drivers/net/ethernet/ibm/ibmvnic.c

   1 // SPDX-License-Identifier: GPL-2.0-or-later
   2 /**************************************************************************/
   3 /*                                                                        */
   4 /*  IBM System i and System p Virtual NIC Device Driver                   */
   5 /*  Copyright (C) 2014 IBM Corp.                                          */
   6 /*  Santiago Leon ([email protected])                                  */
   7 /*  Thomas Falcon ([email protected])                           */
   8 /*  John Allen ([email protected])                                */
   9 /*                                                                        */
  10 /*                                                                        */
  11 /* This module contains the implementation of a virtual ethernet device   */
  12 /* for use with IBM i/p Series LPAR Linux. It utilizes the logical LAN    */
  13 /* option of the RS/6000 Platform Architecture to interface with virtual  */
  14 /* ethernet NICs that are presented to the partition by the hypervisor.   */
  15 /*                                                                         */
  16 /* Messages are passed between the VNIC driver and the VNIC server using  */
  17 /* Command/Response Queues (CRQs) and sub CRQs (sCRQs). CRQs are used to  */
  18 /* issue and receive commands that initiate communication with the server */
  19 /* on driver initialization. Sub CRQs (sCRQs) are similar to CRQs, but    */
  20 /* are used by the driver to notify the server that a packet is           */
  21 /* ready for transmission or that a buffer has been added to receive a    */
  22 /* packet. Subsequently, sCRQs are used by the server to notify the       */
  23 /* driver that a packet transmission has been completed or that a packet  */
  24 /* has been received and placed in a waiting buffer.                      */
  25 /*                                                                        */
  26 /* In lieu of a more conventional "on-the-fly" DMA mapping strategy in    */
  27 /* which skbs are DMA mapped and immediately unmapped when the transmit   */
  28 /* or receive has been completed, the VNIC driver is required to use      */
  29 /* "long term mapping". This entails that large, continuous DMA mapped    */
  30 /* buffers are allocated on driver initialization and these buffers are   */
  31 /* then continuously reused to pass skbs to and from the VNIC server.     */
  32 /*                                                                        */
  33 /**************************************************************************/
  34
  35 #include <linux/module.h>
  36 #include <linux/moduleparam.h>
  37 #include <linux/types.h>
  38 #include <linux/errno.h>
  39 #include <linux/completion.h>
  40 #include <linux/ioport.h>
  41 #include <linux/dma-mapping.h>
  42 #include <linux/kernel.h>
  43 #include <linux/netdevice.h>
  44 #include <linux/etherdevice.h>
  45 #include <linux/skbuff.h>
  46 #include <linux/init.h>
  47 #include <linux/delay.h>
  48 #include <linux/mm.h>
  49 #include <linux/ethtool.h>
  50 #include <linux/proc_fs.h>
  51 #include <linux/if_arp.h>
  52 #include <linux/in.h>
  53 #include <linux/ip.h>
  54 #include <linux/ipv6.h>
  55 #include <linux/irq.h>
  56 #include <linux/kthread.h>
  57 #include <linux/seq_file.h>
  58 #include <linux/interrupt.h>
  59 #include <net/net_namespace.h>
  60 #include <asm/hvcall.h>
  61 #include <linux/atomic.h>
  62 #include <asm/vio.h>
  63 #include <asm/iommu.h>
  64 #include <linux/uaccess.h>
  65 #include <asm/firmware.h>
  66 #include <linux/workqueue.h>
  67 #include <linux/if_vlan.h>
  68 #include <linux/utsname.h>
  69
  70 #include "ibmvnic.h"
  71
  72 static const char ibmvnic_driver_name[] = "ibmvnic";
  73 static const char ibmvnic_driver_string[] = "IBM System i/p Virtual NIC Driver";
  74
  75 MODULE_AUTHOR("Santiago Leon");
