drivers/net/ethernet/ibm/ibmvnic.c

   1 /**************************************************************************/
   2 /*                                                                        */
   3 /*  IBM System i and System p Virtual NIC Device Driver                   */
   4 /*  Copyright (C) 2014 IBM Corp.                                          */
   5 /*  Santiago Leon (santi_leon@yahoo.com)                                  */
   6 /*  Thomas Falcon (tlfalcon@linux.vnet.ibm.com)                           */
   7 /*  John Allen (jallen@linux.vnet.ibm.com)                                */
   8 /*                                                                        */
   9 /*  This program is free software; you can redistribute it and/or modify  */
  10 /*  it under the terms of the GNU General Public License as published by  */
  11 /*  the Free Software Foundation; either version 2 of the License, or     */
  12 /*  (at your option) any later version.                                   */
  13 /*                                                                        */
  14 /*  This program is distributed in the hope that it will be useful,       */
  15 /*  but WITHOUT ANY WARRANTY; without even the implied warranty of        */
  16 /*  MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the         */
  17 /*  GNU General Public License for more details.                          */
  18 /*                                                                        */
  19 /*  You should have received a copy of the GNU General Public License     */
  20 /*  along with this program.                                              */
  21 /*                                                                        */
  22 /* This module contains the implementation of a virtual ethernet device   */
  23 /* for use with IBM i/p Series LPAR Linux. It utilizes the logical LAN    */
  24 /* option of the RS/6000 Platform Architecture to interface with virtual  */
  25 /* ethernet NICs that are presented to the partition by the hypervisor.   */
  26 /*                                                                         */
  27 /* Messages are passed between the VNIC driver and the VNIC server using  */
  28 /* Command/Response Queues (CRQs) and sub CRQs (sCRQs). CRQs are used to  */
  29 /* issue and receive commands that initiate communication with the server */
  30 /* on driver initialization. Sub CRQs (sCRQs) are similar to CRQs, but    */
  31 /* are used by the driver to notify the server that a packet is           */
  32 /* ready for transmission or that a buffer has been added to receive a    */
  33 /* packet. Subsequently, sCRQs are used by the server to notify the       */
  34 /* driver that a packet transmission has been completed or that a packet  */
  35 /* has been received and placed in a waiting buffer.                      */
  36 /*                                                                        */
  37 /* In lieu of a more conventional "on-the-fly" DMA mapping strategy in    */
  38 /* which skbs are DMA mapped and immediately unmapped when the transmit   */
  39 /* or receive has been completed, the VNIC driver is required to use      */
  40 /* "long term mapping". This entails that large, continuous DMA mapped    */
  41 /* buffers are allocated on driver initialization and these buffers are   */
  42 /* then continuously reused to pass skbs to and from the VNIC server.     */
  43 /*                                                                        */
  44 /**************************************************************************/
  45
  46 #include <linux/module.h>
  47 #include <linux/moduleparam.h>
  48 #include <linux/types.h>
  49 #include <linux/errno.h>
  50 #include <linux/completion.h>
  51 #include <linux/ioport.h>
  52 #include <linux/dma-mapping.h>
  53 #include <linux/kernel.h>
  54 #include <linux/netdevice.h>
  55 #include <linux/etherdevice.h>
  56 #include <linux/skbuff.h>
  57 #include <linux/init.h>
  58 #include <linux/delay.h>
  59 #include <linux/mm.h>
  60 #include <linux/ethtool.h>
  61 #include <linux/proc_fs.h>
  62 #include <linux/if_arp.h>
  63 #include <linux/in.h>
  64 #include <linux/ip.h>
  65 #include <linux/ipv6.h>
  66 #include <linux/irq.h>
  67 #include <linux/kthread.h>
  68 #include <linux/seq_file.h>
  69 #include <linux/interrupt.h>
  70 #include <net/net_namespace.h>
  71 #include <asm/hvcall.h>
  72 #include <linux/atomic.h>
  73 #include <asm/vio.h>
  74 #include <asm/iommu.h>
  75 #include <linux/uaccess.h>
  76 #include <asm/firmware.h>
  77 #include <linux/workqueue.h>
  78 #include <linux/if_vlan.h>
  79 #include <linux/utsname.h>
  80
  81 #include "ibmvnic.h"
  82
  83 static const char ibmvnic_driver_name[] = "ibmvnic";
  84 static const char ibmvnic_driver_string[] = "IBM System i/p Virtual NIC Driver";
  85
  86 MODULE_AUTHOR("Santiago Leon");
  87 MODULE_DESCRIPTION("IBM System i/p Virtual NIC Driver");
  88 MODULE_LICENSE("GPL");
  89 MODULE_VERSION(IBMVNIC_DRIVER_VERSION);
  90
  91 static int ibmvnic_version = IBMVNIC_INITIAL_VERSION;
  92 static int ibmvnic_remove(struct vio_dev *);
  93 static void release_sub_crqs(struct ibmvnic_adapter *, bool);
  94 static int ibmvnic_reset_crq(struct ibmvnic_adapter *);
  95 static int ibmvnic_send_crq_init(struct ibmvnic_adapter *);
  96 static int ibmvnic_reenable_crq_queue(struct ibmvnic_adapter *);
  97 static int ibmvnic_send_crq(struct ibmvnic_adapter *, union ibmvnic_crq *);
  98 static int send_subcrq(struct ibmvnic_adapter *adapter, u64 remote_handle,
  99                        union sub_crq *sub_crq);
 100 static int send_subcrq_indirect(struct ibmvnic_adapter *, u64, u64, u64);
 101 static irqreturn_t ibmvnic_interrupt_rx(int irq, void *instance);
 102 static int enable_scrq_irq(struct ibmvnic_adapter *,
 103                            struct ibmvnic_sub_crq_queue *);
 104 static int disable_scrq_irq(struct ibmvnic_adapter *,
 105                             struct ibmvnic_sub_crq_queue *);
 106 static int pending_scrq(struct ibmvnic_adapter *,
 107                         struct ibmvnic_sub_crq_queue *);
 108 static union sub_crq *ibmvnic_next_scrq(struct ibmvnic_adapter *,
 109                                         struct ibmvnic_sub_crq_queue *);
 110 static int ibmvnic_poll(struct napi_struct *napi, int data);
 111 static void send_map_query(struct ibmvnic_adapter *adapter);
 112 static int send_request_map(struct ibmvnic_adapter *, dma_addr_t, __be32, u8);
 113 static int send_request_unmap(struct ibmvnic_adapter *, u8);
 114 static int send_login(struct ibmvnic_adapter *adapter);
 115 static void send_cap_queries(struct ibmvnic_adapter *adapter);
 116 static int init_sub_crqs(struct ibmvnic_adapter *);
 117 static int init_sub_crq_irqs(struct ibmvnic_adapter *adapter);
 118 static int ibmvnic_init(struct ibmvnic_adapter *);
 119 static int ibmvnic_reset_init(struct ibmvnic_adapter *);
 120 static void release_crq_queue(struct ibmvnic_adapter *);
 121 static int __ibmvnic_set_mac(struct net_device *netdev, struct sockaddr *p);
 122 static int init_crq_queue(struct ibmvnic_adapter *adapter);
 123
 124 struct ibmvnic_stat {
 125         char name[ETH_GSTRING_LEN];
 126         int offset;
 127 };
 128
 129 #define IBMVNIC_STAT_OFF(stat) (offsetof(struct ibmvnic_adapter, stats) + \
 130                              offsetof(struct ibmvnic_statistics, stat))
 131 #define IBMVNIC_GET_STAT(a, off) (*((u64 *)(((unsigned long)(a)) + off)))
 132
 133 static const struct ibmvnic_stat ibmvnic_stats[] = {
 134         {"rx_packets", IBMVNIC_STAT_OFF(rx_packets)},
 135         {"rx_bytes", IBMVNIC_STAT_OFF(rx_bytes)},
 136         {"tx_packets", IBMVNIC_STAT_OFF(tx_packets)},
 137         {"tx_bytes", IBMVNIC_STAT_OFF(tx_bytes)},
 138         {"ucast_tx_packets", IBMVNIC_STAT_OFF(ucast_tx_packets)},
 139         {"ucast_rx_packets", IBMVNIC_STAT_OFF(ucast_rx_packets)},
 140         {"mcast_tx_packets", IBMVNIC_STAT_OFF(mcast_tx_packets)},
 141         {"mcast_rx_packets", IBMVNIC_STAT_OFF(mcast_rx_packets)},
 142         {"bcast_tx_packets", IBMVNIC_STAT_OFF(bcast_tx_packets)},
 143         {"bcast_rx_packets", IBMVNIC_STAT_OFF(bcast_rx_packets)},
 144         {"align_errors", IBMVNIC_STAT_OFF(align_errors)},
 145         {"fcs_errors", IBMVNIC_STAT_OFF(fcs_errors)},
 146         {"single_collision_frames", IBMVNIC_STAT_OFF(single_collision_frames)},
 147         {"multi_collision_frames", IBMVNIC_STAT_OFF(multi_collision_frames)},
 148         {"sqe_test_errors", IBMVNIC_STAT_OFF(sqe_test_errors)},
 149         {"deferred_tx", IBMVNIC_STAT_OFF(deferred_tx)},
 150         {"late_collisions", IBMVNIC_STAT_OFF(late_collisions)},
 151         {"excess_collisions", IBMVNIC_STAT_OFF(excess_collisions)},
 152         {"internal_mac_tx_errors", IBMVNIC_STAT_OFF(internal_mac_tx_errors)},
 153         {"carrier_sense", IBMVNIC_STAT_OFF(carrier_sense)},
 154         {"too_long_frames", IBMVNIC_STAT_OFF(too_long_frames)},
 155         {"internal_mac_rx_errors", IBMVNIC_STAT_OFF(internal_mac_rx_errors)},
 156 };
 157
 158 static long h_reg_sub_crq(unsigned long unit_address, unsigned long token,
 159                           unsigned long length, unsigned long *number,
 160                           unsigned long *irq)
 161 {
 162         unsigned long retbuf[PLPAR_HCALL_BUFSIZE];
 163         long rc;
 164
 165         rc = plpar_hcall(H_REG_SUB_CRQ, retbuf, unit_address, token, length);
 166         *number = retbuf[0];
 167         *irq = retbuf[1];
 168
 169         return rc;
 170 }
 171
 172 static int alloc_long_term_buff(struct ibmvnic_adapter *adapter,
 173                                 struct ibmvnic_long_term_buff *ltb, int size)
 174 {
 175         struct device *dev = &adapter->vdev->dev;
 176         int rc;
 177
 178         ltb->size = size;
 179         ltb->buff = dma_alloc_coherent(dev, ltb->size, &ltb->addr,
 180                                        GFP_KERNEL);
 181
 182         if (!ltb->buff) {
 183                 dev_err(dev, "Couldn't alloc long term buffer\n");
 184                 return -ENOMEM;
 185         }
 186         ltb->map_id = adapter->map_id;
 187         adapter->map_id++;
 188
 189         init_completion(&adapter->fw_done);
 190         rc = send_request_map(adapter, ltb->addr,
 191                               ltb->size, ltb->map_id);
 192         if (rc) {
 193                 dma_free_coherent(dev, ltb->size, ltb->buff, ltb->addr);
 194                 return rc;
 195         }
 196         wait_for_completion(&adapter->fw_done);
 197
 198         if (adapter->fw_done_rc) {
 199                 dev_err(dev, "Couldn't map long term buffer,rc = %d\n",
 200                         adapter->fw_done_rc);
 201                 dma_free_coherent(dev, ltb->size, ltb->buff, ltb->addr);
 202                 return -1;
 203         }
 204         return 0;
 205 }
 206
 207 static void free_long_term_buff(struct ibmvnic_adapter *adapter,
 208                                 struct ibmvnic_long_term_buff *ltb)
 209 {
 210         struct device *dev = &adapter->vdev->dev;
 211
 212         if (!ltb->buff)
 213                 return;
 214
 215         if (adapter->reset_reason != VNIC_RESET_FAILOVER &&
 216             adapter->reset_reason != VNIC_RESET_MOBILITY)
 217                 send_request_unmap(adapter, ltb->map_id);
 218         dma_free_coherent(dev, ltb->size, ltb->buff, ltb->addr);
 219 }
 220
 221 static int reset_long_term_buff(struct ibmvnic_adapter *adapter,
 222                                 struct ibmvnic_long_term_buff *ltb)
 223 {
 224         int rc;
 225
 226         memset(ltb->buff, 0, ltb->size);
 227
 228         init_completion(&adapter->fw_done);
 229         rc = send_request_map(adapter, ltb->addr, ltb->size, ltb->map_id);
 230         if (rc)
 231                 return rc;
 232         wait_for_completion(&adapter->fw_done);
 233
 234         if (adapter->fw_done_rc) {
 235                 dev_info(&adapter->vdev->dev,
 236                          "Reset failed, attempting to free and reallocate buffer\n");
 237                 free_long_term_buff(adapter, ltb);
 238                 return alloc_long_term_buff(adapter, ltb, ltb->size);
 239         }
 240         return 0;
 241 }
 242
 243 static void deactivate_rx_pools(struct ibmvnic_adapter *adapter)
 244 {
 245         int i;
 246
 247         for (i = 0; i < be32_to_cpu(adapter->login_rsp_buf->num_rxadd_subcrqs);
 248              i++)
 249                 adapter->rx_pool[i].active = 0;
 250 }
 251
 252 static void replenish_rx_pool(struct ibmvnic_adapter *adapter,
 253                               struct ibmvnic_rx_pool *pool)
 254 {
 255         int count = pool->size - atomic_read(&pool->available);
 256         struct device *dev = &adapter->vdev->dev;
 257         int buffers_added = 0;
 258         unsigned long lpar_rc;
 259         union sub_crq sub_crq;
 260         struct sk_buff *skb;
 261         unsigned int offset;
 262         dma_addr_t dma_addr;
 263         unsigned char *dst;
 264         u64 *handle_array;
 265         int shift = 0;
 266         int index;
 267         int i;
 268
 269         if (!pool->active)
 270                 return;
 271
 272         handle_array = (u64 *)((u8 *)(adapter->login_rsp_buf) +
 273                                       be32_to_cpu(adapter->login_rsp_buf->
 274                                       off_rxadd_subcrqs));
 275
 276         for (i = 0; i < count; ++i) {
 277                 skb = alloc_skb(pool->buff_size, GFP_ATOMIC);
 278                 if (!skb) {
 279                         dev_err(dev, "Couldn't replenish rx buff\n");
 280                         adapter->replenish_no_mem++;
 281                         break;
 282                 }
 283
 284                 index = pool->free_map[pool->next_free];
 285
 286                 if (pool->rx_buff[index].skb)
 287                         dev_err(dev, "Inconsistent free_map!\n");
 288
 289                 /* Copy the skb to the long term mapped DMA buffer */
 290                 offset = index * pool->buff_size;
 291                 dst = pool->long_term_buff.buff + offset;
 292                 memset(dst, 0, pool->buff_size);
 293                 dma_addr = pool->long_term_buff.addr + offset;
 294                 pool->rx_buff[index].data = dst;
 295
 296                 pool->free_map[pool->next_free] = IBMVNIC_INVALID_MAP;
 297                 pool->rx_buff[index].dma = dma_addr;
 298                 pool->rx_buff[index].skb = skb;
 299                 pool->rx_buff[index].pool_index = pool->index;
 300                 pool->rx_buff[index].size = pool->buff_size;
 301
 302                 memset(&sub_crq, 0, sizeof(sub_crq));
 303                 sub_crq.rx_add.first = IBMVNIC_CRQ_CMD;
 304                 sub_crq.rx_add.correlator =
 305                     cpu_to_be64((u64)&pool->rx_buff[index]);
 306                 sub_crq.rx_add.ioba = cpu_to_be32(dma_addr);
 307                 sub_crq.rx_add.map_id = pool->long_term_buff.map_id;
 308
 309                 /* The length field of the sCRQ is defined to be 24 bits so the
 310                  * buffer size needs to be left shifted by a byte before it is
 311                  * converted to big endian to prevent the last byte from being
 312                  * truncated.
 313                  */
 314 #ifdef __LITTLE_ENDIAN__
 315                 shift = 8;
 316 #endif
 317                 sub_crq.rx_add.len = cpu_to_be32(pool->buff_size << shift);
 318
 319                 lpar_rc = send_subcrq(adapter, handle_array[pool->index],
 320                                       &sub_crq);
 321                 if (lpar_rc != H_SUCCESS)
 322                         goto failure;
 323
 324                 buffers_added++;
 325                 adapter->replenish_add_buff_success++;
 326                 pool->next_free = (pool->next_free + 1) % pool->size;
 327         }
 328         atomic_add(buffers_added, &pool->available);
 329         return;
 330
 331 failure:
 332         if (lpar_rc != H_PARAMETER && lpar_rc != H_CLOSED)
 333                 dev_err_ratelimited(dev, "rx: replenish packet buffer failed\n");
 334         pool->free_map[pool->next_free] = index;
 335         pool->rx_buff[index].skb = NULL;
 336
 337         dev_kfree_skb_any(skb);
 338         adapter->replenish_add_buff_failure++;
 339         atomic_add(buffers_added, &pool->available);
 340
 341         if (lpar_rc == H_CLOSED || adapter->failover_pending) {
 342                 /* Disable buffer pool replenishment and report carrier off if
 343                  * queue is closed or pending failover.
 344                  * Firmware guarantees that a signal will be sent to the
 345                  * driver, triggering a reset.
 346                  */
 347                 deactivate_rx_pools(adapter);
 348                 netif_carrier_off(adapter->netdev);
 349         }
 350 }
 351
 352 static void replenish_pools(struct ibmvnic_adapter *adapter)
 353 {
 354         int i;
 355
 356         adapter->replenish_task_cycles++;
 357         for (i = 0; i < be32_to_cpu(adapter->login_rsp_buf->num_rxadd_subcrqs);
 358              i++) {
 359                 if (adapter->rx_pool[i].active)
 360                         replenish_rx_pool(adapter, &adapter->rx_pool[i]);
 361         }
 362 }
 363
 364 static void release_stats_buffers(struct ibmvnic_adapter *adapter)
 365 {
 366         kfree(adapter->tx_stats_buffers);
 367         kfree(adapter->rx_stats_buffers);
 368         adapter->tx_stats_buffers = NULL;
 369         adapter->rx_stats_buffers = NULL;
 370 }
 371
 372 static int init_stats_buffers(struct ibmvnic_adapter *adapter)
 373 {
 374         adapter->tx_stats_buffers =
 375                                 kcalloc(IBMVNIC_MAX_QUEUES,
 376                                         sizeof(struct ibmvnic_tx_queue_stats),
 377                                         GFP_KERNEL);
 378         if (!adapter->tx_stats_buffers)
 379                 return -ENOMEM;
 380
 381         adapter->rx_stats_buffers =
 382                                 kcalloc(IBMVNIC_MAX_QUEUES,
 383                                         sizeof(struct ibmvnic_rx_queue_stats),
 384                                         GFP_KERNEL);
 385         if (!adapter->rx_stats_buffers)
 386                 return -ENOMEM;
 387
 388         return 0;
 389 }
 390
 391 static void release_stats_token(struct ibmvnic_adapter *adapter)
 392 {
 393         struct device *dev = &adapter->vdev->dev;
 394
 395         if (!adapter->stats_token)
 396                 return;
 397
 398         dma_unmap_single(dev, adapter->stats_token,
 399                          sizeof(struct ibmvnic_statistics),
 400                          DMA_FROM_DEVICE);
 401         adapter->stats_token = 0;
 402 }
 403
 404 static int init_stats_token(struct ibmvnic_adapter *adapter)
 405 {
 406         struct device *dev = &adapter->vdev->dev;
 407         dma_addr_t stok;
 408
 409         stok = dma_map_single(dev, &adapter->stats,
 410                               sizeof(struct ibmvnic_statistics),
 411                               DMA_FROM_DEVICE);
 412         if (dma_mapping_error(dev, stok)) {
 413                 dev_err(dev, "Couldn't map stats buffer\n");
 414                 return -1;
 415         }
 416
 417         adapter->stats_token = stok;
 418         netdev_dbg(adapter->netdev, "Stats token initialized (%llx)\n", stok);
 419         return 0;
 420 }
 421
 422 static int reset_rx_pools(struct ibmvnic_adapter *adapter)
 423 {
 424         struct ibmvnic_rx_pool *rx_pool;
 425         int rx_scrqs;
 426         int i, j, rc;
 427         u64 *size_array;
 428
 429         size_array = (u64 *)((u8 *)(adapter->login_rsp_buf) +
 430                 be32_to_cpu(adapter->login_rsp_buf->off_rxadd_buff_size));
 431
 432         rx_scrqs = be32_to_cpu(adapter->login_rsp_buf->num_rxadd_subcrqs);
 433         for (i = 0; i < rx_scrqs; i++) {
 434                 rx_pool = &adapter->rx_pool[i];
 435
 436                 netdev_dbg(adapter->netdev, "Re-setting rx_pool[%d]\n", i);
 437
 438                 if (rx_pool->buff_size != be64_to_cpu(size_array[i])) {
 439                         free_long_term_buff(adapter, &rx_pool->long_term_buff);
 440                         rx_pool->buff_size = be64_to_cpu(size_array[i]);
 441                         alloc_long_term_buff(adapter, &rx_pool->long_term_buff,
 442                                              rx_pool->size *
 443                                              rx_pool->buff_size);
 444                 } else {
 445                         rc = reset_long_term_buff(adapter,
 446                                                   &rx_pool->long_term_buff);
 447                 }
 448
 449                 if (rc)
 450                         return rc;
 451
 452                 for (j = 0; j < rx_pool->size; j++)
 453                         rx_pool->free_map[j] = j;
 454
 455                 memset(rx_pool->rx_buff, 0,
 456                        rx_pool->size * sizeof(struct ibmvnic_rx_buff));
 457
 458                 atomic_set(&rx_pool->available, 0);
 459                 rx_pool->next_alloc = 0;
 460                 rx_pool->next_free = 0;
 461                 rx_pool->active = 1;
 462         }
 463
 464         return 0;
 465 }
 466
 467 static void release_rx_pools(struct ibmvnic_adapter *adapter)
 468 {
 469         struct ibmvnic_rx_pool *rx_pool;
 470         int i, j;
 471
 472         if (!adapter->rx_pool)
 473                 return;
 474
 475         for (i = 0; i < adapter->num_active_rx_pools; i++) {
 476                 rx_pool = &adapter->rx_pool[i];
 477
 478                 netdev_dbg(adapter->netdev, "Releasing rx_pool[%d]\n", i);
 479
 480                 kfree(rx_pool->free_map);
 481                 free_long_term_buff(adapter, &rx_pool->long_term_buff);
 482
 483                 if (!rx_pool->rx_buff)
 484                         continue;
 485
 486                 for (j = 0; j < rx_pool->size; j++) {
 487                         if (rx_pool->rx_buff[j].skb) {
 488                                 dev_kfree_skb_any(rx_pool->rx_buff[j].skb);
 489                                 rx_pool->rx_buff[j].skb = NULL;
 490                         }
 491                 }
 492
 493                 kfree(rx_pool->rx_buff);
 494         }
 495
 496         kfree(adapter->rx_pool);
 497         adapter->rx_pool = NULL;
 498         adapter->num_active_rx_pools = 0;
 499 }
 500
 501 static int init_rx_pools(struct net_device *netdev)
 502 {
 503         struct ibmvnic_adapter *adapter = netdev_priv(netdev);
 504         struct device *dev = &adapter->vdev->dev;
 505         struct ibmvnic_rx_pool *rx_pool;
 506         int rxadd_subcrqs;
 507         u64 *size_array;
 508         int i, j;
 509
 510         rxadd_subcrqs =
 511                 be32_to_cpu(adapter->login_rsp_buf->num_rxadd_subcrqs);
 512         size_array = (u64 *)((u8 *)(adapter->login_rsp_buf) +
 513                 be32_to_cpu(adapter->login_rsp_buf->off_rxadd_buff_size));
 514
 515         adapter->rx_pool = kcalloc(rxadd_subcrqs,
 516                                    sizeof(struct ibmvnic_rx_pool),
 517                                    GFP_KERNEL);
 518         if (!adapter->rx_pool) {
 519                 dev_err(dev, "Failed to allocate rx pools\n");
 520                 return -1;
 521         }
 522
 523         adapter->num_active_rx_pools = rxadd_subcrqs;
 524
 525         for (i = 0; i < rxadd_subcrqs; i++) {
 526                 rx_pool = &adapter->rx_pool[i];
 527
 528                 netdev_dbg(adapter->netdev,
 529                            "Initializing rx_pool[%d], %lld buffs, %lld bytes each\n",
 530                            i, adapter->req_rx_add_entries_per_subcrq,
 531                            be64_to_cpu(size_array[i]));
 532
 533                 rx_pool->size = adapter->req_rx_add_entries_per_subcrq;
 534                 rx_pool->index = i;
 535                 rx_pool->buff_size = be64_to_cpu(size_array[i]);
 536                 rx_pool->active = 1;
 537
 538                 rx_pool->free_map = kcalloc(rx_pool->size, sizeof(int),
 539                                             GFP_KERNEL);
 540                 if (!rx_pool->free_map) {
 541                         release_rx_pools(adapter);
 542                         return -1;
 543                 }
 544
 545                 rx_pool->rx_buff = kcalloc(rx_pool->size,
 546                                            sizeof(struct ibmvnic_rx_buff),
 547                                            GFP_KERNEL);
 548                 if (!rx_pool->rx_buff) {
 549                         dev_err(dev, "Couldn't alloc rx buffers\n");
 550                         release_rx_pools(adapter);
 551                         return -1;
 552                 }
 553
 554                 if (alloc_long_term_buff(adapter, &rx_pool->long_term_buff,
 555                                          rx_pool->size * rx_pool->buff_size)) {
 556                         release_rx_pools(adapter);
 557                         return -1;
 558                 }
 559
 560                 for (j = 0; j < rx_pool->size; ++j)
 561                         rx_pool->free_map[j] = j;
 562
 563                 atomic_set(&rx_pool->available, 0);
 564                 rx_pool->next_alloc = 0;
 565                 rx_pool->next_free = 0;
 566         }
 567
 568         return 0;
 569 }
 570
 571 static int reset_one_tx_pool(struct ibmvnic_adapter *adapter,
