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                 if (&adapter->napi[i]) {
 777                         netdev_dbg(adapter->netdev,
 778                                    "Releasing napi[%d]\n", i);
 779                         netif_napi_del(&adapter->napi[i]);
 780                 }
 781         }
 782
 783         kfree(adapter->napi);
 784         adapter->napi = NULL;
 785         adapter->num_active_rx_napi = 0;
 786         adapter->napi_enabled = false;
 787 }
 788
 789 static int ibmvnic_login(struct net_device *netdev)
 790 {
 791         struct ibmvnic_adapter *adapter = netdev_priv(netdev);
 792         unsigned long timeout = msecs_to_jiffies(30000);
 793         int retry_count = 0;
 794         bool retry;
 795         int rc;
 796
 797         do {
 798                 retry = false;
 799                 if (retry_count > IBMVNIC_MAX_QUEUES) {
 800                         netdev_warn(netdev, "Login attempts exceeded\n");
 801                         return -1;
 802                 }
 803
 804                 adapter->init_done_rc = 0;
 805                 reinit_completion(&adapter->init_done);
 806                 rc = send_login(adapter);
 807                 if (rc) {
 808                         netdev_warn(netdev, "Unable to login\n");
 809                         return rc;
 810                 }
 811
 812                 if (!wait_for_completion_timeout(&adapter->init_done,
 813                                                  timeout)) {
 814                         netdev_warn(netdev, "Login timed out\n");
 815                         return -1;
 816                 }
 817
 818                 if (adapter->init_done_rc == PARTIALSUCCESS) {
 819                         retry_count++;
 820                         release_sub_crqs(adapter, 1);
 821
 822                         retry = true;
 823                         netdev_dbg(netdev,
 824                                    "Received partial success, retrying...\n");
 825                         adapter->init_done_rc = 0;
 826                         reinit_completion(&adapter->init_done);
 827                         send_cap_queries(adapter);
 828                         if (!wait_for_completion_timeout(&adapter->init_done,
 829                                                          timeout)) {
 830                                 netdev_warn(netdev,
 831                                             "Capabilities query timed out\n");
 832                                 return -1;
 833                         }
 834
 835                         rc = init_sub_crqs(adapter);
 836                         if (rc) {
 837                                 netdev_warn(netdev,
 838                                             "SCRQ initialization failed\n");
 839                                 return -1;
 840                         }
 841
 842                         rc = init_sub_crq_irqs(adapter);
 843                         if (rc) {
 844                                 netdev_warn(netdev,
 845                                             "SCRQ irq initialization failed\n");
 846                                 return -1;
 847                         }
 848                 } else if (adapter->init_done_rc) {
 849                         netdev_warn(netdev, "Adapter login failed\n");
 850                         return -1;
 851                 }
 852         } while (retry);
 853
 854         /* handle pending MAC address changes after successful login */
 855         if (adapter->mac_change_pending) {
 856                 __ibmvnic_set_mac(netdev, &adapter->desired.mac);
 857                 adapter->mac_change_pending = false;
 858         }
 859
 860         return 0;
 861 }
 862
 863 static void release_login_buffer(struct ibmvnic_adapter *adapter)
 864 {
 865         kfree(adapter->login_buf);
 866         adapter->login_buf = NULL;
 867 }
 868
 869 static void release_login_rsp_buffer(struct ibmvnic_adapter *adapter)
 870 {
 871         kfree(adapter->login_rsp_buf);
 872         adapter->login_rsp_buf = NULL;
 873 }
 874
 875 static void release_resources(struct ibmvnic_adapter *adapter)
 876 {
 877         release_vpd_data(adapter);
 878
 879         release_tx_pools(adapter);
 880         release_rx_pools(adapter);
 881
 882         release_napi(adapter);
 883         release_login_rsp_buffer(adapter);
 884 }
 885
 886 static int set_link_state(struct ibmvnic_adapter *adapter, u8 link_state)
 887 {
 888         struct net_device *netdev = adapter->netdev;
 889         unsigned long timeout = msecs_to_jiffies(30000);
 890         union ibmvnic_crq crq;
 891         bool resend;
 892         int rc;
 893
 894         netdev_dbg(netdev, "setting link state %d\n", link_state);
 895
 896         memset(&crq, 0, sizeof(crq));
 897         crq.logical_link_state.first = IBMVNIC_CRQ_CMD;
 898         crq.logical_link_state.cmd = LOGICAL_LINK_STATE;
 899         crq.logical_link_state.link_state = link_state;
 900
 901         do {
 902                 resend = false;
 903
 904                 reinit_completion(&adapter->init_done);
 905                 rc = ibmvnic_send_crq(adapter, &crq);
 906                 if (rc) {
 907                         netdev_err(netdev, "Failed to set link state\n");
 908                         return rc;
 909                 }
 910
 911                 if (!wait_for_completion_timeout(&adapter->init_done,
 912                                                  timeout)) {
 913                         netdev_err(netdev, "timeout setting link state\n");
 914                         return -1;
 915                 }
 916
 917                 if (adapter->init_done_rc == 1) {
 918                         /* Partuial success, delay and re-send */
 919                         mdelay(1000);
 920                         resend = true;
 921                 } else if (adapter->init_done_rc) {
 922                         netdev_warn(netdev, "Unable to set link state, rc=%d\n",
 923                                     adapter->init_done_rc);
 924                         return adapter->init_done_rc;
 925                 }
 926         } while (resend);
 927
 928         return 0;
 929 }
 930
 931 static int set_real_num_queues(struct net_device *netdev)
 932 {
 933         struct ibmvnic_adapter *adapter = netdev_priv(netdev);
 934         int rc;
 935
 936         netdev_dbg(netdev, "Setting real tx/rx queues (%llx/%llx)\n",
 937                    adapter->req_tx_queues, adapter->req_rx_queues);
 938
 939         rc = netif_set_real_num_tx_queues(netdev, adapter->req_tx_queues);
 940         if (rc) {
 941                 netdev_err(netdev, "failed to set the number of tx queues\n");
 942                 return rc;
 943         }
 944
 945         rc = netif_set_real_num_rx_queues(netdev, adapter->req_rx_queues);
 946         if (rc)
 947                 netdev_err(netdev, "failed to set the number of rx queues\n");
 948
 949         return rc;
 950 }
 951
 952 static int ibmvnic_get_vpd(struct ibmvnic_adapter *adapter)
 953 {
 954         struct device *dev = &adapter->vdev->dev;
 955         union ibmvnic_crq crq;
 956         int len = 0;
 957         int rc;
 958
 959         if (adapter->vpd->buff)
 960                 len = adapter->vpd->len;
 961
 962         init_completion(&adapter->fw_done);
 963         crq.get_vpd_size.first = IBMVNIC_CRQ_CMD;
 964         crq.get_vpd_size.cmd = GET_VPD_SIZE;
 965         rc = ibmvnic_send_crq(adapter, &crq);
 966         if (rc)
 967                 return rc;
 968         wait_for_completion(&adapter->fw_done);
 969
 970         if (!adapter->vpd->len)
 971                 return -ENODATA;
 972
 973         if (!adapter->vpd->buff)
 974                 adapter->vpd->buff = kzalloc(adapter->vpd->len, GFP_KERNEL);
 975         else if (adapter->vpd->len != len)
 976                 adapter->vpd->buff =
 977                         krealloc(adapter->vpd->buff,
 978                                  adapter->vpd->len, GFP_KERNEL);
 979
 980         if (!adapter->vpd->buff) {
 981                 dev_err(dev, "Could allocate VPD buffer\n");
 982                 return -ENOMEM;
 983         }
 984
 985         adapter->vpd->dma_addr =
 986                 dma_map_single(dev, adapter->vpd->buff, adapter->vpd->len,
 987                                DMA_FROM_DEVICE);
 988         if (dma_mapping_error(dev, adapter->vpd->dma_addr)) {
 989                 dev_err(dev, "Could not map VPD buffer\n");
 990                 kfree(adapter->vpd->buff);
 991                 adapter->vpd->buff = NULL;
 992                 return -ENOMEM;
 993         }
 994
 995         reinit_completion(&adapter->fw_done);
 996         crq.get_vpd.first = IBMVNIC_CRQ_CMD;
 997         crq.get_vpd.cmd = GET_VPD;
 998         crq.get_vpd.ioba = cpu_to_be32(adapter->vpd->dma_addr);
 999         crq.get_vpd.len = cpu_to_be32((u32)adapter->vpd->len);
1000         rc = ibmvnic_send_crq(adapter, &crq);
1001         if (rc) {
1002                 kfree(adapter->vpd->buff);
1003                 adapter->vpd->buff = NULL;
1004                 return rc;
1005         }
1006         wait_for_completion(&adapter->fw_done);
1007
1008         return 0;
1009 }
1010
1011 static int init_resources(struct ibmvnic_adapter *adapter)
1012 {
1013         struct net_device *netdev = adapter->netdev;
1014         int rc;
1015
1016         rc = set_real_num_queues(netdev);
1017         if (rc)
1018                 return rc;
1019
1020         adapter->vpd = kzalloc(sizeof(*adapter->vpd), GFP_KERNEL);
1021         if (!adapter->vpd)
1022                 return -ENOMEM;
1023
1024         /* Vital Product Data (VPD) */
1025         rc = ibmvnic_get_vpd(adapter);
1026         if (rc) {
1027                 netdev_err(netdev, "failed to initialize Vital Product Data (VPD)\n");
1028                 return rc;
1029         }
1030
1031         adapter->map_id = 1;
1032
1033         rc = init_napi(adapter);
1034         if (rc)
1035                 return rc;
1036
1037         send_map_query(adapter);
1038
1039         rc = init_rx_pools(netdev);
1040         if (rc)
1041                 return rc;
1042
1043         rc = init_tx_pools(netdev);
1044         return rc;
1045 }
1046
1047 static int __ibmvnic_open(struct net_device *netdev)
1048 {
1049         struct ibmvnic_adapter *adapter = netdev_priv(netdev);
1050         enum vnic_state prev_state = adapter->state;
1051         int i, rc;
1052
1053         adapter->state = VNIC_OPENING;
1054         replenish_pools(adapter);
1055         ibmvnic_napi_enable(adapter);
1056
1057         /* We're ready to receive frames, enable the sub-crq interrupts and
1058          * set the logical link state to up
1059          */
1060         for (i = 0; i < adapter->req_rx_queues; i++) {
1061                 netdev_dbg(netdev, "Enabling rx_scrq[%d] irq\n", i);
1062                 if (prev_state == VNIC_CLOSED)
1063                         enable_irq(adapter->rx_scrq[i]->irq);
1064                 enable_scrq_irq(adapter, adapter->rx_scrq[i]);
1065         }
1066
1067         for (i = 0; i < adapter->req_tx_queues; i++) {
1068                 netdev_dbg(netdev, "Enabling tx_scrq[%d] irq\n", i);
1069                 if (prev_state == VNIC_CLOSED)
1070                         enable_irq(adapter->tx_scrq[i]->irq);
1071                 enable_scrq_irq(adapter, adapter->tx_scrq[i]);
1072         }
1073
1074         rc = set_link_state(adapter, IBMVNIC_LOGICAL_LNK_UP);
1075         if (rc) {
1076                 for (i = 0; i < adapter->req_rx_queues; i++)
1077                         napi_disable(&adapter->napi[i]);
1078                 release_resources(adapter);
1079                 return rc;
1080         }
1081
1082         netif_tx_start_all_queues(netdev);
1083
1084         if (prev_state == VNIC_CLOSED) {
1085                 for (i = 0; i < adapter->req_rx_queues; i++)
1086                         napi_schedule(&adapter->napi[i]);
1087         }
1088
1089         adapter->state = VNIC_OPEN;
1090         return rc;
1091 }
1092
1093 static int ibmvnic_open(struct net_device *netdev)
1094 {
1095         struct ibmvnic_adapter *adapter = netdev_priv(netdev);
1096         int rc;
1097
1098         /* If device failover is pending, just set device state and return.