  76 MODULE_DESCRIPTION("IBM System i/p Virtual NIC Driver");
  77 MODULE_LICENSE("GPL");
  78 MODULE_VERSION(IBMVNIC_DRIVER_VERSION);
  79
  80 static int ibmvnic_version = IBMVNIC_INITIAL_VERSION;
  81 static void release_sub_crqs(struct ibmvnic_adapter *, bool);
  82 static int ibmvnic_reset_crq(struct ibmvnic_adapter *);
  83 static int ibmvnic_send_crq_init(struct ibmvnic_adapter *);
  84 static int ibmvnic_reenable_crq_queue(struct ibmvnic_adapter *);
  85 static int ibmvnic_send_crq(struct ibmvnic_adapter *, union ibmvnic_crq *);
  86 static int send_subcrq_indirect(struct ibmvnic_adapter *, u64, u64, u64);
  87 static irqreturn_t ibmvnic_interrupt_rx(int irq, void *instance);
  88 static int enable_scrq_irq(struct ibmvnic_adapter *,
  89                            struct ibmvnic_sub_crq_queue *);
  90 static int disable_scrq_irq(struct ibmvnic_adapter *,
  91                             struct ibmvnic_sub_crq_queue *);
  92 static int pending_scrq(struct ibmvnic_adapter *,
  93                         struct ibmvnic_sub_crq_queue *);
  94 static union sub_crq *ibmvnic_next_scrq(struct ibmvnic_adapter *,
  95                                         struct ibmvnic_sub_crq_queue *);
  96 static int ibmvnic_poll(struct napi_struct *napi, int data);
  97 static void send_query_map(struct ibmvnic_adapter *adapter);
  98 static int send_request_map(struct ibmvnic_adapter *, dma_addr_t, u32, u8);
  99 static int send_request_unmap(struct ibmvnic_adapter *, u8);
 100 static int send_login(struct ibmvnic_adapter *adapter);
 101 static void send_query_cap(struct ibmvnic_adapter *adapter);
 102 static int init_sub_crqs(struct ibmvnic_adapter *);
 103 static int init_sub_crq_irqs(struct ibmvnic_adapter *adapter);
 104 static int ibmvnic_reset_init(struct ibmvnic_adapter *, bool reset);
 105 static void release_crq_queue(struct ibmvnic_adapter *);
 106 static int __ibmvnic_set_mac(struct net_device *, u8 *);
 107 static int init_crq_queue(struct ibmvnic_adapter *adapter);
 108 static int send_query_phys_parms(struct ibmvnic_adapter *adapter);
 109 static void ibmvnic_tx_scrq_clean_buffer(struct ibmvnic_adapter *adapter,
 110                                          struct ibmvnic_sub_crq_queue *tx_scrq);
 111 static void free_long_term_buff(struct ibmvnic_adapter *adapter,
 112                                 struct ibmvnic_long_term_buff *ltb);
 113
 114 struct ibmvnic_stat {
 115         char name[ETH_GSTRING_LEN];
 116         int offset;
 117 };
 118
 119 #define IBMVNIC_STAT_OFF(stat) (offsetof(struct ibmvnic_adapter, stats) + \
 120                              offsetof(struct ibmvnic_statistics, stat))
 121 #define IBMVNIC_GET_STAT(a, off) (*((u64 *)(((unsigned long)(a)) + (off))))
 122
 123 static const struct ibmvnic_stat ibmvnic_stats[] = {
 124         {"rx_packets", IBMVNIC_STAT_OFF(rx_packets)},
 125         {"rx_bytes", IBMVNIC_STAT_OFF(rx_bytes)},
 126         {"tx_packets", IBMVNIC_STAT_OFF(tx_packets)},
 127         {"tx_bytes", IBMVNIC_STAT_OFF(tx_bytes)},
 128         {"ucast_tx_packets", IBMVNIC_STAT_OFF(ucast_tx_packets)},
 129         {"ucast_rx_packets", IBMVNIC_STAT_OFF(ucast_rx_packets)},
 130         {"mcast_tx_packets", IBMVNIC_STAT_OFF(mcast_tx_packets)},
 131         {"mcast_rx_packets", IBMVNIC_STAT_OFF(mcast_rx_packets)},
 132         {"bcast_tx_packets", IBMVNIC_STAT_OFF(bcast_tx_packets)},
 133         {"bcast_rx_packets", IBMVNIC_STAT_OFF(bcast_rx_packets)},
 134         {"align_errors", IBMVNIC_STAT_OFF(align_errors)},
 135         {"fcs_errors", IBMVNIC_STAT_OFF(fcs_errors)},
 136         {"single_collision_frames", IBMVNIC_STAT_OFF(single_collision_frames)},
 137         {"multi_collision_frames", IBMVNIC_STAT_OFF(multi_collision_frames)},
 138         {"sqe_test_errors", IBMVNIC_STAT_OFF(sqe_test_errors)},
 139         {"deferred_tx", IBMVNIC_STAT_OFF(deferred_tx)},
 140         {"late_collisions", IBMVNIC_STAT_OFF(late_collisions)},
