buf))[i]);
4246
4247         netdev_dbg(adapter->netdev, "ipv4_chksum = %d\n", buf->ipv4_chksum);
4248         netdev_dbg(adapter->netdev, "ipv6_chksum = %d\n", buf->ipv6_chksum);
4249         netdev_dbg(adapter->netdev, "tcp_ipv4_chksum = %d\n",
4250                    buf->tcp_ipv4_chksum);
4251         netdev_dbg(adapter->netdev, "tcp_ipv6_chksum = %d\n",
4252                    buf->tcp_ipv6_chksum);
4253         netdev_dbg(adapter->netdev, "udp_ipv4_chksum = %d\n",
4254                    buf->udp_ipv4_chksum);
4255         netdev_dbg(adapter->netdev, "udp_ipv6_chksum = %d\n",
4256                    buf->udp_ipv6_chksum);
4257         netdev_dbg(adapter->netdev, "large_tx_ipv4 = %d\n",
4258                    buf->large_tx_ipv4);
4259         netdev_dbg(adapter->netdev, "large_tx_ipv6 = %d\n",
4260                    buf->large_tx_ipv6);
4261         netdev_dbg(adapter->netdev, "large_rx_ipv4 = %d\n",
4262                    buf->large_rx_ipv4);
4263         netdev_dbg(adapter->netdev, "large_rx_ipv6 = %d\n",
4264                    buf->large_rx_ipv6);
4265         netdev_dbg(adapter->netdev, "max_ipv4_hdr_sz = %d\n",
4266                    buf->max_ipv4_header_size);
4267         netdev_dbg(adapter->netdev, "max_ipv6_hdr_sz = %d\n",
4268                    buf->max_ipv6_header_size);
4269         netdev_dbg(adapter->netdev, "max_tcp_hdr_size = %d\n",
4270                    buf->max_tcp_header_size);
4271         netdev_dbg(adapter->netdev, "max_udp_hdr_size = %d\n",
4272                    buf->max_udp_header_size);
4273         netdev_dbg(adapter->netdev, "max_large_tx_size = %d\n",
4274                    buf->max_large_tx_size);
4275         netdev_dbg(adapter->netdev, "max_large_rx_size = %d\n",
4276                    buf->max_large_rx_size);
4277         netdev_dbg(adapter->netdev, "ipv6_ext_hdr = %d\n",
4278                    buf->ipv6_extension_header);
4279         netdev_dbg(adapter->netdev, "tcp_pseudosum_req = %d\n",
4280                    buf->tcp_pseudosum_req);
4281         netdev_dbg(adapter->netdev, "num_ipv6_ext_hd = %d\n",
4282                    buf->num_ipv6_ext_headers);
4283         netdev_dbg(adapter->netdev, "off_ipv6_ext_hd = %d\n",
4284                    buf->off_ipv6_ext_headers);
4285
4286         send_control_ip_offload(adapter);
4287 }
4288
4289 static const char *ibmvnic_fw_err_cause(u16 cause)
4290 {
4291         switch (cause) {
4292         case ADAPTER_PROBLEM:
4293                 return "adapter problem";
4294         case BUS_PROBLEM:
4295                 return "bus problem";
4296         case FW_PROBLEM:
4297                 return "firmware problem";
4298         case DD_PROBLEM:
4299                 return "device driver problem";
4300         case EEH_RECOVERY:
4301                 return "EEH recovery";
4302         case FW_UPDATED:
4303                 return "firmware updated";
4304         case LOW_MEMORY:
4305                 return "low Memory";
4306         default:
4307                 return "unknown";
4308         }
4309 }
4310
4311 static void handle_error_indication(union ibmvnic_crq *crq,
4312                                     struct ibmvnic_adapter *adapter)
4313 {
4314         struct device *dev = &adapter->vdev->dev;
4315         u16 cause;
4316
4317         cause = be16_to_cpu(crq->error_indication.error_cause);
4318
4319         dev_warn_ratelimited(dev,
4320                              "Firmware reports %serror, cause: %s. Starting recovery...\n",
4321                              crq->error_indication.flags
4322                                 & IBMVNIC_FATAL_ERROR ? "FATAL " : "",
4323                              ibmvnic_fw_err_cause(cause));
4324
4325         if (crq->error_indication.flags & IBMVNIC_FATAL_ERROR)
4326                 ibmvnic_reset(adapter, VNIC_RESET_FATAL);
4327         else
4328                 ibmvnic_reset(adapter, VNIC_RESET_NON_FATAL);
4329 }
4330
4331 static int handle_change_mac_rsp(union ibmvnic_crq *crq,
4332                                  struct ibmvnic_adapter *adapter)
4333 {
4334         struct net_device *netdev = adapter->netdev;
4335         struct device *dev = &adapter->vdev->dev;
4336         long rc;
4337
4338         rc = crq->change_mac_addr_rsp.rc.code;
4339         if (rc) {
4340                 dev_err(dev, "Error %ld in CHANGE_MAC_ADDR_RSP\n", rc);
4341                 goto out;
4342         }
4343         /* crq->change_mac_addr.mac_addr is the requested one
4344          * crq->change_mac_addr_rsp.mac_addr is the returned valid one.
4345          */
4346         ether_addr_copy(netdev->dev_addr,
4347                         &crq->change_mac_addr_rsp.mac_addr[0]);
4348         ether_addr_copy(adapter->mac_addr,
4349                         &crq->change_mac_addr_rsp.mac_addr[0]);
4350 out:
4351         complete(&adapter->fw_done);
4352         return rc;
4353 }
4354
4355 static void handle_request_cap_rsp(union ibmvnic_crq *crq,
4356                                    struct ibmvnic_adapter *adapter)
4357 {
4358         struct device *dev = &adapter->vdev->dev;
4359         u64 *req_value;
4360         char *name;
4361
4362         atomic_dec(&adapter->running_cap_crqs);
4363         netdev_dbg(adapter->netdev, "Outstanding request-caps: %d\n",
4364                    atomic_read(&adapter->running_cap_crqs));
4365         switch (be16_to_cpu(crq->request_capability_rsp.capability)) {
4366         case REQ_TX_QUEUES:
4367                 req_value = &adapter->req_tx_queues;
4368                 name = "tx";
4369                 break;
4370         case REQ_RX_QUEUES:
4371                 req_value = &adapter->req_rx_queues;
4372                 name = "rx";
4373                 break;
4374         case REQ_RX_ADD_QUEUES:
4375                 req_value = &adapter->req_rx_add_queues;
4376                 name = "rx_add";
4377                 break;
4378         case REQ_TX_ENTRIES_PER_SUBCRQ:
4379                 req_value = &adapter->req_tx_entries_per_subcrq;
4380                 name = "tx_entries_per_subcrq";
4381                 break;
4382         case REQ_RX_ADD_ENTRIES_PER_SUBCRQ:
4383                 req_value = &adapter->req_rx_add_entries_per_subcrq;
4384                 name = "rx_add_entries_per_subcrq";
4385                 break;
4386         case REQ_MTU:
4387                 req_value = &adapter->req_mtu;
4388                 name = "mtu";
4389                 break;
4390         case PROMISC_REQUESTED:
4391                 req_value = &adapter->promisc;
4392                 name = "promisc";
4393                 break;
4394         default:
4395                 dev_err(dev, "Got invalid cap request rsp %d\n",
4396                         crq->request_capability.capability);
4397                 return;
4398         }
4399
4400         switch (crq->request_capability_rsp.rc.code) {
4401         case SUCCESS:
4402                 break;
4403         case PARTIALSUCCESS:
4404                 dev_info(dev, "req=%lld, rsp=%ld in %s queue, retrying.\n",
4405                          *req_value,
4406                          (long int)be64_to_cpu(crq->request_capability_rsp.