 572                              struct ibmvnic_tx_pool *tx_pool)
 573 {
 574         int rc, i;
 575
 576         rc = reset_long_term_buff(adapter, &tx_pool->long_term_buff);
 577         if (rc)
 578                 return rc;
 579
 580         memset(tx_pool->tx_buff, 0,
 581                tx_pool->num_buffers *
 582                sizeof(struct ibmvnic_tx_buff));
 583
 584         for (i = 0; i < tx_pool->num_buffers; i++)
 585                 tx_pool->free_map[i] = i;
 586
 587         tx_pool->consumer_index = 0;
 588         tx_pool->producer_index = 0;
 589
 590         return 0;
 591 }
 592
 593 static int reset_tx_pools(struct ibmvnic_adapter *adapter)
 594 {
 595         int tx_scrqs;
 596         int i, rc;
 597
 598         tx_scrqs = be32_to_cpu(adapter->login_rsp_buf->num_txsubm_subcrqs);
 599         for (i = 0; i < tx_scrqs; i++) {
 600                 rc = reset_one_tx_pool(adapter, &adapter->tso_pool[i]);
 601                 if (rc)
 602                         return rc;
 603                 rc = reset_one_tx_pool(adapter, &adapter->tx_pool[i]);
 604                 if (rc)
 605                         return rc;
 606         }
 607
 608         return 0;
 609 }
 610
 611 static void release_vpd_data(struct ibmvnic_adapter *adapter)
 612 {
 613         if (!adapter->vpd)
 614                 return;
 615
 616         kfree(adapter->vpd->buff);
 617         kfree(adapter->vpd);
 618
 619         adapter->vpd = NULL;
 620 }
 621
 622 static void release_one_tx_pool(struct ibmvnic_adapter *adapter,
 623                                 struct ibmvnic_tx_pool *tx_pool)
 624 {
 625         kfree(tx_pool->tx_buff);
 626         kfree(tx_pool->free_map);
 627         free_long_term_buff(adapter, &tx_pool->long_term_buff);
 628 }
 629
 630 static void release_tx_pools(struct ibmvnic_adapter *adapter)
 631 {
 632         int i;
 633
 634         if (!adapter->tx_pool)
 635                 return;
 636
 637         for (i = 0; i < adapter->num_active_tx_pools; i++) {
 638                 release_one_tx_pool(adapter, &adapter->tx_pool[i]);
 639                 release_one_tx_pool(adapter, &adapter->tso_pool[i]);
 640         }
 641
 642         kfree(adapter->tx_pool);
 643         adapter->tx_pool = NULL;
 644         kfree(adapter->tso_pool);
 645         adapter->tso_pool = NULL;
 646         adapter->num_active_tx_pools = 0;
 647 }
 648
 649 static int init_one_tx_pool(struct net_device *netdev,
 650                             struct ibmvnic_tx_pool *tx_pool,
 651                             int num_entries, int buf_size)
 652 {
 653         struct ibmvnic_adapter *adapter = netdev_priv(netdev);
 654         int i;
 655
 656         tx_pool->tx_buff = kcalloc(num_entries,
 657                                    sizeof(struct ibmvnic_tx_buff),
 658                                    GFP_KERNEL);
 659         if (!tx_pool->tx_buff)
 660                 return -1;
 661
 662         if (alloc_long_term_buff(adapter, &tx_pool->long_term_buff,
 663                                  num_entries * buf_size))
 664                 return -1;
 665
 666         tx_pool->free_map = kcalloc(num_entries, sizeof(int), GFP_KERNEL);
 667         if (!tx_pool->free_map)
 668                 return -1;
 669
 670         for (i = 0; i < num_entries; i++)
 671                 tx_pool->free_map[i] = i;
 672
 673         tx_pool->consumer_index = 0;
 674         tx_pool->producer_index = 0;
 675         tx_pool->num_buffers = num_entries;
 676         tx_pool->buf_size = buf_size;
 677
 678         return 0;
 679 }
 680
 681 static int init_tx_pools(struct net_device *netdev)
 682 {
 683         struct ibmvnic_adapter *adapter = netdev_priv(netdev);
 684         int tx_subcrqs;
 685         int i, rc;
 686
 687         tx_subcrqs = be32_to_cpu(adapter->login_rsp_buf->num_txsubm_subcrqs);
 688         adapter->tx_pool = kcalloc(tx_subcrqs,
 689                                    sizeof(struct ibmvnic_tx_pool), GFP_KERNEL);
 690         if (!adapter->tx_pool)
 691                 return -1;
 692
 693         adapter->tso_pool = kcalloc(tx_subcrqs,
 694                                     sizeof(struct ibmvnic_tx_pool), GFP_KERNEL);
 695         if (!adapter->tso_pool)
 696                 return -1;
 697
 698         adapter->num_active_tx_pools = tx_subcrqs;
 699
 700         for (i = 0; i < tx_subcrqs; i++) {
 701                 rc = init_one_tx_pool(netdev, &adapter->tx_pool[i],
 702                                       adapter->req_tx_entries_per_subcrq,
 703                                       adapter->req_mtu + VLAN_HLEN);
 704                 if (rc) {
 705                         release_tx_pools(adapter);
 706                         return rc;
 707                 }
 708
 709                 init_one_tx_pool(netdev, &adapter->tso_pool[i],
 710                                  IBMVNIC_TSO_BUFS,
 711                                  IBMVNIC_TSO_BUF_SZ);
 712                 if (rc) {
 713                         release_tx_pools(adapter);
 714                         return rc;
 715                 }
 716         }
 717
 718         return 0;
 719 }
 720
 721 static void ibmvnic_napi_enable(struct ibmvnic_adapter *adapter)
 722 {
 723         int i;
 724
 725         if (adapter->napi_enabled)
 726                 return;
 727
 728         for (i = 0; i < adapter->req_rx_queues; i++)
 729                 napi_enable(&adapter->napi[i]);
 730
 731         adapter->napi_enabled = true;
 732 }
 733
 734 static void ibmvnic_napi_disable(struct ibmvnic_adapter *adapter)
 735 {
 736         int i;
 737
 738         if (!adapter->napi_enabled)
 739                 return;
 740
 741         for (i = 0; i < adapter->req_rx_queues; i++) {
 742                 netdev_dbg(adapter->netdev, "Disabling napi[%d]\n", i);
 743                 napi_disable(&adapter->napi[i]);
 744         }
 745
 746         adapter->napi_enabled = false;
 747 }
 748
 749 static int init_napi(struct ibmvnic_adapter *adapter)
 750 {
 751         int i;
 752
 753         adapter->napi = kcalloc(adapter->req_rx_queues,
 754                                 sizeof(struct napi_struct), GFP_KERNEL);
 755         if (!adapter->napi)
 756                 return -ENOMEM;
 757
 758         for (i = 0; i < adapter->req_rx_queues; i++) {
 759                 netdev_dbg(adapter->netdev, "Adding napi[%d]\n", i);
 760                 netif_napi_add(adapter->netdev, &adapter->napi[i],
 761                                ibmvnic_poll, NAPI_POLL_WEIGHT);
 762         }
 763
 764         adapter->num_active_rx_napi = adapter->req_rx_queues;
 765         return 0;
 766 }
 767
 768 static void release_napi(struct ibmvnic_adapter *adapter)
 769 {
 770         int i;
 771
 772         if (!adapter->napi)
 773                 return;
 774
 775         for (i = 0; i < adapter->num_active_rx_napi; i++) {
 776                 netdev_dbg(adapter->netdev, "Releasing napi[%d]\n", i);
 777                 netif_napi_del(&adapter->napi[i]);
 778         }
 779
 780         kfree(adapter->napi);
 781         adapter->napi = NULL;
 782         adapter->num_active_rx_napi = 0;
 783         adapter->napi_enabled = false;
 784 }
 785
 786 static int ibmvnic_login(struct net_device *netdev)
 787 {
 788         struct ibmvnic_adapter *adapter = netdev_priv(netdev);
 789         unsigned long timeout = msecs_to_jiffies(30000);
 790         int retry_count = 0;
 791         bool retry;
 792         int rc;
 793
 794         do {
 795                 retry = false;
 796                 if (retry_count > IBMVNIC_MAX_QUEUES) {
 797                         netdev_warn(netdev, "Login attempts exceeded\n");
 798                         return -1;
 799                 }
 800
 801                 adapter->init_done_rc = 0;
 802                 reinit_completion(&adapter->init_done);
 803                 rc = send_login(adapter);
 804                 if (rc) {
 805                         netdev_warn(netdev, "Unable to login\n");
 806                         return rc;
 807                 }
 808
 809                 if (!wait_for_completion_timeout(&adapter->init_done,
 810                                                  timeout)) {
 811                         netdev_warn(netdev, "Login timed out\n");
 812                         return -1;
 813                 }
 814
 815                 if (adapter->init_done_rc == PARTIALSUCCESS) {
 816                         retry_count++;
 817                         release_sub_crqs(adapter, 1);
 818
 819                         retry = true;
 820                         netdev_dbg(netdev,
 821                                    "Received partial success, retrying...\n");
 822                         adapter->init_done_rc = 0;
 823                         reinit_completion(&adapter->init_done);
 824                         send_cap_queries(adapter);
 825                         if (!wait_for_completion_timeout(&adapter->init_done,
 826                                                          timeout)) {
 827                                 netdev_warn(netdev,
 828                                             "Capabilities query timed out\n");
 829                                 return -1;
 830                         }
 831
 832                         rc = init_sub_crqs(adapter);
 833                         if (rc) {
 834                                 netdev_warn(netdev,
 835                                             "SCRQ initialization failed\n");
 836                                 return -1;
 837                         }
 838
 839                         rc = init_sub_crq_irqs(adapter);
 840                         if (rc) {
 841                                 netdev_warn(netdev,
 842                                             "SCRQ irq initialization failed\n");
 843                                 return -1;
 844                         }
 845                 } else if (adapter->init_done_rc) {
 846                         netdev_warn(netdev, "Adapter login failed\n");
 847                         return -1;
 848                 }
 849         } while (retry);
 850
 851         /* handle pending MAC address changes after successful login */
 852         if (adapter->mac_change_pending) {
 853                 __ibmvnic_set_mac(netdev, &adapter->desired.mac);
 854                 adapter->mac_change_pending = false;
 855         }
 856
 857         return 0;
 858 }
 859
 860 static void release_login_buffer(struct ibmvnic_adapter *adapter)
 861 {
 862         kfree(adapter->login_buf);
 863         adapter->login_buf = NULL;
 864 }
 865
 866 static void release_login_rsp_buffer(struct ibmvnic_adapter *adapter)
 867 {
 868         kfree(adapter->login_rsp_buf);
 869         adapter->login_rsp_buf = NULL;
 870 }
 871
 872 static void release_resources(struct ibmvnic_adapter *adapter)
 873 {
 874         release_vpd_data(adapter);
 875
 876         release_tx_pools(adapter);
 877         release_rx_pools(adapter);
 878
 879         release_napi(adapter);
 880         release_login_rsp_buffer(adapter);
 881 }
 882
 883 static int set_link_state(struct ibmvnic_adapter *adapter, u8 link_state)
 884 {
 885         struct net_device *netdev = adapter->netdev;
 886         unsigned long timeout = msecs_to_jiffies(30000);
 887         union ibmvnic_crq crq;
 888         bool resend;
 889         int rc;
 890
 891         netdev_dbg(netdev, "setting link state %d\n", link_state);
 892
 893         memset(&crq, 0, sizeof(crq));
 894         crq.logical_link_state.first = IBMVNIC_CRQ_CMD;
 895         crq.logical_link_state.cmd = LOGICAL_LINK_STATE;
 896         crq.logical_link_state.link_state = link_state;
 897
 898         do {
 899                 resend = false;
 900
 901                 reinit_completion(&adapter->init_done);
 902                 rc = ibmvnic_send_crq(adapter, &crq);
 903                 if (rc) {
 904                         netdev_err(netdev, "Failed to set link state\n");
 905                         return rc;
 906                 }
 907
 908                 if (!wait_for_completion_timeout(&adapter->init_done,
 909                                                  timeout)) {
 910                         netdev_err(netdev, "timeout setting link state\n");
 911                         return -1;
 912                 }
 913
 914                 if (adapter->init_done_rc == 1) {
 915                         /* Partuial success, delay and re-send */
 916                         mdelay(1000);
 917                         resend = true;
 918                 } else if (adapter->init_done_rc) {
 919                         netdev_warn(netdev, "Unable to set link state, rc=%d\n",
 920                                     adapter->init_done_rc);
 921                         return adapter->init_done_rc;
 922                 }
 923         } while (resend);
 924
 925         return 0;
 926 }
 927
 928 static int set_real_num_queues(struct net_device *netdev)
 929 {
 930         struct ibmvnic_adapter *adapter = netdev_priv(netdev);
 931         int rc;
 932
 933         netdev_dbg(netdev, "Setting real tx/rx queues (%llx/%llx)\n",
 934                    adapter->req_tx_queues, adapter->req_rx_queues);
 935
 936         rc = netif_set_real_num_tx_queues(netdev, adapter->req_tx_queues);
 937         if (rc) {
 938                 netdev_err(netdev, "failed to set the number of tx queues\n");
 939                 return rc;
 940         }
 941
 942         rc = netif_set_real_num_rx_queues(netdev, adapter->req_rx_queues);
 943         if (rc)
 944                 netdev_err(netdev, "failed to set the number of rx queues\n");
 945
 946         return rc;
 947 }
 948
 949 static int ibmvnic_get_vpd(struct ibmvnic_adapter *adapter)
 950 {
 951         struct device *dev = &adapter->vdev->dev;
 952         union ibmvnic_crq crq;
 953         int len = 0;
 954         int rc;
 955
 956         if (adapter->vpd->buff)
 957                 len = adapter->vpd->len;
 958
 959         init_completion(&adapter->fw_done);
 960         crq.get_vpd_size.first = IBMVNIC_CRQ_CMD;
 961         crq.get_vpd_size.cmd = GET_VPD_SIZE;
 962         rc = ibmvnic_send_crq(adapter, &crq);
 963         if (rc)
 964                 return rc;
 965         wait_for_completion(&adapter->fw_done);
 966
 967         if (!adapter->vpd->len)
 968                 return -ENODATA;
 969
 970         if (!adapter->vpd->buff)
 971                 adapter->vpd->buff = kzalloc(adapter->vpd->len, GFP_KERNEL);
 972         else if (adapter->vpd->len != len)
 973                 adapter->vpd->buff =
 974                         krealloc(adapter->vpd->buff,
 975                                  adapter->vpd->len, GFP_KERNEL);
 976
 977         if (!adapter->vpd->buff) {
 978                 dev_err(dev, "Could allocate VPD buffer\n");
 979                 return -ENOMEM;
 980         }
 981
 982         adapter->vpd->dma_addr =
 983                 dma_map_single(dev, adapter->vpd->buff, adapter->vpd->len,
 984                                DMA_FROM_DEVICE);
 985         if (dma_mapping_error(dev, adapter->vpd->dma_addr)) {
 986                 dev_err(dev, "Could not map VPD buffer\n");
 987                 kfree(adapter->vpd->buff);
 988                 adapter->vpd->buff = NULL;
 989                 return -ENOMEM;
 990         }
 991
 992         reinit_completion(&adapter->fw_done);
 993         crq.get_vpd.first = IBMVNIC_CRQ_CMD;
 994         crq.get_vpd.cmd = GET_VPD;
 995         crq.get_vpd.ioba = cpu_to_be32(adapter->vpd->dma_addr);
 996         crq.get_vpd.len = cpu_to_be32((u32)adapter->vpd->len);
 997         rc = ibmvnic_send_crq(adapter, &crq);
 998         if (rc) {
 999                 kfree(adapter->vpd->buff);
1000                 adapter->vpd->buff = NULL;
1001                 return rc;
1002         }
1003         wait_for_completion(&adapter->fw_done);
1004
1005         return 0;
1006 }
1007
1008 static int init_resources(struct ibmvnic_adapter *adapter)
1009 {
1010         struct net_device *netdev = adapter->netdev;
1011         int rc;
1012
1013         rc = set_real_num_queues(netdev);
1014         if (rc)
1015                 return rc;
1016
1017         adapter->vpd = kzalloc(sizeof(*adapter->vpd), GFP_KERNEL);
1018         if (!adapter->vpd)
1019                 return -ENOMEM;
1020
1021         /* Vital Product Data (VPD) */
1022         rc = ibmvnic_get_vpd(adapter);
1023         if (rc) {
1024                 netdev_err(netdev, "failed to initialize Vital Product Data (VPD)\n");
1025                 return rc;
1026         }
1027
1028         adapter->map_id = 1;
1029
1030         rc = init_napi(adapter);
1031         if (rc)
1032                 return rc;
1033
1034         send_map_query(adapter);
1035
1036         rc = init_rx_pools(netdev);
1037         if (rc)
1038                 return rc;
1039
1040         rc = init_tx_pools(netdev);
1041         return rc;
1042 }
1043
1044 static int __ibmvnic_open(struct net_device *netdev)
1045 {
1046         struct ibmvnic_adapter *adapter = netdev_priv(netdev);
1047         enum vnic_state prev_state = adapter->state;
1048         int i, rc;
1049
1050         adapter->state = VNIC_OPENING;
1051         replenish_pools(adapter);
1052         ibmvnic_napi_enable(adapter);
1053
1054         /* We're ready to receive frames, enable the sub-crq interrupts and
1055          * set the logical link state to up
1056          */
1057         for (i = 0; i < adapter->req_rx_queues; i++) {
1058                 netdev_dbg(netdev, "Enabling rx_scrq[%d] irq\n", i);
1059                 if (prev_state == VNIC_CLOSED)
1060                         enable_irq(adapter->rx_scrq[i]->irq);
1061                 enable_scrq_irq(adapter, adapter->rx_scrq[i]);
1062         }
1063
1064         for (i = 0; i < adapter->req_tx_queues; i++) {
1065                 netdev_dbg(netdev, "Enabling tx_scrq[%d] irq\n", i);
1066                 if (prev_state == VNIC_CLOSED)
1067                         enable_irq(adapter->tx_scrq[i]->irq);
1068                 enable_scrq_irq(adapter, adapter->tx_scrq[i]);
1069         }
1070
1071         rc = set_link_state(adapter, IBMVNIC_LOGICAL_LNK_UP);
1072         if (rc) {
1073                 for (i = 0; i < adapter->req_rx_queues; i++)
1074                         napi_disable(&adapter->napi[i]);
1075                 release_resources(adapter);
1076                 return rc;
1077         }
1078
1079         netif_tx_start_all_queues(netdev);
1080
1081         if (prev_state == VNIC_CLOSED) {
1082                 for (i = 0; i < adapter->req_rx_queues; i++)
1083                         napi_schedule(&adapter->napi[i]);
1084         }
1085
1086         adapter->state = VNIC_OPEN;
1087         return rc;
1088 }
1089
1090 static int ibmvnic_open(struct net_device *netdev)
1091 {
1092         struct ibmvnic_adapter *adapter = netdev_priv(netdev);
1093         int rc;
1094
1095         /* If device failover is pending, just set device state and return.
1096          * Device operation will be handled by reset routine.
1097          */
1098         if (adapter->failover_pending) {
1099                 adapter->state = VNIC_OPEN;
1100                 return 0;
1101         }
1102
1103         if (adapter->state != VNIC_CLOSED) {
1104                 rc = ibmvnic_login(netdev);
1105                 if (rc)
1106                         return rc;
1107
1108                 rc = init_resources(adapter);
1109                 if (rc) {
1110                         netdev_err(netdev, "failed to initialize resources\n");
1111                         release_resources(adapter);
1112                         return rc;
1113                 }
1114         }
1115
1116         rc = __ibmvnic_open(netdev);
1117         netif_carrier_on(netdev);
1118
1119         return rc;
1120 }
1121
1122 static void clean_rx_pools(struct ibmvnic_adapter *adapter)
1123 {
1124         struct ibmvnic_rx_pool *rx_pool;
1125         struct ibmvnic_rx_buff *rx_buff;
1126         u64 rx_entries;
1127         int rx_scrqs;
1128         int i, j;
1129
1130         if (!adapter->rx_pool)
1131                 return;
1132
1133         rx_scrqs = adapter->num_active_rx_pools;
1134         rx_entries = adapter->req_rx_add_entries_per_subcrq;
1135
1136         /* Free any remaining skbs in the rx buffer pools */
1137         for (i = 0; i < rx_scrqs; i++) {
1138                 rx_pool = &adapter->rx_pool[i];
1139                 if (!rx_pool || !rx_pool->rx_buff)
1140                         continue;
1141
1142                 netdev_dbg(adapter->netdev, "Cleaning rx_pool[%d]\n", i);
1143                 for (j = 0; j < rx_entries; j++) {
1144                         rx_buff = &rx_pool->rx_buff[j];
1145                         if (rx_buff && rx_buff->skb) {
1146                                 dev_kfree_skb_any(rx_buff->skb);
1147                                 rx_buff->skb = NULL;
1148                         }
1149                 }
1150         }
1151 }
1152
1153 static void clean_one_tx_pool(struct ibmvnic_adapter *adapter,
1154                               struct ibmvnic_tx_pool *tx_pool)
1155 {
1156         struct ibmvnic_tx_buff *tx_buff;
1157         u64 tx_entries;
1158         int i;
1159
1160         if (!tx_pool || !tx_pool->tx_buff)
1161                 return;
1162
1163         tx_entries = tx_pool->num_buffers;
1164
1165         for (i = 0; i < tx_entries; i++) {
1166                 tx_buff = &tx_pool->tx_buff[i];
1167                 if (tx_buff && tx_buff->skb) {
1168                         dev_kfree_skb_any(tx_buff->skb);
1169                         tx_buff->skb = NULL;
1170                 }
1171         }
1172 }
1173
1174 static void clean_tx_pools(struct ibmvnic_adapter *adapter)
1175 {
1176         int tx_scrqs;
1177         int i;
1178
1179         if (!adapter->tx_pool || !adapter->tso_pool)
1180                 return;
1181
1182         tx_scrqs = adapter->num_active_tx_pools;
1183
1184         /* Free any remaining skbs in the tx buffer pools */
1185         for (i = 0; i < tx_scrqs; i++) {
1186                 netdev_dbg(adapter->netdev, "Cleaning tx_pool[%d]\n", i);
1187                 clean_one_tx_pool(adapter, &adapter->tx_pool[i]);
1188                 clean_one_tx_pool(adapter, &adapter->tso_pool[i]);
1189         }
1190 }
1191
1192 static void ibmvnic_disable_irqs(struct ibmvnic_adapter *adapter)
1193 {
1194         struct net_device *netdev = adapter->netdev;
1195         int i;
1196
1197         if (adapter->tx_scrq) {
1198                 for (i = 0; i < adapter->req_tx_queues; i++)
1199                         if (adapter->tx_scrq[i]->irq) {
1200                                 netdev_dbg(netdev,
1201                                            "Disabling tx_scrq[%d] irq\n", i);
1202                                 disable_scrq_irq(adapter, adapter->tx_scrq[i]);
1203                                 disable_irq(adapter->tx_scrq[i]->irq);
1204                         }
1205         }
1206
1207         if (adapter->rx_scrq) {
1208                 for (i = 0; i < adapter->req_rx_queues; i++) {
1209                         if (adapter->rx_scrq[i]->irq) {
1210                                 netdev_dbg(netdev,
1211                                            "Disabling rx_scrq[%d] irq\n", i);
1212                                 disable_scrq_irq(adapter, adapter->rx_scrq[i]);
1213                                 disable_irq(adapter->rx_scrq[i]->irq);
1214                         }
1215                 }
1216         }
1217 }
1218
1219 static void ibmvnic_cleanup(struct net_device *netdev)
1220 {
1221         struct ibmvnic_adapter *adapter = netdev_priv(netdev);
1222
1223         /* ensure that transmissions are stopped if called by do_reset */
1224         if (adapter->resetting)
1225                 netif_tx_disable(netdev);
1226         else
1227                 netif_tx_stop_all_queues(netdev);
1228
1229         ibmvnic_napi_disable(adapter);
1230         ibmvnic_disable_irqs(adapter);
1231
1232         clean_rx_pools(adapter);
1233         clean_tx_pools(adapter);
1234 }
1235
1236 static int __ibmvnic_close(struct net_device *netdev)
1237 {
1238         struct ibmvnic_adapter *adapter = netdev_priv(netdev);
1239         int rc = 0;
1240
1241         adapter->state = VNIC_CLOSING;
1242         rc = set_link_state(adapter, IBMVNIC_LOGICAL_LNK_DN);
1243         if (rc)
1244                 return rc;
1245         adapter->state = VNIC_CLOSED;
1246         return 0;
1247 }
1248
1249 static int ibmvnic_close(struct net_device *netdev)
1250 {
1251         struct ibmvnic_adapter *adapter = netdev_priv(netdev);
1252         int rc;
1253
1254         /* If device failover is pending, just set device state and return.
1255          * Device operation will be handled by reset routine.
1256          */
1257         if (adapter->failover_pending) {
1258                 adapter->state = VNIC_CLOSED;
1259                 return 0;
1260         }
1261
1262         rc = __ibmvnic_close(netdev);
1263         ibmvnic_cleanup(netdev);
1264
1265         return rc;
1266 }
1267
1268 /**
1269  * build_hdr_data - creates L2/L3/L4 header data buffer
1270  * @hdr_field - bitfield determining needed headers
1271  * @skb - socket buffer
1272  * @hdr_len - array of header lengths
1273  * @tot_len - total length of data
1274  *
1275  * Reads hdr_field to determine which headers are needed by firmware.
1276  * Builds a buffer containing these headers.  Saves individual header
1277  * lengths and total buffer length to be used to build descriptors.