 141         {"excess_collisions", IBMVNIC_STAT_OFF(excess_collisions)},
 142         {"internal_mac_tx_errors", IBMVNIC_STAT_OFF(internal_mac_tx_errors)},
 143         {"carrier_sense", IBMVNIC_STAT_OFF(carrier_sense)},
 144         {"too_long_frames", IBMVNIC_STAT_OFF(too_long_frames)},
 145         {"internal_mac_rx_errors", IBMVNIC_STAT_OFF(internal_mac_rx_errors)},
 146 };
 147
 148 static int send_crq_init_complete(struct ibmvnic_adapter *adapter)
 149 {
 150         union ibmvnic_crq crq;
 151
 152         memset(&crq, 0, sizeof(crq));
 153         crq.generic.first = IBMVNIC_CRQ_INIT_CMD;
 154         crq.generic.cmd = IBMVNIC_CRQ_INIT_COMPLETE;
 155
 156         return ibmvnic_send_crq(adapter, &crq);
 157 }
 158
 159 static int send_version_xchg(struct ibmvnic_adapter *adapter)
 160 {
 161         union ibmvnic_crq crq;
 162
 163         memset(&crq, 0, sizeof(crq));
 164         crq.version_exchange.first = IBMVNIC_CRQ_CMD;
 165         crq.version_exchange.cmd = VERSION_EXCHANGE;
 166         crq.version_exchange.version = cpu_to_be16(ibmvnic_version);
 167
 168         return ibmvnic_send_crq(adapter, &crq);
 169 }
 170
 171 static long h_reg_sub_crq(unsigned long unit_address, unsigned long token,
 172                           unsigned long length, unsigned long *number,
 173                           unsigned long *irq)
 174 {
 175         unsigned long retbuf[PLPAR_HCALL_BUFSIZE];
 176         long rc;
 177
 178         rc = plpar_hcall(H_REG_SUB_CRQ, retbuf, unit_address, token, length);
 179         *number = retbuf[0];
 180         *irq = retbuf[1];
 181
 182         return rc;
 183 }
 184
 185 /**
 186  * ibmvnic_wait_for_completion - Check device state and wait for completion
 187  * @adapter: private device data
 188  * @comp_done: completion structure to wait for
 189  * @timeout: time to wait in milliseconds
 190  *
 191  * Wait for a completion signal or until the timeout limit is reached
 192  * while checking that the device is still active.
 193  */
 194 static int ibmvnic_wait_for_completion(struct ibmvnic_adapter *adapter,
 195                                        struct completion *comp_done,
 196                                        unsigned long timeout)
 197 {
 198         struct net_device *netdev;
 199         unsigned long div_timeout;
 200         u8 retry;
 201
 202         netdev = adapter->netdev;
 203         retry = 5;
 204         div_timeout = msecs_to_jiffies(timeout / retry);
 205         while (true) {
 206                 if (!adapter->crq.active) {
 207                         netdev_err(netdev, "Device down!\n");
 208                         return -ENODEV;
 209                 }
 210                 if (!retry--)
 211                         break;
 212                 if (wait_for_completion_timeout(comp_done, div_timeout))
 213                         return 0;
 214         }
 215         netdev_err(netdev, "Operation timed out.\n");
 216         return -ETIMEDOUT;
 217 }
 218
 219 /**
 220  * reuse_ltb() - Check if a long term buffer can be reused
 221  * @ltb:  The long term buffer to be checked
 222  * @size: The size of the long term buffer.
 223  *
 224  * An LTB can be reused unless its size has changed.
 225  *
 226  * Return: Return true if the LTB can be reused, false otherwise.
 227  */
 228 static bool reuse_ltb(struct ibmvnic_long_term_buff *ltb, int size)
 229 {
 230         return (ltb->buff && ltb->size == size);
 231 }
 232
 233 /**
 234  * alloc_long_term_buff() - Allocate a long term buffer (LTB)
 235  *
 236  * @adapter: ibmvnic adapter associated to the LTB
 237  * @ltb:     container object for the LTB
 238  * @size:    size of the LTB
 239  *
 240  * Allocate an LTB of the specified size and notify VIOS.
 241  *
 242  * If the given @ltb already has the correct size, reuse it. Otherwise if
 243  * its non-NULL, free it. Then allocate a new one of the correct size.
 244  * Notify the VIOS either way since we may now be working with a new VIOS.