4407                                                number), name);
4408
4409                 if (be16_to_cpu(crq->request_capability_rsp.capability) ==
4410                     REQ_MTU) {
4411                         pr_err("mtu of %llu is not supported. Reverting.\n",
4412                                *req_value);
4413                         *req_value = adapter->fallback.mtu;
4414                 } else {
4415                         *req_value =
4416                                 be64_to_cpu(crq->request_capability_rsp.number);
4417                 }
4418
4419                 send_request_cap(adapter, 1);
4420                 return;
4421         default:
4422                 dev_err(dev, "Error %d in request cap rsp\n",
4423                         crq->request_capability_rsp.rc.code);
4424                 return;
4425         }
4426
4427         /* Done receiving requested capabilities, query IP offload support */
4428         if (atomic_read(&adapter->running_cap_crqs) == 0) {
4429                 adapter->wait_capability = false;
4430                 send_query_ip_offload(adapter);
4431         }
4432 }
4433
4434 static int handle_login_rsp(union ibmvnic_crq *login_rsp_crq,
4435                             struct ibmvnic_adapter *adapter)
4436 {
4437         struct device *dev = &adapter->vdev->dev;
4438         struct net_device *netdev = adapter->netdev;
4439         struct ibmvnic_login_rsp_buffer *login_rsp = adapter->login_rsp_buf;
4440         struct ibmvnic_login_buffer *login = adapter->login_buf;
4441         u64 *tx_handle_array;
4442         u64 *rx_handle_array;
4443         int num_tx_pools;
4444         int num_rx_pools;
4445         u64 *size_array;
4446         u32 rsp_len;
4447         int i;
4448
4449         /* CHECK: Test/set of login_pending does not need to be atomic
4450          * because only ibmvnic_tasklet tests/clears this.
4451          */
4452         if (!adapter->login_pending) {
4453                 netdev_warn(netdev, "Ignoring unexpected login response\n");
4454                 return 0;
4455         }
4456         adapter->login_pending = false;
4457
4458         /* If the number of queues requested can't be allocated by the
4459          * server, the login response will return with code 1. We will need
4460          * to resend the login buffer with fewer queues requested.
4461          */
4462         if (login_rsp_crq->generic.rc.code) {
4463                 adapter->init_done_rc = login_rsp_crq->generic.rc.code;
4464                 complete(&adapter->init_done);
4465                 return 0;
4466         }
4467
4468         if (adapter->failover_pending) {
4469                 adapter->init_done_rc = -EAGAIN;
4470                 netdev_dbg(netdev, "Failover pending, ignoring login response\n");
4471                 complete(&adapter->init_done);
4472                 /* login response buffer will be released on reset */
4473                 return 0;
4474         }
4475
4476         netdev->mtu = adapter->req_mtu - ETH_HLEN;
4477
4478         netdev_dbg(adapter->netdev, "Login Response Buffer:\n");
4479         for (i = 0; i < (adapter->login_rsp_buf_sz - 1) / 8 + 1; i++) {
4480                 netdev_dbg(adapter->netdev, "%016lx\n",
4481                            ((unsigned long int *)(adapter->login_rsp_buf))[i]);
4482         }
4483
4484         /* Sanity checks */
4485         if (login->num_txcomp_subcrqs != login_rsp->num_txsubm_subcrqs ||
4486             (be32_to_cpu(login->num_rxcomp_subcrqs) *
4487              adapter->req_rx_add_queues !=
4488              be32_to_cpu(login_rsp->num_rxadd_subcrqs))) {
4489                 dev_err(dev, "FATAL: Inconsistent login and login rsp\n");
4490                 ibmvnic_reset(adapter, VNIC_RESET_FATAL);
4491                 return -EIO;
4492         }
4493
4494         rsp_len = be32_to_cpu(login_rsp->len);
4495         if (be32_to_cpu(login->login_rsp_len) < rsp_len ||
4496             rsp_len <= be32_to_cpu(login_rsp->off_txsubm_subcrqs) ||
4497             rsp_len <= be32_to_cpu(login_rsp->off_rxadd_subcrqs) ||
4498             rsp_len <= be32_to_cpu(login_rsp->off_rxadd_buff_size) ||
4499             rsp_len <= be32_to_cpu(login_rsp->off_supp_tx_desc)) {
4500                 /* This can happen if a login request times out and there are
4501                  * 2 outstanding login requests sent, the LOGIN_RSP crq
4502                  * could have been for the older login request. So we are
4503                  * parsing the newer response buffer which may be incomplete
4504                  */
4505                 dev_err(dev, "FATAL: Login rsp offsets/lengths invalid\n");
4506                 ibmvnic_reset(adapter, VNIC_RESET_FATAL);
4507                 return -EIO;
4508         }
4509
4510         size_array = (u64 *)((u8 *)(adapter->login_rsp_buf) +
4511                 be32_to_cpu(adapter->login_rsp_buf->off_rxadd_buff_size));
4512         /* variable buffer sizes are not supported, so just read the
4513          * first entry.