1278  */
1279 static int build_hdr_data(u8 hdr_field, struct sk_buff *skb,
1280                           int *hdr_len, u8 *hdr_data)
1281 {
1282         int len = 0;
1283         u8 *hdr;
1284
1285         if (skb_vlan_tagged(skb) && !skb_vlan_tag_present(skb))
1286                 hdr_len[0] = sizeof(struct vlan_ethhdr);
1287         else
1288                 hdr_len[0] = sizeof(struct ethhdr);
1289
1290         if (skb->protocol == htons(ETH_P_IP)) {
1291                 hdr_len[1] = ip_hdr(skb)->ihl * 4;
1292                 if (ip_hdr(skb)->protocol == IPPROTO_TCP)
1293                         hdr_len[2] = tcp_hdrlen(skb);
1294                 else if (ip_hdr(skb)->protocol == IPPROTO_UDP)
1295                         hdr_len[2] = sizeof(struct udphdr);
1296         } else if (skb->protocol == htons(ETH_P_IPV6)) {
1297                 hdr_len[1] = sizeof(struct ipv6hdr);
1298                 if (ipv6_hdr(skb)->nexthdr == IPPROTO_TCP)
1299                         hdr_len[2] = tcp_hdrlen(skb);
1300                 else if (ipv6_hdr(skb)->nexthdr == IPPROTO_UDP)
1301                         hdr_len[2] = sizeof(struct udphdr);
1302         } else if (skb->protocol == htons(ETH_P_ARP)) {
1303                 hdr_len[1] = arp_hdr_len(skb->dev);
1304                 hdr_len[2] = 0;
1305         }
1306
1307         memset(hdr_data, 0, 120);
1308         if ((hdr_field >> 6) & 1) {
1309                 hdr = skb_mac_header(skb);
1310                 memcpy(hdr_data, hdr, hdr_len[0]);
1311                 len += hdr_len[0];
1312         }
1313
1314         if ((hdr_field >> 5) & 1) {
1315                 hdr = skb_network_header(skb);
1316                 memcpy(hdr_data + len, hdr, hdr_len[1]);
1317                 len += hdr_len[1];
1318         }
1319
1320         if ((hdr_field >> 4) & 1) {
1321                 hdr = skb_transport_header(skb);
1322                 memcpy(hdr_data + len, hdr, hdr_len[2]);
1323                 len += hdr_len[2];
1324         }
1325         return len;
1326 }
1327
1328 /**
1329  * create_hdr_descs - create header and header extension descriptors
1330  * @hdr_field - bitfield determining needed headers
1331  * @data - buffer containing header data
1332  * @len - length of data buffer
1333  * @hdr_len - array of individual header lengths
1334  * @scrq_arr - descriptor array
1335  *
1336  * Creates header and, if needed, header extension descriptors and
1337  * places them in a descriptor array, scrq_arr
1338  */
1339
1340 static int create_hdr_descs(u8 hdr_field, u8 *hdr_data, int len, int *hdr_len,
1341                             union sub_crq *scrq_arr)
1342 {
1343         union sub_crq hdr_desc;
1344         int tmp_len = len;
1345         int num_descs = 0;
1346         u8 *data, *cur;
1347         int tmp;
1348
1349         while (tmp_len > 0) {
1350                 cur = hdr_data + len - tmp_len;
1351
1352                 memset(&hdr_desc, 0, sizeof(hdr_desc));
1353                 if (cur != hdr_data) {
1354                         data = hdr_desc.hdr_ext.data;
1355                         tmp = tmp_len > 29 ? 29 : tmp_len;
1356                         hdr_desc.hdr_ext.first = IBMVNIC_CRQ_CMD;
1357                         hdr_desc.hdr_ext.type = IBMVNIC_HDR_EXT_DESC;
1358                         hdr_desc.hdr_ext.len = tmp;
1359                 } else {
1360                         data = hdr_desc.hdr.data;
1361                         tmp = tmp_len > 24 ? 24 : tmp_len;
1362                         hdr_desc.hdr.first = IBMVNIC_CRQ_CMD;
1363                         hdr_desc.hdr.type = IBMVNIC_HDR_DESC;
1364                         hdr_desc.hdr.len = tmp;
1365                         hdr_desc.hdr.l2_len = (u8)hdr_len[0];
1366                         hdr_desc.hdr.l3_len = cpu_to_be16((u16)hdr_len[1]);
1367                         hdr_desc.hdr.l4_len = (u8)hdr_len[2];
1368                         hdr_desc.hdr.flag = hdr_field << 1;
1369                 }
1370                 memcpy(data, cur, tmp);
1371                 tmp_len -= tmp;
1372                 *scrq_arr = hdr_desc;
1373                 scrq_arr++;
1374                 num_descs++;
1375         }
1376
1377         return num_descs;
1378 }
1379
1380 /**
1381  * build_hdr_descs_arr - build a header descriptor array
1382  * @skb - socket buffer
1383  * @num_entries - number of descriptors to be sent
1384  * @subcrq - first TX descriptor
1385  * @hdr_field - bit field determining which headers will be sent
1386  *
1387  * This function will build a TX descriptor array with applicable
1388  * L2/L3/L4 packet header descriptors to be sent by send_subcrq_indirect.
1389  */
1390
1391 static void build_hdr_descs_arr(struct ibmvnic_tx_buff *txbuff,
1392                                 int *num_entries, u8 hdr_field)
1393 {
1394         int hdr_len[3] = {0, 0, 0};
1395         int tot_len;
1396         u8 *hdr_data = txbuff->hdr_data;
1397
1398         tot_len = build_hdr_data(hdr_field, txbuff->skb, hdr_len,
1399                                  txbuff->hdr_data);
1400         *num_entries += create_hdr_descs(hdr_field, hdr_data, tot_len, hdr_len,
1401                          txbuff->indir_arr + 1);
1402 }
1403
1404 static int ibmvnic_xmit_workarounds(struct sk_buff *skb,
1405                                     struct net_device *netdev)
1406 {
1407         /* For some backing devices, mishandling of small packets
1408          * can result in a loss of connection or TX stall. Device
1409          * architects recommend that no packet should be smaller
1410          * than the minimum MTU value provided to the driver, so
1411          * pad any packets to that length
1412          */
1413         if (skb->len < netdev->min_mtu)
1414                 return skb_put_padto(skb, netdev->min_mtu);
1415
1416         return 0;
1417 }
1418
1419 static netdev_tx_t ibmvnic_xmit(struct sk_buff *skb, struct net_device *netdev)
1420 {
1421         struct ibmvnic_adapter *adapter = netdev_priv(netdev);
1422         int queue_num = skb_get_queue_mapping(skb);
1423         u8 *hdrs = (u8 *)&adapter->tx_rx_desc_req;
1424         struct device *dev = &adapter->vdev->dev;
1425         struct ibmvnic_tx_buff *tx_buff = NULL;
1426         struct ibmvnic_sub_crq_queue *tx_scrq;
1427         struct ibmvnic_tx_pool *tx_pool;
1428         unsigned int tx_send_failed = 0;
1429         unsigned int tx_map_failed = 0;
1430         unsigned int tx_dropped = 0;
1431         unsigned int tx_packets = 0;
1432         unsigned int tx_bytes = 0;
1433         dma_addr_t data_dma_addr;
1434         struct netdev_queue *txq;
1435         unsigned long lpar_rc;
1436         union sub_crq tx_crq;
1437         unsigned int offset;
1438         int num_entries = 1;
1439         unsigned char *dst;
1440         u64 *handle_array;
1441         int index = 0;
1442         u8 proto = 0;
1443         netdev_tx_t ret = NETDEV_TX_OK;
1444
1445         if (adapter->resetting) {
1446                 if (!netif_subqueue_stopped(netdev, skb))
1447                         netif_stop_subqueue(netdev, queue_num);
1448                 dev_kfree_skb_any(skb);
1449
1450                 tx_send_failed++;
1451                 tx_dropped++;
1452                 ret = NETDEV_TX_OK;
1453                 goto out;
1454         }
1455
1456         if (ibmvnic_xmit_workarounds(skb, netdev)) {
1457                 tx_dropped++;
1458                 tx_send_failed++;
1459                 ret = NETDEV_TX_OK;
1460                 goto out;
1461         }
1462         if (skb_is_gso(skb))
1463                 tx_pool = &adapter->tso_pool[queue_num];
1464         else
1465                 tx_pool = &adapter->tx_pool[queue_num];
1466
1467         tx_scrq = adapter->tx_scrq[queue_num];
1468         txq = netdev_get_tx_queue(netdev, skb_get_queue_mapping(skb));
1469         handle_array = (u64 *)((u8 *)(adapter->login_rsp_buf) +
1470                 be32_to_cpu(adapter->login_rsp_buf->off_txsubm_subcrqs));
1471
1472         index = tx_pool->free_map[tx_pool->consumer_index];
1473
1474         if (index == IBMVNIC_INVALID_MAP) {
1475                 dev_kfree_skb_any(skb);
1476                 tx_send_failed++;
1477                 tx_dropped++;
1478                 ret = NETDEV_TX_OK;
1479                 goto out;
1480         }
1481
1482         tx_pool->free_map[tx_pool->consumer_index] = IBMVNIC_INVALID_MAP;
1483
1484         offset = index * tx_pool->buf_size;
1485         dst = tx_pool->long_term_buff.buff + offset;
1486         memset(dst, 0, tx_pool->buf_size);
1487         data_dma_addr = tx_pool->long_term_buff.addr + offset;
1488
1489         if (skb_shinfo(skb)->nr_frags) {
1490                 int cur, i;
1491
1492                 /* Copy the head */
1493                 skb_copy_from_linear_data(skb, dst, skb_headlen(skb));
1494                 cur = skb_headlen(skb);
1495
1496                 /* Copy the frags */
1497                 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
1498                         const skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
1499
1500                         memcpy(dst + cur,
1501                                page_address(skb_frag_page(frag)) +
1502                                frag->page_offset, skb_frag_size(frag));
1503                         cur += skb_frag_size(frag);
1504                 }
1505         } else {
1506                 skb_copy_from_linear_data(skb, dst, skb->len);
1507         }
1508
1509         tx_pool->consumer_index =
1510             (tx_pool->consumer_index + 1) % tx_pool->num_buffers;
1511
1512         tx_buff = &tx_pool->tx_buff[index];
1513         tx_buff->skb = skb;
1514         tx_buff->data_dma[0] = data_dma_addr;
1515         tx_buff->data_len[0] = skb->len;
1516         tx_buff->index = index;
1517         tx_buff->pool_index = queue_num;
1518         tx_buff->last_frag = true;
1519
1520         memset(&tx_crq, 0, sizeof(tx_crq));
1521         tx_crq.v1.first = IBMVNIC_CRQ_CMD;
1522         tx_crq.v1.type = IBMVNIC_TX_DESC;
1523         tx_crq.v1.n_crq_elem = 1;
1524         tx_crq.v1.n_sge = 1;
1525         tx_crq.v1.flags1 = IBMVNIC_TX_COMP_NEEDED;
1526
1527         if (skb_is_gso(skb))
1528                 tx_crq.v1.correlator =
1529                         cpu_to_be32(index | IBMVNIC_TSO_POOL_MASK);
1530         else
1531                 tx_crq.v1.correlator = cpu_to_be32(index);
1532         tx_crq.v1.dma_reg = cpu_to_be16(tx_pool->long_term_buff.map_id);
1533         tx_crq.v1.sge_len = cpu_to_be32(skb->len);
1534         tx_crq.v1.ioba = cpu_to_be64(data_dma_addr);
1535
1536         if (adapter->vlan_header_insertion && skb_vlan_tag_present(skb)) {
1537                 tx_crq.v1.flags2 |= IBMVNIC_TX_VLAN_INSERT;
1538                 tx_crq.v1.vlan_id = cpu_to_be16(skb->vlan_tci);
1539         }
1540
1541         if (skb->protocol == htons(ETH_P_IP)) {
1542                 tx_crq.v1.flags1 |= IBMVNIC_TX_PROT_IPV4;
1543                 proto = ip_hdr(skb)->protocol;
1544         } else if (skb->protocol == htons(ETH_P_IPV6)) {
1545                 tx_crq.v1.flags1 |= IBMVNIC_TX_PROT_IPV6;
1546                 proto = ipv6_hdr(skb)->nexthdr;
1547         }
1548
1549         if (proto == IPPROTO_TCP)
1550                 tx_crq.v1.flags1 |= IBMVNIC_TX_PROT_TCP;
1551         else if (proto == IPPROTO_UDP)
1552                 tx_crq.v1.flags1 |= IBMVNIC_TX_PROT_UDP;
1553
1554         if (skb->ip_summed == CHECKSUM_PARTIAL) {
1555                 tx_crq.v1.flags1 |= IBMVNIC_TX_CHKSUM_OFFLOAD;
1556                 hdrs += 2;
1557         }
1558         if (skb_is_gso(skb)) {
1559                 tx_crq.v1.flags1 |= IBMVNIC_TX_LSO;
1560                 tx_crq.v1.mss = cpu_to_be16(skb_shinfo(skb)->gso_size);
1561                 hdrs += 2;
1562         }
1563         /* determine if l2/3/4 headers are sent to firmware */
1564         if ((*hdrs >> 7) & 1) {
1565                 build_hdr_descs_arr(tx_buff, &num_entries, *hdrs);
1566                 tx_crq.v1.n_crq_elem = num_entries;
1567                 tx_buff->num_entries = num_entries;
1568                 tx_buff->indir_arr[0] = tx_crq;
1569                 tx_buff->indir_dma = dma_map_single(dev, tx_buff->indir_arr,
1570                                                     sizeof(tx_buff->indir_arr),
1571                                                     DMA_TO_DEVICE);
1572                 if (dma_mapping_error(dev, tx_buff->indir_dma)) {
1573                         dev_kfree_skb_any(skb);
1574                         tx_buff->skb = NULL;
1575                         if (!firmware_has_feature(FW_FEATURE_CMO))
1576                                 dev_err(dev, "tx: unable to map descriptor array\n");
1577                         tx_map_failed++;
1578                         tx_dropped++;
1579                         ret = NETDEV_TX_OK;
1580                         goto tx_err_out;
1581                 }
1582                 lpar_rc = send_subcrq_indirect(adapter, handle_array[queue_num],
1583                                                (u64)tx_buff->indir_dma,
1584                                                (u64)num_entries);
1585         } else {
1586                 tx_buff->num_entries = num_entries;
1587                 lpar_rc = send_subcrq(adapter, handle_array[queue_num],
1588                                       &tx_crq);
1589         }
1590         if (lpar_rc != H_SUCCESS) {
1591                 if (lpar_rc != H_CLOSED && lpar_rc != H_PARAMETER)
1592                         dev_err_ratelimited(dev, "tx: send failed\n");
1593                 dev_kfree_skb_any(skb);
1594                 tx_buff->skb = NULL;
1595
1596                 if (lpar_rc == H_CLOSED || adapter->failover_pending) {
1597                         /* Disable TX and report carrier off if queue is closed
1598                          * or pending failover.
1599                          * Firmware guarantees that a signal will be sent to the
1600                          * driver, triggering a reset or some other action.
1601                          */
1602                         netif_tx_stop_all_queues(netdev);
1603                         netif_carrier_off(netdev);
1604                 }
1605
1606                 tx_send_failed++;
1607                 tx_dropped++;
1608                 ret = NETDEV_TX_OK;
1609                 goto tx_err_out;
1610         }
1611
1612         if (atomic_add_return(num_entries, &tx_scrq->used)
1613                                         >= adapter->req_tx_entries_per_subcrq) {
1614                 netdev_dbg(netdev, "Stopping queue %d\n", queue_num);
1615                 netif_stop_subqueue(netdev, queue_num);
1616         }
1617
1618         tx_packets++;
1619         tx_bytes += skb->len;
1620         txq->trans_start = jiffies;
1621         ret = NETDEV_TX_OK;
1622         goto out;
1623
1624 tx_err_out:
1625         /* roll back consumer index and map array*/
1626         if (tx_pool->consumer_index == 0)
1627                 tx_pool->consumer_index =
1628                         tx_pool->num_buffers - 1;
1629         else
1630                 tx_pool->consumer_index--;
1631         tx_pool->free_map[tx_pool->consumer_index] = index;
1632 out:
1633         netdev->stats.tx_dropped += tx_dropped;
1634         netdev->stats.tx_bytes += tx_bytes;
1635         netdev->stats.tx_packets += tx_packets;
1636         adapter->tx_send_failed += tx_send_failed;
1637         adapter->tx_map_failed += tx_map_failed;
1638         adapter->tx_stats_buffers[queue_num].packets += tx_packets;
1639         adapter->tx_stats_buffers[queue_num].bytes += tx_bytes;
1640         adapter->tx_stats_buffers[queue_num].dropped_packets += tx_dropped;
1641
1642         return ret;
1643 }
1644
1645 static void ibmvnic_set_multi(struct net_device *netdev)
1646 {
1647         struct ibmvnic_adapter *adapter = netdev_priv(netdev);
1648         struct netdev_hw_addr *ha;
1649         union ibmvnic_crq crq;
1650
1651         memset(&crq, 0, sizeof(crq));
1652         crq.request_capability.first = IBMVNIC_CRQ_CMD;
1653         crq.request_capability.cmd = REQUEST_CAPABILITY;
1654
1655         if (netdev->flags & IFF_PROMISC) {
1656                 if (!adapter->promisc_supported)
1657                         return;
1658         } else {
1659                 if (netdev->flags & IFF_ALLMULTI) {
1660                         /* Accept all multicast */
1661                         memset(&crq, 0, sizeof(crq));
1662                         crq.multicast_ctrl.first = IBMVNIC_CRQ_CMD;
1663                         crq.multicast_ctrl.cmd = MULTICAST_CTRL;
1664                         crq.multicast_ctrl.flags = IBMVNIC_ENABLE_ALL;
1665                         ibmvnic_send_crq(adapter, &crq);
1666                 } else if (netdev_mc_empty(netdev)) {
1667                         /* Reject all multicast */
1668                         memset(&crq, 0, sizeof(crq));
1669                         crq.multicast_ctrl.first = IBMVNIC_CRQ_CMD;
1670                         crq.multicast_ctrl.cmd = MULTICAST_CTRL;
1671                         crq.multicast_ctrl.flags = IBMVNIC_DISABLE_ALL;
1672                         ibmvnic_send_crq(adapter, &crq);
1673                 } else {
1674                         /* Accept one or more multicast(s) */
1675                         netdev_for_each_mc_addr(ha, netdev) {
1676                                 memset(&crq, 0, sizeof(crq));
1677                                 crq.multicast_ctrl.first = IBMVNIC_CRQ_CMD;
1678                                 crq.multicast_ctrl.cmd = MULTICAST_CTRL;
1679                                 crq.multicast_ctrl.flags = IBMVNIC_ENABLE_MC;
1680                                 ether_addr_copy(&crq.multicast_ctrl.mac_addr[0],
1681                                                 ha->addr);
1682                                 ibmvnic_send_crq(adapter, &crq);
1683                         }
1684                 }
1685         }
1686 }
1687
1688 static int __ibmvnic_set_mac(struct net_device *netdev, struct sockaddr *p)
1689 {
1690         struct ibmvnic_adapter *adapter = netdev_priv(netdev);
1691         struct sockaddr *addr = p;
1692         union ibmvnic_crq crq;
1693         int rc;
1694
1695         if (!is_valid_ether_addr(addr->sa_data))
1696                 return -EADDRNOTAVAIL;
1697
1698         memset(&crq, 0, sizeof(crq));
1699         crq.change_mac_addr.first = IBMVNIC_CRQ_CMD;
1700         crq.change_mac_addr.cmd = CHANGE_MAC_ADDR;
1701         ether_addr_copy(&crq.change_mac_addr.mac_addr[0], addr->sa_data);
1702
1703         init_completion(&adapter->fw_done);
1704         rc = ibmvnic_send_crq(adapter, &crq);
1705         if (rc)
1706                 return rc;
1707         wait_for_completion(&adapter->fw_done);
1708         /* netdev->dev_addr is changed in handle_change_mac_rsp function */
1709         return adapter->fw_done_rc ? -EIO : 0;
1710 }
1711
1712 static int ibmvnic_set_mac(struct net_device *netdev, void *p)
1713 {
1714         struct ibmvnic_adapter *adapter = netdev_priv(netdev);
1715         struct sockaddr *addr = p;
1716         int rc;
1717
1718         if (adapter->state == VNIC_PROBED) {
1719                 memcpy(&adapter->desired.mac, addr, sizeof(struct sockaddr));
1720                 adapter->mac_change_pending = true;
1721                 return 0;
1722         }
1723
1724         rc = __ibmvnic_set_mac(netdev, addr);
1725
1726         return rc;
1727 }
1728
1729 /**
1730  * do_reset returns zero if we are able to keep processing reset events, or
1731  * non-zero if we hit a fatal error and must halt.
1732  */
1733 static int do_reset(struct ibmvnic_adapter *adapter,
1734                     struct ibmvnic_rwi *rwi, u32 reset_state)
1735 {
1736         u64 old_num_rx_queues, old_num_tx_queues;
1737         u64 old_num_rx_slots, old_num_tx_slots;
1738         struct net_device *netdev = adapter->netdev;
1739         int i, rc;
1740
1741         netdev_dbg(adapter->netdev, "Re-setting driver (%d)\n",
1742                    rwi->reset_reason);
1743
1744         netif_carrier_off(netdev);
1745         adapter->reset_reason = rwi->reset_reason;
1746
1747         old_num_rx_queues = adapter->req_rx_queues;
1748         old_num_tx_queues = adapter->req_tx_queues;
1749         old_num_rx_slots = adapter->req_rx_add_entries_per_subcrq;
1750         old_num_tx_slots = adapter->req_tx_entries_per_subcrq;
1751
1752         ibmvnic_cleanup(netdev);
1753
1754         if (adapter->reset_reason != VNIC_RESET_MOBILITY &&
1755             adapter->reset_reason != VNIC_RESET_FAILOVER) {
1756                 rc = __ibmvnic_close(netdev);
1757                 if (rc)
1758                         return rc;
1759         }
1760
1761         if (adapter->reset_reason == VNIC_RESET_CHANGE_PARAM ||
1762             adapter->wait_for_reset) {
1763                 release_resources(adapter);
1764                 release_sub_crqs(adapter, 1);
1765                 release_crq_queue(adapter);
1766         }
1767
1768         if (adapter->reset_reason != VNIC_RESET_NON_FATAL) {
1769                 /* remove the closed state so when we call open it appears
1770                  * we are coming from the probed state.
1771                  */
1772                 adapter->state = VNIC_PROBED;
1773
1774                 if (adapter->wait_for_reset) {
1775                         rc = init_crq_queue(adapter);
1776                 } else if (adapter->reset_reason == VNIC_RESET_MOBILITY) {
1777                         rc = ibmvnic_reenable_crq_queue(adapter);
1778                         release_sub_crqs(adapter, 1);
1779                 } else {
1780                         rc = ibmvnic_reset_crq(adapter);
1781                         if (!rc)
1782                                 rc = vio_enable_interrupts(adapter->vdev);
1783                 }
1784
1785                 if (rc) {
1786                         netdev_err(adapter->netdev,
1787                                    "Couldn't initialize crq. rc=%d\n", rc);
1788                         return rc;
1789                 }
1790
1791                 rc = ibmvnic_reset_init(adapter);
1792                 if (rc)
1793                         return IBMVNIC_INIT_FAILED;
1794
1795                 /* If the adapter was in PROBE state prior to the reset,
1796                  * exit here.
1797                  */
1798                 if (reset_state == VNIC_PROBED)
1799                         return 0;
1800
1801                 rc = ibmvnic_login(netdev);
1802                 if (rc) {
1803                         adapter->state = reset_state;
1804                         return rc;
1805                 }
1806
1807                 if (adapter->reset_reason == VNIC_RESET_CHANGE_PARAM ||
1808                     adapter->wait_for_reset) {
1809                         rc = init_resources(adapter);
1810                         if (rc)
1811                                 return rc;
1812                 } else if (adapter->req_rx_queues != old_num_rx_queues ||
1813                            adapter->req_tx_queues != old_num_tx_queues ||
1814                            adapter->req_rx_add_entries_per_subcrq !=
1815                                                         old_num_rx_slots ||
1816                            adapter->req_tx_entries_per_subcrq !=
1817                                                         old_num_tx_slots) {
1818                         release_rx_pools(adapter);
1819                         release_tx_pools(adapter);
1820                         release_napi(adapter);
1821                         release_vpd_data(adapter);
1822
1823                         rc = init_resources(adapter);
1824                         if (rc)
1825                                 return rc;
1826
1827                 } else {
1828                         rc = reset_tx_pools(adapter);
1829                         if (rc)
1830                                 return rc;
1831
1832                         rc = reset_rx_pools(adapter);
1833                         if (rc)
1834                                 return rc;
1835                 }
1836                 ibmvnic_disable_irqs(adapter);
1837         }
1838         adapter->state = VNIC_CLOSED;
1839
1840         if (reset_state == VNIC_CLOSED)
1841                 return 0;
1842
1843         rc = __ibmvnic_open(netdev);
1844         if (rc) {
1845                 if (list_empty(&adapter->rwi_list))
1846                         adapter->state = VNIC_CLOSED;
1847                 else
1848                         adapter->state = reset_state;
1849
1850                 return 0;
1851         }
1852
1853         /* kick napi */
1854         for (i = 0; i < adapter->req_rx_queues; i++)
1855                 napi_schedule(&adapter->napi[i]);
1856
1857         if (adapter->reset_reason != VNIC_RESET_FAILOVER &&
1858             adapter->reset_reason != VNIC_RESET_CHANGE_PARAM)
1859                 call_netdevice_notifiers(NETDEV_NOTIFY_PEERS, netdev);
1860
1861         netif_carrier_on(netdev);
1862
1863         return 0;
1864 }
1865
1866 static int do_hard_reset(struct ibmvnic_adapter *adapter,
1867                          struct ibmvnic_rwi *rwi, u32 reset_state)
1868 {
1869         struct net_device *netdev = adapter->netdev;
1870         int rc;
1871
1872         netdev_dbg(adapter->netdev, "Hard resetting driver (%d)\n",
1873                    rwi->reset_reason);
1874
1875         netif_carrier_off(netdev);
1876         adapter->reset_reason = rwi->reset_reason;
1877
1878         ibmvnic_cleanup(netdev);
1879         release_resources(adapter);
1880         release_sub_crqs(adapter, 0);
1881         release_crq_queue(adapter);
1882
1883         /* remove the closed state so when we call open it appears
1884          * we are coming from the probed state.
1885          */
1886         adapter->state = VNIC_PROBED;
1887
1888         rc = init_crq_queue(adapter);
1889         if (rc) {
1890                 netdev_err(adapter->netdev,
1891                            "Couldn't initialize crq. rc=%d\n", rc);
1892                 return rc;
1893         }
1894
1895         rc = ibmvnic_init(adapter);
1896         if (rc)
1897                 return rc;
1898
1899         /* If the adapter was in PROBE state prior to the reset,
1900          * exit here.