1099          * Device operation will be handled by reset routine.
1100          */
1101         if (adapter->failover_pending) {
1102                 adapter->state = VNIC_OPEN;
1103                 return 0;
1104         }
1105
1106         if (adapter->state != VNIC_CLOSED) {
1107                 rc = ibmvnic_login(netdev);
1108                 if (rc)
1109                         return rc;
1110
1111                 rc = init_resources(adapter);
1112                 if (rc) {
1113                         netdev_err(netdev, "failed to initialize resources\n");
1114                         release_resources(adapter);
1115                         return rc;
1116                 }
1117         }
1118
1119         rc = __ibmvnic_open(netdev);
1120         netif_carrier_on(netdev);
1121
1122         return rc;
1123 }
1124
1125 static void clean_rx_pools(struct ibmvnic_adapter *adapter)
1126 {
1127         struct ibmvnic_rx_pool *rx_pool;
1128         struct ibmvnic_rx_buff *rx_buff;
1129         u64 rx_entries;
1130         int rx_scrqs;
1131         int i, j;
1132
1133         if (!adapter->rx_pool)
1134                 return;
1135
1136         rx_scrqs = adapter->num_active_rx_pools;
1137         rx_entries = adapter->req_rx_add_entries_per_subcrq;
1138
1139         /* Free any remaining skbs in the rx buffer pools */
1140         for (i = 0; i < rx_scrqs; i++) {
1141                 rx_pool = &adapter->rx_pool[i];
1142                 if (!rx_pool || !rx_pool->rx_buff)
1143                         continue;
1144
1145                 netdev_dbg(adapter->netdev, "Cleaning rx_pool[%d]\n", i);
1146                 for (j = 0; j < rx_entries; j++) {
1147                         rx_buff = &rx_pool->rx_buff[j];
1148                         if (rx_buff && rx_buff->skb) {
1149                                 dev_kfree_skb_any(rx_buff->skb);
1150                                 rx_buff->skb = NULL;
1151                         }
1152                 }
1153         }
1154 }
1155
1156 static void clean_one_tx_pool(struct ibmvnic_adapter *adapter,
1157                               struct ibmvnic_tx_pool *tx_pool)
1158 {
1159         struct ibmvnic_tx_buff *tx_buff;
1160         u64 tx_entries;
1161         int i;
1162
1163         if (!tx_pool || !tx_pool->tx_buff)
1164                 return;
1165
1166         tx_entries = tx_pool->num_buffers;
1167
1168         for (i = 0; i < tx_entries; i++) {
1169                 tx_buff = &tx_pool->tx_buff[i];
1170                 if (tx_buff && tx_buff->skb) {
1171                         dev_kfree_skb_any(tx_buff->skb);
1172                         tx_buff->skb = NULL;
1173                 }
1174         }
1175 }
1176
1177 static void clean_tx_pools(struct ibmvnic_adapter *adapter)
1178 {
1179         int tx_scrqs;
1180         int i;
1181
1182         if (!adapter->tx_pool || !adapter->tso_pool)
1183                 return;
1184
1185         tx_scrqs = adapter->num_active_tx_pools;
1186
1187         /* Free any remaining skbs in the tx buffer pools */
1188         for (i = 0; i < tx_scrqs; i++) {
1189                 netdev_dbg(adapter->netdev, "Cleaning tx_pool[%d]\n", i);
1190                 clean_one_tx_pool(adapter, &adapter->tx_pool[i]);
1191                 clean_one_tx_pool(adapter, &adapter->tso_pool[i]);
1192         }
1193 }
1194
1195 static void ibmvnic_disable_irqs(struct ibmvnic_adapter *adapter)
1196 {
1197         struct net_device *netdev = adapter->netdev;
1198         int i;
1199
1200         if (adapter->tx_scrq) {
1201                 for (i = 0; i < adapter->req_tx_queues; i++)
1202                         if (adapter->tx_scrq[i]->irq) {
1203                                 netdev_dbg(netdev,
1204                                            "Disabling tx_scrq[%d] irq\n", i);
1205                                 disable_scrq_irq(adapter, adapter->tx_scrq[i]);
1206                                 disable_irq(adapter->tx_scrq[i]->irq);
1207                         }
1208         }
1209
1210         if (adapter->rx_scrq) {
1211                 for (i = 0; i < adapter->req_rx_queues; i++) {
1212                         if (adapter->rx_scrq[i]->irq) {
1213                                 netdev_dbg(netdev,
1214                                            "Disabling rx_scrq[%d] irq\n", i);
1215                                 disable_scrq_irq(adapter, adapter->rx_scrq[i]);
1216                                 disable_irq(adapter->rx_scrq[i]->irq);
1217                         }
1218                 }
1219         }
1220 }
1221
1222 static void ibmvnic_cleanup(struct net_device *netdev)
1223 {
1224         struct ibmvnic_adapter *adapter = netdev_priv(netdev);
1225
1226         /* ensure that transmissions are stopped if called by do_reset */
1227         if (adapter->resetting)
1228                 netif_tx_disable(netdev);
1229         else
1230                 netif_tx_stop_all_queues(netdev);
1231
1232         ibmvnic_napi_disable(adapter);
1233         ibmvnic_disable_irqs(adapter);
1234
1235         clean_rx_pools(adapter);
1236         clean_tx_pools(adapter);
1237 }
1238
1239 static int __ibmvnic_close(struct net_device *netdev)
1240 {
1241         struct ibmvnic_adapter *adapter = netdev_priv(netdev);
1242         int rc = 0;
1243
1244         adapter->state = VNIC_CLOSING;
1245         rc = set_link_state(adapter, IBMVNIC_LOGICAL_LNK_DN);
1246         if (rc)
1247                 return rc;
1248         adapter->state = VNIC_CLOSED;
1249         return 0;
1250 }
1251
1252 static int ibmvnic_close(struct net_device *netdev)
1253 {
1254         struct ibmvnic_adapter *adapter = netdev_priv(netdev);
1255         int rc;
1256
1257         /* If device failover is pending, just set device state and return.
1258          * Device operation will be handled by reset routine.
1259          */
1260         if (adapter->failover_pending) {
1261                 adapter->state = VNIC_CLOSED;
1262                 return 0;
1263         }
1264
1265         rc = __ibmvnic_close(netdev);
1266         ibmvnic_cleanup(netdev);
1267
1268         return rc;
1269 }
1270
1271 /**
1272  * build_hdr_data - creates L2/L3/L4 header data buffer
1273  * @hdr_field - bitfield determining needed headers
1274  * @skb - socket buffer
1275  * @hdr_len - array of header lengths
1276  * @tot_len - total length of data
1277  *
1278  * Reads hdr_field to determine which headers are needed by firmware.
1279  * Builds a buffer containing these headers.  Saves individual header
1280  * lengths and total buffer length to be used to build descriptors.
1281  */
1282 static int build_hdr_data(u8 hdr_field, struct sk_buff *skb,
1283                           int *hdr_len, u8 *hdr_data)
1284 {
1285         int len = 0;
1286         u8 *hdr;
1287
1288         if (skb_vlan_tagged(skb) && !skb_vlan_tag_present(skb))
1289                 hdr_len[0] = sizeof(struct vlan_ethhdr);
1290         else
1291                 hdr_len[0] = sizeof(struct ethhdr);
1292
1293         if (skb->protocol == htons(ETH_P_IP)) {
1294                 hdr_len[1] = ip_hdr(skb)->ihl * 4;
1295                 if (ip_hdr(skb)->protocol == IPPROTO_TCP)
1296                         hdr_len[2] = tcp_hdrlen(skb);
1297                 else if (ip_hdr(skb)->protocol == IPPROTO_UDP)
1298                         hdr_len[2] = sizeof(struct udphdr);
1299         } else if (skb->protocol == htons(ETH_P_IPV6)) {
1300                 hdr_len[1] = sizeof(struct ipv6hdr);
1301                 if (ipv6_hdr(skb)->nexthdr == IPPROTO_TCP)
1302                         hdr_len[2] = tcp_hdrlen(skb);
1303                 else if (ipv6_hdr(skb)->nexthdr == IPPROTO_UDP)
1304                         hdr_len[2] = sizeof(struct udphdr);
1305         } else if (skb->protocol == htons(ETH_P_ARP)) {
1306                 hdr_len[1] = arp_hdr_len(skb->dev);
1307                 hdr_len[2] = 0;
1308         }
1309
1310         memset(hdr_data, 0, 120);
1311         if ((hdr_field >> 6) & 1) {
1312                 hdr = skb_mac_header(skb);
1313                 memcpy(hdr_data, hdr, hdr_len[0]);
1314                 len += hdr_len[0];
1315         }
1316
1317         if ((hdr_field >> 5) & 1) {
1318                 hdr = skb_network_header(skb);
1319                 memcpy(hdr_data + len, hdr, hdr_len[1]);
1320                 len += hdr_len[1];
1321         }
1322
1323         if ((hdr_field >> 4) & 1) {
1324                 hdr = skb_transport_header(skb);
1325                 memcpy(hdr_data + len, hdr, hdr_len[2]);
1326                 len += hdr_len[2];
1327         }
1328         return len;
1329 }
1330
1331 /**
1332  * create_hdr_descs - create header and header extension descriptors
1333  * @hdr_field - bitfield determining needed headers
1334  * @data - buffer containing header data
1335  * @len - length of data buffer
1336  * @hdr_len - array of individual header lengths
1337  * @scrq_arr - descriptor array
1338  *
1339  * Creates header and, if needed, header extension descriptors and
1340  * places them in a descriptor array, scrq_arr
1341  */
1342
1343 static int create_hdr_descs(u8 hdr_field, u8 *hdr_data, int len, int *hdr_len,
1344                             union sub_crq *scrq_arr)
1345 {
1346         union sub_crq hdr_desc;
1347         int tmp_len = len;
1348         int num_descs = 0;
1349         u8 *data, *cur;
1350         int tmp;
1351
1352         while (tmp_len > 0) {
1353                 cur = hdr_data + len - tmp_len;
1354
1355                 memset(&hdr_desc, 0, sizeof(hdr_desc));
1356                 if (cur != hdr_data) {
1357                         data = hdr_desc.hdr_ext.data;
1358                         tmp = tmp_len > 29 ? 29 : tmp_len;
1359                         hdr_desc.hdr_ext.first = IBMVNIC_CRQ_CMD;
1360                         hdr_desc.hdr_ext.type = IBMVNIC_HDR_EXT_DESC;
1361                         hdr_desc.hdr_ext.len = tmp;
1362                 } else {
1363                         data = hdr_desc.hdr.data;
1364                         tmp = tmp_len > 24 ? 24 : tmp_len;
1365                         hdr_desc.hdr.first = IBMVNIC_CRQ_CMD;
1366                         hdr_desc.hdr.type = IBMVNIC_HDR_DESC;
1367                         hdr_desc.hdr.len = tmp;
1368                         hdr_desc.hdr.l2_len = (u8)hdr_len[0];
1369                         hdr_desc.hdr.l3_len = cpu_to_be16((u16)hdr_len[1]);
1370                         hdr_desc.hdr.l4_len = (u8)hdr_len[2];
1371                         hdr_desc.hdr.flag = hdr_field << 1;
1372                 }
1373                 memcpy(data, cur, tmp);
1374                 tmp_len -= tmp;
1375                 *scrq_arr = hdr_desc;
1376                 scrq_arr++;
1377                 num_descs++;
1378         }
1379
1380         return num_descs;
1381 }
1382
1383 /**
1384  * build_hdr_descs_arr - build a header descriptor array
1385  * @skb - socket buffer
1386  * @num_entries - number of descriptors to be sent
1387  * @subcrq - first TX descriptor
1388  * @hdr_field - bit field determining which headers will be sent
1389  *
1390  * This function will build a TX descriptor array with applicable
1391  * L2/L3/L4 packet header descriptors to be sent by send_subcrq_indirect.