4514          */
4515         adapter->cur_rx_buf_sz = be64_to_cpu(size_array[0]);
4516
4517         num_tx_pools = be32_to_cpu(adapter->login_rsp_buf->num_txsubm_subcrqs);
4518         num_rx_pools = be32_to_cpu(adapter->login_rsp_buf->num_rxadd_subcrqs);
4519
4520         tx_handle_array = (u64 *)((u8 *)(adapter->login_rsp_buf) +
4521                                   be32_to_cpu(adapter->login_rsp_buf->off_txsubm_subcrqs));
4522         rx_handle_array = (u64 *)((u8 *)(adapter->login_rsp_buf) +
4523                                   be32_to_cpu(adapter->login_rsp_buf->off_rxadd_subcrqs));
4524
4525         for (i = 0; i < num_tx_pools; i++)
4526                 adapter->tx_scrq[i]->handle = tx_handle_array[i];
4527
4528         for (i = 0; i < num_rx_pools; i++)
4529                 adapter->rx_scrq[i]->handle = rx_handle_array[i];
4530
4531         adapter->num_active_tx_scrqs = num_tx_pools;
4532         adapter->num_active_rx_scrqs = num_rx_pools;
4533         release_login_rsp_buffer(adapter);
4534         release_login_buffer(adapter);
4535         complete(&adapter->init_done);
4536
4537         return 0;
4538 }
4539
4540 static void handle_request_unmap_rsp(union ibmvnic_crq *crq,
4541                                      struct ibmvnic_adapter *adapter)
4542 {
4543         struct device *dev = &adapter->vdev->dev;
4544         long rc;
4545
4546         rc = crq->request_unmap_rsp.rc.code;
4547         if (rc)
4548                 dev_err(dev, "Error %ld in REQUEST_UNMAP_RSP\n", rc);
4549 }
4550
4551 static void handle_query_map_rsp(union ibmvnic_crq *crq,
4552                                  struct ibmvnic_adapter *adapter)
4553 {
4554         struct net_device *netdev = adapter->netdev;
4555         struct device *dev = &adapter->vdev->dev;
4556         long rc;
4557
4558         rc = crq->query_map_rsp.rc.code;
4559         if (rc) {
4560                 dev_err(dev, "Error %ld in QUERY_MAP_RSP\n", rc);
4561                 return;
4562         }
4563         netdev_dbg(netdev, "page_size = %d\ntot_pages = %d\nfree_pages = %d\n",
4564                    crq->query_map_rsp.page_size, crq->query_map_rsp.tot_pages,
4565                    crq->query_map_rsp.free_pages);
4566 }
4567
4568 static void handle_query_cap_rsp(union ibmvnic_crq *crq,
4569                                  struct ibmvnic_adapter *adapter)
4570 {
4571         struct net_device *netdev = adapter->netdev;
4572         struct device *dev = &adapter->vdev->dev;
4573         long rc;
4574
4575         atomic_dec(&adapter->running_cap_crqs);
4576         netdev_dbg(netdev, "Outstanding queries: %d\n",
4577                    atomic_read(&adapter->running_cap_crqs));
4578         rc = crq->query_capability.rc.code;
4579         if (rc) {
4580                 dev_err(dev, "Error %ld in QUERY_CAP_RSP\n", rc);
4581                 goto out;
4582         }
4583
4584         switch (be16_to_cpu(crq->query_capability.capability)) {
4585         case MIN_TX_QUEUES:
4586                 adapter->min_tx_queues =
4587                     be64_to_cpu(crq->query_capability.number);
4588                 netdev_dbg(netdev, "min_tx_queues = %lld\n",
4589                            adapter->min_tx_queues);
4590                 break;
4591         case MIN_RX_QUEUES:
4592                 adapter->min_rx_queues =
4593                     be64_to_cpu(crq->query_capability.number);
4594                 netdev_dbg(netdev, "min_rx_queues = %lld\n",
4595                            adapter->min_rx_queues);
4596                 break;
4597         case MIN_RX_ADD_QUEUES:
4598                 adapter->min_rx_add_queues =
4599                     be64_to_cpu(crq->query_capability.number);
4600                 netdev_dbg(netdev, "min_rx_add_queues = %lld\n",
4601                            adapter->min_rx_add_queues);
4602                 break;
4603         case MAX_TX_QUEUES:
4604                 adapter->max_tx_queues =
4605                     be64_to_cpu(crq->query_capability.number);
4606                 netdev_dbg(netdev, "max_tx_queues = %lld\n",
4607                            adapter->max_tx_queues);
4608                 break;
4609         case MAX_RX_QUEUES:
4610                 adapter->max_rx_queues =
4611                     be64_to_cpu(crq->query_capability.number);
4612                 netdev_dbg(netdev, "max_rx_queues = %lld\n",
4613                            adapter->max_rx_queues);
4614                 break;
4615         case MAX_RX_ADD_QUEUES:
4616                 adapter->max_rx_add_queues =
4617                     be64_to_cpu(crq->query_capability.number);
4618                 netdev_dbg(netdev, "max_rx_add_queues = %lld\n",
4619                            adapter->max_rx_add_queues);
4620                 break;
4621         case MIN_TX_ENTRIES_PER_SUBCRQ:
4622                 adapter->min_tx_entries_per_subcrq =
4623                     be64_to_cpu(crq->query_capability.number);
4624                 netdev_dbg(netdev, "min_tx_entries_per_subcrq = %lld\n",
4625                            adapter->min_tx_entries_per_subcrq);
4626                 break;
4627         case MIN_RX_ADD_ENTRIES_PER_SUBCRQ:
4628                 adapter->min_rx_add_entries_per_subcrq =
4629                     be64_to_cpu(crq->query_capability.number);
4630                 netdev_dbg(netdev, "min_rx_add_entrs_per_subcrq = %lld\n",
4631                            adapter->min_rx_add_entries_per_subcrq);
4632                 break;
4633         case MAX_TX_ENTRIES_PER_SUBCRQ:
4634                 adapter->max_tx_entries_per_subcrq =
4635                     be64_to_cpu(crq->query_capability.number);
4636                 netdev_dbg(netdev, "max_tx_entries_per_subcrq = %lld\n",
4637                            adapter->max_tx_entries_per_subcrq);
4638                 break;
4639         case MAX_RX_ADD_ENTRIES_PER_SUBCRQ:
4640                 adapter->max_rx_add_entries_per_subcrq =
4641                     be64_to_cpu(crq->query_capability.number);
4642                 netdev_dbg(netdev, "max_rx_add_entrs_per_subcrq = %lld\n",
4643                            adapter->max_rx_add_entries_per_subcrq);
4644                 break;
4645         case TCP_IP_OFFLOAD:
4646                 adapter->tcp_ip_offload =
4647                     be64_to_cpu(crq->query_capability.number);
4648                 netdev_dbg(netdev, "tcp_ip_offload = %lld\n",
4649                            adapter->tcp_ip_offload);
4650                 break;
4651         case PROMISC_SUPPORTED:
4652                 adapter->promisc_supported =
4653                     be64_to_cpu(crq->query_capability.number);
4654                 netdev_dbg(netdev, "promisc_supported = %lld\n",
4655                            adapter->promisc_supported);
4656                 break;
4657         case MIN_MTU:
4658                 adapter->min_mtu = be64_to_cpu(crq->query_capability.number);
4659                 netdev->min_mtu = adapter->min_mtu - ETH_HLEN;
4660                 netdev_dbg(netdev, "min_mtu = %lld\n", adapter->min_mtu);
4661                 break;
4662         case MAX_MTU:
4663                 adapter->max_mtu = be64_to_cpu(crq->query_capability.number);
4664                 netdev->max_mtu = adapter->max_mtu - ETH_HLEN;
4665                 netdev_dbg(netdev, "max_mtu = %lld\n", adapter->max_mtu);
4666                 break;
4667         case MAX_MULTICAST_FILTERS:
4668                 adapter->max_multicast_filters =
4669                     be64_to_cpu(crq->query_capability.number);
4670                 netdev_dbg(netdev, "max_multicast_filters = %lld\n",
4671                            adapter->max_multicast_filters);
4672                 break;
4673         case VLAN_HEADER_INSERTION:
4674                 adapter->vlan_header_insertion =