1901          */
1902         if (reset_state == VNIC_PROBED)
1903                 return 0;
1904
1905         rc = ibmvnic_login(netdev);
1906         if (rc) {
1907                 adapter->state = VNIC_PROBED;
1908                 return 0;
1909         }
1910
1911         rc = init_resources(adapter);
1912         if (rc)
1913                 return rc;
1914
1915         ibmvnic_disable_irqs(adapter);
1916         adapter->state = VNIC_CLOSED;
1917
1918         if (reset_state == VNIC_CLOSED)
1919                 return 0;
1920
1921         rc = __ibmvnic_open(netdev);
1922         if (rc) {
1923                 if (list_empty(&adapter->rwi_list))
1924                         adapter->state = VNIC_CLOSED;
1925                 else
1926                         adapter->state = reset_state;
1927
1928                 return 0;
1929         }
1930
1931         netif_carrier_on(netdev);
1932
1933         return 0;
1934 }
1935
1936 static struct ibmvnic_rwi *get_next_rwi(struct ibmvnic_adapter *adapter)
1937 {
1938         struct ibmvnic_rwi *rwi;
1939         unsigned long flags;
1940
1941         spin_lock_irqsave(&adapter->rwi_lock, flags);
1942
1943         if (!list_empty(&adapter->rwi_list)) {
1944                 rwi = list_first_entry(&adapter->rwi_list, struct ibmvnic_rwi,
1945                                        list);
1946                 list_del(&rwi->list);
1947         } else {
1948                 rwi = NULL;
1949         }
1950
1951         spin_unlock_irqrestore(&adapter->rwi_lock, flags);
1952         return rwi;
1953 }
1954
1955 static void free_all_rwi(struct ibmvnic_adapter *adapter)
1956 {
1957         struct ibmvnic_rwi *rwi;
1958
1959         rwi = get_next_rwi(adapter);
1960         while (rwi) {
1961                 kfree(rwi);
1962                 rwi = get_next_rwi(adapter);
1963         }
1964 }
1965
1966 static void __ibmvnic_reset(struct work_struct *work)
1967 {
1968         struct ibmvnic_rwi *rwi;
1969         struct ibmvnic_adapter *adapter;
1970         struct net_device *netdev;
1971         bool we_lock_rtnl = false;
1972         u32 reset_state;
1973         int rc = 0;
1974
1975         adapter = container_of(work, struct ibmvnic_adapter, ibmvnic_reset);
1976         netdev = adapter->netdev;
1977
1978         /* netif_set_real_num_xx_queues needs to take rtnl lock here
1979          * unless wait_for_reset is set, in which case the rtnl lock
1980          * has already been taken before initializing the reset
1981          */
1982         if (!adapter->wait_for_reset) {
1983                 rtnl_lock();
1984                 we_lock_rtnl = true;
1985         }
1986         reset_state = adapter->state;
1987
1988         rwi = get_next_rwi(adapter);
1989         while (rwi) {
1990                 if (adapter->force_reset_recovery) {
1991                         adapter->force_reset_recovery = false;
1992                         rc = do_hard_reset(adapter, rwi, reset_state);
1993                 } else {
1994                         rc = do_reset(adapter, rwi, reset_state);
1995                 }
1996                 kfree(rwi);
1997                 if (rc && rc != IBMVNIC_INIT_FAILED &&
1998                     !adapter->force_reset_recovery)
1999                         break;
2000
2001                 rwi = get_next_rwi(adapter);
2002         }
2003
2004         if (adapter->wait_for_reset) {
2005                 adapter->wait_for_reset = false;
2006                 adapter->reset_done_rc = rc;
2007                 complete(&adapter->reset_done);
2008         }
2009
2010         if (rc) {
2011                 netdev_dbg(adapter->netdev, "Reset failed\n");
2012                 free_all_rwi(adapter);
2013         }
2014
2015         adapter->resetting = false;
2016         if (we_lock_rtnl)
2017                 rtnl_unlock();
2018 }
2019
2020 static int ibmvnic_reset(struct ibmvnic_adapter *adapter,
2021                          enum ibmvnic_reset_reason reason)
2022 {
2023         struct list_head *entry, *tmp_entry;
2024         struct ibmvnic_rwi *rwi, *tmp;
2025         struct net_device *netdev = adapter->netdev;
2026         unsigned long flags;
2027         int ret;
2028
2029         if (adapter->state == VNIC_REMOVING ||
2030             adapter->state == VNIC_REMOVED ||
2031             adapter->failover_pending) {
2032                 ret = EBUSY;
2033                 netdev_dbg(netdev, "Adapter removing or pending failover, skipping reset\n");
2034                 goto err;
2035         }
2036
2037         if (adapter->state == VNIC_PROBING) {
2038                 netdev_warn(netdev, "Adapter reset during probe\n");
2039                 ret = adapter->init_done_rc = EAGAIN;
2040                 goto err;
2041         }
2042
2043         spin_lock_irqsave(&adapter->rwi_lock, flags);
2044
2045         list_for_each(entry, &adapter->rwi_list) {
2046                 tmp = list_entry(entry, struct ibmvnic_rwi, list);
2047                 if (tmp->reset_reason == reason) {
2048                         netdev_dbg(netdev, "Skipping matching reset\n");
2049                         spin_unlock_irqrestore(&adapter->rwi_lock, flags);
2050                         ret = EBUSY;
2051                         goto err;
2052                 }
2053         }
2054
2055         rwi = kzalloc(sizeof(*rwi), GFP_ATOMIC);
2056         if (!rwi) {
2057                 spin_unlock_irqrestore(&adapter->rwi_lock, flags);
2058                 ibmvnic_close(netdev);
2059                 ret = ENOMEM;
2060                 goto err;
2061         }
2062         /* if we just received a transport event,
2063          * flush reset queue and process this reset
2064          */
2065         if (adapter->force_reset_recovery && !list_empty(&adapter->rwi_list)) {
2066                 list_for_each_safe(entry, tmp_entry, &adapter->rwi_list)
2067                         list_del(entry);
2068         }
2069         rwi->reset_reason = reason;
2070         list_add_tail(&rwi->list, &adapter->rwi_list);
2071         spin_unlock_irqrestore(&adapter->rwi_lock, flags);
2072         adapter->resetting = true;
2073         netdev_dbg(adapter->netdev, "Scheduling reset (reason %d)\n", reason);
2074         schedule_work(&adapter->ibmvnic_reset);
2075
2076         return 0;
2077 err:
2078         if (adapter->wait_for_reset)
2079                 adapter->wait_for_reset = false;
2080         return -ret;
2081 }
2082
2083 static void ibmvnic_tx_timeout(struct net_device *dev)
2084 {
2085         struct ibmvnic_adapter *adapter = netdev_priv(dev);
2086
2087         ibmvnic_reset(adapter, VNIC_RESET_TIMEOUT);
2088 }
2089
2090 static void remove_buff_from_pool(struct ibmvnic_adapter *adapter,
2091                                   struct ibmvnic_rx_buff *rx_buff)
2092 {
2093         struct ibmvnic_rx_pool *pool = &adapter->rx_pool[rx_buff->pool_index];
2094
2095         rx_buff->skb = NULL;
2096
2097         pool->free_map[pool->next_alloc] = (int)(rx_buff - pool->rx_buff);
2098         pool->next_alloc = (pool->next_alloc + 1) % pool->size;
2099
2100         atomic_dec(&pool->available);
2101 }
2102
2103 static int ibmvnic_poll(struct napi_struct *napi, int budget)
2104 {
2105         struct net_device *netdev = napi->dev;
2106         struct ibmvnic_adapter *adapter = netdev_priv(netdev);
2107         int scrq_num = (int)(napi - adapter->napi);
2108         int frames_processed = 0;
2109
2110 restart_poll:
2111         while (frames_processed < budget) {
2112                 struct sk_buff *skb;
2113                 struct ibmvnic_rx_buff *rx_buff;
2114                 union sub_crq *next;
2115                 u32 length;
2116                 u16 offset;
2117                 u8 flags = 0;
2118
2119                 if (unlikely(adapter->resetting &&
2120                              adapter->reset_reason != VNIC_RESET_NON_FATAL)) {
2121                         enable_scrq_irq(adapter, adapter->rx_scrq[scrq_num]);
2122                         napi_complete_done(napi, frames_processed);
2123                         return frames_processed;
2124                 }
2125
2126                 if (!pending_scrq(adapter, adapter->rx_scrq[scrq_num]))
2127                         break;
2128                 next = ibmvnic_next_scrq(adapter, adapter->rx_scrq[scrq_num]);
2129                 rx_buff =
2130                     (struct ibmvnic_rx_buff *)be64_to_cpu(next->
2131                                                           rx_comp.correlator);
2132                 /* do error checking */
2133                 if (next->rx_comp.rc) {
2134                         netdev_dbg(netdev, "rx buffer returned with rc %x\n",
2135                                    be16_to_cpu(next->rx_comp.rc));
2136                         /* free the entry */
2137                         next->rx_comp.first = 0;
2138                         dev_kfree_skb_any(rx_buff->skb);
2139                         remove_buff_from_pool(adapter, rx_buff);
2140                         continue;
2141                 } else if (!rx_buff->skb) {
2142                         /* free the entry */
2143                         next->rx_comp.first = 0;
2144                         remove_buff_from_pool(adapter, rx_buff);
2145                         continue;
2146                 }
2147
2148                 length = be32_to_cpu(next->rx_comp.len);
2149                 offset = be16_to_cpu(next->rx_comp.off_frame_data);
2150                 flags = next->rx_comp.flags;
2151                 skb = rx_buff->skb;
2152                 skb_copy_to_linear_data(skb, rx_buff->data + offset,
2153                                         length);
2154
2155                 /* VLAN Header has been stripped by the system firmware and
2156                  * needs to be inserted by the driver
2157                  */
2158                 if (adapter->rx_vlan_header_insertion &&
2159                     (flags & IBMVNIC_VLAN_STRIPPED))
2160                         __vlan_hwaccel_put_tag(skb, htons(ETH_P_8021Q),
2161                                                ntohs(next->rx_comp.vlan_tci));
2162
2163                 /* free the entry */
2164                 next->rx_comp.first = 0;
2165                 remove_buff_from_pool(adapter, rx_buff);
2166
2167                 skb_put(skb, length);
2168                 skb->protocol = eth_type_trans(skb, netdev);
2169                 skb_record_rx_queue(skb, scrq_num);
2170
2171                 if (flags & IBMVNIC_IP_CHKSUM_GOOD &&
2172                     flags & IBMVNIC_TCP_UDP_CHKSUM_GOOD) {
2173                         skb->ip_summed = CHECKSUM_UNNECESSARY;
2174                 }
2175
2176                 length = skb->len;
2177                 napi_gro_receive(napi, skb); /* send it up */
2178                 netdev->stats.rx_packets++;
2179                 netdev->stats.rx_bytes += length;
2180                 adapter->rx_stats_buffers[scrq_num].packets++;
2181                 adapter->rx_stats_buffers[scrq_num].bytes += length;
2182                 frames_processed++;
2183         }
2184
2185         if (adapter->state != VNIC_CLOSING)
2186                 replenish_rx_pool(adapter, &adapter->rx_pool[scrq_num]);
2187
2188         if (frames_processed < budget) {
2189                 enable_scrq_irq(adapter, adapter->rx_scrq[scrq_num]);
2190                 napi_complete_done(napi, frames_processed);
2191                 if (pending_scrq(adapter, adapter->rx_scrq[scrq_num]) &&
2192                     napi_reschedule(napi)) {
2193                         disable_scrq_irq(adapter, adapter->rx_scrq[scrq_num]);
2194                         goto restart_poll;
2195                 }
2196         }
2197         return frames_processed;
2198 }
2199
2200 static int wait_for_reset(struct ibmvnic_adapter *adapter)
2201 {
2202         int rc, ret;
2203
2204         adapter->fallback.mtu = adapter->req_mtu;
2205         adapter->fallback.rx_queues = adapter->req_rx_queues;
2206         adapter->fallback.tx_queues = adapter->req_tx_queues;
2207         adapter->fallback.rx_entries = adapter->req_rx_add_entries_per_subcrq;
2208         adapter->fallback.tx_entries = adapter->req_tx_entries_per_subcrq;
2209
2210         init_completion(&adapter->reset_done);
2211         adapter->wait_for_reset = true;
2212         rc = ibmvnic_reset(adapter, VNIC_RESET_CHANGE_PARAM);
2213         if (rc)
2214                 return rc;
2215         wait_for_completion(&adapter->reset_done);
2216
2217         ret = 0;
2218         if (adapter->reset_done_rc) {
2219                 ret = -EIO;
2220                 adapter->desired.mtu = adapter->fallback.mtu;
2221                 adapter->desired.rx_queues = adapter->fallback.rx_queues;
2222                 adapter->desired.tx_queues = adapter->fallback.tx_queues;
2223                 adapter->desired.rx_entries = adapter->fallback.rx_entries;
2224                 adapter->desired.tx_entries = adapter->fallback.tx_entries;
2225
2226                 init_completion(&adapter->reset_done);
2227                 adapter->wait_for_reset = true;
2228                 rc = ibmvnic_reset(adapter, VNIC_RESET_CHANGE_PARAM);
2229                 if (rc)
2230                         return ret;
2231                 wait_for_completion(&adapter->reset_done);
2232         }
2233         adapter->wait_for_reset = false;
2234
2235         return ret;
2236 }
2237
2238 static int ibmvnic_change_mtu(struct net_device *netdev, int new_mtu)
2239 {
2240         struct ibmvnic_adapter *adapter = netdev_priv(netdev);
2241
2242         adapter->desired.mtu = new_mtu + ETH_HLEN;
2243
2244         return wait_for_reset(adapter);
2245 }
2246
2247 static netdev_features_t ibmvnic_features_check(struct sk_buff *skb,
2248                                                 struct net_device *dev,
2249                                                 netdev_features_t features)
2250 {
2251         /* Some backing hardware adapters can not
2252          * handle packets with a MSS less than 224
2253          * or with only one segment.
2254          */
2255         if (skb_is_gso(skb)) {
2256                 if (skb_shinfo(skb)->gso_size < 224 ||
2257                     skb_shinfo(skb)->gso_segs == 1)
2258                         features &= ~NETIF_F_GSO_MASK;
2259         }
2260
2261         return features;
2262 }
2263
2264 static const struct net_device_ops ibmvnic_netdev_ops = {
2265         .ndo_open               = ibmvnic_open,
2266         .ndo_stop               = ibmvnic_close,
2267         .ndo_start_xmit         = ibmvnic_xmit,
2268         .ndo_set_rx_mode        = ibmvnic_set_multi,
2269         .ndo_set_mac_address    = ibmvnic_set_mac,
2270         .ndo_validate_addr      = eth_validate_addr,
2271         .ndo_tx_timeout         = ibmvnic_tx_timeout,
2272         .ndo_change_mtu         = ibmvnic_change_mtu,
2273         .ndo_features_check     = ibmvnic_features_check,
2274 };
2275
2276 /* ethtool functions */
2277
2278 static int ibmvnic_get_link_ksettings(struct net_device *netdev,
2279                                       struct ethtool_link_ksettings *cmd)
2280 {
2281         u32 supported, advertising;
2282
2283         supported = (SUPPORTED_1000baseT_Full | SUPPORTED_Autoneg |
2284                           SUPPORTED_FIBRE);
2285         advertising = (ADVERTISED_1000baseT_Full | ADVERTISED_Autoneg |
2286                             ADVERTISED_FIBRE);
2287         cmd->base.speed = SPEED_1000;
2288         cmd->base.duplex = DUPLEX_FULL;
2289         cmd->base.port = PORT_FIBRE;
2290         cmd->base.phy_address = 0;
2291         cmd->base.autoneg = AUTONEG_ENABLE;
2292
2293         ethtool_convert_legacy_u32_to_link_mode(cmd->link_modes.supported,
2294                                                 supported);
2295         ethtool_convert_legacy_u32_to_link_mode(cmd->link_modes.advertising,
2296                                                 advertising);
2297
2298         return 0;
2299 }
2300
2301 static void ibmvnic_get_drvinfo(struct net_device *netdev,
2302                                 struct ethtool_drvinfo *info)
2303 {
2304         struct ibmvnic_adapter *adapter = netdev_priv(netdev);
2305
2306         strlcpy(info->driver, ibmvnic_driver_name, sizeof(info->driver));
2307         strlcpy(info->version, IBMVNIC_DRIVER_VERSION, sizeof(info->version));
2308         strlcpy(info->fw_version, adapter->fw_version,
2309                 sizeof(info->fw_version));
2310 }
2311
2312 static u32 ibmvnic_get_msglevel(struct net_device *netdev)
2313 {
2314         struct ibmvnic_adapter *adapter = netdev_priv(netdev);
2315
2316         return adapter->msg_enable;
2317 }
2318
2319 static void ibmvnic_set_msglevel(struct net_device *netdev, u32 data)
2320 {
2321         struct ibmvnic_adapter *adapter = netdev_priv(netdev);
2322
2323         adapter->msg_enable = data;
2324 }
2325
2326 static u32 ibmvnic_get_link(struct net_device *netdev)
2327 {
2328         struct ibmvnic_adapter *adapter = netdev_priv(netdev);
2329
2330         /* Don't need to send a query because we request a logical link up at
2331          * init and then we wait for link state indications
2332          */
2333         return adapter->logical_link_state;
2334 }
2335
2336 static void ibmvnic_get_ringparam(struct net_device *netdev,
2337                                   struct ethtool_ringparam *ring)
2338 {
2339         struct ibmvnic_adapter *adapter = netdev_priv(netdev);
2340
2341         if (adapter->priv_flags & IBMVNIC_USE_SERVER_MAXES) {
2342                 ring->rx_max_pending = adapter->max_rx_add_entries_per_subcrq;
2343                 ring->tx_max_pending = adapter->max_tx_entries_per_subcrq;
2344         } else {
2345                 ring->rx_max_pending = IBMVNIC_MAX_QUEUE_SZ;
2346                 ring->tx_max_pending = IBMVNIC_MAX_QUEUE_SZ;
2347         }
2348         ring->rx_mini_max_pending = 0;
2349         ring->rx_jumbo_max_pending = 0;
2350         ring->rx_pending = adapter->req_rx_add_entries_per_subcrq;
2351         ring->tx_pending = adapter->req_tx_entries_per_subcrq;
2352         ring->rx_mini_pending = 0;
2353         ring->rx_jumbo_pending = 0;
2354 }
2355
2356 static int ibmvnic_set_ringparam(struct net_device *netdev,
2357                                  struct ethtool_ringparam *ring)
2358 {
2359         struct ibmvnic_adapter *adapter = netdev_priv(netdev);
2360         int ret;
2361
2362         ret = 0;
2363         adapter->desired.rx_entries = ring->rx_pending;
2364         adapter->desired.tx_entries = ring->tx_pending;
2365
2366         ret = wait_for_reset(adapter);
2367
2368         if (!ret &&
2369             (adapter->req_rx_add_entries_per_subcrq != ring->rx_pending ||
2370              adapter->req_tx_entries_per_subcrq != ring->tx_pending))
2371                 netdev_info(netdev,
2372                             "Could not match full ringsize request. Requested: RX %d, TX %d; Allowed: RX %llu, TX %llu\n",
2373                             ring->rx_pending, ring->tx_pending,
2374                             adapter->req_rx_add_entries_per_subcrq,
2375                             adapter->req_tx_entries_per_subcrq);
2376         return ret;
2377 }
2378
2379 static void ibmvnic_get_channels(struct net_device *netdev,
2380                                  struct ethtool_channels *channels)
2381 {
2382         struct ibmvnic_adapter *adapter = netdev_priv(netdev);
2383
2384         if (adapter->priv_flags & IBMVNIC_USE_SERVER_MAXES) {
2385                 channels->max_rx = adapter->max_rx_queues;
2386                 channels->max_tx = adapter->max_tx_queues;
2387         } else {
2388                 channels->max_rx = IBMVNIC_MAX_QUEUES;
2389                 channels->max_tx = IBMVNIC_MAX_QUEUES;
2390         }
2391
2392         channels->max_other = 0;
2393         channels->max_combined = 0;
2394         channels->rx_count = adapter->req_rx_queues;
2395         channels->tx_count = adapter->req_tx_queues;
2396         channels->other_count = 0;
2397         channels->combined_count = 0;
2398 }
2399
2400 static int ibmvnic_set_channels(struct net_device *netdev,
2401                                 struct ethtool_channels *channels)
2402 {
2403         struct ibmvnic_adapter *adapter = netdev_priv(netdev);
2404         int ret;
2405
2406         ret = 0;
2407         adapter->desired.rx_queues = channels->rx_count;
2408         adapter->desired.tx_queues = channels->tx_count;
2409
2410         ret = wait_for_reset(adapter);
2411
2412         if (!ret &&
2413             (adapter->req_rx_queues != channels->rx_count ||
2414              adapter->req_tx_queues != channels->tx_count))
2415                 netdev_info(netdev,
2416                             "Could not match full channels request. Requested: RX %d, TX %d; Allowed: RX %llu, TX %llu\n",
2417                             channels->rx_count, channels->tx_count,
2418                             adapter->req_rx_queues, adapter->req_tx_queues);
2419         return ret;
2420
2421 }
2422
2423 static void ibmvnic_get_strings(struct net_device *dev, u32 stringset, u8 *data)
2424 {
2425         struct ibmvnic_adapter *adapter = netdev_priv(dev);
2426         int i;
2427
2428         switch (stringset) {
2429         case ETH_SS_STATS:
2430                 for (i = 0; i < ARRAY_SIZE(ibmvnic_stats);
2431                                 i++, data += ETH_GSTRING_LEN)
2432                         memcpy(data, ibmvnic_stats[i].name, ETH_GSTRING_LEN);
2433
2434                 for (i = 0; i < adapter->req_tx_queues; i++) {
2435                         snprintf(data, ETH_GSTRING_LEN, "tx%d_packets", i);
2436                         data += ETH_GSTRING_LEN;
2437
2438                         snprintf(data, ETH_GSTRING_LEN, "tx%d_bytes", i);
2439                         data += ETH_GSTRING_LEN;
2440
2441                         snprintf(data, ETH_GSTRING_LEN,
2442                                  "tx%d_dropped_packets", i);
2443                         data += ETH_GSTRING_LEN;
2444                 }
2445
2446                 for (i = 0; i < adapter->req_rx_queues; i++) {
2447                         snprintf(data, ETH_GSTRING_LEN, "rx%d_packets", i);
2448                         data += ETH_GSTRING_LEN;
2449
2450                         snprintf(data, ETH_GSTRING_LEN, "rx%d_bytes", i);
2451                         data += ETH_GSTRING_LEN;
2452
2453                         snprintf(data, ETH_GSTRING_LEN, "rx%d_interrupts", i);
2454                         data += ETH_GSTRING_LEN;
2455                 }
2456                 break;
2457
2458         case ETH_SS_PRIV_FLAGS:
2459                 for (i = 0; i < ARRAY_SIZE(ibmvnic_priv_flags); i++)
2460                         strcpy(data + i * ETH_GSTRING_LEN,
2461                                ibmvnic_priv_flags[i]);
2462                 break;
2463         default:
2464                 return;
2465         }
2466 }
2467
2468 static int ibmvnic_get_sset_count(struct net_device *dev, int sset)
2469 {
2470         struct ibmvnic_adapter *adapter = netdev_priv(dev);
2471
2472         switch (sset) {
2473         case ETH_SS_STATS:
2474                 return ARRAY_SIZE(ibmvnic_stats) +
2475                        adapter->req_tx_queues * NUM_TX_STATS +
2476                        adapter->req_rx_queues * NUM_RX_STATS;
2477         case ETH_SS_PRIV_FLAGS:
2478                 return ARRAY_SIZE(ibmvnic_priv_flags);
2479         default:
2480                 return -EOPNOTSUPP;
2481         }
2482 }
2483
2484 static void ibmvnic_get_ethtool_stats(struct net_device *dev,
2485                                       struct ethtool_stats *stats, u64 *data)
2486 {
2487         struct ibmvnic_adapter *adapter = netdev_priv(dev);
2488         union ibmvnic_crq crq;
2489         int i, j;
2490         int rc;
2491
2492         memset(&crq, 0, sizeof(crq));
2493         crq.request_statistics.first = IBMVNIC_CRQ_CMD;
2494         crq.request_statistics.cmd = REQUEST_STATISTICS;
2495         crq.request_statistics.ioba = cpu_to_be32(adapter->stats_token);
2496         crq.request_statistics.len =
2497             cpu_to_be32(sizeof(struct ibmvnic_statistics));
2498
2499         /* Wait for data to be written */
2500         init_completion(&adapter->stats_done);
2501         rc = ibmvnic_send_crq(adapter, &crq);
2502         if (rc)
2503                 return;
2504         wait_for_completion(&adapter->stats_done);
2505
2506         for (i = 0; i < ARRAY_SIZE(ibmvnic_stats); i++)
2507                 data[i] = be64_to_cpu(IBMVNIC_GET_STAT(adapter,
2508                                                 ibmvnic_stats[i].offset));
2509
2510         for (j = 0; j < adapter->req_tx_queues; j++) {
2511                 data[i] = adapter->tx_stats_buffers[j].packets;
2512                 i++;
2513                 data[i] = adapter->tx_stats_buffers[j].bytes;
2514                 i++;
2515                 data[i] = adapter->tx_stats_buffers[j].dropped_packets;
2516                 i++;
2517         }
2518
2519         for (j = 0; j < adapter->req_rx_queues; j++) {
2520                 data[i] = adapter->rx_stats_buffers[j].packets;
2521                 i++;
2522                 data[i] = adapter->rx_stats_buffers[j].bytes;
2523                 i++;
2524                 data[i] = adapter->rx_stats_buffers[j].interrupts;
2525                 i++;
2526         }
2527 }
2528
2529 static u32 ibmvnic_get_priv_flags(struct net_device *netdev)
2530 {
2531         struct ibmvnic_adapter *adapter = netdev_priv(netdev);
2532
2533         return adapter->priv_flags;
2534 }
2535
2536 static int ibmvnic_set_priv_flags(struct net_device *netdev, u32 flags)
2537 {
2538         struct ibmvnic_adapter *adapter = netdev_priv(netdev);
2539         bool which_maxes = !!(flags & IBMVNIC_USE_SERVER_MAXES);
2540
2541         if (which_maxes)
2542                 adapter->priv_flags |= IBMVNIC_USE_SERVER_MAXES;
2543         else
2544                 adapter->priv_flags &= ~IBMVNIC_USE_SERVER_MAXES;
2545
2546         return 0;
2547 }
2548 static const struct ethtool_ops ibmvnic_ethtool_ops = {
2549         .get_drvinfo            = ibmvnic_get_drvinfo,
2550         .get_msglevel           = ibmvnic_get_msglevel,
2551         .set_msglevel           = ibmvnic_set_msglevel,
2552         .get_link               = ibmvnic_get_link,
2553         .get_ringparam          = ibmvnic_get_ringparam,
2554         .set_ringparam          = ibmvnic_set_ringparam,
2555         .get_channels           = ibmvnic_get_channels,
2556         .set_channels           = ibmvnic_set_channels,
2557         .get_strings            = ibmvnic_get_strings,
2558         .get_sset_count         = ibmvnic_get_sset_count,
2559         .get_ethtool_stats      = ibmvnic_get_ethtool_stats,
2560         .get_link_ksettings     = ibmvnic_get_link_ksettings,
2561         .get_priv_flags         = ibmvnic_get_priv_flags,
2562         .set_priv_flags         = ibmvnic_set_priv_flags,
2563 };
2564
2565 /* Routines for managing CRQs/sCRQs  */
2566
2567 static int reset_one_sub_crq_queue(struct ibmvnic_adapter *adapter,
2568                                    struct ibmvnic_sub_crq_queue *scrq)
2569 {
2570         int rc;
2571
2572         if (scrq->irq) {
2573                 free_irq(scrq->irq, scrq);
2574                 irq_dispose_mapping(scrq->irq);
2575                 scrq->irq = 0;
2576         }
2577
2578         memset(scrq->msgs, 0, 4 * PAGE_SIZE);
2579         atomic_set(&scrq->used, 0);
2580         scrq->cur = 0;
2581
2582         rc = h_reg_sub_crq(adapter->vdev->unit_address, scrq->msg_token,
2583                            4 * PAGE_SIZE, &scrq->crq_num, &scrq->hw_irq);
2584         return rc;
2585 }
2586
2587 static int reset_sub_crq_queues(struct ibmvnic_adapter *adapter)
2588 {
2589         int i, rc;
2590
2591         for (i = 0; i < adapter->req_tx_queues; i++) {
2592                 netdev_dbg(adapter->netdev, "Re-setting tx_scrq[%d]\n", i);
2593                 rc = reset_one_sub_crq_queue(adapter, adapter->tx_scrq[i]);
2594                 if (rc)
2595                         return rc;
2596         }
2597
2598         for (i = 0; i < adapter->req_rx_queues; i++) {
2599                 netdev_dbg(adapter->netdev, "Re-setting rx_scrq[%d]\n", i);
2600                 rc = reset_one_sub_crq_queue(adapter, adapter->rx_scrq[i]);
2601                 if (rc)
2602                         return rc;
2603         }
2604
2605         return rc;
2606 }
2607
2608 static void release_sub_crq_queue(struct ibmvnic_adapter *adapter,
2609                                   struct ibmvnic_sub_crq_queue *scrq,
2610                                   bool do_h_free)
2611 {
2612         struct device *dev = &adapter->vdev->dev;
2613         long rc;
2614
2615         netdev_dbg(adapter->netdev, "Releasing sub-CRQ\n");
2616
2617         if (do_h_free) {
2618                 /* Close the sub-crqs */
2619                 do {
2620                         rc = plpar_hcall_norets(H_FREE_SUB_CRQ,
2621                                                 adapter->vdev->unit_address,
2622                                                 scrq->crq_num);
2623                 } while (rc == H_BUSY || H_IS_LONG_BUSY(rc));
2624
2625                 if (rc) {
2626                         netdev_err(adapter->netdev,
2627                                    "Failed to release sub-CRQ %16lx, rc = %ld\n",
2628                                    scrq->crq_num, rc);
2629                 }
2630         }
2631
2632         dma_unmap_single(dev, scrq->msg_token, 4 * PAGE_SIZE,
2633                          DMA_BIDIRECTIONAL);
2634         free_pages((unsigned long)scrq->msgs, 2);
2635         kfree(scrq);
2636 }
2637
2638 static struct ibmvnic_sub_crq_queue *init_sub_crq_queue(struct ibmvnic_adapter
2639                                                         *adapter)
2640 {
2641         struct device *dev = &adapter->vdev->dev;
2642         struct ibmvnic_sub_crq_queue *scrq;
2643         int rc;
2644
2645         scrq = kzalloc(sizeof(*scrq), GFP_KERNEL);
2646         if (!scrq)
2647                 return NULL;
2648
2649         scrq->msgs =
2650                 (union sub_crq *)__get_free_pages(GFP_KERNEL | __GFP_ZERO, 2);