1392  */
1393
1394 static void build_hdr_descs_arr(struct ibmvnic_tx_buff *txbuff,
1395                                 int *num_entries, u8 hdr_field)
1396 {
1397         int hdr_len[3] = {0, 0, 0};
1398         int tot_len;
1399         u8 *hdr_data = txbuff->hdr_data;
1400
1401         tot_len = build_hdr_data(hdr_field, txbuff->skb, hdr_len,
1402                                  txbuff->hdr_data);
1403         *num_entries += create_hdr_descs(hdr_field, hdr_data, tot_len, hdr_len,
1404                          txbuff->indir_arr + 1);
1405 }
1406
1407 static int ibmvnic_xmit_workarounds(struct sk_buff *skb,
1408                                     struct net_device *netdev)
1409 {
1410         /* For some backing devices, mishandling of small packets
1411          * can result in a loss of connection or TX stall. Device
1412          * architects recommend that no packet should be smaller
1413          * than the minimum MTU value provided to the driver, so
1414          * pad any packets to that length
1415          */
1416         if (skb->len < netdev->min_mtu)
1417                 return skb_put_padto(skb, netdev->min_mtu);
1418
1419         return 0;
1420 }
1421
1422 static netdev_tx_t ibmvnic_xmit(struct sk_buff *skb, struct net_device *netdev)
1423 {
1424         struct ibmvnic_adapter *adapter = netdev_priv(netdev);
1425         int queue_num = skb_get_queue_mapping(skb);
1426         u8 *hdrs = (u8 *)&adapter->tx_rx_desc_req;
1427         struct device *dev = &adapter->vdev->dev;
1428         struct ibmvnic_tx_buff *tx_buff = NULL;
1429         struct ibmvnic_sub_crq_queue *tx_scrq;
1430         struct ibmvnic_tx_pool *tx_pool;
1431         unsigned int tx_send_failed = 0;
1432         unsigned int tx_map_failed = 0;
1433         unsigned int tx_dropped = 0;
1434         unsigned int tx_packets = 0;
1435         unsigned int tx_bytes = 0;
1436         dma_addr_t data_dma_addr;
1437         struct netdev_queue *txq;
1438         unsigned long lpar_rc;
1439         union sub_crq tx_crq;
1440         unsigned int offset;
1441         int num_entries = 1;
1442         unsigned char *dst;
1443         u64 *handle_array;
1444         int index = 0;
1445         u8 proto = 0;
1446         netdev_tx_t ret = NETDEV_TX_OK;
1447
1448         if (adapter->resetting) {
1449                 if (!netif_subqueue_stopped(netdev, skb))
1450                         netif_stop_subqueue(netdev, queue_num);
1451                 dev_kfree_skb_any(skb);
1452
1453                 tx_send_failed++;
1454                 tx_dropped++;
1455                 ret = NETDEV_TX_OK;
1456                 goto out;
1457         }
1458
1459         if (ibmvnic_xmit_workarounds(skb, netdev)) {
1460                 tx_dropped++;
1461                 tx_send_failed++;
1462                 ret = NETDEV_TX_OK;
1463                 goto out;
1464         }
1465         if (skb_is_gso(skb))
1466                 tx_pool = &adapter->tso_pool[queue_num];
1467         else
1468                 tx_pool = &adapter->tx_pool[queue_num];
1469
1470         tx_scrq = adapter->tx_scrq[queue_num];
1471         txq = netdev_get_tx_queue(netdev, skb_get_queue_mapping(skb));
1472         handle_array = (u64 *)((u8 *)(adapter->login_rsp_buf) +
1473                 be32_to_cpu(adapter->login_rsp_buf->off_txsubm_subcrqs));
1474
1475         index = tx_pool->free_map[tx_pool->consumer_index];
1476
1477         if (index == IBMVNIC_INVALID_MAP) {
1478                 dev_kfree_skb_any(skb);
1479                 tx_send_failed++;
1480                 tx_dropped++;
1481                 ret = NETDEV_TX_OK;
1482                 goto out;
1483         }
1484
1485         tx_pool->free_map[tx_pool->consumer_index] = IBMVNIC_INVALID_MAP;
1486
1487         offset = index * tx_pool->buf_size;
1488         dst = tx_pool->long_term_buff.buff + offset;
1489         memset(dst, 0, tx_pool->buf_size);
1490         data_dma_addr = tx_pool->long_term_buff.addr + offset;
1491
1492         if (skb_shinfo(skb)->nr_frags) {
1493                 int cur, i;
1494
1495                 /* Copy the head */
1496                 skb_copy_from_linear_data(skb, dst, skb_headlen(skb));
1497                 cur = skb_headlen(skb);
1498
1499                 /* Copy the frags */
1500                 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
1501                         const skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
1502
1503                         memcpy(dst + cur,
1504                                page_address(skb_frag_page(frag)) +
1505                                frag->page_offset, skb_frag_size(frag));
1506                         cur += skb_frag_size(frag);
1507                 }
1508         } else {
1509                 skb_copy_from_linear_data(skb, dst, skb->len);
1510         }
1511
1512         tx_pool->consumer_index =
1513             (tx_pool->consumer_index + 1) % tx_pool->num_buffers;
1514
1515         tx_buff = &tx_pool->tx_buff[index];
1516         tx_buff->skb = skb;
1517         tx_buff->data_dma[0] = data_dma_addr;
1518         tx_buff->data_len[0] = skb->len;
1519         tx_buff->index = index;
1520         tx_buff->pool_index = queue_num;
1521         tx_buff->last_frag = true;
1522
1523         memset(&tx_crq, 0, sizeof(tx_crq));
1524         tx_crq.v1.first = IBMVNIC_CRQ_CMD;
1525         tx_crq.v1.type = IBMVNIC_TX_DESC;
1526         tx_crq.v1.n_crq_elem = 1;
1527         tx_crq.v1.n_sge = 1;
1528         tx_crq.v1.flags1 = IBMVNIC_TX_COMP_NEEDED;
1529
1530         if (skb_is_gso(skb))
1531                 tx_crq.v1.correlator =
1532                         cpu_to_be32(index | IBMVNIC_TSO_POOL_MASK);
1533         else
1534                 tx_crq.v1.correlator = cpu_to_be32(index);
1535         tx_crq.v1.dma_reg = cpu_to_be16(tx_pool->long_term_buff.map_id);
1536         tx_crq.v1.sge_len = cpu_to_be32(skb->len);
1537         tx_crq.v1.ioba = cpu_to_be64(data_dma_addr);
1538
1539         if (adapter->vlan_header_insertion && skb_vlan_tag_present(skb)) {
1540                 tx_crq.v1.flags2 |= IBMVNIC_TX_VLAN_INSERT;
1541                 tx_crq.v1.vlan_id = cpu_to_be16(skb->vlan_tci);
1542         }
1543
1544         if (skb->protocol == htons(ETH_P_IP)) {
1545                 tx_crq.v1.flags1 |= IBMVNIC_TX_PROT_IPV4;
1546                 proto = ip_hdr(skb)->protocol;
1547         } else if (skb->protocol == htons(ETH_P_IPV6)) {
1548                 tx_crq.v1.flags1 |= IBMVNIC_TX_PROT_IPV6;
1549                 proto = ipv6_hdr(skb)->nexthdr;
1550         }
1551
1552         if (proto == IPPROTO_TCP)
1553                 tx_crq.v1.flags1 |= IBMVNIC_TX_PROT_TCP;
1554         else if (proto == IPPROTO_UDP)
1555                 tx_crq.v1.flags1 |= IBMVNIC_TX_PROT_UDP;
1556
1557         if (skb->ip_summed == CHECKSUM_PARTIAL) {
1558                 tx_crq.v1.flags1 |= IBMVNIC_TX_CHKSUM_OFFLOAD;
1559                 hdrs += 2;
1560         }
1561         if (skb_is_gso(skb)) {
1562                 tx_crq.v1.flags1 |= IBMVNIC_TX_LSO;
1563                 tx_crq.v1.mss = cpu_to_be16(skb_shinfo(skb)->gso_size);
1564                 hdrs += 2;
1565         }
1566         /* determine if l2/3/4 headers are sent to firmware */
1567         if ((*hdrs >> 7) & 1) {
1568                 build_hdr_descs_arr(tx_buff, &num_entries, *hdrs);
1569                 tx_crq.v1.n_crq_elem = num_entries;
1570                 tx_buff->num_entries = num_entries;
1571                 tx_buff->indir_arr[0] = tx_crq;
1572                 tx_buff->indir_dma = dma_map_single(dev, tx_buff->indir_arr,
1573                                                     sizeof(tx_buff->indir_arr),
1574                                                     DMA_TO_DEVICE);
1575                 if (dma_mapping_error(dev, tx_buff->indir_dma)) {
1576                         dev_kfree_skb_any(skb);
1577                         tx_buff->skb = NULL;
1578                         if (!firmware_has_feature(FW_FEATURE_CMO))
1579                                 dev_err(dev, "tx: unable to map descriptor array\n");
1580                         tx_map_failed++;
1581                         tx_dropped++;
1582                         ret = NETDEV_TX_OK;
1583                         goto tx_err_out;
1584                 }
1585                 lpar_rc = send_subcrq_indirect(adapter, handle_array[queue_num],
1586                                                (u64)tx_buff->indir_dma,
1587                                                (u64)num_entries);
1588         } else {
1589                 tx_buff->num_entries = num_entries;
1590                 lpar_rc = send_subcrq(adapter, handle_array[queue_num],
1591                                       &tx_crq);
1592         }
1593         if (lpar_rc != H_SUCCESS) {
1594                 if (lpar_rc != H_CLOSED && lpar_rc != H_PARAMETER)
1595                         dev_err_ratelimited(dev, "tx: send failed\n");
1596                 dev_kfree_skb_any(skb);
1597                 tx_buff->skb = NULL;
1598
1599                 if (lpar_rc == H_CLOSED || adapter->failover_pending) {
1600                         /* Disable TX and report carrier off if queue is closed
1601                          * or pending failover.
1602                          * Firmware guarantees that a signal will be sent to the
1603                          * driver, triggering a reset or some other action.