4675                     be64_to_cpu(crq->query_capability.number);
4676                 if (adapter->vlan_header_insertion)
4677                         netdev->features |= NETIF_F_HW_VLAN_STAG_TX;
4678                 netdev_dbg(netdev, "vlan_header_insertion = %lld\n",
4679                            adapter->vlan_header_insertion);
4680                 break;
4681         case RX_VLAN_HEADER_INSERTION:
4682                 adapter->rx_vlan_header_insertion =
4683                     be64_to_cpu(crq->query_capability.number);
4684                 netdev_dbg(netdev, "rx_vlan_header_insertion = %lld\n",
4685                            adapter->rx_vlan_header_insertion);
4686                 break;
4687         case MAX_TX_SG_ENTRIES:
4688                 adapter->max_tx_sg_entries =
4689                     be64_to_cpu(crq->query_capability.number);
4690                 netdev_dbg(netdev, "max_tx_sg_entries = %lld\n",
4691                            adapter->max_tx_sg_entries);
4692                 break;
4693         case RX_SG_SUPPORTED:
4694                 adapter->rx_sg_supported =
4695                     be64_to_cpu(crq->query_capability.number);
4696                 netdev_dbg(netdev, "rx_sg_supported = %lld\n",
4697                            adapter->rx_sg_supported);
4698                 break;
4699         case OPT_TX_COMP_SUB_QUEUES:
4700                 adapter->opt_tx_comp_sub_queues =
4701                     be64_to_cpu(crq->query_capability.number);
4702                 netdev_dbg(netdev, "opt_tx_comp_sub_queues = %lld\n",
4703                            adapter->opt_tx_comp_sub_queues);
4704                 break;
4705         case OPT_RX_COMP_QUEUES:
4706                 adapter->opt_rx_comp_queues =
4707                     be64_to_cpu(crq->query_capability.number);
4708                 netdev_dbg(netdev, "opt_rx_comp_queues = %lld\n",
4709                            adapter->opt_rx_comp_queues);
4710                 break;
4711         case OPT_RX_BUFADD_Q_PER_RX_COMP_Q:
4712                 adapter->opt_rx_bufadd_q_per_rx_comp_q =
4713                     be64_to_cpu(crq->query_capability.number);
4714                 netdev_dbg(netdev, "opt_rx_bufadd_q_per_rx_comp_q = %lld\n",
4715                            adapter->opt_rx_bufadd_q_per_rx_comp_q);
4716                 break;
4717         case OPT_TX_ENTRIES_PER_SUBCRQ:
4718                 adapter->opt_tx_entries_per_subcrq =
4719                     be64_to_cpu(crq->query_capability.number);
4720                 netdev_dbg(netdev, "opt_tx_entries_per_subcrq = %lld\n",
4721                            adapter->opt_tx_entries_per_subcrq);
4722                 break;
4723         case OPT_RXBA_ENTRIES_PER_SUBCRQ:
4724                 adapter->opt_rxba_entries_per_subcrq =
4725                     be64_to_cpu(crq->query_capability.number);
4726                 netdev_dbg(netdev, "opt_rxba_entries_per_subcrq = %lld\n",
4727                            adapter->opt_rxba_entries_per_subcrq);
4728                 break;
4729         case TX_RX_DESC_REQ:
4730                 adapter->tx_rx_desc_req = crq->query_capability.number;
4731                 netdev_dbg(netdev, "tx_rx_desc_req = %llx\n",
4732                            adapter->tx_rx_desc_req);
4733                 break;
4734
4735         default:
4736                 netdev_err(netdev, "Got invalid cap rsp %d\n",
4737                            crq->query_capability.capability);
4738         }
4739
4740 out:
4741         if (atomic_read(&adapter->running_cap_crqs) == 0) {
4742                 adapter->wait_capability = false;
4743                 send_request_cap(adapter, 0);
4744         }
4745 }
4746
4747 static int send_query_phys_parms(struct ibmvnic_adapter *adapter)
4748 {
4749         union ibmvnic_crq crq;
4750         int rc;
4751
4752         memset(&crq, 0, sizeof(crq));
4753         crq.query_phys_parms.first = IBMVNIC_CRQ_CMD;
4754         crq.query_phys_parms.cmd = QUERY_PHYS_PARMS;
4755
4756         mutex_lock(&adapter->fw_lock);
4757         adapter->fw_done_rc = 0;
4758         reinit_completion(&adapter->fw_done);
4759
4760         rc = ibmvnic_send_crq(adapter, &crq);
4761         if (rc) {
4762                 mutex_unlock(&adapter->fw_lock);
4763                 return rc;
4764         }
4765
4766         rc = ibmvnic_wait_for_completion(adapter, &adapter->fw_done, 10000);
4767         if (rc) {
4768                 mutex_unlock(&adapter->fw_lock);
4769                 return rc;
4770         }
4771
4772         mutex_unlock(&adapter->fw_lock);
4773         return adapter->fw_done_rc ? -EIO : 0;
4774 }
4775
4776 static int handle_query_phys_parms_rsp(union ibmvnic_crq *crq,
4777                                        struct ibmvnic_adapter *adapter)
4778 {
4779         struct net_device *netdev = adapter->netdev;
4780         int rc;
4781         __be32 rspeed = cpu_to_be32(crq->query_phys_parms_rsp.speed);
4782
4783         rc = crq->query_phys_parms_rsp.rc.code;
4784         if (rc) {
4785                 netdev_err(netdev, "Error %d in QUERY_PHYS_PARMS\n", rc);
4786                 return rc;
4787         }
4788         switch (rspeed) {
4789         case IBMVNIC_10MBPS:
4790                 adapter->speed = SPEED_10;
4791                 break;
4792         case IBMVNIC_100MBPS:
4793                 adapter->speed = SPEED_100;
4794                 break;
4795         case IBMVNIC_1GBPS:
4796                 adapter->speed = SPEED_1000;
4797                 break;
4798         case IBMVNIC_10GBPS:
4799                 adapter->speed = SPEED_10000;
4800                 break;
4801         case IBMVNIC_25GBPS:
4802                 adapter->speed = SPEED_25000;
4803                 break;
4804         case IBMVNIC_40GBPS:
4805                 adapter->speed = SPEED_40000;
4806                 break;
4807         case IBMVNIC_50GBPS:
4808                 adapter->speed = SPEED_50000;
4809                 break;
4810         case IBMVNIC_100GBPS:
4811                 adapter->speed = SPEED_100000;
4812                 break;
4813         case IBMVNIC_200GBPS:
4814                 adapter->speed = SPEED_200000;
4815                 break;
4816         default:
4817                 if (netif_carrier_ok(netdev))
4818                         netdev_warn(netdev, "Unknown speed 0x%08x\n", rspeed);
4819                 adapter->speed = SPEED_UNKNOWN;
4820         }
4821         if (crq->query_phys_parms_rsp.flags1 & IBMVNIC_FULL_DUPLEX)
4822                 adapter->duplex = DUPLEX_FULL;
4823         else if (crq->query_phys_parms_rsp.flags1 & IBMVNIC_HALF_DUPLEX)
4824                 adapter->duplex = DUPLEX_HALF;
4825         else
4826                 adapter->duplex = DUPLEX_UNKNOWN;
4827
4828         return rc;
4829 }
4830
4831 static void ibmvnic_handle_crq(union ibmvnic_crq *crq,
4832                                struct ibmvnic_adapter *adapter)
4833 {
4834         struct ibmvnic_generic_crq *gen_crq = &crq->generic;
4835         struct net_device *netdev = adapter->netdev;
4836         struct device *dev = &adapter->vdev->dev;
4837         u64 *u64_crq = (u64 *)crq;
4838         long rc;
4839
4840         netdev_dbg(netdev, "Handling CRQ: %016lx %016lx\n",
4841                    (unsigned long int)cpu_to_be64(u64_crq[0]),
4842                    (unsigned long int)cpu_to_be64(u64_crq[1]));
4843         switch (gen_crq->first) {
4844         case IBMVNIC_CRQ_INIT_RSP:
4845                 switch (gen_crq->cmd) {
4846                 case IBMVNIC_CRQ_INIT:
4847                         dev_info(dev, "Partner initialized\n");
4848                         adapter->from_passive_init = true;
4849                         /* Discard any stale login responses from prev reset.
4850                          * CHECK: should we clear even on INIT_COMPLETE?