2651         if (!scrq->msgs) {
2652                 dev_warn(dev, "Couldn't allocate crq queue messages page\n");
2653                 goto zero_page_failed;
2654         }
2655
2656         scrq->msg_token = dma_map_single(dev, scrq->msgs, 4 * PAGE_SIZE,
2657                                          DMA_BIDIRECTIONAL);
2658         if (dma_mapping_error(dev, scrq->msg_token)) {
2659                 dev_warn(dev, "Couldn't map crq queue messages page\n");
2660                 goto map_failed;
2661         }
2662
2663         rc = h_reg_sub_crq(adapter->vdev->unit_address, scrq->msg_token,
2664                            4 * PAGE_SIZE, &scrq->crq_num, &scrq->hw_irq);
2665
2666         if (rc == H_RESOURCE)
2667                 rc = ibmvnic_reset_crq(adapter);
2668
2669         if (rc == H_CLOSED) {
2670                 dev_warn(dev, "Partner adapter not ready, waiting.\n");
2671         } else if (rc) {
2672                 dev_warn(dev, "Error %d registering sub-crq\n", rc);
2673                 goto reg_failed;
2674         }
2675
2676         scrq->adapter = adapter;
2677         scrq->size = 4 * PAGE_SIZE / sizeof(*scrq->msgs);
2678         spin_lock_init(&scrq->lock);
2679
2680         netdev_dbg(adapter->netdev,
2681                    "sub-crq initialized, num %lx, hw_irq=%lx, irq=%x\n",
2682                    scrq->crq_num, scrq->hw_irq, scrq->irq);
2683
2684         return scrq;
2685
2686 reg_failed:
2687         dma_unmap_single(dev, scrq->msg_token, 4 * PAGE_SIZE,
2688                          DMA_BIDIRECTIONAL);
2689 map_failed:
2690         free_pages((unsigned long)scrq->msgs, 2);
2691 zero_page_failed:
2692         kfree(scrq);
2693
2694         return NULL;
2695 }
2696
2697 static void release_sub_crqs(struct ibmvnic_adapter *adapter, bool do_h_free)
2698 {
2699         int i;
2700
2701         if (adapter->tx_scrq) {
2702                 for (i = 0; i < adapter->num_active_tx_scrqs; i++) {
2703                         if (!adapter->tx_scrq[i])
2704                                 continue;
2705
2706                         netdev_dbg(adapter->netdev, "Releasing tx_scrq[%d]\n",
2707                                    i);
2708                         if (adapter->tx_scrq[i]->irq) {
2709                                 free_irq(adapter->tx_scrq[i]->irq,
2710                                          adapter->tx_scrq[i]);
2711                                 irq_dispose_mapping(adapter->tx_scrq[i]->irq);
2712                                 adapter->tx_scrq[i]->irq = 0;
2713                         }
2714
2715                         release_sub_crq_queue(adapter, adapter->tx_scrq[i],
2716                                               do_h_free);
2717                 }
2718
2719                 kfree(adapter->tx_scrq);
2720                 adapter->tx_scrq = NULL;
2721                 adapter->num_active_tx_scrqs = 0;
2722         }
2723
2724         if (adapter->rx_scrq) {
2725                 for (i = 0; i < adapter->num_active_rx_scrqs; i++) {
2726                         if (!adapter->rx_scrq[i])
2727                                 continue;
2728
2729                         netdev_dbg(adapter->netdev, "Releasing rx_scrq[%d]\n",
2730                                    i);
2731                         if (adapter->rx_scrq[i]->irq) {
2732                                 free_irq(adapter->rx_scrq[i]->irq,
2733                                          adapter->rx_scrq[i]);
2734                                 irq_dispose_mapping(adapter->rx_scrq[i]->irq);
2735                                 adapter->rx_scrq[i]->irq = 0;
2736                         }
2737
2738                         release_sub_crq_queue(adapter, adapter->rx_scrq[i],
2739                                               do_h_free);
2740                 }
2741
2742                 kfree(adapter->rx_scrq);
2743                 adapter->rx_scrq = NULL;
2744                 adapter->num_active_rx_scrqs = 0;
2745         }
2746 }
2747
2748 static int disable_scrq_irq(struct ibmvnic_adapter *adapter,
2749                             struct ibmvnic_sub_crq_queue *scrq)
2750 {
2751         struct device *dev = &adapter->vdev->dev;
2752         unsigned long rc;
2753
2754         rc = plpar_hcall_norets(H_VIOCTL, adapter->vdev->unit_address,
2755                                 H_DISABLE_VIO_INTERRUPT, scrq->hw_irq, 0, 0);
2756         if (rc)
2757                 dev_err(dev, "Couldn't disable scrq irq 0x%lx. rc=%ld\n",
2758                         scrq->hw_irq, rc);
2759         return rc;
2760 }
2761
2762 static int enable_scrq_irq(struct ibmvnic_adapter *adapter,
2763                            struct ibmvnic_sub_crq_queue *scrq)
2764 {
2765         struct device *dev = &adapter->vdev->dev;
2766         unsigned long rc;
2767
2768         if (scrq->hw_irq > 0x100000000ULL) {
2769                 dev_err(dev, "bad hw_irq = %lx\n", scrq->hw_irq);
2770                 return 1;
2771         }
2772
2773         if (adapter->resetting &&
2774             adapter->reset_reason == VNIC_RESET_MOBILITY) {
2775                 u64 val = (0xff000000) | scrq->hw_irq;
2776
2777                 rc = plpar_hcall_norets(H_EOI, val);
2778                 if (rc)
2779                         dev_err(dev, "H_EOI FAILED irq 0x%llx. rc=%ld\n",
2780                                 val, rc);
2781         }
2782
2783         rc = plpar_hcall_norets(H_VIOCTL, adapter->vdev->unit_address,
2784                                 H_ENABLE_VIO_INTERRUPT, scrq->hw_irq, 0, 0);
2785         if (rc)
2786                 dev_err(dev, "Couldn't enable scrq irq 0x%lx. rc=%ld\n",
2787                         scrq->hw_irq, rc);
2788         return rc;
2789 }
2790
2791 static int ibmvnic_complete_tx(struct ibmvnic_adapter *adapter,
2792                                struct ibmvnic_sub_crq_queue *scrq)
2793 {
2794         struct device *dev = &adapter->vdev->dev;
2795         struct ibmvnic_tx_pool *tx_pool;
2796         struct ibmvnic_tx_buff *txbuff;
2797         union sub_crq *next;
2798         int index;
2799         int i, j;
2800         u8 *first;
2801
2802 restart_loop:
2803         while (pending_scrq(adapter, scrq)) {
2804                 unsigned int pool = scrq->pool_index;
2805                 int num_entries = 0;
2806
2807                 next = ibmvnic_next_scrq(adapter, scrq);
2808                 for (i = 0; i < next->tx_comp.num_comps; i++) {
2809                         if (next->tx_comp.rcs[i]) {
2810                                 dev_err(dev, "tx error %x\n",
2811                                         next->tx_comp.rcs[i]);
2812                                 continue;
2813                         }
2814                         index = be32_to_cpu(next->tx_comp.correlators[i]);
2815                         if (index & IBMVNIC_TSO_POOL_MASK) {
2816                                 tx_pool = &adapter->tso_pool[pool];
2817                                 index &= ~IBMVNIC_TSO_POOL_MASK;
2818                         } else {
2819                                 tx_pool = &adapter->tx_pool[pool];
2820                         }
2821
2822                         txbuff = &tx_pool->tx_buff[index];
2823
2824                         for (j = 0; j < IBMVNIC_MAX_FRAGS_PER_CRQ; j++) {
2825                                 if (!txbuff->data_dma[j])
2826                                         continue;
2827
2828                                 txbuff->data_dma[j] = 0;
2829                         }
2830                         /* if sub_crq was sent indirectly */
2831                         first = &txbuff->indir_arr[0].generic.first;
2832                         if (*first == IBMVNIC_CRQ_CMD) {
2833                                 dma_unmap_single(dev, txbuff->indir_dma,
2834                                                  sizeof(txbuff->indir_arr),
2835                                                  DMA_TO_DEVICE);
2836                                 *first = 0;
2837                         }
2838
2839                         if (txbuff->last_frag) {
2840                                 dev_kfree_skb_any(txbuff->skb);
2841                                 txbuff->skb = NULL;
2842                         }
2843
2844                         num_entries += txbuff->num_entries;
2845
2846                         tx_pool->free_map[tx_pool->producer_index] = index;
2847                         tx_pool->producer_index =
2848                                 (tx_pool->producer_index + 1) %
2849                                         tx_pool->num_buffers;
2850                 }
2851                 /* remove tx_comp scrq*/
2852                 next->tx_comp.first = 0;
2853
2854                 if (atomic_sub_return(num_entries, &scrq->used) <=
2855                     (adapter->req_tx_entries_per_subcrq / 2) &&
2856                     __netif_subqueue_stopped(adapter->netdev,
2857                                              scrq->pool_index)) {
2858                         netif_wake_subqueue(adapter->netdev, scrq->pool_index);
2859                         netdev_dbg(adapter->netdev, "Started queue %d\n",
2860                                    scrq->pool_index);
2861                 }
2862         }
2863
2864         enable_scrq_irq(adapter, scrq);
2865
2866         if (pending_scrq(adapter, scrq)) {
2867                 disable_scrq_irq(adapter, scrq);
2868                 goto restart_loop;
2869         }
2870
2871         return 0;
2872 }
2873
2874 static irqreturn_t ibmvnic_interrupt_tx(int irq, void *instance)
2875 {
2876         struct ibmvnic_sub_crq_queue *scrq = instance;
2877         struct ibmvnic_adapter *adapter = scrq->adapter;
2878
2879         disable_scrq_irq(adapter, scrq);
2880         ibmvnic_complete_tx(adapter, scrq);
2881
2882         return IRQ_HANDLED;
2883 }
2884
2885 static irqreturn_t ibmvnic_interrupt_rx(int irq, void *instance)
2886 {
2887         struct ibmvnic_sub_crq_queue *scrq = instance;
2888         struct ibmvnic_adapter *adapter = scrq->adapter;
2889
2890         /* When booting a kdump kernel we can hit pending interrupts
2891          * prior to completing driver initialization.
2892          */
2893         if (unlikely(adapter->state != VNIC_OPEN))
2894                 return IRQ_NONE;
2895
2896         adapter->rx_stats_buffers[scrq->scrq_num].interrupts++;
2897
2898         if (napi_schedule_prep(&adapter->napi[scrq->scrq_num])) {
2899                 disable_scrq_irq(adapter, scrq);
2900                 __napi_schedule(&adapter->napi[scrq->scrq_num]);
2901         }
2902
2903         return IRQ_HANDLED;
2904 }
2905
2906 static int init_sub_crq_irqs(struct ibmvnic_adapter *adapter)
2907 {
2908         struct device *dev = &adapter->vdev->dev;
2909         struct ibmvnic_sub_crq_queue *scrq;
2910         int i = 0, j = 0;
2911         int rc = 0;
2912
2913         for (i = 0; i < adapter->req_tx_queues; i++) {
2914                 netdev_dbg(adapter->netdev, "Initializing tx_scrq[%d] irq\n",
2915                            i);
2916                 scrq = adapter->tx_scrq[i];
2917                 scrq->irq = irq_create_mapping(NULL, scrq->hw_irq);
2918
2919                 if (!scrq->irq) {
2920                         rc = -EINVAL;
2921                         dev_err(dev, "Error mapping irq\n");
2922                         goto req_tx_irq_failed;
2923                 }
2924
2925                 rc = request_irq(scrq->irq, ibmvnic_interrupt_tx,
2926                                  0, "ibmvnic_tx", scrq);
2927
2928                 if (rc) {
2929                         dev_err(dev, "Couldn't register tx irq 0x%x. rc=%d\n",
2930                                 scrq->irq, rc);
2931                         irq_dispose_mapping(scrq->irq);
2932                         goto req_tx_irq_failed;
2933                 }
2934         }
2935
2936         for (i = 0; i < adapter->req_rx_queues; i++) {
2937                 netdev_dbg(adapter->netdev, "Initializing rx_scrq[%d] irq\n",
2938                            i);
2939                 scrq = adapter->rx_scrq[i];
2940                 scrq->irq = irq_create_mapping(NULL, scrq->hw_irq);
2941                 if (!scrq->irq) {
2942                         rc = -EINVAL;
2943                         dev_err(dev, "Error mapping irq\n");
2944                         goto req_rx_irq_failed;
2945                 }
2946                 rc = request_irq(scrq->irq, ibmvnic_interrupt_rx,
2947                                  0, "ibmvnic_rx", scrq);
2948                 if (rc) {
2949                         dev_err(dev, "Couldn't register rx irq 0x%x. rc=%d\n",
2950                                 scrq->irq, rc);
2951                         irq_dispose_mapping(scrq->irq);
2952                         goto req_rx_irq_failed;
2953                 }
2954         }
2955         return rc;
2956
2957 req_rx_irq_failed:
2958         for (j = 0; j < i; j++) {
2959                 free_irq(adapter->rx_scrq[j]->irq, adapter->rx_scrq[j]);
2960                 irq_dispose_mapping(adapter->rx_scrq[j]->irq);
2961         }
2962         i = adapter->req_tx_queues;
2963 req_tx_irq_failed:
2964         for (j = 0; j < i; j++) {
2965                 free_irq(adapter->tx_scrq[j]->irq, adapter->tx_scrq[j]);
2966                 irq_dispose_mapping(adapter->rx_scrq[j]->irq);
2967         }
2968         release_sub_crqs(adapter, 1);
2969         return rc;
2970 }
2971
2972 static int init_sub_crqs(struct ibmvnic_adapter *adapter)
2973 {
2974         struct device *dev = &adapter->vdev->dev;
2975         struct ibmvnic_sub_crq_queue **allqueues;
2976         int registered_queues = 0;
2977         int total_queues;
2978         int more = 0;
2979         int i;
2980
2981         total_queues = adapter->req_tx_queues + adapter->req_rx_queues;
2982
2983         allqueues = kcalloc(total_queues, sizeof(*allqueues), GFP_KERNEL);
2984         if (!allqueues)
2985                 return -1;
2986
2987         for (i = 0; i < total_queues; i++) {
2988                 allqueues[i] = init_sub_crq_queue(adapter);
2989                 if (!allqueues[i]) {
2990                         dev_warn(dev, "Couldn't allocate all sub-crqs\n");
2991                         break;
2992                 }
2993                 registered_queues++;
2994         }
2995
2996         /* Make sure we were able to register the minimum number of queues */
2997         if (registered_queues <
2998             adapter->min_tx_queues + adapter->min_rx_queues) {
2999                 dev_err(dev, "Fatal: Couldn't init  min number of sub-crqs\n");
3000                 goto tx_failed;
3001         }
3002
3003         /* Distribute the failed allocated queues*/
3004         for (i = 0; i < total_queues - registered_queues + more ; i++) {
3005                 netdev_dbg(adapter->netdev, "Reducing number of queues\n");
3006                 switch (i % 3) {
3007                 case 0:
3008                         if (adapter->req_rx_queues > adapter->min_rx_queues)
3009                                 adapter->req_rx_queues--;
3010                         else
3011                                 more++;
3012                         break;
3013                 case 1:
3014                         if (adapter->req_tx_queues > adapter->min_tx_queues)
3015                                 adapter->req_tx_queues--;
3016                         else
3017                                 more++;
3018                         break;
3019                 }
3020         }
3021
3022         adapter->tx_scrq = kcalloc(adapter->req_tx_queues,
3023                                    sizeof(*adapter->tx_scrq), GFP_KERNEL);
3024         if (!adapter->tx_scrq)
3025                 goto tx_failed;
3026
3027         for (i = 0; i < adapter->req_tx_queues; i++) {
3028                 adapter->tx_scrq[i] = allqueues[i];
3029                 adapter->tx_scrq[i]->pool_index = i;
3030                 adapter->num_active_tx_scrqs++;
3031         }
3032
3033         adapter->rx_scrq = kcalloc(adapter->req_rx_queues,
3034                                    sizeof(*adapter->rx_scrq), GFP_KERNEL);
3035         if (!adapter->rx_scrq)
3036                 goto rx_failed;
3037
3038         for (i = 0; i < adapter->req_rx_queues; i++) {
3039                 adapter->rx_scrq[i] = allqueues[i + adapter->req_tx_queues];
3040                 adapter->rx_scrq[i]->scrq_num = i;
3041                 adapter->num_active_rx_scrqs++;
3042         }
3043
3044         kfree(allqueues);
3045         return 0;
3046
3047 rx_failed:
3048         kfree(adapter->tx_scrq);
3049         adapter->tx_scrq = NULL;
3050 tx_failed:
3051         for (i = 0; i < registered_queues; i++)
3052                 release_sub_crq_queue(adapter, allqueues[i], 1);
3053         kfree(allqueues);
3054         return -1;
3055 }
3056
3057 static void ibmvnic_send_req_caps(struct ibmvnic_adapter *adapter, int retry)
3058 {
3059         struct device *dev = &adapter->vdev->dev;
3060         union ibmvnic_crq crq;
3061         int max_entries;
3062
3063         if (!retry) {
3064                 /* Sub-CRQ entries are 32 byte long */
3065                 int entries_page = 4 * PAGE_SIZE / (sizeof(u64) * 4);
3066
3067                 if (adapter->min_tx_entries_per_subcrq > entries_page ||
3068                     adapter->min_rx_add_entries_per_subcrq > entries_page) {
3069                         dev_err(dev, "Fatal, invalid entries per sub-crq\n");
3070                         return;
3071                 }
3072
3073                 if (adapter->desired.mtu)
3074                         adapter->req_mtu = adapter->desired.mtu;
3075                 else
3076                         adapter->req_mtu = adapter->netdev->mtu + ETH_HLEN;
3077
3078                 if (!adapter->desired.tx_entries)
3079                         adapter->desired.tx_entries =
3080                                         adapter->max_tx_entries_per_subcrq;
3081                 if (!adapter->desired.rx_entries)
3082                         adapter->desired.rx_entries =
3083                                         adapter->max_rx_add_entries_per_subcrq;
3084
3085                 max_entries = IBMVNIC_MAX_LTB_SIZE /
3086                               (adapter->req_mtu + IBMVNIC_BUFFER_HLEN);
3087
3088                 if ((adapter->req_mtu + IBMVNIC_BUFFER_HLEN) *
3089                         adapter->desired.tx_entries > IBMVNIC_MAX_LTB_SIZE) {
3090                         adapter->desired.tx_entries = max_entries;
3091                 }
3092
3093                 if ((adapter->req_mtu + IBMVNIC_BUFFER_HLEN) *
3094                         adapter->desired.rx_entries > IBMVNIC_MAX_LTB_SIZE) {
3095                         adapter->desired.rx_entries = max_entries;
3096                 }
3097
3098                 if (adapter->desired.tx_entries)
3099                         adapter->req_tx_entries_per_subcrq =
3100                                         adapter->desired.tx_entries;
3101                 else
3102                         adapter->req_tx_entries_per_subcrq =
3103                                         adapter->max_tx_entries_per_subcrq;
3104
3105                 if (adapter->desired.rx_entries)
3106                         adapter->req_rx_add_entries_per_subcrq =
3107                                         adapter->desired.rx_entries;
3108                 else
3109                         adapter->req_rx_add_entries_per_subcrq =
3110                                         adapter->max_rx_add_entries_per_subcrq;
3111
3112                 if (adapter->desired.tx_queues)
3113                         adapter->req_tx_queues =
3114                                         adapter->desired.tx_queues;
3115                 else
3116                         adapter->req_tx_queues =
3117                                         adapter->opt_tx_comp_sub_queues;
3118
3119                 if (adapter->desired.rx_queues)
3120                         adapter->req_rx_queues =
3121                                         adapter->desired.rx_queues;
3122                 else
3123                         adapter->req_rx_queues =
3124                                         adapter->opt_rx_comp_queues;
3125
3126                 adapter->req_rx_add_queues = adapter->max_rx_add_queues;
3127         }
3128
3129         memset(&crq, 0, sizeof(crq));
3130         crq.request_capability.first = IBMVNIC_CRQ_CMD;
3131         crq.request_capability.cmd = REQUEST_CAPABILITY;
3132
3133         crq.request_capability.capability = cpu_to_be16(REQ_TX_QUEUES);
3134         crq.request_capability.number = cpu_to_be64(adapter->req_tx_queues);
3135         atomic_inc(&adapter->running_cap_crqs);
3136         ibmvnic_send_crq(adapter, &crq);
3137
3138         crq.request_capability.capability = cpu_to_be16(REQ_RX_QUEUES);
3139         crq.request_capability.number = cpu_to_be64(adapter->req_rx_queues);
3140         atomic_inc(&adapter->running_cap_crqs);
3141         ibmvnic_send_crq(adapter, &crq);
3142
3143         crq.request_capability.capability = cpu_to_be16(REQ_RX_ADD_QUEUES);
3144         crq.request_capability.number = cpu_to_be64(adapter->req_rx_add_queues);
3145         atomic_inc(&adapter->running_cap_crqs);
3146         ibmvnic_send_crq(adapter, &crq);
3147
3148         crq.request_capability.capability =
3149             cpu_to_be16(REQ_TX_ENTRIES_PER_SUBCRQ);
3150         crq.request_capability.number =
3151             cpu_to_be64(adapter->req_tx_entries_per_subcrq);
3152         atomic_inc(&adapter->running_cap_crqs);
3153         ibmvnic_send_crq(adapter, &crq);
3154
3155         crq.request_capability.capability =
3156             cpu_to_be16(REQ_RX_ADD_ENTRIES_PER_SUBCRQ);
3157         crq.request_capability.number =
3158             cpu_to_be64(adapter->req_rx_add_entries_per_subcrq);
3159         atomic_inc(&adapter->running_cap_crqs);
3160         ibmvnic_send_crq(adapter, &crq);
3161
3162         crq.request_capability.capability = cpu_to_be16(REQ_MTU);
3163         crq.request_capability.number = cpu_to_be64(adapter->req_mtu);
3164         atomic_inc(&adapter->running_cap_crqs);
3165         ibmvnic_send_crq(adapter, &crq);
3166
3167         if (adapter->netdev->flags & IFF_PROMISC) {
3168                 if (adapter->promisc_supported) {
3169                         crq.request_capability.capability =
3170                             cpu_to_be16(PROMISC_REQUESTED);
3171                         crq.request_capability.number = cpu_to_be64(1);
3172                         atomic_inc(&adapter->running_cap_crqs);
3173                         ibmvnic_send_crq(adapter, &crq);
3174                 }
3175         } else {
3176                 crq.request_capability.capability =
3177                     cpu_to_be16(PROMISC_REQUESTED);
3178                 crq.request_capability.number = cpu_to_be64(0);
3179                 atomic_inc(&adapter->running_cap_crqs);
3180                 ibmvnic_send_crq(adapter, &crq);
3181         }
3182 }
3183
3184 static int pending_scrq(struct ibmvnic_adapter *adapter,
3185                         struct ibmvnic_sub_crq_queue *scrq)
3186 {
3187         union sub_crq *entry = &scrq->msgs[scrq->cur];
3188
3189         if (entry->generic.first & IBMVNIC_CRQ_CMD_RSP)
3190                 return 1;
3191         else
3192                 return 0;
3193 }
3194
3195 static union sub_crq *ibmvnic_next_scrq(struct ibmvnic_adapter *adapter,
3196                                         struct ibmvnic_sub_crq_queue *scrq)
3197 {
3198         union sub_crq *entry;
3199         unsigned long flags;
3200
3201         spin_lock_irqsave(&scrq->lock, flags);
3202         entry = &scrq->msgs[scrq->cur];
3203         if (entry->generic.first & IBMVNIC_CRQ_CMD_RSP) {
3204                 if (++scrq->cur == scrq->size)
3205                         scrq->cur = 0;
3206         } else {
3207                 entry = NULL;
3208         }
3209         spin_unlock_irqrestore(&scrq->lock, flags);
3210
3211         return entry;
3212 }
3213
3214 static union ibmvnic_crq *ibmvnic_next_crq(struct ibmvnic_adapter *adapter)
3215 {
3216         struct ibmvnic_crq_queue *queue = &adapter->crq;
3217         union ibmvnic_crq *crq;
3218
3219         crq = &queue->msgs[queue->cur];
3220         if (crq->generic.first & IBMVNIC_CRQ_CMD_RSP) {
3221                 if (++queue->cur == queue->size)
3222                         queue->cur = 0;
3223         } else {
3224                 crq = NULL;
3225         }
3226
3227         return crq;
3228 }
3229
3230 static void print_subcrq_error(struct device *dev, int rc, const char *func)
3231 {
3232         switch (rc) {
3233         case H_PARAMETER:
3234                 dev_warn_ratelimited(dev,
3235                                      "%s failed: Send request is malformed or adapter failover pending. (rc=%d)\n",
3236                                      func, rc);
3237                 break;
3238         case H_CLOSED:
3239                 dev_warn_ratelimited(dev,
3240                                      "%s failed: Backing queue closed. Adapter is down or failover pending. (rc=%d)\n",
3241                                      func, rc);
3242                 break;
3243         default:
3244                 dev_err_ratelimited(dev, "%s failed: (rc=%d)\n", func, rc);
3245                 break;
3246         }
3247 }
3248
3249 static int send_subcrq(struct ibmvnic_adapter *adapter, u64 remote_handle,
3250                        union sub_crq *sub_crq)
3251 {
3252         unsigned int ua = adapter->vdev->unit_address;
3253         struct device *dev = &adapter->vdev->dev;
3254         u64 *u64_crq = (u64 *)sub_crq;
3255         int rc;
3256
3257         netdev_dbg(adapter->netdev,
3258                    "Sending sCRQ %016lx: %016lx %016lx %016lx %016lx\n",
3259                    (unsigned long int)cpu_to_be64(remote_handle),
3260                    (unsigned long int)cpu_to_be64(u64_crq[0]),
3261                    (unsigned long int)cpu_to_be64(u64_crq[1]),
3262                    (unsigned long int)cpu_to_be64(u64_crq[2]),
3263                    (unsigned long int)cpu_to_be64(u64_crq[3]));
3264
3265         /* Make sure the hypervisor sees the complete request */
3266         mb();
3267
3268         rc = plpar_hcall_norets(H_SEND_SUB_CRQ, ua,
3269                                 cpu_to_be64(remote_handle),
3270                                 cpu_to_be64(u64_crq[0]),
3271                                 cpu_to_be64(u64_crq[1]),
3272                                 cpu_to_be64(u64_crq[2]),
3273                                 cpu_to_be64(u64_crq[3]));
3274
3275         if (rc)
3276                 print_subcrq_error(dev, rc, __func__);
3277
3278         return rc;
3279 }
3280
3281 static int send_subcrq_indirect(struct ibmvnic_adapter *adapter,
3282                                 u64 remote_handle, u64 ioba, u64 num_entries)
3283 {
3284         unsigned int ua = adapter->vdev->unit_address;
3285         struct device *dev = &adapter->vdev->dev;
3286         int rc;
3287
3288         /* Make sure the hypervisor sees the complete request */
3289         mb();
3290         rc = plpar_hcall_norets(H_SEND_SUB_CRQ_INDIRECT, ua,
3291                                 cpu_to_be64(remote_handle),
3292                                 ioba, num_entries);
3293
3294         if (rc)
3295                 print_subcrq_error(dev, rc, __func__);
3296
3297         return rc;
3298 }
3299
3300 static int ibmvnic_send_crq(struct ibmvnic_adapter *adapter,
3301                             union ibmvnic_crq *crq)
3302 {
3303         unsigned int ua = adapter->vdev->unit_address;
3304         struct device *dev = &adapter->vdev->dev;
3305         u64 *u64_crq = (u64 *)crq;
3306         int rc;
3307
3308         netdev_dbg(adapter->netdev, "Sending CRQ: %016lx %016lx\n",
3309                    (unsigned long int)cpu_to_be64(u64_crq[0]),
3310                    (unsigned long int)cpu_to_be64(u64_crq[1]));
3311
3312         if (!adapter->crq.active &&
3313             crq->generic.first != IBMVNIC_CRQ_INIT_CMD) {
3314                 dev_warn(dev, "Invalid request detected while CRQ is inactive, possible device state change during reset\n");
3315                 return -EINVAL;
3316         }
3317
3318         /* Make sure the hypervisor sees the complete request */
3319         mb();
3320
3321         rc = plpar_hcall_norets(H_SEND_CRQ, ua,
3322                                 cpu_to_be64(u64_crq[0]),
3323                                 cpu_to_be64(u64_crq[1]));
3324
3325         if (rc) {
3326                 if (rc == H_CLOSED) {
3327                         dev_warn(dev, "CRQ Queue closed\n");
3328                         if (adapter->resetting)
3329                                 ibmvnic_reset(adapter, VNIC_RESET_FATAL);
3330                 }
3331
3332                 dev_warn(dev, "Send error (rc=%d)\n", rc);
3333         }
3334
3335         return rc;
3336 }
3337
3338 static int ibmvnic_send_crq_init(struct ibmvnic_adapter *adapter)
3339 {
3340         union ibmvnic_crq crq;
3341
3342         memset(&crq, 0, sizeof(crq));
3343         crq.generic.first = IBMVNIC_CRQ_INIT_CMD;
3344         crq.generic.cmd = IBMVNIC_CRQ_INIT;
3345         netdev_dbg(adapter->netdev, "Sending CRQ init\n");
3346
3347         return ibmvnic_send_crq(adapter, &crq);
3348 }
3349
3350 static int send_version_xchg(struct ibmvnic_adapter *adapter)
3351 {
3352         union ibmvnic_crq crq;
3353
3354         memset(&crq, 0, sizeof(crq));
3355         crq.version_exchange.first = IBMVNIC_CRQ_CMD;
3356         crq.version_exchange.cmd = VERSION_EXCHANGE;
3357         crq.version_exchange.version = cpu_to_be16(ibmvnic_version);
3358
3359         return ibmvnic_send_crq(adapter, &crq);
3360 }
3361
3362 struct vnic_login_client_data {
3363         u8      type;
3364         __be16  len;
3365         char    name[];
3366 } __packed;
3367
3368 static int vnic_client_data_len(struct ibmvnic_adapter *adapter)
3369 {
3370         int len;
3371
3372         /* Calculate the amount of buffer space needed for the
3373          * vnic client data in the login buffer. There are four entries,
3374          * OS name, LPAR name, device name, and a null last entry.