1604                          */
1605                         netif_tx_stop_all_queues(netdev);
1606                         netif_carrier_off(netdev);
1607                 }
1608
1609                 tx_send_failed++;
1610                 tx_dropped++;
1611                 ret = NETDEV_TX_OK;
1612                 goto tx_err_out;
1613         }
1614
1615         if (atomic_add_return(num_entries, &tx_scrq->used)
1616                                         >= adapter->req_tx_entries_per_subcrq) {
1617                 netdev_dbg(netdev, "Stopping queue %d\n", queue_num);
1618                 netif_stop_subqueue(netdev, queue_num);
1619         }
1620
1621         tx_packets++;
1622         tx_bytes += skb->len;
1623         txq->trans_start = jiffies;
1624         ret = NETDEV_TX_OK;
1625         goto out;
1626
1627 tx_err_out:
1628         /* roll back consumer index and map array*/
1629         if (tx_pool->consumer_index == 0)
1630                 tx_pool->consumer_index =
1631                         tx_pool->num_buffers - 1;
1632         else
1633                 tx_pool->consumer_index--;
1634         tx_pool->free_map[tx_pool->consumer_index] = index;
1635 out:
1636         netdev->stats.tx_dropped += tx_dropped;
1637         netdev->stats.tx_bytes += tx_bytes;
1638         netdev->stats.tx_packets += tx_packets;
1639         adapter->tx_send_failed += tx_send_failed;
1640         adapter->tx_map_failed += tx_map_failed;
1641         adapter->tx_stats_buffers[queue_num].packets += tx_packets;
1642         adapter->tx_stats_buffers[queue_num].bytes += tx_bytes;
1643         adapter->tx_stats_buffers[queue_num].dropped_packets += tx_dropped;
1644
1645         return ret;
1646 }
1647
1648 static void ibmvnic_set_multi(struct net_device *netdev)
1649 {
1650         struct ibmvnic_adapter *adapter = netdev_priv(netdev);
1651         struct netdev_hw_addr *ha;
1652         union ibmvnic_crq crq;
1653
1654         memset(&crq, 0, sizeof(crq));
1655         crq.request_capability.first = IBMVNIC_CRQ_CMD;
1656         crq.request_capability.cmd = REQUEST_CAPABILITY;
1657
1658         if (netdev->flags & IFF_PROMISC) {
1659                 if (!adapter->promisc_supported)
1660                         return;
1661         } else {
1662                 if (netdev->flags & IFF_ALLMULTI) {
1663                         /* Accept all multicast */
1664                         memset(&crq, 0, sizeof(crq));
1665                         crq.multicast_ctrl.first = IBMVNIC_CRQ_CMD;
1666                         crq.multicast_ctrl.cmd = MULTICAST_CTRL;
1667                         crq.multicast_ctrl.flags = IBMVNIC_ENABLE_ALL;
1668                         ibmvnic_send_crq(adapter, &crq);
1669                 } else if (netdev_mc_empty(netdev)) {
1670                         /* Reject all multicast */
1671                         memset(&crq, 0, sizeof(crq));
1672                         crq.multicast_ctrl.first = IBMVNIC_CRQ_CMD;
1673                         crq.multicast_ctrl.cmd = MULTICAST_CTRL;
1674                         crq.multicast_ctrl.flags = IBMVNIC_DISABLE_ALL;
1675                         ibmvnic_send_crq(adapter, &crq);
1676                 } else {
1677                         /* Accept one or more multicast(s) */
1678                         netdev_for_each_mc_addr(ha, netdev) {
1679                                 memset(&crq, 0, sizeof(crq));
1680                                 crq.multicast_ctrl.first = IBMVNIC_CRQ_CMD;
1681                                 crq.multicast_ctrl.cmd = MULTICAST_CTRL;
1682                                 crq.multicast_ctrl.flags = IBMVNIC_ENABLE_MC;
1683                                 ether_addr_copy(&crq.multicast_ctrl.mac_addr[0],
1684                                                 ha->addr);
1685                                 ibmvnic_send_crq(adapter, &crq);
1686                         }
1687                 }
1688         }
1689 }
1690
1691 static int __ibmvnic_set_mac(struct net_device *netdev, struct sockaddr *p)
1692 {
1693         struct ibmvnic_adapter *adapter = netdev_priv(netdev);
1694         struct sockaddr *addr = p;
1695         union ibmvnic_crq crq;
1696         int rc;
1697
1698         if (!is_valid_ether_addr(addr->sa_data))
1699                 return -EADDRNOTAVAIL;
1700
1701         memset(&crq, 0, sizeof(crq));
1702         crq.change_mac_addr.first = IBMVNIC_CRQ_CMD;
1703         crq.change_mac_addr.cmd = CHANGE_MAC_ADDR;
1704         ether_addr_copy(&crq.change_mac_addr.mac_addr[0], addr->sa_data);
1705
1706         init_completion(&adapter->fw_done);
1707         rc = ibmvnic_send_crq(adapter, &crq);
1708         if (rc)
1709                 return rc;
1710         wait_for_completion(&adapter->fw_done);
1711         /* netdev->dev_addr is changed in handle_change_mac_rsp function */
1712         return adapter->fw_done_rc ? -EIO : 0;
1713 }
1714
1715 static int ibmvnic_set_mac(struct net_device *netdev, void *p)
1716 {
1717         struct ibmvnic_adapter *adapter = netdev_priv(netdev);
1718         struct sockaddr *addr = p;
1719         int rc;
1720
1721         if (adapter->state == VNIC_PROBED) {
1722                 memcpy(&adapter->desired.mac, addr, sizeof(struct sockaddr));
1723                 adapter->mac_change_pending = true;
1724                 return 0;
1725         }
1726
1727         rc = __ibmvnic_set_mac(netdev, addr);
1728
1729         return rc;
1730 }
1731
1732 /**
1733  * do_reset returns zero if we are able to keep processing reset events, or
1734  * non-zero if we hit a fatal error and must halt.
1735  */
1736 static int do_reset(struct ibmvnic_adapter *adapter,
1737                     struct ibmvnic_rwi *rwi, u32 reset_state)
1738 {
1739         u64 old_num_rx_queues, old_num_tx_queues;
1740         struct net_device *netdev = adapter->netdev;
1741         int i, rc;
1742
1743         netdev_dbg(adapter->netdev, "Re-setting driver (%d)\n",
1744                    rwi->reset_reason);
1745
1746         netif_carrier_off(netdev);
1747         adapter->reset_reason = rwi->reset_reason;
1748
1749         old_num_rx_queues = adapter->req_rx_queues;
1750         old_num_tx_queues = adapter->req_tx_queues;
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                         release_rx_pools(adapter);
1815                         release_tx_pools(adapter);
1816                         release_napi(adapter);
1817                         release_vpd_data(adapter);
1818
1819                         rc = init_resources(adapter);
1820                         if (rc)
1821                                 return rc;
1822
1823                 } else {
1824                         rc = reset_tx_pools(adapter);
1825                         if (rc)
1826                                 return rc;
1827
1828                         rc = reset_rx_pools(adapter);
1829                         if (rc)
1830                                 return rc;
1831                 }
1832                 ibmvnic_disable_irqs(adapter);
1833         }
1834         adapter->state = VNIC_CLOSED;
1835
1836         if (reset_state == VNIC_CLOSED)
1837                 return 0;
1838
1839         rc = __ibmvnic_open(netdev);
1840         if (rc) {
1841                 if (list_empty(&adapter->rwi_list))
1842                         adapter->state = VNIC_CLOSED;
1843                 else
1844                         adapter->state = reset_state;
1845
1846                 return 0;
1847         }
1848
1849         /* kick napi */
1850         for (i = 0; i < adapter->req_rx_queues; i++)
1851                 napi_schedule(&adapter->napi[i]);
1852
1853         if (adapter->reset_reason != VNIC_RESET_FAILOVER &&
1854             adapter->reset_reason != VNIC_RESET_CHANGE_PARAM)
1855                 netdev_notify_peers(netdev);
1856
1857         netif_carrier_on(netdev);
1858
1859         return 0;
1860 }
1861
1862 static int do_hard_reset(struct ibmvnic_adapter *adapter,
1863                          struct ibmvnic_rwi *rwi, u32 reset_state)
1864 {
1865         struct net_device *netdev = adapter->netdev;
1866         int rc;
1867
1868         netdev_dbg(adapter->netdev, "Hard resetting driver (%d)\n",
1869                    rwi->reset_reason);
1870
1871         netif_carrier_off(netdev);
1872         adapter->reset_reason = rwi->reset_reason;
1873
1874         ibmvnic_cleanup(netdev);
1875         release_resources(adapter);
1876         release_sub_crqs(adapter, 0);
1877         release_crq_queue(adapter);
1878
1879         /* remove the closed state so when we call open it appears
1880          * we are coming from the probed state.
1881          */
1882         adapter->state = VNIC_PROBED;
1883
1884         rc = init_crq_queue(adapter);
1885         if (rc) {
1886                 netdev_err(adapter->netdev,
1887                            "Couldn't initialize crq. rc=%d\n", rc);
1888                 return rc;
1889         }
1890
1891         rc = ibmvnic_init(adapter);
1892         if (rc)
1893                 return rc;
1894
1895         /* If the adapter was in PROBE state prior to the reset,
1896          * exit here.