4851                          */
4852                         adapter->login_pending = false;
4853
4854                         if (!completion_done(&adapter->init_done)) {
4855                                 complete(&adapter->init_done);
4856                                 adapter->init_done_rc = -EIO;
4857                         }
4858                         rc = ibmvnic_reset(adapter, VNIC_RESET_FAILOVER);
4859                         if (rc && rc != -EBUSY) {
4860                                 /* We were unable to schedule the failover
4861                                  * reset either because the adapter was still
4862                                  * probing (eg: during kexec) or we could not
4863                                  * allocate memory. Clear the failover_pending
4864                                  * flag since no one else will. We ignore
4865                                  * EBUSY because it means either FAILOVER reset
4866                                  * is already scheduled or the adapter is
4867                                  * being removed.
4868                                  */
4869                                 netdev_err(netdev,
4870                                            "Error %ld scheduling failover reset\n",
4871                                            rc);
4872                                 adapter->failover_pending = false;
4873                         }
4874                         break;
4875                 case IBMVNIC_CRQ_INIT_COMPLETE:
4876                         dev_info(dev, "Partner initialization complete\n");
4877                         adapter->crq.active = true;
4878                         send_version_xchg(adapter);
4879                         break;
4880                 default:
4881                         dev_err(dev, "Unknown crq cmd: %d\n", gen_crq->cmd);
4882                 }
4883                 return;
4884         case IBMVNIC_CRQ_XPORT_EVENT:
4885                 netif_carrier_off(netdev);
4886                 adapter->crq.active = false;
4887                 /* terminate any thread waiting for a response
4888                  * from the device
4889                  */
4890                 if (!completion_done(&adapter->fw_done)) {
4891                         adapter->fw_done_rc = -EIO;
4892                         complete(&adapter->fw_done);
4893                 }
4894
4895                 /* if we got here during crq-init, retry crq-init */
4896                 if (!completion_done(&adapter->init_done)) {
4897                         adapter->init_done_rc = -EAGAIN;
4898                         complete(&adapter->init_done);
4899                 }
4900
4901                 if (!completion_done(&adapter->stats_done))
4902                         complete(&adapter->stats_done);
4903                 if (test_bit(0, &adapter->resetting))
4904                         adapter->force_reset_recovery = true;
4905                 if (gen_crq->cmd == IBMVNIC_PARTITION_MIGRATED) {
4906                         dev_info(dev, "Migrated, re-enabling adapter\n");
4907                         ibmvnic_reset(adapter, VNIC_RESET_MOBILITY);
4908                 } else if (gen_crq->cmd == IBMVNIC_DEVICE_FAILOVER) {
4909                         dev_info(dev, "Backing device failover detected\n");
4910                         adapter->failover_pending = true;
4911                 } else {
4912                         /* The adapter lost the connection */
4913                         dev_err(dev, "Virtual Adapter failed (rc=%d)\n",
4914                                 gen_crq->cmd);
4915                         ibmvnic_reset(adapter, VNIC_RESET_FATAL);
4916                 }
4917                 return;
4918         case IBMVNIC_CRQ_CMD_RSP:
4919                 break;
4920         default:
4921                 dev_err(dev, "Got an invalid msg type 0x%02x\n",
4922                         gen_crq->first);
4923                 return;
4924         }
4925
4926         switch (gen_crq->cmd) {
4927         case VERSION_EXCHANGE_RSP:
4928                 rc = crq->version_exchange_rsp.rc.code;
4929                 if (rc) {
4930                         dev_err(dev, "Error %ld in VERSION_EXCHG_RSP\n", rc);
4931                         break;
4932                 }
4933                 ibmvnic_version =
4934                             be16_to_cpu(crq->version_exchange_rsp.version);
4935                 dev_info(dev, "Partner protocol version is %d\n",
4936                          ibmvnic_version);
4937                 send_query_cap(adapter);
4938                 break;
4939         case QUERY_CAPABILITY_RSP:
4940                 handle_query_cap_rsp(crq, adapter);
4941                 break;
4942         case QUERY_MAP_RSP:
4943                 handle_query_map_rsp(crq, adapter);
4944                 break;
4945         case REQUEST_MAP_RSP:
4946                 adapter->fw_done_rc = crq->request_map_rsp.rc.code;
4947                 complete(&adapter->fw_done);
4948                 break;
4949         case REQUEST_UNMAP_RSP:
4950                 handle_request_unmap_rsp(crq, adapter);
4951                 break;
4952         case REQUEST_CAPABILITY_RSP:
4953                 handle_request_cap_rsp(crq, adapter);
4954                 break;
4955         case LOGIN_RSP:
4956                 netdev_dbg(netdev, "Got Login Response\n");
4957                 handle_login_rsp(crq, adapter);
4958                 break;
4959         case LOGICAL_LINK_STATE_RSP:
4960                 netdev_dbg(netdev,
4961                            "Got Logical Link State Response, state: %d rc: %d\n",
4962                            crq->logical_link_state_rsp.link_state,
4963                            crq->logical_link_state_rsp.rc.code);
4964                 adapter->logical_link_state =
4965                     crq->logical_link_state_rsp.link_state;
4966                 adapter->init_done_rc = crq->logical_link_state_rsp.rc.code;
4967                 complete(&adapter->init_done);
4968                 break;
4969         case LINK_STATE_INDICATION:
4970                 netdev_dbg(netdev, "Got Logical Link State Indication\n");
4971                 adapter->phys_link_state =
4972                     crq->link_state_indication.phys_link_state;
4973                 adapter->logical_link_state =
4974                     crq->link_state_indication.logical_link_state;
4975                 if (adapter->phys_link_state && adapter->logical_link_state)
4976                         netif_carrier_on(netdev);
4977                 else
4978                         netif_carrier_off(netdev);
4979                 break;
4980         case CHANGE_MAC_ADDR_RSP:
4981                 netdev_dbg(netdev, "Got MAC address change Response\n");
4982                 adapter->fw_done_rc = handle_change_mac_rsp(crq, adapter);
4983                 break;
4984         case ERROR_INDICATION:
4985                 netdev_dbg(netdev, "Got Error Indication\n");
4986                 handle_error_indication(crq, adapter);
4987                 break;
4988         case REQUEST_STATISTICS_RSP:
4989                 netdev_dbg(netdev, "Got Statistics Response\n");
4990                 complete(&adapter->stats_done);
4991                 break;
4992         case QUERY_IP_OFFLOAD_RSP:
4993                 netdev_dbg(netdev, "Got Query IP offload Response\n");
4994                 handle_query_ip_offload_rsp(adapter);
4995                 break;
4996         case MULTICAST_CTRL_RSP:
4997                 netdev_dbg(netdev, "Got multicast control Response\n");
4998                 break;
4999         case CONTROL_IP_OFFLOAD_RSP:
5000                 netdev_dbg(netdev, "Got Control IP offload Response\n");
5001                 dma_unmap_single(dev, adapter->ip_offload_ctrl_tok,
5002                                  sizeof(adapter->ip_offload_ctrl),
5003                                  DMA_TO_DEVICE);
5004                 complete(&adapter->init_done);
5005                 break;
5006         case COLLECT_FW_TRACE_RSP:
5007                 netdev_dbg(netdev, "Got Collect firmware trace Response\n");
5008                 complete(&adapter->fw_done);
5009                 break;
5010         case GET_VPD_SIZE_RSP:
5011                 handle_vpd_size_rsp(crq, adapter);
5012                 break;
5013         case GET_VPD_RSP:
5014                 handle_vpd_rsp(crq, adapter);
5015                 break;
5016         case QUERY_PHYS_PARMS_RSP:
5017                 adapter->fw_done_rc = handle_query_phys_parms_rsp(crq, adapter);
5018                 complete(&adapter->fw_done);
5019                 break;
5020         default:
5021                 netdev_err(netdev, "Got an invalid cmd type 0x%02x\n",
5022                            gen_crq->cmd);
5023         }
5024 }
5025
5026 static irqreturn_t ibmvnic_interrupt(int irq, void *instance)
5027 {
5028         struct ibmvnic_adapter *adapter = instance;
5029
5030         tasklet_schedule(&adapter->tasklet);
5031         return IRQ_HANDLED;
5032 }
5033
5034 static void ibmvnic_tasklet(struct tasklet_struct *t)
5035 {
5036         struct ibmvnic_adapter *adapter = from_tasklet(adapter, t, tasklet);
5037         struct ibmvnic_crq_queue *queue = &adapter->crq;
5038         union ibmvnic_crq *crq;
5039         unsigned long flags;
5040         bool done = false;
5041
5042         spin_lock_irqsave(&queue->lock, flags);
5043         while (!done) {
5044                 /* Pull all the valid messages off the CRQ */
5045                 while ((crq = ibmvnic_next_crq(adapter)) != NULL) {
5046                         /* This barrier makes sure ibmvnic_next_crq()'s
5047                          * crq->generic.first & IBMVNIC_CRQ_CMD_RSP is loaded
5048                          * before ibmvnic_handle_crq()'s
5049                          * switch(gen_crq->first) and switch(gen_crq->cmd).