3375          */
3376         len = 4 * sizeof(struct vnic_login_client_data);
3377         len += 6; /* "Linux" plus NULL */
3378         len += strlen(utsname()->nodename) + 1;
3379         len += strlen(adapter->netdev->name) + 1;
3380
3381         return len;
3382 }
3383
3384 static void vnic_add_client_data(struct ibmvnic_adapter *adapter,
3385                                  struct vnic_login_client_data *vlcd)
3386 {
3387         const char *os_name = "Linux";
3388         int len;
3389
3390         /* Type 1 - LPAR OS */
3391         vlcd->type = 1;
3392         len = strlen(os_name) + 1;
3393         vlcd->len = cpu_to_be16(len);
3394         strncpy(vlcd->name, os_name, len);
3395         vlcd = (struct vnic_login_client_data *)(vlcd->name + len);
3396
3397         /* Type 2 - LPAR name */
3398         vlcd->type = 2;
3399         len = strlen(utsname()->nodename) + 1;
3400         vlcd->len = cpu_to_be16(len);
3401         strncpy(vlcd->name, utsname()->nodename, len);
3402         vlcd = (struct vnic_login_client_data *)(vlcd->name + len);
3403
3404         /* Type 3 - device name */
3405         vlcd->type = 3;
3406         len = strlen(adapter->netdev->name) + 1;
3407         vlcd->len = cpu_to_be16(len);
3408         strncpy(vlcd->name, adapter->netdev->name, len);
3409 }
3410
3411 static int send_login(struct ibmvnic_adapter *adapter)
3412 {
3413         struct ibmvnic_login_rsp_buffer *login_rsp_buffer;
3414         struct ibmvnic_login_buffer *login_buffer;
3415         struct device *dev = &adapter->vdev->dev;
3416         dma_addr_t rsp_buffer_token;
3417         dma_addr_t buffer_token;
3418         size_t rsp_buffer_size;
3419         union ibmvnic_crq crq;
3420         size_t buffer_size;
3421         __be64 *tx_list_p;
3422         __be64 *rx_list_p;
3423         int client_data_len;
3424         struct vnic_login_client_data *vlcd;
3425         int i;
3426
3427         if (!adapter->tx_scrq || !adapter->rx_scrq) {
3428                 netdev_err(adapter->netdev,
3429                            "RX or TX queues are not allocated, device login failed\n");
3430                 return -1;
3431         }
3432
3433         release_login_rsp_buffer(adapter);
3434         client_data_len = vnic_client_data_len(adapter);
3435
3436         buffer_size =
3437             sizeof(struct ibmvnic_login_buffer) +
3438             sizeof(u64) * (adapter->req_tx_queues + adapter->req_rx_queues) +
3439             client_data_len;
3440
3441         login_buffer = kzalloc(buffer_size, GFP_ATOMIC);
3442         if (!login_buffer)
3443                 goto buf_alloc_failed;
3444
3445         buffer_token = dma_map_single(dev, login_buffer, buffer_size,
3446                                       DMA_TO_DEVICE);
3447         if (dma_mapping_error(dev, buffer_token)) {
3448                 dev_err(dev, "Couldn't map login buffer\n");
3449                 goto buf_map_failed;
3450         }
3451
3452         rsp_buffer_size = sizeof(struct ibmvnic_login_rsp_buffer) +
3453                           sizeof(u64) * adapter->req_tx_queues +
3454                           sizeof(u64) * adapter->req_rx_queues +
3455                           sizeof(u64) * adapter->req_rx_queues +
3456                           sizeof(u8) * IBMVNIC_TX_DESC_VERSIONS;
3457
3458         login_rsp_buffer = kmalloc(rsp_buffer_size, GFP_ATOMIC);
3459         if (!login_rsp_buffer)
3460                 goto buf_rsp_alloc_failed;
3461
3462         rsp_buffer_token = dma_map_single(dev, login_rsp_buffer,
3463                                           rsp_buffer_size, DMA_FROM_DEVICE);
3464         if (dma_mapping_error(dev, rsp_buffer_token)) {
3465                 dev_err(dev, "Couldn't map login rsp buffer\n");
3466                 goto buf_rsp_map_failed;
3467         }
3468
3469         adapter->login_buf = login_buffer;
3470         adapter->login_buf_token = buffer_token;
3471         adapter->login_buf_sz = buffer_size;
3472         adapter->login_rsp_buf = login_rsp_buffer;
3473         adapter->login_rsp_buf_token = rsp_buffer_token;
3474         adapter->login_rsp_buf_sz = rsp_buffer_size;
3475
3476         login_buffer->len = cpu_to_be32(buffer_size);
3477         login_buffer->version = cpu_to_be32(INITIAL_VERSION_LB);
3478         login_buffer->num_txcomp_subcrqs = cpu_to_be32(adapter->req_tx_queues);
3479         login_buffer->off_txcomp_subcrqs =
3480             cpu_to_be32(sizeof(struct ibmvnic_login_buffer));
3481         login_buffer->num_rxcomp_subcrqs = cpu_to_be32(adapter->req_rx_queues);
3482         login_buffer->off_rxcomp_subcrqs =
3483             cpu_to_be32(sizeof(struct ibmvnic_login_buffer) +
3484                         sizeof(u64) * adapter->req_tx_queues);
3485         login_buffer->login_rsp_ioba = cpu_to_be32(rsp_buffer_token);
3486         login_buffer->login_rsp_len = cpu_to_be32(rsp_buffer_size);
3487
3488         tx_list_p = (__be64 *)((char *)login_buffer +
3489                                       sizeof(struct ibmvnic_login_buffer));
3490         rx_list_p = (__be64 *)((char *)login_buffer +
3491                                       sizeof(struct ibmvnic_login_buffer) +
3492                                       sizeof(u64) * adapter->req_tx_queues);
3493
3494         for (i = 0; i < adapter->req_tx_queues; i++) {
3495                 if (adapter->tx_scrq[i]) {
3496                         tx_list_p[i] = cpu_to_be64(adapter->tx_scrq[i]->
3497                                                    crq_num);
3498                 }
3499         }
3500
3501         for (i = 0; i < adapter->req_rx_queues; i++) {
3502                 if (adapter->rx_scrq[i]) {
3503                         rx_list_p[i] = cpu_to_be64(adapter->rx_scrq[i]->
3504                                                    crq_num);
3505                 }
3506         }
3507
3508         /* Insert vNIC login client data */
3509         vlcd = (struct vnic_login_client_data *)
3510                 ((char *)rx_list_p + (sizeof(u64) * adapter->req_rx_queues));
3511         login_buffer->client_data_offset =
3512                         cpu_to_be32((char *)vlcd - (char *)login_buffer);
3513         login_buffer->client_data_len = cpu_to_be32(client_data_len);
3514
3515         vnic_add_client_data(adapter, vlcd);
3516
3517         netdev_dbg(adapter->netdev, "Login Buffer:\n");
3518         for (i = 0; i < (adapter->login_buf_sz - 1) / 8 + 1; i++) {
3519                 netdev_dbg(adapter->netdev, "%016lx\n",
3520                            ((unsigned long int *)(adapter->login_buf))[i]);
3521         }
3522
3523         memset(&crq, 0, sizeof(crq));
3524         crq.login.first = IBMVNIC_CRQ_CMD;
3525         crq.login.cmd = LOGIN;
3526         crq.login.ioba = cpu_to_be32(buffer_token);
3527         crq.login.len = cpu_to_be32(buffer_size);
3528         ibmvnic_send_crq(adapter, &crq);
3529
3530         return 0;
3531
3532 buf_rsp_map_failed:
3533         kfree(login_rsp_buffer);
3534 buf_rsp_alloc_failed:
3535         dma_unmap_single(dev, buffer_token, buffer_size, DMA_TO_DEVICE);
3536 buf_map_failed:
3537         kfree(login_buffer);
3538 buf_alloc_failed:
3539         return -1;
3540 }
3541
3542 static int send_request_map(struct ibmvnic_adapter *adapter, dma_addr_t addr,
3543                             u32 len, u8 map_id)
3544 {
3545         union ibmvnic_crq crq;
3546
3547         memset(&crq, 0, sizeof(crq));
3548         crq.request_map.first = IBMVNIC_CRQ_CMD;
3549         crq.request_map.cmd = REQUEST_MAP;
3550         crq.request_map.map_id = map_id;
3551         crq.request_map.ioba = cpu_to_be32(addr);
3552         crq.request_map.len = cpu_to_be32(len);
3553         return ibmvnic_send_crq(adapter, &crq);
3554 }
3555
3556 static int send_request_unmap(struct ibmvnic_adapter *adapter, u8 map_id)
3557 {
3558         union ibmvnic_crq crq;
3559
3560         memset(&crq, 0, sizeof(crq));
3561         crq.request_unmap.first = IBMVNIC_CRQ_CMD;
3562         crq.request_unmap.cmd = REQUEST_UNMAP;
3563         crq.request_unmap.map_id = map_id;
3564         return ibmvnic_send_crq(adapter, &crq);
3565 }
3566
3567 static void send_map_query(struct ibmvnic_adapter *adapter)
3568 {
3569         union ibmvnic_crq crq;
3570
3571         memset(&crq, 0, sizeof(crq));
3572         crq.query_map.first = IBMVNIC_CRQ_CMD;
3573         crq.query_map.cmd = QUERY_MAP;
3574         ibmvnic_send_crq(adapter, &crq);
3575 }
3576
3577 /* Send a series of CRQs requesting various capabilities of the VNIC server */
3578 static void send_cap_queries(struct ibmvnic_adapter *adapter)
3579 {
3580         union ibmvnic_crq crq;
3581
3582         atomic_set(&adapter->running_cap_crqs, 0);
3583         memset(&crq, 0, sizeof(crq));
3584         crq.query_capability.first = IBMVNIC_CRQ_CMD;
3585         crq.query_capability.cmd = QUERY_CAPABILITY;
3586
3587         crq.query_capability.capability = cpu_to_be16(MIN_TX_QUEUES);
3588         atomic_inc(&adapter->running_cap_crqs);
3589         ibmvnic_send_crq(adapter, &crq);
3590
3591         crq.query_capability.capability = cpu_to_be16(MIN_RX_QUEUES);
3592         atomic_inc(&adapter->running_cap_crqs);
3593         ibmvnic_send_crq(adapter, &crq);
3594
3595         crq.query_capability.capability = cpu_to_be16(MIN_RX_ADD_QUEUES);
3596         atomic_inc(&adapter->running_cap_crqs);
3597         ibmvnic_send_crq(adapter, &crq);
3598
3599         crq.query_capability.capability = cpu_to_be16(MAX_TX_QUEUES);
3600         atomic_inc(&adapter->running_cap_crqs);
3601         ibmvnic_send_crq(adapter, &crq);
3602
3603         crq.query_capability.capability = cpu_to_be16(MAX_RX_QUEUES);
3604         atomic_inc(&adapter->running_cap_crqs);
3605         ibmvnic_send_crq(adapter, &crq);
3606
3607         crq.query_capability.capability = cpu_to_be16(MAX_RX_ADD_QUEUES);
3608         atomic_inc(&adapter->running_cap_crqs);
3609         ibmvnic_send_crq(adapter, &crq);
3610
3611         crq.query_capability.capability =
3612             cpu_to_be16(MIN_TX_ENTRIES_PER_SUBCRQ);
3613         atomic_inc(&adapter->running_cap_crqs);
3614         ibmvnic_send_crq(adapter, &crq);
3615
3616         crq.query_capability.capability =
3617             cpu_to_be16(MIN_RX_ADD_ENTRIES_PER_SUBCRQ);
3618         atomic_inc(&adapter->running_cap_crqs);
3619         ibmvnic_send_crq(adapter, &crq);
3620
3621         crq.query_capability.capability =
3622             cpu_to_be16(MAX_TX_ENTRIES_PER_SUBCRQ);
3623         atomic_inc(&adapter->running_cap_crqs);
3624         ibmvnic_send_crq(adapter, &crq);
3625
3626         crq.query_capability.capability =
3627             cpu_to_be16(MAX_RX_ADD_ENTRIES_PER_SUBCRQ);
3628         atomic_inc(&adapter->running_cap_crqs);
3629         ibmvnic_send_crq(adapter, &crq);
3630
3631         crq.query_capability.capability = cpu_to_be16(TCP_IP_OFFLOAD);
3632         atomic_inc(&adapter->running_cap_crqs);
3633         ibmvnic_send_crq(adapter, &crq);
3634
3635         crq.query_capability.capability = cpu_to_be16(PROMISC_SUPPORTED);
3636         atomic_inc(&adapter->running_cap_crqs);
3637         ibmvnic_send_crq(adapter, &crq);
3638
3639         crq.query_capability.capability = cpu_to_be16(MIN_MTU);
3640         atomic_inc(&adapter->running_cap_crqs);
3641         ibmvnic_send_crq(adapter, &crq);
3642
3643         crq.query_capability.capability = cpu_to_be16(MAX_MTU);
3644         atomic_inc(&adapter->running_cap_crqs);
3645         ibmvnic_send_crq(adapter, &crq);
3646
3647         crq.query_capability.capability = cpu_to_be16(MAX_MULTICAST_FILTERS);
3648         atomic_inc(&adapter->running_cap_crqs);
3649         ibmvnic_send_crq(adapter, &crq);
3650
3651         crq.query_capability.capability = cpu_to_be16(VLAN_HEADER_INSERTION);
3652         atomic_inc(&adapter->running_cap_crqs);
3653         ibmvnic_send_crq(adapter, &crq);
3654
3655         crq.query_capability.capability = cpu_to_be16(RX_VLAN_HEADER_INSERTION);
3656         atomic_inc(&adapter->running_cap_crqs);
3657         ibmvnic_send_crq(adapter, &crq);
3658
3659         crq.query_capability.capability = cpu_to_be16(MAX_TX_SG_ENTRIES);
3660         atomic_inc(&adapter->running_cap_crqs);
3661         ibmvnic_send_crq(adapter, &crq);
3662
3663         crq.query_capability.capability = cpu_to_be16(RX_SG_SUPPORTED);
3664         atomic_inc(&adapter->running_cap_crqs);
3665         ibmvnic_send_crq(adapter, &crq);
3666
3667         crq.query_capability.capability = cpu_to_be16(OPT_TX_COMP_SUB_QUEUES);
3668         atomic_inc(&adapter->running_cap_crqs);
3669         ibmvnic_send_crq(adapter, &crq);
3670
3671         crq.query_capability.capability = cpu_to_be16(OPT_RX_COMP_QUEUES);
3672         atomic_inc(&adapter->running_cap_crqs);
3673         ibmvnic_send_crq(adapter, &crq);
3674
3675         crq.query_capability.capability =
3676                         cpu_to_be16(OPT_RX_BUFADD_Q_PER_RX_COMP_Q);
3677         atomic_inc(&adapter->running_cap_crqs);
3678         ibmvnic_send_crq(adapter, &crq);
3679
3680         crq.query_capability.capability =
3681                         cpu_to_be16(OPT_TX_ENTRIES_PER_SUBCRQ);
3682         atomic_inc(&adapter->running_cap_crqs);
3683         ibmvnic_send_crq(adapter, &crq);
3684
3685         crq.query_capability.capability =
3686                         cpu_to_be16(OPT_RXBA_ENTRIES_PER_SUBCRQ);
3687         atomic_inc(&adapter->running_cap_crqs);
3688         ibmvnic_send_crq(adapter, &crq);
3689
3690         crq.query_capability.capability = cpu_to_be16(TX_RX_DESC_REQ);
3691         atomic_inc(&adapter->running_cap_crqs);
3692         ibmvnic_send_crq(adapter, &crq);
3693 }
3694
3695 static void handle_vpd_size_rsp(union ibmvnic_crq *crq,
3696                                 struct ibmvnic_adapter *adapter)
3697 {
3698         struct device *dev = &adapter->vdev->dev;
3699
3700         if (crq->get_vpd_size_rsp.rc.code) {
3701                 dev_err(dev, "Error retrieving VPD size, rc=%x\n",
3702                         crq->get_vpd_size_rsp.rc.code);
3703                 complete(&adapter->fw_done);
3704                 return;
3705         }
3706
3707         adapter->vpd->len = be64_to_cpu(crq->get_vpd_size_rsp.len);
3708         complete(&adapter->fw_done);
3709 }
3710
3711 static void handle_vpd_rsp(union ibmvnic_crq *crq,
3712                            struct ibmvnic_adapter *adapter)
3713 {
3714         struct device *dev = &adapter->vdev->dev;
3715         unsigned char *substr = NULL;
3716         u8 fw_level_len = 0;
3717
3718         memset(adapter->fw_version, 0, 32);
3719
3720         dma_unmap_single(dev, adapter->vpd->dma_addr, adapter->vpd->len,
3721                          DMA_FROM_DEVICE);
3722
3723         if (crq->get_vpd_rsp.rc.code) {
3724                 dev_err(dev, "Error retrieving VPD from device, rc=%x\n",
3725                         crq->get_vpd_rsp.rc.code);
3726                 goto complete;
3727         }
3728
3729         /* get the position of the firmware version info
3730          * located after the ASCII 'RM' substring in the buffer
3731          */
3732         substr = strnstr(adapter->vpd->buff, "RM", adapter->vpd->len);
3733         if (!substr) {
3734                 dev_info(dev, "Warning - No FW level has been provided in the VPD buffer by the VIOS Server\n");
3735                 goto complete;
3736         }
3737
3738         /* get length of firmware level ASCII substring */
3739         if ((substr + 2) < (adapter->vpd->buff + adapter->vpd->len)) {
3740                 fw_level_len = *(substr + 2);
3741         } else {
3742                 dev_info(dev, "Length of FW substr extrapolated VDP buff\n");
3743                 goto complete;
3744         }
3745
3746         /* copy firmware version string from vpd into adapter */
3747         if ((substr + 3 + fw_level_len) <
3748             (adapter->vpd->buff + adapter->vpd->len)) {
3749                 strncpy((char *)adapter->fw_version, substr + 3, fw_level_len);
3750         } else {
3751                 dev_info(dev, "FW substr extrapolated VPD buff\n");
3752         }
3753
3754 complete:
3755         if (adapter->fw_version[0] == '\0')
3756                 strncpy((char *)adapter->fw_version, "N/A", 3 * sizeof(char));
3757         complete(&adapter->fw_done);
3758 }
3759
3760 static void handle_query_ip_offload_rsp(struct ibmvnic_adapter *adapter)
3761 {
3762         struct device *dev = &adapter->vdev->dev;
3763         struct ibmvnic_query_ip_offload_buffer *buf = &adapter->ip_offload_buf;
3764         union ibmvnic_crq crq;
3765         int i;
3766
3767         dma_unmap_single(dev, adapter->ip_offload_tok,
3768                          sizeof(adapter->ip_offload_buf), DMA_FROM_DEVICE);
3769
3770         netdev_dbg(adapter->netdev, "Query IP Offload Buffer:\n");
3771         for (i = 0; i < (sizeof(adapter->ip_offload_buf) - 1) / 8 + 1; i++)
3772                 netdev_dbg(adapter->netdev, "%016lx\n",
3773                            ((unsigned long int *)(buf))[i]);
3774
3775         netdev_dbg(adapter->netdev, "ipv4_chksum = %d\n", buf->ipv4_chksum);
3776         netdev_dbg(adapter->netdev, "ipv6_chksum = %d\n", buf->ipv6_chksum);
3777         netdev_dbg(adapter->netdev, "tcp_ipv4_chksum = %d\n",
3778                    buf->tcp_ipv4_chksum);
3779         netdev_dbg(adapter->netdev, "tcp_ipv6_chksum = %d\n",
3780                    buf->tcp_ipv6_chksum);
3781         netdev_dbg(adapter->netdev, "udp_ipv4_chksum = %d\n",
3782                    buf->udp_ipv4_chksum);
3783         netdev_dbg(adapter->netdev, "udp_ipv6_chksum = %d\n",
3784                    buf->udp_ipv6_chksum);
3785         netdev_dbg(adapter->netdev, "large_tx_ipv4 = %d\n",
3786                    buf->large_tx_ipv4);
3787         netdev_dbg(adapter->netdev, "large_tx_ipv6 = %d\n",
3788                    buf->large_tx_ipv6);
3789         netdev_dbg(adapter->netdev, "large_rx_ipv4 = %d\n",
3790                    buf->large_rx_ipv4);
3791         netdev_dbg(adapter->netdev, "large_rx_ipv6 = %d\n",
3792                    buf->large_rx_ipv6);
3793         netdev_dbg(adapter->netdev, "max_ipv4_hdr_sz = %d\n",
3794                    buf->max_ipv4_header_size);
3795         netdev_dbg(adapter->netdev, "max_ipv6_hdr_sz = %d\n",
3796                    buf->max_ipv6_header_size);
3797         netdev_dbg(adapter->netdev, "max_tcp_hdr_size = %d\n",
3798                    buf->max_tcp_header_size);
3799         netdev_dbg(adapter->netdev, "max_udp_hdr_size = %d\n",
3800                    buf->max_udp_header_size);
3801         netdev_dbg(adapter->netdev, "max_large_tx_size = %d\n",
3802                    buf->max_large_tx_size);
3803         netdev_dbg(adapter->netdev, "max_large_rx_size = %d\n",
3804                    buf->max_large_rx_size);
3805         netdev_dbg(adapter->netdev, "ipv6_ext_hdr = %d\n",
3806                    buf->ipv6_extension_header);
3807         netdev_dbg(adapter->netdev, "tcp_pseudosum_req = %d\n",
3808                    buf->tcp_pseudosum_req);
3809         netdev_dbg(adapter->netdev, "num_ipv6_ext_hd = %d\n",
3810                    buf->num_ipv6_ext_headers);
3811         netdev_dbg(adapter->netdev, "off_ipv6_ext_hd = %d\n",
3812                    buf->off_ipv6_ext_headers);
3813
3814         adapter->ip_offload_ctrl_tok =
3815             dma_map_single(dev, &adapter->ip_offload_ctrl,
3816                            sizeof(adapter->ip_offload_ctrl), DMA_TO_DEVICE);
3817
3818         if (dma_mapping_error(dev, adapter->ip_offload_ctrl_tok)) {
3819                 dev_err(dev, "Couldn't map ip offload control buffer\n");
3820                 return;
3821         }
3822
3823         adapter->ip_offload_ctrl.len =
3824             cpu_to_be32(sizeof(adapter->ip_offload_ctrl));
3825         adapter->ip_offload_ctrl.version = cpu_to_be32(INITIAL_VERSION_IOB);
3826         adapter->ip_offload_ctrl.ipv4_chksum = buf->ipv4_chksum;
3827         adapter->ip_offload_ctrl.ipv6_chksum = buf->ipv6_chksum;
3828         adapter->ip_offload_ctrl.tcp_ipv4_chksum = buf->tcp_ipv4_chksum;
3829         adapter->ip_offload_ctrl.udp_ipv4_chksum = buf->udp_ipv4_chksum;
3830         adapter->ip_offload_ctrl.tcp_ipv6_chksum = buf->tcp_ipv6_chksum;
3831         adapter->ip_offload_ctrl.udp_ipv6_chksum = buf->udp_ipv6_chksum;
3832         adapter->ip_offload_ctrl.large_tx_ipv4 = buf->large_tx_ipv4;
3833         adapter->ip_offload_ctrl.large_tx_ipv6 = buf->large_tx_ipv6;
3834
3835         /* large_rx disabled for now, additional features needed */
3836         adapter->ip_offload_ctrl.large_rx_ipv4 = 0;
3837         adapter->ip_offload_ctrl.large_rx_ipv6 = 0;
3838
3839         adapter->netdev->features = NETIF_F_SG | NETIF_F_GSO;
3840
3841         if (buf->tcp_ipv4_chksum || buf->udp_ipv4_chksum)
3842                 adapter->netdev->features |= NETIF_F_IP_CSUM;
3843
3844         if (buf->tcp_ipv6_chksum || buf->udp_ipv6_chksum)
3845                 adapter->netdev->features |= NETIF_F_IPV6_CSUM;
3846
3847         if ((adapter->netdev->features &
3848             (NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM)))
3849                 adapter->netdev->features |= NETIF_F_RXCSUM;
3850
3851         if (buf->large_tx_ipv4)
3852                 adapter->netdev->features |= NETIF_F_TSO;
3853         if (buf->large_tx_ipv6)
3854                 adapter->netdev->features |= NETIF_F_TSO6;
3855
3856         adapter->netdev->hw_features |= adapter->netdev->features;
3857
3858         memset(&crq, 0, sizeof(crq));
3859         crq.control_ip_offload.first = IBMVNIC_CRQ_CMD;
3860         crq.control_ip_offload.cmd = CONTROL_IP_OFFLOAD;
3861         crq.control_ip_offload.len =
3862             cpu_to_be32(sizeof(adapter->ip_offload_ctrl));
3863         crq.control_ip_offload.ioba = cpu_to_be32(adapter->ip_offload_ctrl_tok);
3864         ibmvnic_send_crq(adapter, &crq);
3865 }
3866
3867 static const char *ibmvnic_fw_err_cause(u16 cause)
3868 {
3869         switch (cause) {
3870         case ADAPTER_PROBLEM:
3871                 return "adapter problem";
3872         case BUS_PROBLEM:
3873                 return "bus problem";
3874         case FW_PROBLEM:
3875                 return "firmware problem";
3876         case DD_PROBLEM:
3877                 return "device driver problem";
3878         case EEH_RECOVERY:
3879                 return "EEH recovery";
3880         case FW_UPDATED:
3881                 return "firmware updated";
3882         case LOW_MEMORY:
3883                 return "low Memory";
3884         default:
3885                 return "unknown";
3886         }
3887 }
3888