1897          */
1898         if (reset_state == VNIC_PROBED)
1899                 return 0;
1900
1901         rc = ibmvnic_login(netdev);
1902         if (rc) {
1903                 adapter->state = VNIC_PROBED;
1904                 return 0;
1905         }
1906
1907         rc = init_resources(adapter);
1908         if (rc)
1909                 return rc;
1910
1911         ibmvnic_disable_irqs(adapter);
1912         adapter->state = VNIC_CLOSED;
1913
1914         if (reset_state == VNIC_CLOSED)
1915                 return 0;
1916
1917         rc = __ibmvnic_open(netdev);
1918         if (rc) {
1919                 if (list_empty(&adapter->rwi_list))
1920                         adapter->state = VNIC_CLOSED;
1921                 else
1922                         adapter->state = reset_state;
1923
1924                 return 0;
1925         }
1926
1927         netif_carrier_on(netdev);
1928
1929         return 0;
1930 }
1931
1932 static struct ibmvnic_rwi *get_next_rwi(struct ibmvnic_adapter *adapter)
1933 {
1934         struct ibmvnic_rwi *rwi;
1935
1936         mutex_lock(&adapter->rwi_lock);
1937
1938         if (!list_empty(&adapter->rwi_list)) {
1939                 rwi = list_first_entry(&adapter->rwi_list, struct ibmvnic_rwi,
1940                                        list);
1941                 list_del(&rwi->list);
1942         } else {
1943                 rwi = NULL;
1944         }
1945
1946         mutex_unlock(&adapter->rwi_lock);
1947         return rwi;
1948 }
1949
1950 static void free_all_rwi(struct ibmvnic_adapter *adapter)
1951 {
1952         struct ibmvnic_rwi *rwi;
1953
1954         rwi = get_next_rwi(adapter);
1955         while (rwi) {
1956                 kfree(rwi);
1957                 rwi = get_next_rwi(adapter);
1958         }
1959 }
1960
1961 static void __ibmvnic_reset(struct work_struct *work)
1962 {
1963         struct ibmvnic_rwi *rwi;
1964         struct ibmvnic_adapter *adapter;
1965         struct net_device *netdev;
1966         bool we_lock_rtnl = false;
1967         u32 reset_state;
1968         int rc = 0;
1969
1970         adapter = container_of(work, struct ibmvnic_adapter, ibmvnic_reset);
1971         netdev = adapter->netdev;
1972
1973         /* netif_set_real_num_xx_queues needs to take rtnl lock here
1974          * unless wait_for_reset is set, in which case the rtnl lock
1975          * has already been taken before initializing the reset
1976          */
1977         if (!adapter->wait_for_reset) {
1978                 rtnl_lock();
1979                 we_lock_rtnl = true;
1980         }
1981         reset_state = adapter->state;
1982
1983         rwi = get_next_rwi(adapter);
1984         while (rwi) {
1985                 if (adapter->force_reset_recovery) {
1986                         adapter->force_reset_recovery = false;
1987                         rc = do_hard_reset(adapter, rwi, reset_state);
1988                 } else {
1989                         rc = do_reset(adapter, rwi, reset_state);
1990                 }
1991                 kfree(rwi);
1992                 if (rc && rc != IBMVNIC_INIT_FAILED &&
1993                     !adapter->force_reset_recovery)
1994                         break;
1995
1996                 rwi = get_next_rwi(adapter);
1997         }
1998
1999         if (adapter->wait_for_reset) {
2000                 adapter->wait_for_reset = false;
2001                 adapter->reset_done_rc = rc;
2002                 complete(&adapter->reset_done);
2003         }
2004
2005         if (rc) {
2006                 netdev_dbg(adapter->netdev, "Reset failed\n");
2007                 free_all_rwi(adapter);
2008         }
2009
2010         adapter->resetting = false;
2011         if (we_lock_rtnl)
2012                 rtnl_unlock();
2013 }
2014
2015 static int ibmvnic_reset(struct ibmvnic_adapter *adapter,
2016                          enum ibmvnic_reset_reason reason)
2017 {
2018         struct list_head *entry, *tmp_entry;
2019         struct ibmvnic_rwi *rwi, *tmp;
2020         struct net_device *netdev = adapter->netdev;
2021         int ret;
2022
2023         if (adapter->state == VNIC_REMOVING ||
2024             adapter->state == VNIC_REMOVED ||
2025             adapter->failover_pending) {
2026                 ret = EBUSY;
2027                 netdev_dbg(netdev, "Adapter removing or pending failover, skipping reset\n");
2028                 goto err;
2029         }
2030
2031         if (adapter->state == VNIC_PROBING) {
2032                 netdev_warn(netdev, "Adapter reset during probe\n");
2033                 ret = adapter->init_done_rc = EAGAIN;
2034                 goto err;
2035         }
2036
2037         mutex_lock(&adapter->rwi_lock);
2038
2039         list_for_each(entry, &adapter->rwi_list) {
2040                 tmp = list_entry(entry, struct ibmvnic_rwi, list);
2041                 if (tmp->reset_reason == reason) {
2042                         netdev_dbg(netdev, "Skipping matching reset\n");
2043                         mutex_unlock(&adapter->rwi_lock);
2044                         ret = EBUSY;
2045                         goto err;
2046                 }
2047         }
2048
2049         rwi = kzalloc(sizeof(*rwi), GFP_KERNEL);
2050         if (!rwi) {
2051                 mutex_unlock(&adapter->rwi_lock);
2052                 ibmvnic_close(netdev);
2053                 ret = ENOMEM;
2054                 goto err;
2055         }
2056         /* if we just received a transport event,
2057          * flush reset queue and process this reset
2058          */
2059         if (adapter->force_reset_recovery && !list_empty(&adapter->rwi_list)) {
2060                 list_for_each_safe(entry, tmp_entry, &adapter->rwi_list)
2061                         list_del(entry);
2062         }
2063         rwi->reset_reason = reason;
2064         list_add_tail(&rwi->list, &adapter->rwi_list);
2065         mutex_unlock(&adapter->rwi_lock);
2066         adapter->resetting = true;
2067         netdev_dbg(adapter->netdev, "Scheduling reset (reason %d)\n", reason);
2068         schedule_work(&adapter->ibmvnic_reset);
2069
2070         return 0;
2071 err:
2072         if (adapter->wait_for_reset)
2073                 adapter->wait_for_reset = false;
2074         return -ret;
2075 }
2076
2077 static void ibmvnic_tx_timeout(struct net_device *dev)
2078 {
2079         struct ibmvnic_adapter *adapter = netdev_priv(dev);
2080
2081         ibmvnic_reset(adapter, VNIC_RESET_TIMEOUT);
2082 }
2083
2084 static void remove_buff_from_pool(struct ibmvnic_adapter *adapter,
2085                                   struct ibmvnic_rx_buff *rx_buff)
2086 {
2087         struct ibmvnic_rx_pool *pool = &adapter->rx_pool[rx_buff->pool_index];
2088
2089         rx_buff->skb = NULL;
2090
2091         pool->free_map[pool->next_alloc] = (int)(rx_buff - pool->rx_buff);
2092         pool->next_alloc = (pool->next_alloc + 1) % pool->size;
2093
2094         atomic_dec(&pool->available);
2095 }
2096
2097 static int ibmvnic_poll(struct napi_struct *napi, int budget)
2098 {
2099         struct net_device *netdev = napi->dev;
2100         struct ibmvnic_adapter *adapter = netdev_priv(netdev);
2101         int scrq_num = (int)(napi - adapter->napi);
2102         int frames_processed = 0;
2103
2104 restart_poll:
2105         while (frames_processed < budget) {
2106                 struct sk_buff *skb;
2107                 struct ibmvnic_rx_buff *rx_buff;
2108                 union sub_crq *next;
2109                 u32 length;
2110                 u16 offset;
2111                 u8 flags = 0;
2112
2113                 if (unlikely(adapter->resetting &&
2114                              adapter->reset_reason != VNIC_RESET_NON_FATAL)) {
2115                         enable_scrq_irq(adapter, adapter->rx_scrq[scrq_num]);
2116                         napi_complete_done(napi, frames_processed);
2117                         return frames_processed;
2118                 }
2119
2120                 if (!pending_scrq(adapter, adapter->rx_scrq[scrq_num]))
2121                         break;
2122                 next = ibmvnic_next_scrq(adapter, adapter->rx_scrq[scrq_num]);
2123                 rx_buff =
2124                     (struct ibmvnic_rx_buff *)be64_to_cpu(next->
2125                                                           rx_comp.correlator);
2126                 /* do error checking */
2127                 if (next->rx_comp.rc) {
2128                         netdev_dbg(netdev, "rx buffer returned with rc %x\n",
2129                                    be16_to_cpu(next->rx_comp.rc));
2130                         /* free the entry */
2131                         next->rx_comp.first = 0;
2132                         dev_kfree_skb_any(rx_buff->skb);
2133                         remove_buff_from_pool(adapter, rx_buff);
2134                         continue;
2135                 } else if (!rx_buff->skb) {
2136                         /* free the entry */
2137                         next->rx_comp.first = 0;
2138                         remove_buff_from_pool(adapter, rx_buff);
2139                         continue;
2140                 }
2141
2142                 length = be32_to_cpu(next->rx_comp.len);
2143                 offset = be16_to_cpu(next->rx_comp.off_frame_data);
2144                 flags = next->rx_comp.flags;
2145                 skb = rx_buff->skb;
2146                 skb_copy_to_linear_data(skb, rx_buff->data + offset,
2147                                         length);
2148
2149                 /* VLAN Header has been stripped by the system firmware and
2150                  * needs to be inserted by the driver
2151                  */
2152                 if (adapter->rx_vlan_header_insertion &&
2153                     (flags & IBMVNIC_VLAN_STRIPPED))
2154                         __vlan_hwaccel_put_tag(skb, htons(ETH_P_8021Q),
2155                                                ntohs(next->rx_comp.vlan_tci));
2156
2157                 /* free the entry */
2158                 next->rx_comp.first = 0;
2159                 remove_buff_from_pool(adapter, rx_buff);
2160
2161                 skb_put(skb, length);
2162                 skb->protocol = eth_type_trans(skb, netdev);
2163                 skb_record_rx_queue(skb, scrq_num);
2164
2165                 if (flags & IBMVNIC_IP_CHKSUM_GOOD &&
2166                     flags & IBMVNIC_TCP_UDP_CHKSUM_GOOD) {
2167                         skb->ip_summed = CHECKSUM_UNNECESSARY;
2168                 }
2169
2170                 length = skb->len;
2171                 napi_gro_receive(napi, skb); /* send it up */
2172                 netdev->stats.rx_packets++;
2173                 netdev->stats.rx_bytes += length;
2174                 adapter->rx_stats_buffers[scrq_num].packets++;
2175                 adapter->rx_stats_buffers[scrq_num].bytes += length;
2176                 frames_processed++;
2177         }
2178
2179         if (adapter->state != VNIC_CLOSING)
2180                 replenish_rx_pool(adapter, &adapter->rx_pool[scrq_num]);
2181
2182         if (frames_processed < budget) {
2183                 enable_scrq_irq(adapter, adapter->rx_scrq[scrq_num]);
2184                 napi_complete_done(napi, frames_processed);
2185                 if (pending_scrq(adapter, adapter->rx_scrq[scrq_num]) &&
2186                     napi_reschedule(napi)) {
2187                         disable_scrq_irq(adapter, adapter->rx_scrq[scrq_num]);
2188                         goto restart_poll;
2189                 }
2190         }
2191         return frames_processed;
2192 }
2193
2194 static int wait_for_reset(struct ibmvnic_adapter *adapter)
2195 {
2196         int rc, ret;
2197
2198         adapter->fallback.mtu = adapter->req_mtu;
2199         adapter->fallback.rx_queues = adapter->req_rx_queues;
2200         adapter->fallback.tx_queues = adapter->req_tx_queues;
2201         adapter->fallback.rx_entries = adapter->req_rx_add_entries_per_subcrq;
2202         adapter->fallback.tx_entries = adapter->req_tx_entries_per_subcrq;
2203
2204         init_completion(&adapter->reset_done);
2205         adapter->wait_for_reset = true;
2206         rc = ibmvnic_reset(adapter, VNIC_RESET_CHANGE_PARAM);
2207         if (rc)
2208                 return rc;
2209         wait_for_completion(&adapter->reset_done);
2210
2211         ret = 0;
2212         if (adapter->reset_done_rc) {
2213                 ret = -EIO;
2214                 adapter->desired.mtu = adapter->fallback.mtu;
2215                 adapter->desired.rx_queues = adapter->fallback.rx_queues;
2216                 adapter->desired.tx_queues = adapter->fallback.tx_queues;
2217                 adapter->desired.rx_entries = adapter->fallback.rx_entries;
2218                 adapter->desired.tx_entries = adapter->fallback.tx_entries;
2219
2220                 init_completion(&adapter->reset_done);
2221                 adapter->wait_for_reset = true;
2222                 rc = ibmvnic_reset(adapter, VNIC_RESET_CHANGE_PARAM);
2223                 if (rc)
2224                         return ret;
2225                 wait_for_completion(&adapter->reset_done);
2226         }
2227         adapter->wait_for_reset = false;
2228
2229         return ret;
2230 }
2231
2232 static int ibmvnic_change_mtu(struct net_device *netdev, int new_mtu)
2233 {
2234         struct ibmvnic_adapter *adapter = netdev_priv(netdev);
2235
2236         adapter->desired.mtu = new_mtu + ETH_HLEN;
2237
2238         return wait_for_reset(adapter);
2239 }
2240
2241 static netdev_features_t ibmvnic_features_check(struct sk_buff *skb,
2242                                                 struct net_device *dev,
2243                                                 netdev_features_t features)
2244 {
2245         /* Some backing hardware adapters can not
2246          * handle packets with a MSS less than 224
2247          * or with only one segment.