5050                          */
5051                         dma_rmb();
5052                         ibmvnic_handle_crq(crq, adapter);
5053                         crq->generic.first = 0;
5054                 }
5055         }
5056         /* if capabilities CRQ's were sent in this tasklet, the following
5057          * tasklet must wait until all responses are received
5058          */
5059         if (atomic_read(&adapter->running_cap_crqs) != 0)
5060                 adapter->wait_capability = true;
5061         spin_unlock_irqrestore(&queue->lock, flags);
5062 }
5063
5064 static int ibmvnic_reenable_crq_queue(struct ibmvnic_adapter *adapter)
5065 {
5066         struct vio_dev *vdev = adapter->vdev;
5067         int rc;
5068
5069         do {
5070                 rc = plpar_hcall_norets(H_ENABLE_CRQ, vdev->unit_address);
5071         } while (rc == H_IN_PROGRESS || rc == H_BUSY || H_IS_LONG_BUSY(rc));
5072
5073         if (rc)
5074                 dev_err(&vdev->dev, "Error enabling adapter (rc=%d)\n", rc);
5075
5076         return rc;
5077 }
5078
5079 static int ibmvnic_reset_crq(struct ibmvnic_adapter *adapter)
5080 {
5081         struct ibmvnic_crq_queue *crq = &adapter->crq;
5082         struct device *dev = &adapter->vdev->dev;
5083         struct vio_dev *vdev = adapter->vdev;
5084         int rc;
5085
5086         /* Close the CRQ */
5087         do {
5088                 rc = plpar_hcall_norets(H_FREE_CRQ, vdev->unit_address);
5089         } while (rc == H_BUSY || H_IS_LONG_BUSY(rc));
5090
5091         /* Clean out the queue */
5092         if (!crq->msgs)
5093                 return -EINVAL;
5094
5095         memset(crq->msgs, 0, PAGE_SIZE);
5096         crq->cur = 0;
5097         crq->active = false;
5098
5099         /* And re-open it again */
5100         rc = plpar_hcall_norets(H_REG_CRQ, vdev->unit_address,
5101                                 crq->msg_token, PAGE_SIZE);
5102
5103         if (rc == H_CLOSED)
5104                 /* Adapter is good, but other end is not ready */
5105                 dev_warn(dev, "Partner adapter not ready\n");
5106         else if (rc != 0)
5107                 dev_warn(dev, "Couldn't register crq (rc=%d)\n", rc);
5108
5109         return rc;
5110 }
5111
5112 static void release_crq_queue(struct ibmvnic_adapter *adapter)
5113 {
5114         struct ibmvnic_crq_queue *crq = &adapter->crq;
5115         struct vio_dev *vdev = adapter->vdev;
5116         long rc;
5117
5118         if (!crq->msgs)
5119                 return;
5120
5121         netdev_dbg(adapter->netdev, "Releasing CRQ\n");
5122         free_irq(vdev->irq, adapter);
5123         tasklet_kill(&adapter->tasklet);
5124         do {
5125                 rc = plpar_hcall_norets(H_FREE_CRQ, vdev->unit_address);
5126         } while (rc == H_BUSY || H_IS_LONG_BUSY(rc));
5127
5128         dma_unmap_single(&vdev->dev, crq->msg_token, PAGE_SIZE,
5129                          DMA_BIDIRECTIONAL);
5130         free_page((unsigned long)crq->msgs);
5131         crq->msgs = NULL;
5132         crq->active = false;
5133 }
5134
5135 static int init_crq_queue(struct ibmvnic_adapter *adapter)
5136 {
5137         struct ibmvnic_crq_queue *crq = &adapter->crq;
5138         struct device *dev = &adapter->vdev->dev;
5139         struct vio_dev *vdev = adapter->vdev;
5140         int rc, retrc = -ENOMEM;
5141
5142         if (crq->msgs)
5143                 return 0;
5144
5145         crq->msgs = (union ibmvnic_crq *)get_zeroed_page(GFP_KERNEL);
5146         /* Should we allocate more than one page? */
5147
5148         if (!crq->msgs)
5149                 return -ENOMEM;
5150
5151         crq->size = PAGE_SIZE / sizeof(*crq->msgs);
5152         crq->msg_token = dma_map_single(dev, crq->msgs, PAGE_SIZE,
5153                                         DMA_BIDIRECTIONAL);
5154         if (dma_mapping_error(dev, crq->msg_token))
5155                 goto map_failed;
5156
5157         rc = plpar_hcall_norets(H_REG_CRQ, vdev->unit_address,
5158                                 crq->msg_token, PAGE_SIZE);
5159
5160         if (rc == H_RESOURCE)
5161                 /* maybe kexecing and resource is busy. try a reset */
5162                 rc = ibmvnic_reset_crq(adapter);
5163         retrc = rc;
5164
5165         if (rc == H_CLOSED) {
5166                 dev_warn(dev, "Partner adapter not ready\n");
5167         } else if (rc) {
5168                 dev_warn(dev, "Error %d opening adapter\n", rc);
5169                 goto reg_crq_failed;
5170         }
5171
5172         retrc = 0;
5173
5174         tasklet_setup(&adapter->tasklet, (void *)ibmvnic_tasklet);
5175
5176         netdev_dbg(adapter->netdev, "registering irq 0x%x\n", vdev->irq);
5177         snprintf(crq->name, sizeof(crq->name), "ibmvnic-%x",
5178                  adapter->vdev->unit_address);
5179         rc = request_irq(vdev->irq, ibmvnic_interrupt, 0, crq->name, adapter);
5180         if (rc) {
5181                 dev_err(dev, "Couldn't register irq 0x%x. rc=%d\n",
5182                         vdev->irq, rc);
5183                 goto req_irq_failed;
5184         }
5185
5186         rc = vio_enable_interrupts(vdev);
5187         if (rc) {
5188                 dev_err(dev, "Error %d enabling interrupts\n", rc);
5189                 goto req_irq_failed;
5190         }
5191
5192         crq->cur = 0;
5193         spin_lock_init(&crq->lock);
5194
5195         /* process any CRQs that were queued before we enabled interrupts */
5196         tasklet_schedule(&adapter->tasklet);
5197
5198         return retrc;
5199
5200 req_irq_failed:
5201         tasklet_kill(&adapter->tasklet);
5202         do {
5203                 rc = plpar_hcall_norets(H_FREE_CRQ, vdev->unit_address);
5204         } while (rc == H_BUSY || H_IS_LONG_BUSY(rc));
5205 reg_crq_failed:
5206         dma_unmap_single(dev, crq->msg_token, PAGE_SIZE, DMA_BIDIRECTIONAL);
5207 map_failed:
5208         free_page((unsigned long)crq->msgs);
5209         crq->msgs = NULL;
5210         return retrc;
5211 }
5212
5213 static int ibmvnic_reset_init(struct ibmvnic_adapter *adapter, bool reset)
5214 {
5215         struct device *dev = &adapter->vdev->dev;
5216         unsigned long timeout = msecs_to_jiffies(20000);
5217         u64 old_num_rx_queues = adapter->req_rx_queues;
5218         u64 old_num_tx_queues = adapter->req_tx_queues;
5219         int rc;
5220
5221         adapter->from_passive_init = false;
5222
5223         if (reset)
5224                 reinit_completion(&adapter->init_done);
5225
5226         adapter->init_done_rc = 0;
5227         rc = ibmvnic_send_crq_init(adapter);
5228         if (rc) {
5229                 dev_err(dev, "Send crq init failed with error %d\n", rc);
5230                 return rc;
5231         }
5232
5233         if (!wait_for_completion_timeout(&adapter->init_done, timeout)) {
5234                 dev_err(dev, "Initialization sequence timed out\n");
5235                 return -1;
5236         }
5237
5238         if (adapter->init_done_rc) {