3889 static void handle_error_indication(union ibmvnic_crq *crq,
3890                                     struct ibmvnic_adapter *adapter)
3891 {
3892         struct device *dev = &adapter->vdev->dev;
3893         u16 cause;
3894
3895         cause = be16_to_cpu(crq->error_indication.error_cause);
3896
3897         dev_warn_ratelimited(dev,
3898                              "Firmware reports %serror, cause: %s. Starting recovery...\n",
3899                              crq->error_indication.flags
3900                                 & IBMVNIC_FATAL_ERROR ? "FATAL " : "",
3901                              ibmvnic_fw_err_cause(cause));
3902
3903         if (crq->error_indication.flags & IBMVNIC_FATAL_ERROR)
3904                 ibmvnic_reset(adapter, VNIC_RESET_FATAL);
3905         else
3906                 ibmvnic_reset(adapter, VNIC_RESET_NON_FATAL);
3907 }
3908
3909 static int handle_change_mac_rsp(union ibmvnic_crq *crq,
3910                                  struct ibmvnic_adapter *adapter)
3911 {
3912         struct net_device *netdev = adapter->netdev;
3913         struct device *dev = &adapter->vdev->dev;
3914         long rc;
3915
3916         rc = crq->change_mac_addr_rsp.rc.code;
3917         if (rc) {
3918                 dev_err(dev, "Error %ld in CHANGE_MAC_ADDR_RSP\n", rc);
3919                 goto out;
3920         }
3921         memcpy(netdev->dev_addr, &crq->change_mac_addr_rsp.mac_addr[0],
3922                ETH_ALEN);
3923 out:
3924         complete(&adapter->fw_done);
3925         return rc;
3926 }
3927
3928 static void handle_request_cap_rsp(union ibmvnic_crq *crq,
3929                                    struct ibmvnic_adapter *adapter)
3930 {
3931         struct device *dev = &adapter->vdev->dev;
3932         u64 *req_value;
3933         char *name;
3934
3935         atomic_dec(&adapter->running_cap_crqs);
3936         switch (be16_to_cpu(crq->request_capability_rsp.capability)) {
3937         case REQ_TX_QUEUES:
3938                 req_value = &adapter->req_tx_queues;
3939                 name = "tx";
3940                 break;
3941         case REQ_RX_QUEUES:
3942                 req_value = &adapter->req_rx_queues;
3943                 name = "rx";
3944                 break;
3945         case REQ_RX_ADD_QUEUES:
3946                 req_value = &adapter->req_rx_add_queues;
3947                 name = "rx_add";
3948                 break;
3949         case REQ_TX_ENTRIES_PER_SUBCRQ:
3950                 req_value = &adapter->req_tx_entries_per_subcrq;
3951                 name = "tx_entries_per_subcrq";
3952                 break;
3953         case REQ_RX_ADD_ENTRIES_PER_SUBCRQ:
3954                 req_value = &adapter->req_rx_add_entries_per_subcrq;
3955                 name = "rx_add_entries_per_subcrq";
3956                 break;
3957         case REQ_MTU:
3958                 req_value = &adapter->req_mtu;
3959                 name = "mtu";
3960                 break;
3961         case PROMISC_REQUESTED:
3962                 req_value = &adapter->promisc;
3963                 name = "promisc";
3964                 break;
3965         default:
3966                 dev_err(dev, "Got invalid cap request rsp %d\n",
3967                         crq->request_capability.capability);
3968                 return;
3969         }
3970
3971         switch (crq->request_capability_rsp.rc.code) {
3972         case SUCCESS:
3973                 break;
3974         case PARTIALSUCCESS:
3975                 dev_info(dev, "req=%lld, rsp=%ld in %s queue, retrying.\n",
3976                          *req_value,
3977                          (long int)be64_to_cpu(crq->request_capability_rsp.
3978                                                number), name);
3979
3980                 if (be16_to_cpu(crq->request_capability_rsp.capability) ==
3981                     REQ_MTU) {
3982                         pr_err("mtu of %llu is not supported. Reverting.\n",
3983                                *req_value);
3984                         *req_value = adapter->fallback.mtu;
3985                 } else {
3986                         *req_value =
3987                                 be64_to_cpu(crq->request_capability_rsp.number);
3988                 }
3989
3990                 ibmvnic_send_req_caps(adapter, 1);
3991                 return;
3992         default:
3993                 dev_err(dev, "Error %d in request cap rsp\n",
3994                         crq->request_capability_rsp.rc.code);
3995                 return;
3996         }
3997
3998         /* Done receiving requested capabilities, query IP offload support */
3999         if (atomic_read(&adapter->running_cap_crqs) == 0) {
4000                 union ibmvnic_crq newcrq;
4001                 int buf_sz = sizeof(struct ibmvnic_query_ip_offload_buffer);
4002                 struct ibmvnic_query_ip_offload_buffer *ip_offload_buf =
4003                     &adapter->ip_offload_buf;
4004
4005                 adapter->wait_capability = false;
4006                 adapter->ip_offload_tok = dma_map_single(dev, ip_offload_buf,
4007                                                          buf_sz,
4008                                                          DMA_FROM_DEVICE);
4009
4010                 if (dma_mapping_error(dev, adapter->ip_offload_tok)) {
4011                         if (!firmware_has_feature(FW_FEATURE_CMO))
4012                                 dev_err(dev, "Couldn't map offload buffer\n");
4013                         return;
4014                 }
4015
4016                 memset(&newcrq, 0, sizeof(newcrq));
4017                 newcrq.query_ip_offload.first = IBMVNIC_CRQ_CMD;
4018                 newcrq.query_ip_offload.cmd = QUERY_IP_OFFLOAD;
4019                 newcrq.query_ip_offload.len = cpu_to_be32(buf_sz);
4020                 newcrq.query_ip_offload.ioba =
4021                     cpu_to_be32(adapter->ip_offload_tok);
4022
4023                 ibmvnic_send_crq(adapter, &newcrq);
4024         }
4025 }
4026
4027 static int handle_login_rsp(union ibmvnic_crq *login_rsp_crq,
4028                             struct ibmvnic_adapter *adapter)
4029 {
4030         struct device *dev = &adapter->vdev->dev;
4031         struct net_device *netdev = adapter->netdev;
4032         struct ibmvnic_login_rsp_buffer *login_rsp = adapter->login_rsp_buf;
4033         struct ibmvnic_login_buffer *login = adapter->login_buf;
4034         int i;
4035
4036         dma_unmap_single(dev, adapter->login_buf_token, adapter->login_buf_sz,
4037                          DMA_TO_DEVICE);
4038         dma_unmap_single(dev, adapter->login_rsp_buf_token,
4039                          adapter->login_rsp_buf_sz, DMA_FROM_DEVICE);
4040
4041         /* If the number of queues requested can't be allocated by the
4042          * server, the login response will return with code 1. We will need
4043          * to resend the login buffer with fewer queues requested.
4044          */
4045         if (login_rsp_crq->generic.rc.code) {
4046                 adapter->init_done_rc = login_rsp_crq->generic.rc.code;
4047                 complete(&adapter->init_done);
4048                 return 0;
4049         }
4050
4051         netdev->mtu = adapter->req_mtu - ETH_HLEN;
4052
4053         netdev_dbg(adapter->netdev, "Login Response Buffer:\n");
4054         for (i = 0; i < (adapter->login_rsp_buf_sz - 1) / 8 + 1; i++) {
4055                 netdev_dbg(adapter->netdev, "%016lx\n",
4056                            ((unsigned long int *)(adapter->login_rsp_buf))[i]);
4057         }
4058
4059         /* Sanity checks */
4060         if (login->num_txcomp_subcrqs != login_rsp->num_txsubm_subcrqs ||
4061             (be32_to_cpu(login->num_rxcomp_subcrqs) *
4062              adapter->req_rx_add_queues !=
4063              be32_to_cpu(login_rsp->num_rxadd_subcrqs))) {
4064                 dev_err(dev, "FATAL: Inconsistent login and login rsp\n");
4065                 ibmvnic_remove(adapter->vdev);
4066                 return -EIO;
4067         }
4068         release_login_buffer(adapter);
4069         complete(&adapter->init_done);
4070
4071         return 0;
4072 }
4073
4074 static void handle_request_unmap_rsp(union ibmvnic_crq *crq,
4075                                      struct ibmvnic_adapter *adapter)
4076 {
4077         struct device *dev = &adapter->vdev->dev;
4078         long rc;
4079
4080         rc = crq->request_unmap_rsp.rc.code;
4081         if (rc)
4082                 dev_err(dev, "Error %ld in REQUEST_UNMAP_RSP\n", rc);
4083 }
4084
4085 static void handle_query_map_rsp(union ibmvnic_crq *crq,
4086                                  struct ibmvnic_adapter *adapter)
4087 {
4088         struct net_device *netdev = adapter->netdev;
4089         struct device *dev = &adapter->vdev->dev;
4090         long rc;
4091
4092         rc = crq->query_map_rsp.rc.code;
4093         if (rc) {
4094                 dev_err(dev, "Error %ld in QUERY_MAP_RSP\n", rc);
4095                 return;
4096         }
4097         netdev_dbg(netdev, "page_size = %d\ntot_pages = %d\nfree_pages = %d\n",
4098                    crq->query_map_rsp.page_size, crq->query_map_rsp.tot_pages,
4099                    crq->query_map_rsp.free_pages);
4100 }
4101
4102 static void handle_query_cap_rsp(union ibmvnic_crq *crq,
4103                                  struct ibmvnic_adapter *adapter)
4104 {
4105         struct net_device *netdev = adapter->netdev;
4106         struct device *dev = &adapter->vdev->dev;
4107         long rc;
4108
4109         atomic_dec(&adapter->running_cap_crqs);
4110         netdev_dbg(netdev, "Outstanding queries: %d\n",
4111                    atomic_read(&adapter->running_cap_crqs));
4112         rc = crq->query_capability.rc.code;
4113         if (rc) {
4114                 dev_err(dev, "Error %ld in QUERY_CAP_RSP\n", rc);
4115                 goto out;
4116         }
4117
4118         switch (be16_to_cpu(crq->query_capability.capability)) {
4119         case MIN_TX_QUEUES:
4120                 adapter->min_tx_queues =
4121                     be64_to_cpu(crq->query_capability.number);
4122                 netdev_dbg(netdev, "min_tx_queues = %lld\n",
4123                            adapter->min_tx_queues);
4124                 break;
4125         case MIN_RX_QUEUES:
4126                 adapter->min_rx_queues =
4127                     be64_to_cpu(crq->query_capability.number);
4128                 netdev_dbg(netdev, "min_rx_queues = %lld\n",
4129                            adapter->min_rx_queues);
4130                 break;
4131         case MIN_RX_ADD_QUEUES:
4132                 adapter->min_rx_add_queues =
4133                     be64_to_cpu(crq->query_capability.number);
4134                 netdev_dbg(netdev, "min_rx_add_queues = %lld\n",
4135                            adapter->min_rx_add_queues);
4136                 break;
4137         case MAX_TX_QUEUES:
4138                 adapter->max_tx_queues =
4139                     be64_to_cpu(crq->query_capability.number);
4140                 netdev_dbg(netdev, "max_tx_queues = %lld\n",
4141                            adapter->max_tx_queues);
4142                 break;
4143         case MAX_RX_QUEUES:
4144                 adapter->max_rx_queues =
4145                     be64_to_cpu(crq->query_capability.number);
4146                 netdev_dbg(netdev, "max_rx_queues = %lld\n",
4147                            adapter->max_rx_queues);
4148                 break;
4149         case MAX_RX_ADD_QUEUES:
4150                 adapter->max_rx_add_queues =
4151                     be64_to_cpu(crq->query_capability.number);
4152                 netdev_dbg(netdev, "max_rx_add_queues = %lld\n",
4153                            adapter->max_rx_add_queues);
4154                 break;
4155         case MIN_TX_ENTRIES_PER_SUBCRQ:
4156                 adapter->min_tx_entries_per_subcrq =
4157                     be64_to_cpu(crq->query_capability.number);
4158                 netdev_dbg(netdev, "min_tx_entries_per_subcrq = %lld\n",
4159                            adapter->min_tx_entries_per_subcrq);
4160                 break;
4161         case MIN_RX_ADD_ENTRIES_PER_SUBCRQ:
4162                 adapter->min_rx_add_entries_per_subcrq =
4163                     be64_to_cpu(crq->query_capability.number);
4164                 netdev_dbg(netdev, "min_rx_add_entrs_per_subcrq = %lld\n",
4165                            adapter->min_rx_add_entries_per_subcrq);
4166                 break;
4167         case MAX_TX_ENTRIES_PER_SUBCRQ:
4168                 adapter->max_tx_entries_per_subcrq =
4169                     be64_to_cpu(crq->query_capability.number);
4170                 netdev_dbg(netdev, "max_tx_entries_per_subcrq = %lld\n",
4171                            adapter->max_tx_entries_per_subcrq);
4172                 break;
4173         case MAX_RX_ADD_ENTRIES_PER_SUBCRQ:
4174                 adapter->max_rx_add_entries_per_subcrq =
4175                     be64_to_cpu(crq->query_capability.number);
4176                 netdev_dbg(netdev, "max_rx_add_entrs_per_subcrq = %lld\n",
4177                            adapter->max_rx_add_entries_per_subcrq);
4178                 break;
4179         case TCP_IP_OFFLOAD:
4180                 adapter->tcp_ip_offload =
4181                     be64_to_cpu(crq->query_capability.number);
4182                 netdev_dbg(netdev, "tcp_ip_offload = %lld\n",
4183                            adapter->tcp_ip_offload);
4184                 break;
4185         case PROMISC_SUPPORTED:
4186                 adapter->promisc_supported =
4187                     be64_to_cpu(crq->query_capability.number);
4188                 netdev_dbg(netdev, "promisc_supported = %lld\n",
4189                            adapter->promisc_supported);
4190                 break;
4191         case MIN_MTU:
4192                 adapter->min_mtu = be64_to_cpu(crq->query_capability.number);
4193                 netdev->min_mtu = adapter->min_mtu - ETH_HLEN;
4194                 netdev_dbg(netdev, "min_mtu = %lld\n", adapter->min_mtu);
4195                 break;
4196         case MAX_MTU:
4197                 adapter->max_mtu = be64_to_cpu(crq->query_capability.number);
4198                 netdev->max_mtu = adapter->max_mtu - ETH_HLEN;
4199                 netdev_dbg(netdev, "max_mtu = %lld\n", adapter->max_mtu);
4200                 break;
4201         case MAX_MULTICAST_FILTERS:
4202                 adapter->max_multicast_filters =
4203                     be64_to_cpu(crq->query_capability.number);
4204                 netdev_dbg(netdev, "max_multicast_filters = %lld\n",
4205                            adapter->max_multicast_filters);
4206                 break;
4207         case VLAN_HEADER_INSERTION:
4208                 adapter->vlan_header_insertion =
4209                     be64_to_cpu(crq->query_capability.number);
4210                 if (adapter->vlan_header_insertion)
4211                         netdev->features |= NETIF_F_HW_VLAN_STAG_TX;
4212                 netdev_dbg(netdev, "vlan_header_insertion = %lld\n",
4213                            adapter->vlan_header_insertion);
4214                 break;
4215         case RX_VLAN_HEADER_INSERTION:
4216                 adapter->rx_vlan_header_insertion =
4217                     be64_to_cpu(crq->query_capability.number);
4218                 netdev_dbg(netdev, "rx_vlan_header_insertion = %lld\n",
4219                            adapter->rx_vlan_header_insertion);
4220                 break;
4221         case MAX_TX_SG_ENTRIES:
4222                 adapter->max_tx_sg_entries =
4223                     be64_to_cpu(crq->query_capability.number);
4224                 netdev_dbg(netdev, "max_tx_sg_entries = %lld\n",
4225                            adapter->max_tx_sg_entries);
4226                 break;
4227         case RX_SG_SUPPORTED:
4228                 adapter->rx_sg_supported =
4229                     be64_to_cpu(crq->query_capability.number);
4230                 netdev_dbg(netdev, "rx_sg_supported = %lld\n",
4231                            adapter->rx_sg_supported);
4232                 break;
4233         case OPT_TX_COMP_SUB_QUEUES:
4234                 adapter->opt_tx_comp_sub_queues =
4235                     be64_to_cpu(crq->query_capability.number);
4236                 netdev_dbg(netdev, "opt_tx_comp_sub_queues = %lld\n",
4237                            adapter->opt_tx_comp_sub_queues);
4238                 break;
4239         case OPT_RX_COMP_QUEUES:
4240                 adapter->opt_rx_comp_queues =
4241                     be64_to_cpu(crq->query_capability.number);
4242                 netdev_dbg(netdev, "opt_rx_comp_queues = %lld\n",
4243                            adapter->opt_rx_comp_queues);
4244                 break;
4245         case OPT_RX_BUFADD_Q_PER_RX_COMP_Q:
4246                 adapter->opt_rx_bufadd_q_per_rx_comp_q =
4247                     be64_to_cpu(crq->query_capability.number);
4248                 netdev_dbg(netdev, "opt_rx_bufadd_q_per_rx_comp_q = %lld\n",
4249                            adapter->opt_rx_bufadd_q_per_rx_comp_q);
4250                 break;
4251         case OPT_TX_ENTRIES_PER_SUBCRQ:
4252                 adapter->opt_tx_entries_per_subcrq =
4253                     be64_to_cpu(crq->query_capability.number);
4254                 netdev_dbg(netdev, "opt_tx_entries_per_subcrq = %lld\n",
4255                            adapter->opt_tx_entries_per_subcrq);
4256                 break;
4257         case OPT_RXBA_ENTRIES_PER_SUBCRQ:
4258                 adapter->opt_rxba_entries_per_subcrq =
4259                     be64_to_cpu(crq->query_capability.number);
4260                 netdev_dbg(netdev, "opt_rxba_entries_per_subcrq = %lld\n",
4261                            adapter->opt_rxba_entries_per_subcrq);
4262                 break;
4263         case TX_RX_DESC_REQ:
4264                 adapter->tx_rx_desc_req = crq->query_capability.number;
4265                 netdev_dbg(netdev, "tx_rx_desc_req = %llx\n",
4266                            adapter->tx_rx_desc_req);
4267                 break;
4268
4269         default:
4270                 netdev_err(netdev, "Got invalid cap rsp %d\n",
4271                            crq->query_capability.capability);
4272         }
4273
4274 out:
4275         if (atomic_read(&adapter->running_cap_crqs) == 0) {
4276                 adapter->wait_capability = false;
4277                 ibmvnic_send_req_caps(adapter, 0);
4278         }
4279 }
4280
4281 static void ibmvnic_handle_crq(union ibmvnic_crq *crq,
4282                                struct ibmvnic_adapter *adapter)
4283 {
4284         struct ibmvnic_generic_crq *gen_crq = &crq->generic;
4285         struct net_device *netdev = adapter->netdev;
4286         struct device *dev = &adapter->vdev->dev;
4287         u64 *u64_crq = (u64 *)crq;
4288         long rc;
4289
4290         netdev_dbg(netdev, "Handling CRQ: %016lx %016lx\n",
4291                    (unsigned long int)cpu_to_be64(u64_crq[0]),
4292                    (unsigned long int)cpu_to_be64(u64_crq[1]));
4293         switch (gen_crq->first) {
4294         case IBMVNIC_CRQ_INIT_RSP:
4295                 switch (gen_crq->cmd) {
4296                 case IBMVNIC_CRQ_INIT:
4297                         dev_info(dev, "Partner initialized\n");
4298                         adapter->from_passive_init = true;
4299                         adapter->failover_pending = false;
4300                         if (!completion_done(&adapter->init_done)) {
4301                                 complete(&adapter->init_done);
4302                                 adapter->init_done_rc = -EIO;
4303                         }
4304                         ibmvnic_reset(adapter, VNIC_RESET_FAILOVER);
4305                         break;
4306                 case IBMVNIC_CRQ_INIT_COMPLETE:
4307                         dev_info(dev, "Partner initialization complete\n");
4308                         adapter->crq.active = true;
4309                         send_version_xchg(adapter);
4310                         break;
4311                 default:
4312                         dev_err(dev, "Unknown crq cmd: %d\n", gen_crq->cmd);
4313                 }
4314                 return;
4315         case IBMVNIC_CRQ_XPORT_EVENT:
4316                 netif_carrier_off(netdev);
4317                 adapter->crq.active = false;
4318                 if (adapter->resetting)
4319                         adapter->force_reset_recovery = true;
4320                 if (gen_crq->cmd == IBMVNIC_PARTITION_MIGRATED) {
4321                         dev_info(dev, "Migrated, re-enabling adapter\n");
4322                         ibmvnic_reset(adapter, VNIC_RESET_MOBILITY);
4323                 } else if (gen_crq->cmd == IBMVNIC_DEVICE_FAILOVER) {
4324                         dev_info(dev, "Backing device failover detected\n");
4325                         adapter->failover_pending = true;
4326                 } else {
4327                         /* The adapter lost the connection */
4328                         dev_err(dev, "Virtual Adapter failed (rc=%d)\n",
4329                                 gen_crq->cmd);
4330                         ibmvnic_reset(adapter, VNIC_RESET_FATAL);
4331                 }
4332                 return;
4333         case IBMVNIC_CRQ_CMD_RSP:
4334                 break;
4335         default:
4336                 dev_err(dev, "Got an invalid msg type 0x%02x\n",
4337                         gen_crq->first);
4338                 return;
4339         }
4340
4341         switch (gen_crq->cmd) {
4342         case VERSION_EXCHANGE_RSP:
4343                 rc = crq->version_exchange_rsp.rc.code;
4344                 if (rc) {
4345                         dev_err(dev, "Error %ld in VERSION_EXCHG_RSP\n", rc);
4346                         break;
4347                 }
4348                 dev_info(dev, "Partner protocol version is %d\n",
4349                          crq->version_exchange_rsp.version);
4350                 if (be16_to_cpu(crq->version_exchange_rsp.version) <
4351                     ibmvnic_version)
4352                         ibmvnic_version =
4353                             be16_to_cpu(crq->version_exchange_rsp.version);
4354                 send_cap_queries(adapter);
4355                 break;
4356         case QUERY_CAPABILITY_RSP:
4357                 handle_query_cap_rsp(crq, adapter);
4358                 break;
4359         case QUERY_MAP_RSP:
4360                 handle_query_map_rsp(crq, adapter);
4361                 break;
4362         case REQUEST_MAP_RSP:
4363                 adapter->fw_done_rc = crq->request_map_rsp.rc.code;
4364                 complete(&adapter->fw_done);
4365                 break;
4366         case REQUEST_UNMAP_RSP:
4367                 handle_request_unmap_rsp(crq, adapter);
4368                 break;
4369         case REQUEST_CAPABILITY_RSP:
4370                 handle_request_cap_rsp(crq, adapter);
4371                 break;
4372         case LOGIN_RSP:
4373                 netdev_dbg(netdev, "Got Login Response\n");
4374                 handle_login_rsp(crq, adapter);
4375                 break;
4376         case LOGICAL_LINK_STATE_RSP:
4377                 netdev_dbg(netdev,
4378                            "Got Logical Link State Response, state: %d rc: %d\n",
4379                            crq->logical_link_state_rsp.link_state,
4380                            crq->logical_link_state_rsp.rc.code);
4381                 adapter->logical_link_state =
4382                     crq->logical_link_state_rsp.link_state;
4383                 adapter->init_done_rc = crq->logical_link_state_rsp.rc.code;
4384                 complete(&adapter->init_done);
4385                 break;
4386         case LINK_STATE_INDICATION:
4387                 netdev_dbg(netdev, "Got Logical Link State Indication\n");
4388                 adapter->phys_link_state =
4389                     crq->link_state_indication.phys_link_state;
4390                 adapter->logical_link_state =
4391                     crq->link_state_indication.logical_link_state;
4392                 break;
4393         case CHANGE_MAC_ADDR_RSP:
4394                 netdev_dbg(netdev, "Got MAC address change Response\n");
4395                 adapter->fw_done_rc = handle_change_mac_rsp(crq, adapter);
4396                 break;
4397         case ERROR_INDICATION:
4398                 netdev_dbg(netdev, "Got Error Indication\n");
4399                 handle_error_indication(crq, adapter);
4400                 break;
4401         case REQUEST_STATISTICS_RSP:
4402                 netdev_dbg(netdev, "Got Statistics Response\n");
4403                 complete(&adapter->stats_done);
4404                 break;
4405         case QUERY_IP_OFFLOAD_RSP:
4406                 netdev_dbg(netdev, "Got Query IP offload Response\n");
4407                 handle_query_ip_offload_rsp(adapter);
4408                 break;
4409         case MULTICAST_CTRL_RSP:
4410                 netdev_dbg(netdev, "Got multicast control Response\n");
4411                 break;
4412         case CONTROL_IP_OFFLOAD_RSP:
4413                 netdev_dbg(netdev, "Got Control IP offload Response\n");
4414                 dma_unmap_single(dev, adapter->ip_offload_ctrl_tok,
4415                                  sizeof(adapter->ip_offload_ctrl),
4416                                  DMA_TO_DEVICE);
4417                 complete(&adapter->init_done);
4418                 break;
4419         case COLLECT_FW_TRACE_RSP:
4420                 netdev_dbg(netdev, "Got Collect firmware trace Response\n");
4421                 complete(&adapter->fw_done);
4422                 break;
4423         case GET_VPD_SIZE_RSP:
4424                 handle_vpd_size_rsp(crq, adapter);
4425                 break;
4426         case GET_VPD_RSP:
4427                 handle_vpd_rsp(crq, adapter);
4428                 break;
4429         default:
4430                 netdev_err(netdev, "Got an invalid cmd type 0x%02x\n",
4431                            gen_crq->cmd);
4432         }
4433 }
4434
4435 static irqreturn_t ibmvnic_interrupt(int irq, void *instance)
4436 {
4437         struct ibmvnic_adapter *adapter = instance;
4438
4439         tasklet_schedule(&adapter->tasklet);
4440         return IRQ_HANDLED;
4441 }
4442
4443 static void ibmvnic_tasklet(void *data)
4444 {
4445         struct ibmvnic_adapter *adapter = data;
4446         struct ibmvnic_crq_queue *queue = &adapter->crq;
4447         union ibmvnic_crq *crq;
4448         unsigned long flags;
4449         bool done = false;
4450
4451         spin_lock_irqsave(&queue->lock, flags);
4452         while (!done) {
4453                 /* Pull all the valid messages off the CRQ */
4454                 while ((crq = ibmvnic_next_crq(adapter)) != NULL) {
4455                         ibmvnic_handle_crq(crq, adapter);
4456                         crq->generic.first = 0;
4457                 }
4458
4459                 /* remain in tasklet until all
4460                  * capabilities responses are received
4461                  */
4462                 if (!adapter->wait_capability)
4463                         done = true;
4464         }
4465         /* if capabilities CRQ's were sent in this tasklet, the following
4466          * tasklet must wait until all responses are received
4467          */
4468         if (atomic_read(&adapter->running_cap_crqs) != 0)
4469                 adapter->wait_capability = true;
4470         spin_unlock_irqrestore(&queue->lock, flags);
4471 }
4472
4473 static int ibmvnic_reenable_crq_queue(struct ibmvnic_adapter *adapter)
4474 {
4475         struct vio_dev *vdev = adapter->vdev;
4476         int rc;
4477
4478         do {
4479                 rc = plpar_hcall_norets(H_ENABLE_CRQ, vdev->unit_address);
4480         } while (rc == H_IN_PROGRESS || rc == H_BUSY || H_IS_LONG_BUSY(rc));
4481
4482         if (rc)
4483                 dev_err(&vdev->dev, "Error enabling adapter (rc=%d)\n", rc);
4484
4485         return rc;
4486 }
4487
4488 static int ibmvnic_reset_crq(struct ibmvnic_adapter *adapter)
4489 {
4490         struct ibmvnic_crq_queue *crq = &adapter->crq;
4491         struct device *dev = &adapter->vdev->dev;
4492         struct vio_dev *vdev = adapter->vdev;
4493         int rc;
4494
4495         /* Close the CRQ */
4496         do {
4497                 rc = plpar_hcall_norets(H_FREE_CRQ, vdev->unit_address);
4498         } while (rc == H_BUSY || H_IS_LONG_BUSY(rc));
4499
4500         /* Clean out the queue */
4501         memset(crq->msgs, 0, PAGE_SIZE);
4502         crq->cur = 0;
4503         crq->active = false;
4504
4505         /* And re-open it again */
4506         rc = plpar_hcall_norets(H_REG_CRQ, vdev->unit_address,
4507                                 crq->msg_token, PAGE_SIZE);
4508
4509         if (rc == H_CLOSED)
4510                 /* Adapter is good, but other end is not ready */
4511                 dev_warn(dev, "Partner adapter not ready\n");
4512         else if (rc != 0)
4513                 dev_warn(dev, "Couldn't register crq (rc=%d)\n", rc);
4514
4515         return rc;
4516 }
4517
4518 static void release_crq_queue(struct ibmvnic_adapter *adapter)
4519 {
4520         struct ibmvnic_crq_queue *crq = &adapter->crq;
4521         struct vio_dev *vdev = adapter->vdev;
4522         long rc;
4523
4524         if (!crq->msgs)
4525                 return;
4526
4527         netdev_dbg(adapter->netdev, "Releasing CRQ\n");
4528         free_irq(vdev->irq, adapter);
4529         tasklet_kill(&adapter->tasklet);
4530         do {
4531                 rc = plpar_hcall_norets(H_FREE_CRQ, vdev->unit_address);
4532         } while (rc == H_BUSY || H_IS_LONG_BUSY(rc));
4533
4534         dma_unmap_single(&vdev->dev, crq->msg_token, PAGE_SIZE,
4535                          DMA_BIDIRECTIONAL);
4536         free_page((unsigned long)crq->msgs);
4537         crq->msgs = NULL;
4538         crq->active = false;
4539 }
4540
4541 static int init_crq_queue(struct ibmvnic_adapter *adapter)
4542 {
4543         struct ibmvnic_crq_queue *crq = &adapter->crq;
4544         struct device *dev = &adapter->vdev->dev;
4545         struct vio_dev *vdev = adapter->vdev;
4546         int rc, retrc = -ENOMEM;
4547
4548         if (crq->msgs)
4549                 return 0;
4550
4551         crq->msgs = (union ibmvnic_crq *)get_zeroed_page(GFP_KERNEL);
4552         /* Should we allocate more than one page? */
4553
4554         if (!crq->msgs)
4555                 return -ENOMEM;
4556
4557         crq->size = PAGE_SIZE / sizeof(*crq->msgs);
4558         crq->msg_token = dma_map_single(dev, crq->msgs, PAGE_SIZE,
4559                                         DMA_BIDIRECTIONAL);
4560         if (dma_mapping_error(dev, crq->msg_token))
4561                 goto map_failed;
4562
4563         rc = plpar_hcall_norets(H_REG_CRQ, vdev->unit_address,
4564                                 crq->msg_token, PAGE_SIZE);
4565
4566         if (rc == H_RESOURCE)
4567                 /* maybe kexecing and resource is busy. try a reset */
4568                 rc = ibmvnic_reset_crq(adapter);
4569         retrc = rc;
4570
4571         if (rc == H_CLOSED) {
4572                 dev_warn(dev, "Partner adapter not ready\n");
4573         } else if (rc) {
4574                 dev_warn(dev, "Error %d opening adapter\n", rc);
4575                 goto reg_crq_failed;
4576         }
4577
4578         retrc = 0;
4579
4580         tasklet_init(&adapter->tasklet, (void *)ibmvnic_tasklet,
4581                      (unsigned long)adapter);
4582
4583         netdev_dbg(adapter->netdev, "registering irq 0x%x\n", vdev->irq);
4584         rc = request_irq(vdev->irq, ibmvnic_interrupt, 0, IBMVNIC_NAME,
4585                          adapter);
4586         if (rc) {
4587                 dev_err(dev, "Couldn't register irq 0x%x. rc=%d\n",
4588                         vdev->irq, rc);
4589                 goto req_irq_failed;
4590         }
4591
4592         rc = vio_enable_interrupts(vdev);
4593         if (rc) {
4594                 dev_err(dev, "Error %d enabling interrupts\n", rc);
4595                 goto req_irq_failed;
4596         }
4597
4598         crq->cur = 0;
4599         spin_lock_init(&crq->lock);
4600
4601         return retrc;
4602
4603 req_irq_failed:
4604         tasklet_kill(&adapter->tasklet);
4605         do {
4606                 rc = plpar_hcall_norets(H_FREE_CRQ, vdev->unit_address);
4607         } while (rc == H_BUSY || H_IS_LONG_BUSY(rc));
4608 reg_crq_failed:
4609         dma_unmap_single(dev, crq->msg_token, PAGE_SIZE, DMA_BIDIRECTIONAL);
4610 map_failed:
4611         free_page((unsigned long)crq->msgs);
4612         crq->msgs = NULL;
4613         return retrc;
4614 }
4615
4616 static int ibmvnic_reset_init(struct ibmvnic_adapter *adapter)
4617 {
4618         struct device *dev = &adapter->vdev->dev;
4619         unsigned long timeout = msecs_to_jiffies(30000);
4620         u64 old_num_rx_queues, old_num_tx_queues;
4621         int rc;
4622
4623         adapter->from_passive_init = false;
4624
4625         old_num_rx_queues = adapter->req_rx_queues;
4626         old_num_tx_queues = adapter->req_tx_queues;
4627
4628         init_completion(&adapter->init_done);
4629         adapter->init_done_rc = 0;
4630         ibmvnic_send_crq_init(adapter);
4631         if (!wait_for_completion_timeout(&adapter->init_done, timeout)) {
4632                 dev_err(dev, "Initialization sequence timed out\n");
4633                 return -1;
4634         }
4635
4636         if (adapter->init_done_rc) {
4637                 release_crq_queue(adapter);
4638                 return adapter->init_done_rc;
4639         }
4640
4641         if (adapter->from_passive_init) {
4642                 adapter->state = VNIC_OPEN;
4643                 adapter->from_passive_init = false;
4644                 return -1;
4645         }
4646
4647         if (adapter->resetting && !adapter->wait_for_reset &&
4648             adapter->reset_reason != VNIC_RESET_MOBILITY) {
4649                 if (adapter->req_rx_queues != old_num_rx_queues ||
4650                     adapter->req_tx_queues != old_num_tx_queues) {
4651                         release_sub_crqs(adapter, 0);
4652                         rc = init_sub_crqs(adapter);
4653                 } else {
4654                         rc = reset_sub_crq_queues(adapter);
4655                 }
4656         } else {
4657                 rc = init_sub_crqs(adapter);
4658         }
4659
4660         if (rc) {
4661                 dev_err(dev, "Initialization of sub crqs failed\n");
4662                 release_crq_queue(adapter);
4663                 return rc;
4664         }
4665
4666         rc = init_sub_crq_irqs(adapter);
4667         if (rc) {
4668                 dev_err(dev, "Failed to initialize sub crq irqs\n");
4669                 release_crq_queue(adapter);
4670         }
4671
4672         return rc;
4673 }
4674
4675 static int ibmvnic_init(struct ibmvnic_adapter *adapter)
4676 {
4677         struct device *dev = &adapter->vdev->dev;
4678         unsigned long timeout = msecs_to_jiffies(30000);
4679         int rc;
4680
4681         adapter->from_passive_init = false;
4682
4683         init_completion(&adapter->init_done);
4684         adapter->init_done_rc = 0;
4685         ibmvnic_send_crq_init(adapter);
4686         if (!wait_for_completion_timeout(&adapter->init_done, timeout)) {
4687                 dev_err(dev, "Initialization sequence timed out\n");
4688                 return -1;
4689         }
4690
4691         if (adapter->init_done_rc) {
4692                 release_crq_queue(adapter);
4693                 return adapter->init_done_rc;
4694         }
4695
4696         if (adapter->from_passive_init) {
4697                 adapter->state = VNIC_OPEN;
4698                 adapter->from_passive_init = false;
4699                 return -1;
4700         }
4701
4702         rc = init_sub_crqs(adapter);
4703         if (rc) {
4704                 dev_err(dev, "Initialization of sub crqs failed\n");
4705                 release_crq_queue(adapter);
4706                 return rc;
4707         }
4708
4709         rc = init_sub_crq_irqs(adapter);
4710         if (rc) {
4711                 dev_err(dev, "Failed to initialize sub crq irqs\n");
4712                 release_crq_queue(adapter);
4713         }
4714
4715         return rc;
4716 }
4717
4718 static struct device_attribute dev_attr_failover;
4719
4720 static int ibmvnic_probe(struct vio_dev *dev, const struct vio_device_id *id)
4721 {
4722         struct ibmvnic_adapter *adapter;
4723         struct net_device *netdev;
4724         unsigned char *mac_addr_p;
4725         int rc;
4726
4727         dev_dbg(&dev->dev, "entering ibmvnic_probe for UA 0x%x\n",
4728                 dev->unit_address);
4729
4730         mac_addr_p = (unsigned char *)vio_get_attribute(dev,
4731                                                         VETH_MAC_ADDR, NULL);
4732         if (!mac_addr_p) {
4733                 dev_err(&dev->dev,
4734                         "(%s:%3.3d) ERROR: Can't find MAC_ADDR attribute\n",
4735                         __FILE__, __LINE__);
4736                 return 0;
4737         }
4738
4739         netdev = alloc_etherdev_mq(sizeof(struct ibmvnic_adapter),
4740                                    IBMVNIC_MAX_QUEUES);
4741         if (!netdev)
4742                 return -ENOMEM;
4743
4744         adapter = netdev_priv(netdev);
4745         adapter->state = VNIC_PROBING;
4746         dev_set_drvdata(&dev->dev, netdev);
4747         adapter->vdev = dev;
4748         adapter->netdev = netdev;
4749
4750         ether_addr_copy(adapter->mac_addr, mac_addr_p);
4751         ether_addr_copy(netdev->dev_addr, adapter->mac_addr);
4752         netdev->irq = dev->irq;
4753         netdev->netdev_ops = &ibmvnic_netdev_ops;
4754         netdev->ethtool_ops = &ibmvnic_ethtool_ops;
4755         SET_NETDEV_DEV(netdev, &dev->dev);
4756
4757         spin_lock_init(&adapter->stats_lock);
4758
4759         INIT_WORK(&adapter->ibmvnic_reset, __ibmvnic_reset);
4760         INIT_LIST_HEAD(&adapter->rwi_list);
4761         spin_lock_init(&adapter->rwi_lock);
4762         adapter->resetting = false;
4763
4764         adapter->mac_change_pending = false;
4765
4766         do {
4767                 rc = init_crq_queue(adapter);
4768                 if (rc) {
4769                         dev_err(&dev->dev, "Couldn't initialize crq. rc=%d\n",
4770                                 rc);
4771                         goto ibmvnic_init_fail;
4772                 }
4773
4774                 rc = ibmvnic_init(adapter);
4775                 if (rc && rc != EAGAIN)
4776                         goto ibmvnic_init_fail;
4777         } while (rc == EAGAIN);
4778
4779         rc = init_stats_buffers(adapter);
4780         if (rc)
4781                 goto ibmvnic_init_fail;
4782
4783         rc = init_stats_token(adapter);
4784         if (rc)
4785                 goto ibmvnic_stats_fail;
4786
4787         netdev->mtu = adapter->req_mtu - ETH_HLEN;
4788         netdev->min_mtu = adapter->min_mtu - ETH_HLEN;
4789         netdev->max_mtu = adapter->max_mtu - ETH_HLEN;
4790
4791         rc = device_create_file(&dev->dev, &dev_attr_failover);
4792         if (rc)
4793                 goto ibmvnic_dev_file_err;
4794
4795         netif_carrier_off(netdev);
4796         rc = register_netdev(netdev);
4797         if (rc) {
4798                 dev_err(&dev->dev, "failed to register netdev rc=%d\n", rc);
4799                 goto ibmvnic_register_fail;
4800         }
4801         dev_info(&dev->dev, "ibmvnic registered\n");
4802
4803         adapter->state = VNIC_PROBED;
4804
4805         adapter->wait_for_reset = false;
4806
4807         return 0;
4808
4809 ibmvnic_register_fail:
4810         device_remove_file(&dev->dev, &dev_attr_failover);
4811
4812 ibmvnic_dev_file_err:
4813         release_stats_token(adapter);
4814
4815 ibmvnic_stats_fail:
4816         release_stats_buffers(adapter);
4817
4818 ibmvnic_init_fail:
4819         release_sub_crqs(adapter, 1);
4820         release_crq_queue(adapter);
4821         free_netdev(netdev);
4822
4823         return rc;
4824 }
4825
4826 static int ibmvnic_remove(struct vio_dev *dev)
4827 {
4828         struct net_device *netdev = dev_get_drvdata(&dev->dev);
4829         struct ibmvnic_adapter *adapter = netdev_priv(netdev);
4830
4831         adapter->state = VNIC_REMOVING;
4832         rtnl_lock();
4833         unregister_netdevice(netdev);
4834
4835         release_resources(adapter);
4836         release_sub_crqs(adapter, 1);
4837         release_crq_queue(adapter);
4838
4839         release_stats_token(adapter);
4840         release_stats_buffers(adapter);
4841
4842         adapter->state = VNIC_REMOVED;
4843
4844         rtnl_unlock();
4845         device_remove_file(&dev->dev, &dev_attr_failover);
4846         free_netdev(netdev);
4847         dev_set_drvdata(&dev->dev, NULL);
4848
4849         return 0;
4850 }
4851
4852 static ssize_t failover_store(struct device *dev, struct device_attribute *attr,
4853                               const char *buf, size_t count)
4854 {
4855         struct net_device *netdev = dev_get_drvdata(dev);
4856         struct ibmvnic_adapter *adapter = netdev_priv(netdev);
4857         unsigned long retbuf[PLPAR_HCALL_BUFSIZE];
4858         __be64 session_token;
4859         long rc;
4860
4861         if (!sysfs_streq(buf, "1"))
4862                 return -EINVAL;
4863
4864         rc = plpar_hcall(H_VIOCTL, retbuf, adapter->vdev->unit_address,
4865                          H_GET_SESSION_TOKEN, 0, 0, 0);
4866         if (rc) {
4867                 netdev_err(netdev, "Couldn't retrieve session token, rc %ld\n",
4868                            rc);
4869                 return -EINVAL;
4870         }
4871
4872         session_token = (__be64)retbuf[0];
4873         netdev_dbg(netdev, "Initiating client failover, session id %llx\n",
4874                    be64_to_cpu(session_token));
4875         rc = plpar_hcall_norets(H_VIOCTL, adapter->vdev->unit_address,
4876                                 H_SESSION_ERR_DETECTED, session_token, 0, 0);
4877         if (rc) {
4878                 netdev_err(netdev, "Client initiated failover failed, rc %ld\n",
4879                            rc);
4880                 return -EINVAL;
4881         }
4882
4883         return count;
4884 }
4885
4886 static DEVICE_ATTR_WO(failover);
4887
4888 static unsigned long ibmvnic_get_desired_dma(struct vio_dev *vdev)
4889 {
4890         struct net_device *netdev = dev_get_drvdata(&vdev->dev);
4891         struct ibmvnic_adapter *adapter;
4892         struct iommu_table *tbl;
4893         unsigned long ret = 0;
4894         int i;
4895
4896         tbl = get_iommu_table_base(&vdev->dev);
4897
4898         /* netdev inits at probe time along with the structures we need below*/
4899         if (!netdev)
4900                 return IOMMU_PAGE_ALIGN(IBMVNIC_IO_ENTITLEMENT_DEFAULT, tbl);
4901
4902         adapter = netdev_priv(netdev);
4903
4904         ret += PAGE_SIZE; /* the crq message queue */
4905         ret += IOMMU_PAGE_ALIGN(sizeof(struct ibmvnic_statistics), tbl);
4906
4907         for (i = 0; i < adapter->req_tx_queues + adapter->req_rx_queues; i++)
4908                 ret += 4 * PAGE_SIZE; /* the scrq message queue */
4909
4910         for (i = 0; i < be32_to_cpu(adapter->login_rsp_buf->num_rxadd_subcrqs);
4911              i++)
4912                 ret += adapter->rx_pool[i].size *
4913                     IOMMU_PAGE_ALIGN(adapter->rx_pool[i].buff_size, tbl);
4914
4915         return ret;
4916 }
4917
4918 static int ibmvnic_resume(struct device *dev)
4919 {
4920         struct net_device *netdev = dev_get_drvdata(dev);
4921         struct ibmvnic_adapter *adapter = netdev_priv(netdev);
4922
4923         if (adapter->state != VNIC_OPEN)
4924                 return 0;
4925
4926         tasklet_schedule(&adapter->tasklet);
4927
4928         return 0;
4929 }
4930
4931 static const struct vio_device_id ibmvnic_device_table[] = {
4932         {"network", "IBM,vnic"},
4933         {"", "" }
4934 };
4935 MODULE_DEVICE_TABLE(vio, ibmvnic_device_table);
4936
4937 static const struct dev_pm_ops ibmvnic_pm_ops = {
4938         .resume = ibmvnic_resume
4939 };
4940
4941 static struct vio_driver ibmvnic_driver = {
4942         .id_table       = ibmvnic_device_table,
4943         .probe          = ibmvnic_probe,
4944         .remove         = ibmvnic_remove,
4945         .get_desired_dma = ibmvnic_get_desired_dma,
4946         .name           = ibmvnic_driver_name,
4947         .pm             = &ibmvnic_pm_ops,
4948 };
4949
4950 /* module functions */
4951 static int __init ibmvnic_module_init(void)
4952 {
4953         pr_info("%s: %s %s\n", ibmvnic_driver_name, ibmvnic_driver_string,
4954                 IBMVNIC_DRIVER_VERSION);
4955
4956         return vio_register_driver(&ibmvnic_driver);
4957 }
4958
4959 static void __exit ibmvnic_module_exit(void)
4960 {
4961         vio_unregister_driver(&ibmvnic_driver);
4962 }
4963
4964 module_init(ibmvnic_module_init);
4965 module_exit(ibmvnic_module_exit);