2248          */
2249         if (skb_is_gso(skb)) {
2250                 if (skb_shinfo(skb)->gso_size < 224 ||
2251                     skb_shinfo(skb)->gso_segs == 1)
2252                         features &= ~NETIF_F_GSO_MASK;
2253         }
2254
2255         return features;
2256 }
2257
2258 static const struct net_device_ops ibmvnic_netdev_ops = {
2259         .ndo_open               = ibmvnic_open,
2260         .ndo_stop               = ibmvnic_close,
2261         .ndo_start_xmit         = ibmvnic_xmit,
2262         .ndo_set_rx_mode        = ibmvnic_set_multi,
2263         .ndo_set_mac_address    = ibmvnic_set_mac,
2264         .ndo_validate_addr      = eth_validate_addr,
2265         .ndo_tx_timeout         = ibmvnic_tx_timeout,
2266         .ndo_change_mtu         = ibmvnic_change_mtu,
2267         .ndo_features_check     = ibmvnic_features_check,
2268 };
2269
2270 /* ethtool functions */
2271
2272 static int ibmvnic_get_link_ksettings(struct net_device *netdev,
2273                                       struct ethtool_link_ksettings *cmd)
2274 {
2275         u32 supported, advertising;
2276
2277         supported = (SUPPORTED_1000baseT_Full | SUPPORTED_Autoneg |
2278                           SUPPORTED_FIBRE);
2279         advertising = (ADVERTISED_1000baseT_Full | ADVERTISED_Autoneg |
2280                             ADVERTISED_FIBRE);
2281         cmd->base.speed = SPEED_1000;
2282         cmd->base.duplex = DUPLEX_FULL;
2283         cmd->base.port = PORT_FIBRE;
2284         cmd->base.phy_address = 0;
2285         cmd->base.autoneg = AUTONEG_ENABLE;
2286
2287         ethtool_convert_legacy_u32_to_link_mode(cmd->link_modes.supported,
2288                                                 supported);
2289         ethtool_convert_legacy_u32_to_link_mode(cmd->link_modes.advertising,
2290                                                 advertising);
2291
2292         return 0;
2293 }
2294
2295 static void ibmvnic_get_drvinfo(struct net_device *netdev,
2296                                 struct ethtool_drvinfo *info)
2297 {
2298         struct ibmvnic_adapter *adapter = netdev_priv(netdev);
2299
2300         strlcpy(info->driver, ibmvnic_driver_name, sizeof(info->driver));
2301         strlcpy(info->version, IBMVNIC_DRIVER_VERSION, sizeof(info->version));
2302         strlcpy(info->fw_version, adapter->fw_version,
2303                 sizeof(info->fw_version));
2304 }
2305
2306 static u32 ibmvnic_get_msglevel(struct net_device *netdev)
2307 {
2308         struct ibmvnic_adapter *adapter = netdev_priv(netdev);
2309
2310         return adapter->msg_enable;
2311 }
2312
2313 static void ibmvnic_set_msglevel(struct net_device *netdev, u32 data)
2314 {
2315         struct ibmvnic_adapter *adapter = netdev_priv(netdev);
2316
2317         adapter->msg_enable = data;
2318 }
2319
2320 static u32 ibmvnic_get_link(struct net_device *netdev)
2321 {
2322         struct ibmvnic_adapter *adapter = netdev_priv(netdev);
2323
2324         /* Don't need to send a query because we request a logical link up at
2325          * init and then we wait for link state indications
2326          */
2327         return adapter->logical_link_state;
2328 }
2329
2330 static void ibmvnic_get_ringparam(struct net_device *netdev,
2331                                   struct ethtool_ringparam *ring)
2332 {
2333         struct ibmvnic_adapter *adapter = netdev_priv(netdev);
2334
2335         if (adapter->priv_flags & IBMVNIC_USE_SERVER_MAXES) {
2336                 ring->rx_max_pending = adapter->max_rx_add_entries_per_subcrq;
2337                 ring->tx_max_pending = adapter->max_tx_entries_per_subcrq;
2338         } else {
2339                 ring->rx_max_pending = IBMVNIC_MAX_QUEUE_SZ;
2340                 ring->tx_max_pending = IBMVNIC_MAX_QUEUE_SZ;
2341         }
2342         ring->rx_mini_max_pending = 0;
2343         ring->rx_jumbo_max_pending = 0;
2344         ring->rx_pending = adapter->req_rx_add_entries_per_subcrq;
2345         ring->tx_pending = adapter->req_tx_entries_per_subcrq;
2346         ring->rx_mini_pending = 0;
2347         ring->rx_jumbo_pending = 0;
2348 }
2349
2350 static int ibmvnic_set_ringparam(struct net_device *netdev,
2351                                  struct ethtool_ringparam *ring)
2352 {
2353         struct ibmvnic_adapter *adapter = netdev_priv(netdev);
2354         int ret;
2355
2356         ret = 0;
2357         adapter->desired.rx_entries = ring->rx_pending;
2358         adapter->desired.tx_entries = ring->tx_pending;
2359
2360         ret = wait_for_reset(adapter);
2361
2362         if (!ret &&
2363             (adapter->req_rx_add_entries_per_subcrq != ring->rx_pending ||
2364              adapter->req_tx_entries_per_subcrq != ring->tx_pending))
2365                 netdev_info(netdev,
2366                             "Could not match full ringsize request. Requested: RX %d, TX %d; Allowed: RX %llu, TX %llu\n",
2367                             ring->rx_pending, ring->tx_pending,
2368                             adapter->req_rx_add_entries_per_subcrq,
2369                             adapter->req_tx_entries_per_subcrq);
2370         return ret;
2371 }
2372
2373 static void ibmvnic_get_channels(struct net_device *netdev,
2374                                  struct ethtool_channels *channels)
2375 {
2376         struct ibmvnic_adapter *adapter = netdev_priv(netdev);
2377
2378         if (adapter->priv_flags & IBMVNIC_USE_SERVER_MAXES) {
2379                 channels->max_rx = adapter->max_rx_queues;
2380                 channels->max_tx = adapter->max_tx_queues;
2381         } else {
2382                 channels->max_rx = IBMVNIC_MAX_QUEUES;
2383                 channels->max_tx = IBMVNIC_MAX_QUEUES;
2384         }
2385
2386         channels->max_other = 0;
2387         channels->max_combined = 0;
2388         channels->rx_count = adapter->req_rx_queues;
2389         channels->tx_count = adapter->req_tx_queues;
2390         channels->other_count = 0;
2391         channels->combined_count = 0;
2392 }
2393
2394 static int ibmvnic_set_channels(struct net_device *netdev,
2395                                 struct ethtool_channels *channels)
2396 {
2397         struct ibmvnic_adapter *adapter = netdev_priv(netdev);
2398         int ret;
2399
2400         ret = 0;
2401         adapter->desired.rx_queues = channels->rx_count;
2402         adapter->desired.tx_queues = channels->tx_count;
2403
2404         ret = wait_for_reset(adapter);
2405
2406         if (!ret &&
2407             (adapter->req_rx_queues != channels->rx_count ||
2408              adapter->req_tx_queues != channels->tx_count))
2409                 netdev_info(netdev,
2410                             "Could not match full channels request. Requested: RX %d, TX %d; Allowed: RX %llu, TX %llu\n",
2411                             channels->rx_count, channels->tx_count,
2412                             adapter->req_rx_queues, adapter->req_tx_queues);
2413         return ret;
2414
2415 }
2416
2417 static void ibmvnic_get_strings(struct net_device *dev, u32 stringset, u8 *data)
2418 {
2419         struct ibmvnic_adapter *adapter = netdev_priv(dev);
2420         int i;
2421
2422         switch (stringset) {
2423         case ETH_SS_STATS:
2424                 for (i = 0; i < ARRAY_SIZE(ibmvnic_stats);
2425                                 i++, data += ETH_GSTRING_LEN)
2426                         memcpy(data, ibmvnic_stats[i].name, ETH_GSTRING_LEN);
2427
2428                 for (i = 0; i < adapter->req_tx_queues; i++) {
2429                         snprintf(data, ETH_GSTRING_LEN, "tx%d_packets", i);
2430                         data += ETH_GSTRING_LEN;
2431
2432                         snprintf(data, ETH_GSTRING_LEN, "tx%d_bytes", i);
2433                         data += ETH_GSTRING_LEN;
2434
2435                         snprintf(data, ETH_GSTRING_LEN,
2436                                  "tx%d_dropped_packets", i);
2437                         data += ETH_GSTRING_LEN;
2438                 }
2439
2440                 for (i = 0; i < adapter->req_rx_queues; i++) {
2441                         snprintf(data, ETH_GSTRING_LEN, "rx%d_packets", i);
2442                         data += ETH_GSTRING_LEN;
2443
2444                         snprintf(data, ETH_GSTRING_LEN, "rx%d_bytes", i);
2445                         data += ETH_GSTRING_LEN;
2446
2447                         snprintf(data, ETH_GSTRING_LEN, "rx%d_interrupts", i);
2448                         data += ETH_GSTRING_LEN;
2449                 }
2450                 break;
2451
2452         case ETH_SS_PRIV_FLAGS:
2453                 for (i = 0; i < ARRAY_SIZE(ibmvnic_priv_flags); i++)
2454                         strcpy(data + i * ETH_GSTRING_LEN,
2455                                ibmvnic_priv_flags[i]);
2456                 break;
2457         default:
2458                 return;
2459         }
2460 }
2461
2462 static int ibmvnic_get_sset_count(struct net_device *dev, int sset)
2463 {
2464         struct ibmvnic_adapter *adapter = netdev_priv(dev);
2465
2466         switch (sset) {
2467         case ETH_SS_STATS:
2468                 return ARRAY_SIZE(ibmvnic_stats) +
2469                        adapter->req_tx_queues * NUM_TX_STATS +
2470                        adapter->req_rx_queues * NUM_RX_STATS;
2471         case ETH_SS_PRIV_FLAGS:
2472                 return ARRAY_SIZE(ibmvnic_priv_flags);
2473         default:
2474                 return -EOPNOTSUPP;
2475         }
2476 }
2477
2478 static void ibmvnic_get_ethtool_stats(struct net_device *dev,
2479                                       struct ethtool_stats *stats, u64 *data)
2480 {
2481         struct ibmvnic_adapter *adapter = netdev_priv(dev);
2482         union ibmvnic_crq crq;
2483         int i, j;
2484         int rc;
2485
2486         memset(&crq, 0, sizeof(crq));
2487         crq.request_statistics.first = IBMVNIC_CRQ_CMD;
2488         crq.request_statistics.cmd = REQUEST_STATISTICS;
2489         crq.request_statistics.ioba = cpu_to_be32(adapter->stats_token);
2490         crq.request_statistics.len =
2491             cpu_to_be32(sizeof(struct ibmvnic_statistics));
2492
2493         /* Wait for data to be written */
2494         init_completion(&adapter->stats_done);
2495         rc = ibmvnic_send_crq(adapter, &crq);
2496         if (rc)
2497                 return;
2498         wait_for_completion(&adapter->stats_done);
2499
2500         for (i = 0; i < ARRAY_SIZE(ibmvnic_stats); i++)
2501                 data[i] = be64_to_cpu(IBMVNIC_GET_STAT(adapter,
2502                                                 ibmvnic_stats[i].offset));
2503
2504         for (j = 0; j < adapter->req_tx_queues; j++) {
2505                 data[i] = adapter->tx_stats_buffers[j].packets;
2506                 i++;
2507                 data[i] = adapter->tx_stats_buffers[j].bytes;
2508                 i++;
2509                 data[i] = adapter->tx_stats_buffers[j].dropped_packets;
2510                 i++;
2511         }
2512
2513         for (j = 0; j < adapter->req_rx_queues; j++) {
2514                 data[i] = adapter->rx_stats_buffers[j].packets;
2515                 i++;
2516                 data[i] = adapter->rx_stats_buffers[j].bytes;
2517                 i++;