5239                 release_crq_queue(adapter);
5240                 return adapter->init_done_rc;
5241         }
5242
5243         if (adapter->from_passive_init) {
5244                 adapter->state = VNIC_OPEN;
5245                 adapter->from_passive_init = false;
5246                 return -1;
5247         }
5248
5249         if (reset &&
5250             test_bit(0, &adapter->resetting) && !adapter->wait_for_reset &&
5251             adapter->reset_reason != VNIC_RESET_MOBILITY) {
5252                 if (adapter->req_rx_queues != old_num_rx_queues ||
5253                     adapter->req_tx_queues != old_num_tx_queues) {
5254                         release_sub_crqs(adapter, 0);
5255                         rc = init_sub_crqs(adapter);
5256                 } else {
5257                         /* no need to reinitialize completely, but we do
5258                          * need to clean up transmits that were in flight
5259                          * when we processed the reset.  Failure to do so
5260                          * will confound the upper layer, usually TCP, by
5261                          * creating the illusion of transmits that are
5262                          * awaiting completion.
5263                          */
5264                         clean_tx_pools(adapter);
5265
5266                         rc = reset_sub_crq_queues(adapter);
5267                 }
5268         } else {
5269                 rc = init_sub_crqs(adapter);
5270         }
5271
5272         if (rc) {
5273                 dev_err(dev, "Initialization of sub crqs failed\n");
5274                 release_crq_queue(adapter);
5275                 return rc;
5276         }
5277
5278         rc = init_sub_crq_irqs(adapter);
5279         if (rc) {
5280                 dev_err(dev, "Failed to initialize sub crq irqs\n");
5281                 release_crq_queue(adapter);
5282         }
5283
5284         return rc;
5285 }
5286
5287 static struct device_attribute dev_attr_failover;
5288
5289 static int ibmvnic_probe(struct vio_dev *dev, const struct vio_device_id *id)
5290 {
5291         struct ibmvnic_adapter *adapter;
5292         struct net_device *netdev;
5293         unsigned char *mac_addr_p;
5294         int rc;
5295
5296         dev_dbg(&dev->dev, "entering ibmvnic_probe for UA 0x%x\n",
5297                 dev->unit_address);
5298
5299         mac_addr_p = (unsigned char *)vio_get_attribute(dev,
5300                                                         VETH_MAC_ADDR, NULL);
5301         if (!mac_addr_p) {
5302                 dev_err(&dev->dev,
5303                         "(%s:%3.3d) ERROR: Can't find MAC_ADDR attribute\n",
5304                         __FILE__, __LINE__);
5305                 return 0;
5306         }
5307
5308         netdev = alloc_etherdev_mq(sizeof(struct ibmvnic_adapter),
5309                                    IBMVNIC_MAX_QUEUES);
5310         if (!netdev)
5311                 return -ENOMEM;
5312
5313         adapter = netdev_priv(netdev);
5314         adapter->state = VNIC_PROBING;
5315         dev_set_drvdata(&dev->dev, netdev);
5316         adapter->vdev = dev;
5317         adapter->netdev = netdev;
5318         adapter->login_pending = false;
5319
5320         ether_addr_copy(adapter->mac_addr, mac_addr_p);
5321         ether_addr_copy(netdev->dev_addr, adapter->mac_addr);
5322         netdev->irq = dev->irq;
5323         netdev->netdev_ops = &ibmvnic_netdev_ops;
5324         netdev->ethtool_ops = &ibmvnic_ethtool_ops;
5325         SET_NETDEV_DEV(netdev, &dev->dev);
5326
5327         spin_lock_init(&adapter->stats_lock);
5328
5329         INIT_WORK(&adapter->ibmvnic_reset, __ibmvnic_reset);
5330         INIT_DELAYED_WORK(&adapter->ibmvnic_delayed_reset,
5331                           __ibmvnic_delayed_reset);
5332         INIT_LIST_HEAD(&adapter->rwi_list);
5333         spin_lock_init(&adapter->rwi_lock);
5334         spin_lock_init(&adapter->state_lock);
5335         mutex_init(&adapter->fw_lock);
5336         init_completion(&adapter->init_done);
5337         init_completion(&adapter->fw_done);
5338         init_completion(&adapter->reset_done);
5339         init_completion(&adapter->stats_done);
5340         clear_bit(0, &adapter->resetting);
5341
5342         do {
5343                 rc = init_crq_queue(adapter);
5344                 if (rc) {
5345                         dev_err(&dev->dev, "Couldn't initialize crq. rc=%d\n",
5346                                 rc);
5347                         goto ibmvnic_init_fail;
5348                 }
5349
5350                 rc = ibmvnic_reset_init(adapter, false);
5351                 if (rc && rc != EAGAIN)
5352                         goto ibmvnic_init_fail;
5353         } while (rc == EAGAIN);
5354
5355         rc = init_stats_buffers(adapter);
5356         if (rc)
5357                 goto ibmvnic_init_fail;
5358
5359         rc = init_stats_token(adapter);
5360         if (rc)
5361                 goto ibmvnic_stats_fail;
5362
5363         netdev->mtu = adapter->req_mtu - ETH_HLEN;
5364         netdev->min_mtu = adapter->min_mtu - ETH_HLEN;
5365         netdev->max_mtu = adapter->max_mtu - ETH_HLEN;
5366
5367         rc = device_create_file(&dev->dev, &dev_attr_failover);
5368         if (rc)
5369                 goto ibmvnic_dev_file_err;
5370
5371         netif_carrier_off(netdev);
5372
5373         adapter->state = VNIC_PROBED;
5374
5375         adapter->wait_for_reset = false;
5376         adapter->last_reset_time = jiffies;
5377
5378         rc = register_netdev(netdev);
5379         if (rc) {
5380                 dev_err(&dev->dev, "failed to register netdev rc=%d\n", rc);
5381                 goto ibmvnic_register_fail;
5382         }
5383         dev_info(&dev->dev, "ibmvnic registered\n");
5384
5385         return 0;
5386
5387 ibmvnic_register_fail:
5388         device_remove_file(&dev->dev, &dev_attr_failover);
5389
5390 ibmvnic_dev_file_err:
5391         release_stats_token(adapter);
5392
5393 ibmvnic_stats_fail:
5394         release_stats_buffers(adapter);
5395
5396 ibmvnic_init_fail:
5397         release_sub_crqs(adapter, 1);
5398         release_crq_queue(adapter);
5399         mutex_destroy(&adapter->fw_lock);
5400         free_netdev(netdev);
5401
5402         return rc;
5403 }
5404
5405 static int ibmvnic_remove(struct vio_dev *dev)
5406 {
5407         struct net_device *netdev = dev_get_drvdata(&dev->dev);
5408         struct ibmvnic_adapter *adapter = netdev_priv(netdev);
5409         unsigned long flags;
5410
5411         spin_lock_irqsave(&adapter->state_lock, flags);
5412
5413         /* If ibmvnic_reset() is scheduling a reset, wait for it to
5414          * finish. Then, set the state to REMOVING to prevent it from
5415          * scheduling any more work and to have reset functions ignore
5416          * any resets that have already been scheduled. Drop the lock
5417          * after setting state, so __ibmvnic_reset() which is called
5418          * from the flush_work() below, can make progress.
5419          */
5420         spin_lock(&adapter->rwi_lock);
5421         adapter->state = VNIC_REMOVING;
5422         spin_unlock(&adapter->rwi_lock);
5423
5424         spin_unlock_irqrestore(&adapter->state_lock, flags);
5425
5426         flush_work(&adapter->ibmvnic_reset);
5427         flush_delayed_work(&adapter->ibmvnic_delayed_reset);
5428
5429         rtnl_lock();
5430         unregister_netdevice(netdev);
5431
5432         release_resources(adapter);
5433         release_sub_crqs(adapter, 1);
5434         release_crq_queue(adapter);
5435
5436         release_stats_token(adapter);
5437         release_stats_buffers(adapter);
5438
5439         adapter->state = VNIC_REMOVED;
5440
5441         rtnl_unlock();
5442         mutex_destroy(&adapter->fw_lock);
5443         device_remove_file(&dev->dev, &dev_attr_failover);
5444         free_netdev(netdev);
5445         dev_set_drvdata(&dev->dev, NULL);
5446
5447         return 0;
5448 }
5449
5450 static ssize_t failover_store(struct device *dev, struct device_attribute *attr,
5451                               const char *buf, size_t count)
5452 {
5453         struct net_device *netdev = dev_get_drvdata(dev);
5454         struct ibmvnic_adapter *adapter = netdev_priv(netdev);
5455         unsigned long retbuf[PLPAR_HCALL_BUFSIZE];
5456         __be64 session_token;
5457         long rc;
5458
5459         if (!sysfs_streq(buf, "1"))
5460                 return -EINVAL;
5461
5462         rc = plpar_hcall(H_VIOCTL, retbuf, adapter->vdev->unit_address,
5463                          H_GET_SESSION_TOKEN, 0, 0, 0);
5464         if (rc) {
5465                 netdev_err(netdev, "Couldn't retrieve session token, rc %ld\n",
5466                            rc);
5467                 return -EINVAL;
5468         }
5469
5470         session_token = (__be64)retbuf[0];
5471         netdev_dbg(netdev, "Initiating client failover, session id %llx\n",
5472                    be64_to_cpu(session_token));
5473         rc = plpar_hcall_norets(H_VIOCTL, adapter->vdev->unit_address,
5474                                 H_SESSION_ERR_DETECTED, session_token, 0, 0);
5475         if (rc) {
5476                 netdev_err(netdev, "Client initiated failover failed, rc %ld\n",
5477                            rc);
5478                 return -EINVAL;
5479         }
5480
5481         return count;
5482 }
5483
5484 static DEVICE_ATTR_WO(failover);
5485
5486 static unsigned long ibmvnic_get_desired_dma(struct vio_dev *vdev)
5487 {
5488         struct net_device *netdev = dev_get_drvdata(&vdev->dev);
5489         struct ibmvnic_adapter *adapter;
5490         struct iommu_table *tbl;
5491         unsigned long ret = 0;
5492         int i;
5493
5494         tbl = get_iommu_table_base(&vdev->dev);
5495
5496         /* netdev inits at probe time along with the structures we need below*/
5497         if (!netdev)
5498                 return IOMMU_PAGE_ALIGN(IBMVNIC_IO_ENTITLEMENT_DEFAULT, tbl);
5499
5500         adapter = netdev_priv(netdev);
5501
5502         ret += PAGE_SIZE; /* the crq message queue */
5503         ret += IOMMU_PAGE_ALIGN(sizeof(struct ibmvnic_statistics), tbl);
5504
5505         for (i = 0; i < adapter->req_tx_queues + adapter->req_rx_queues; i++)
5506                 ret += 4 * PAGE_SIZE; /* the scrq message queue */
5507
5508         for (i = 0; i < adapter->num_active_rx_pools; i++)
5509                 ret += adapter->rx_pool[i].size *
5510                     IOMMU_PAGE_ALIGN(adapter->rx_pool[i].buff_size, tbl);
5511
5512         return ret;
5513 }
5514
5515 static int ibmvnic_resume(struct device *dev)
5516 {
5517         struct net_device *netdev = dev_get_drvdata(dev);
5518         struct ibmvnic_adapter *adapter = netdev_priv(netdev);
5519
5520         if (adapter->state != VNIC_OPEN)
5521                 return 0;
5522
5523         tasklet_schedule(&adapter->tasklet);
5524
5525         return 0;
5526 }
5527
5528 static const struct vio_device_id ibmvnic_device_table[] = {
5529         {"network", "IBM,vnic"},
5530         {"", "" }
5531 };
5532 MODULE_DEVICE_TABLE(vio, ibmvnic_device_table);
5533
5534 static const struct dev_pm_ops ibmvnic_pm_ops = {
5535         .resume = ibmvnic_resume
5536 };
5537
5538 static struct vio_driver ibmvnic_driver = {
5539         .id_table       = ibmvnic_device_table,
5540         .probe          = ibmvnic_probe,
5541         .remove         = ibmvnic_remove,
5542         .get_desired_dma = ibmvnic_get_desired_dma,
5543         .name           = ibmvnic_driver_name,
5544         .pm             = &ibmvnic_pm_ops,
5545 };
5546
5547 /* module functions */
5548 static int __init ibmvnic_module_init(void)
5549 {
5550         pr_info("%s: %s %s\n", ibmvnic_driver_name, ibmvnic_driver_string,
5551                 IBMVNIC_DRIVER_VERSION);
5552
5553         return vio_register_driver(&ibmvnic_driver);
5554 }
5555
5556 static void __exit ibmvnic_module_exit(void)
5557 {
5558         vio_unregister_driver(&ibmvnic_driver);
5559 }
5560
5561 module_init(ibmvnic_module_init);
5562 module_exit(ibmvnic_module_exit);