2518                 data[i] = adapter->rx_stats_buffers[j].interrupts;
2519                 i++;
2520         }
2521 }
2522
2523 static u32 ibmvnic_get_priv_flags(struct net_device *netdev)
2524 {
2525         struct ibmvnic_adapter *adapter = netdev_priv(netdev);
2526
2527         return adapter->priv_flags;
2528 }
2529
2530 static int ibmvnic_set_priv_flags(struct net_device *netdev, u32 flags)
2531 {
2532         struct ibmvnic_adapter *adapter = netdev_priv(netdev);
2533         bool which_maxes = !!(flags & IBMVNIC_USE_SERVER_MAXES);
2534
2535         if (which_maxes)
2536                 adapter->priv_flags |= IBMVNIC_USE_SERVER_MAXES;
2537         else
2538                 adapter->priv_flags &= ~IBMVNIC_USE_SERVER_MAXES;
2539
2540         return 0;
2541 }
2542 static const struct ethtool_ops ibmvnic_ethtool_ops = {
2543         .get_drvinfo            = ibmvnic_get_drvinfo,
2544         .get_msglevel           = ibmvnic_get_msglevel,
2545         .set_msglevel           = ibmvnic_set_msglevel,
2546         .get_link               = ibmvnic_get_link,
2547         .get_ringparam          = ibmvnic_get_ringparam,
2548         .set_ringparam          = ibmvnic_set_ringparam,
2549         .get_channels           = ibmvnic_get_channels,
2550         .set_channels           = ibmvnic_set_channels,
2551         .get_strings            = ibmvnic_get_strings,
2552         .get_sset_count         = ibmvnic_get_sset_count,
2553         .get_ethtool_stats      = ibmvnic_get_ethtool_stats,
2554         .get_link_ksettings     = ibmvnic_get_link_ksettings,
2555         .get_priv_flags         = ibmvnic_get_priv_flags,
2556         .set_priv_flags         = ibmvnic_set_priv_flags,
2557 };
2558
2559 /* Routines for managing CRQs/sCRQs  */
2560
2561 static int reset_one_sub_crq_queue(struct ibmvnic_adapter *adapter,
2562                                    struct ibmvnic_sub_crq_queue *scrq)
2563 {
2564         int rc;
2565
2566         if (scrq->irq) {
2567                 free_irq(scrq->irq, scrq);
2568                 irq_dispose_mapping(scrq->irq);
2569                 scrq->irq = 0;
2570         }
2571
2572         memset(scrq->msgs, 0, 4 * PAGE_SIZE);
2573         atomic_set(&scrq->used, 0);
2574         scrq->cur = 0;
2575
2576         rc = h_reg_sub_crq(adapter->vdev->unit_address, scrq->msg_token,
2577                            4 * PAGE_SIZE, &scrq->crq_num, &scrq->hw_irq);
2578         return rc;
2579 }
2580
2581 static int reset_sub_crq_queues(struct ibmvnic_adapter *adapter)
2582 {
2583         int i, rc;
2584
2585         for (i = 0; i < adapter->req_tx_queues; i++) {
2586                 netdev_dbg(adapter->netdev, "Re-setting tx_scrq[%d]\n", i);
2587                 rc = reset_one_sub_crq_queue(adapter, adapter->tx_scrq[i]);
2588                 if (rc)
2589                         return rc;
2590         }
2591
2592         for (i = 0; i < adapter->req_rx_queues; i++) {
2593                 netdev_dbg(adapter->netdev, "Re-setting rx_scrq[%d]\n", i);
2594                 rc = reset_one_sub_crq_queue(adapter, adapter->rx_scrq[i]);
2595                 if (rc)
2596                         return rc;
2597         }
2598
2599         return rc;
2600 }
2601
2602 static void release_sub_crq_queue(struct ibmvnic_adapter *adapter,
2603                                   struct ibmvnic_sub_crq_queue *scrq,
2604                                   bool do_h_free)
2605 {
2606         struct device *dev = &adapter->vdev->dev;
2607         long rc;
2608
2609         netdev_dbg(adapter->netdev, "Releasing sub-CRQ\n");
2610
2611         if (do_h_free) {
2612                 /* Close the sub-crqs */
2613                 do {
2614                         rc = plpar_hcall_norets(H_FREE_SUB_CRQ,
2615                                                 adapter->vdev->unit_address,
2616                                                 scrq->crq_num);
2617                 } while (rc == H_BUSY || H_IS_LONG_BUSY(rc));
2618
2619                 if (rc) {
2620                         netdev_err(adapter->netdev,
2621                                    "Failed to release sub-CRQ %16lx, rc = %ld\n",
2622                                    scrq->crq_num, rc);
2623                 }
2624         }
2625
2626         dma_unmap_single(dev, scrq->msg_token, 4 * PAGE_SIZE,
2627                          DMA_BIDIRECTIONAL);
2628         free_pages((unsigned long)scrq->msgs, 2);
2629         kfree(scrq);
2630 }
2631
2632 static struct ibmvnic_sub_crq_queue *init_sub_crq_queue(struct ibmvnic_adapter
2633                                                         *adapter)
2634 {
2635         struct device *dev = &adapter->vdev->dev;
2636         struct ibmvnic_sub_crq_queue *scrq;
2637         int rc;
2638
2639         scrq = kzalloc(sizeof(*scrq), GFP_KERNEL);
2640         if (!scrq)
2641                 return NULL;
2642
2643         scrq->msgs =
2644                 (union sub_crq *)__get_free_pages(GFP_KERNEL | __GFP_ZERO, 2);
2645         if (!scrq->msgs) {
2646                 dev_warn(dev, "Couldn't allocate crq queue messages page\n");
2647                 goto zero_page_failed;
2648         }
2649
2650         scrq->msg_token = dma_map_single(dev, scrq->msgs, 4 * PAGE_SIZE,
2651                                          DMA_BIDIRECTIONAL);
2652         if (dma_mapping_error(dev, scrq->msg_token)) {
2653                 dev_warn(dev, "Couldn't map crq queue messages page\n");
2654                 goto map_failed;
2655         }
2656
2657         rc = h_reg_sub_crq(adapter->vdev->unit_address, scrq->msg_token,
2658                            4 * PAGE_SIZE, &scrq->crq_num, &scrq->hw_irq);
2659
2660         if (rc == H_RESOURCE)
2661                 rc = ibmvnic_reset_crq(adapter);
2662
2663         if (rc == H_CLOSED) {
2664                 dev_warn(dev, "Partner adapter not ready, waiting.\n");
2665         } else if (rc) {
2666                 dev_warn(dev, "Error %d registering sub-crq\n", rc);
2667                 goto reg_failed;
2668         }
2669
2670         scrq->adapter = adapter;
2671         scrq->size = 4 * PAGE_SIZE / sizeof(*scrq->msgs);
2672         spin_lock_init(&scrq->lock);
2673
2674         netdev_dbg(adapter->netdev,
2675                    "sub-crq initialized, num %lx, hw_irq=%lx, irq=%x\n",
2676                    scrq->crq_num, scrq->hw_irq, scrq->irq);
2677
2678         return scrq;
2679
2680 reg_failed:
2681         dma_unmap_single(dev, scrq->msg_token, 4 * PAGE_SIZE,
2682                          DMA_BIDIRECTIONAL);
2683 map_failed:
2684         free_pages((unsigned long)scrq->msgs, 2);
2685 zero_page_failed:
2686         kfree(scrq);
2687
2688         return NULL;
2689 }
2690
2691 static void release_sub_crqs(struct ibmvnic_adapter *adapter, bool do_h_free)
2692 {
2693         int i;
2694
2695         if (adapter->tx_scrq) {
2696                 for (i = 0; i < adapter->num_active_tx_scrqs; i++) {
2697                         if (!adapter->tx_scrq[i])
2698                                 continue;
2699
2700                         netdev_dbg(adapter->netdev, "Releasing tx_scrq[%d]\n",
2701                                    i);
2702                         if (adapter->tx_scrq[i]->irq) {
2703                                 free_irq(adapter->tx_scrq[i]->irq,
2704                                          adapter->tx_scrq[i]);
2705                                 irq_dispose_mapping(adapter->tx_scrq[i]->irq);
2706                                 adapter->tx_scrq[i]->irq = 0;
2707                         }
2708
2709                         release_sub_crq_queue(adapter, adapter->tx_scrq[i],
2710                                               do_h_free);
2711                 }
2712
2713                 kfree(adapter->tx_scrq);
2714                 adapter->tx_scrq = NULL;
2715                 adapter->num_active_tx_scrqs = 0;
2716         }
2717
2718         if (adapter->rx_scrq) {
2719                 for (i = 0; i < adapter->num_active_rx_scrqs; i++) {
2720                         if (!adapter->rx_scrq[i])
2721                                 continue;
2722
2723                         netdev_dbg(adapter->netdev, "Releasing rx_scrq[%d]\n",
2724                                    i);
2725                         if (adapter->rx_scrq[i]->irq) {
2726                                 free_irq(adapter->rx_scrq[i]->irq,
2727                                          adapter->rx_scrq[i]);
2728                                 irq_dispose_mapping(adapter->rx_scrq[i]->irq);
2729                                 adapter->rx_scrq[i]->irq = 0;
2730                         }
2731
2732                         release_sub_crq_queue(adapter, adapter->rx_scrq[i],
2733                                               do_h_free);
2734                 }
2735
2736                 kfree(adapter->rx_scrq);
2737                 adapter->rx_scrq = NULL;
2738                 adapter->num_active_rx_scrqs = 0;
2739         }
2740 }
2741
2742 static int disable_scrq_irq(struct ibmvnic_adapter *adapter,
2743                             struct ibmvnic_sub_crq_queue *scrq)
2744 {
2745         struct device *dev = &adapter->vdev->dev;
2746         unsigned long rc;
2747
2748         rc = plpar_hcall_norets(H_VIOCTL, adapter->vdev->unit_address,
2749                                 H_DISABLE_VIO_INTERRUPT, scrq->hw_irq, 0, 0);
2750         if (rc)
2751                 dev_err(dev, "Couldn't disable scrq irq 0x%lx. rc=%ld\n",
2752                         scrq->hw_irq, rc);
2753         return rc;
2754 }
2755
2756 static int enable_scrq_irq(struct ibmvnic_adapter *adapter,
2757                            struct ibmvnic_sub_crq_queue *scrq)
2758 {
2759         struct device *dev = &adapter->vdev->dev;
2760         unsigned long rc;
2761
2762         if (scrq->hw_irq > 0x100000000ULL) {
2763                 dev_err(dev, "bad hw_irq = %lx\n", scrq->hw_irq);
2764                 return 1;
2765         }
2766
2767         if (adapter->resetting &&
2768             adapter->reset_reason == VNIC_RESET_MOBILITY) {
2769                 u64 val = (0xff000000) | scrq->hw_irq;
2770
2771                 rc = plpar_hcall_norets(H_EOI, val);
2772                 if (rc)
2773                         dev_err(dev, "H_EOI FAILED irq 0x%llx. rc=%ld\n",
2774                                 val, rc);
2775         }
2776
2777         rc = plpar_hcall_norets(H_VIOCTL, adapter->vdev->unit_address,
2778                                 H_ENABLE_VIO_INTERRUPT, scrq->hw_irq, 0, 0);
2779         if (rc)
2780                 dev_err(dev, "Couldn't enable scrq irq 0x%lx. rc=%ld\n",
2781                         scrq->hw_irq, rc);
2782         return rc;
2783 }
2784
2785 static int ibmvnic_complete_tx(struct ibmvnic_adapter *adapter,
2786                                struct ibmvnic_sub_crq_queue *scrq)
2787 {
2788         struct device *dev = &adapter->vdev->dev;
2789         struct ibmvnic_tx_pool *tx_pool;
2790         struct ibmvnic_tx_buff *txbuff;
2791         union sub_crq *next;
2792         int index;
2793         int i, j;
2794         u8 *first;
2795
2796 restart_loop:
2797         while (pending_scrq(adapter, scrq)) {
2798                 unsigned int pool = scrq->pool_index;
2799                 int num_entries = 0;
2800
2801                 next = ibmvnic_next_scrq(adapter, scrq);
2802                 for (i = 0; i < next->tx_comp.num_comps; i++) {
2803                         if (next->tx_comp.rcs[i]) {
2804                                 dev_err(dev, "tx error %x\n",
2805                                         next->tx_comp.rcs[i]);
2806                                 continue;
2807                         }
2808                         index = be32_to_cpu(next->tx_comp.correlators[i]);
2809                         if (index & IBMVNIC_TSO_POOL_MASK) {
2810                                 tx_pool = &adapter->tso_pool[pool];
2811                                 index &= ~IBMVNIC_TSO_POOL_MASK;
2812                         } else {
2813                                 tx_pool = &adapter->tx_pool[pool];
2814                         }
2815
2816                         txbuff = &tx_pool->tx_buff[index];
2817
2818                         for (j = 0; j < IBMVNIC_MAX_FRAGS_PER_CRQ; j++) {
2819                                 if (!txbuff->data_dma[j])
2820                                         continue;
2821
2822                                 txbuff->data_dma[j] = 0;
2823                         }
2824                         /* if sub_crq was sent indirectly */
2825                         first = &txbuff->indir_arr[0].generic.first;
2826                         if (*first == IBMVNIC_CRQ_CMD) {
2827                                 dma_unmap_single(dev, txbuff->indir_dma,
2828                                                  sizeof(txbuff->indir_arr),
2829                                                  DMA_TO_DEVICE);
2830                                 *first = 0;
2831                         }
2832
2833                         if (txbuff->last_frag) {
2834                                 dev_kfree_skb_any(txbuff->skb);
2835                                 txbuff->skb = NULL;
2836                         }
2837
2838                         num_entries += txbuff->num_entries;
2839
2840                         tx_pool->free_map[tx_pool->producer_index] = index;
2841                         tx_pool->producer_index =
2842                                 (tx_pool->producer_index + 1) %
2843                                         tx_pool->num_buffers;
2844                 }
2845                 /* remove tx_comp scrq*/
2846                 next->tx_comp.first = 0;
2847
2848                 if (atomic_sub_return(num_entries, &scrq->used) <=
2849                     (adapter->req_tx_entries_per_subcrq / 2) &&
2850                     __netif_subqueue_stopped(adapter->netdev,
2851                                              scrq->pool_index)) {
2852                         netif_wake_subqueue(adapter->netdev, scrq->pool_index);
2853                         netdev_dbg(adapter->netdev, "Started queue %d\n",
2854                                    scrq->pool_index);
2855                 }
2856         }
2857
2858         enable_scrq_irq(adapter, scrq);
2859
2860         if (pending_scrq(adapter, scrq)) {
2861                 disable_scrq_irq(adapter, scrq);
2862                 goto restart_loop;
2863         }
2864
2865         return 0;
2866 }
2867
2868 static irqreturn_t ibmvnic_interrupt_tx(int irq, void *instance)
2869 {
2870         struct ibmvnic_sub_crq_queue *scrq = instance;
2871         struct ibmvnic_adapter *adapter = scrq->adapter;
2872
2873         disable_scrq_irq(adapter, scrq);
2874         ibmvnic_complete_tx(adapter, scrq);
2875
2876         return IRQ_HANDLED;
2877 }
2878
2879 static irqreturn_t ibmvnic_interrupt_rx(int irq, void *instance)
2880 {
2881         struct ibmvnic_sub_crq_queue *scrq = instance;
2882         struct ibmvnic_adapter *adapter = scrq->adapter;
2883
2884         /* When booting a kdump kernel we can hit pending interrupts
2885          * prior to completing driver initialization.
2886          */
2887         if (unlikely(adapter->state != VNIC_OPEN))
2888                 return IRQ_NONE;
2889
2890         adapter->rx_stats_buffers[scrq->scrq_num].interrupts++;
2891
2892         if (napi_schedule_prep(&adapter->napi[scrq->scrq_num])) {
2893                 disable_scrq_irq(adapter, scrq);
2894                 __napi_schedule(&adapter->napi[scrq->scrq_num]);
2895         }
2896
2897         return IRQ_HANDLED;
2898 }
2899
2900 static int init_sub_crq_irqs(struct ibmvnic_adapter *adapter)
2901 {
2902         struct device *dev = &adapter->vdev->dev;
2903         struct ibmvnic_sub_crq_queue *scrq;
2904         int i = 0, j = 0;
2905         int rc = 0;
2906
2907         for (i = 0; i < adapter->req_tx_queues; i++) {
2908                 netdev_dbg(adapter->netdev, "Initializing tx_scrq[%d] irq\n",
2909                            i);
2910                 scrq = adapter->tx_scrq[i];
2911                 scrq->irq = irq_create_mapping(NULL, scrq->hw_irq);
2912
2913                 if (!scrq->irq) {
2914                         rc = -EINVAL;
2915                         dev_err(dev, "Error mapping irq\n");
2916                         goto req_tx_irq_failed;
2917                 }
2918
2919                 rc = request_irq(scrq->irq, ibmvnic_interrupt_tx,
2920                                  0, "ibmvnic_tx", scrq);
2921
2922                 if (rc) {
2923                         dev_err(dev, "Couldn't register tx irq 0x%x. rc=%d\n",
2924                                 scrq->irq, rc);
2925                         irq_dispose_mapping(scrq->irq);
2926                         goto req_tx_irq_failed;
2927                 }
2928         }
2929
2930         for (i = 0; i < adapter->req_rx_queues; i++) {
2931                 netdev_dbg(adapter->netdev, "Initializing rx_scrq[%d] irq\n",
2932                            i);
2933                 scrq = adapter->rx_scrq[i];
2934                 scrq->irq = irq_create_mapping(NULL, scrq->hw_irq);
2935                 if (!scrq->irq) {
2936                         rc = -EINVAL;
2937                         dev_err(dev, "Error mapping irq\n");
2938                         goto req_rx_irq_failed;
2939                 }
2940                 rc = request_irq(scrq->irq, ibmvnic_interrupt_rx,
2941                                  0, "ibmvnic_rx", scrq);
2942                 if (rc) {
2943                         dev_err(dev, "Couldn't register rx irq 0x%x. rc=%d\n",
2944                                 scrq->irq, rc);
2945                         irq_dispose_mapping(scrq->irq);
2946                         goto req_rx_irq_failed;
2947                 }
2948         }
2949         return rc;
2950
2951 req_rx_irq_failed:
2952         for (j = 0; j < i; j++) {
2953                 free_irq(adapter->rx_scrq[j]->irq, adapter->rx_scrq[j]);
2954                 irq_dispose_mapping(adapter->rx_scrq[j]->irq);
2955         }
2956         i = adapter->req_tx_queues;
2957 req_tx_irq_failed:
2958         for (j = 0; j < i; j++) {
2959                 free_irq(adapter->tx_scrq[j]->irq, adapter->tx_scrq[j]);
2960                 irq_dispose_mapping(adapter->rx_scrq[j]->irq);
2961         }
2962         release_sub_crqs(adapter, 1);
2963         return rc;
2964 }
2965
2966 static int init_sub_crqs(struct ibmvnic_adapter *adapter)
2967 {
2968         struct device *dev = &adapter->vdev->dev;
2969         struct ibmvnic_sub_crq_queue **allqueues;
2970         int registered_queues = 0;
2971         int total_queues;
2972         int more = 0;
2973         int i;
2974
2975         total_queues = adapter->req_tx_queues + adapter->req_rx_queues;
2976
2977         allqueues = kcalloc(total_queues, sizeof(*allqueues), GFP_KERNEL);
2978         if (!allqueues)
2979                 return -1;
2980
2981         for (i = 0; i < total_queues; i++) {
2982                 allqueues[i] = init_sub_crq_queue(adapter);
2983                 if (!allqueues[i]) {
2984                         dev_warn(dev, "Couldn't allocate all sub-crqs\n");
2985                         break;
2986                 }
2987                 registered_queues++;
2988         }
2989
2990         /* Make sure we were able to register the minimum number of queues */
2991         if (registered_queues <
2992             adapter->min_tx_queues + adapter->min_rx_queues) {
2993                 dev_err(dev, "Fatal: Couldn't init  min number of sub-crqs\n");
2994                 goto tx_failed;
2995         }
2996
2997         /* Distribute the failed allocated queues*/
2998         for (i = 0; i < total_queues - registered_queues + more ; i++) {
2999                 netdev_dbg(adapter->netdev, "Reducing number of queues\n");
3000                 switch (i % 3) {
3001                 case 0:
3002                         if (adapter->req_rx_queues > adapter->min_rx_queues)
3003                                 adapter->req_rx_queues--;
3004                         else
3005                                 more++;
3006                         break;
3007                 case 1:
3008                         if (adapter->req_tx_queues > adapter->min_tx_queues)
3009                                 adapter->req_tx_queues--;
3010                         else
3011                                 more++;
3012                         break;
3013                 }
3014         }
3015
3016         adapter->tx_scrq = kcalloc(adapter->req_tx_queues,
3017                                    sizeof(*adapter->tx_scrq), GFP_KERNEL);
3018         if (!adapter->tx_scrq)
3019                 goto tx_failed;
3020
3021         for (i = 0; i < adapter->req_tx_queues; i++) {
3022                 adapter->tx_scrq[i] = allqueues[i];
3023                 adapter->tx_scrq[i]->pool_index = i;
3024                 adapter->num_active_tx_scrqs++;
3025         }
3026
3027         adapter->rx_scrq = kcalloc(adapter->req_rx_queues,
3028                                    sizeof(*adapter->rx_scrq), GFP_KERNEL);
3029         if (!adapter->rx_scrq)
3030                 goto rx_failed;
3031
3032         for (i = 0; i < adapter->req_rx_queues; i++) {
3033                 adapter->rx_scrq[i] = allqueues[i + adapter->req_tx_queues];
3034                 adapter->rx_scrq[i]->scrq_num = i;
3035                 adapter->num_active_rx_scrqs++;
3036         }
3037
3038         kfree(allqueues);
3039         return 0;
3040
3041 rx_failed:
3042         kfree(adapter->tx_scrq);
3043         adapter->tx_scrq = NULL;
3044 tx_failed:
3045         for (i = 0; i < registered_queues; i++)
3046                 release_sub_crq_queue(adapter, allqueues[i], 1);
3047         kfree(allqueues);
3048         return -1;
3049 }
3050
3051 static void ibmvnic_send_req_caps(struct ibmvnic_adapter *adapter, int retry)
3052 {
3053         struct device *dev = &adapter->vdev->dev;
3054         union ibmvnic_crq crq;
3055         int max_entries;
3056
3057         if (!retry) {
3058                 /* Sub-CRQ entries are 32 byte long */
3059                 int entries_page = 4 * PAGE_SIZE / (sizeof(u64) * 4);
3060
3061                 if (adapter->min_tx_entries_per_subcrq > entries_page ||
3062                     adapter->min_rx_add_entries_per_subcrq > entries_page) {
3063                         dev_err(dev, "Fatal, invalid entries per sub-crq\n");
3064                         return;
3065                 }