 245  *
 246  * Allocating larger chunks of memory during resets, specially LPM or under
 247  * low memory situations can cause resets to fail/timeout and for LPAR to
 248  * lose connectivity. So hold onto the LTB even if we fail to communicate
 249  * with the VIOS and reuse it on next open. Free LTB when adapter is closed.
 250  *
 251  * Return: 0 if we were able to allocate the LTB and notify the VIOS and
 252  *         a negative value otherwise.
 253  */
 254 static int alloc_long_term_buff(struct ibmvnic_adapter *adapter,
 255                                 struct ibmvnic_long_term_buff *ltb, int size)
 256 {
 257         struct device *dev = &adapter->vdev->dev;
 258         int rc;
 259
 260         if (!reuse_ltb(ltb, size)) {
 261                 dev_dbg(dev,
 262                         "LTB size changed from 0x%llx to 0x%x, reallocating\n",
 263                          ltb->size, size);
 264                 free_long_term_buff(adapter, ltb);
 265         }
 266
 267         if (ltb->buff) {
 268                 dev_dbg(dev, "Reusing LTB [map %d, size 0x%llx]\n",
 269                         ltb->map_id, ltb->size);
 270         } else {
 271                 ltb->buff = dma_alloc_coherent(dev, size, &ltb->addr,
 272                                                GFP_KERNEL);
 273                 if (!ltb->buff) {
 274                         dev_err(dev, "Couldn't alloc long term buffer\n");
 275                         return -ENOMEM;
 276                 }
 277                 ltb->size = size;
 278
 279                 ltb->map_id = find_first_zero_bit(adapter->map_ids,
 280                                                   MAX_MAP_ID);
 281                 bitmap_set(adapter->map_ids, ltb->map_id, 1);
 282
 283                 dev_dbg(dev,
 284                         "Allocated new LTB [map %d, size 0x%llx]\n",
 285                          ltb->map_id, ltb->size);
 286         }
 287
 288         /* Ensure ltb is zeroed - specially when reusing it. */
 289         memset(ltb->buff, 0, ltb->size);
 290
 291         mutex_lock(&adapter->fw_lock);
 292         adapter->fw_done_rc = 0;
 293         reinit_completion(&adapter->fw_done);
 294
 295         rc = send_request_map(adapter, ltb->addr, ltb->size, ltb->map_id);
 296         if (rc) {
 297                 dev_err(dev, "send_request_map failed, rc = %d\n", rc);
 298                 goto out;
 299         }
 300
 301         rc = ibmvnic_wait_for_completion(adapter, &adapter->fw_done, 10000);
 302         if (rc) {
 303                 dev_err(dev, "LTB map request aborted or timed out, rc = %d\n",
 304                         rc);
 305                 goto out;
 306         }
 307
 308         if (adapter->fw_done_rc) {
 309                 dev_err(dev, "Couldn't map LTB, rc = %d\n",
 310                         adapter->fw_done_rc);
 311                 rc = -1;
 312                 goto out;
 313         }
 314         rc = 0;
 315 out:
 316         /* don't free LTB on communication error - see function header */
 317         mutex_unlock(&adapter->fw_lock);
 318         return rc;
 319 }
 320
 321 static void free_long_term_buff(struct ibmvnic_adapter *adapter,
 322                                 struct ibmvnic_long_term_buff *ltb)
 323 {
 324         struct device *dev = &adapter->vdev->dev;
 325
 326         if (!ltb->buff)
 327                 return;
 328
 329         /* VIOS automatically unmaps the long term buffer at remote
 330          * end for the following resets:
 331          * FAILOVER, MOBILITY, TIMEOUT.
 332          */
 333         if (adapter->reset_reason != VNIC_RESET_FAILOVER &&
 334             adapter->reset_reason != VNIC_RESET_MOBILITY &&
 335             adapter->reset_reason != VNIC_RESET_TIMEOUT)
 336                 send_request_unmap(adapter, ltb->map_id);
 337
 338         dma_free_coherent(dev, ltb->size, ltb->buff, ltb->addr);
 339
 340         ltb->buff = NULL;
 341         /* mark this map_id free */
 342         bitmap_clear(adapter->map_ids, ltb->map_id, 1);
 343         ltb->map_id = 0;
 344 }
 345
 346 static void deactivate_rx_pools(struct ibmvnic_adapter *adapter)
 347 {
 348         int i;
 349
 350         for (i = 0; i < adapter->num_active_rx_pools; i++)
 351                 adapter->rx_pool[i].active = 0;
 352 }
 353
 354 static void replenish_rx_pool(struct ibmvnic_adapter *adapter,
 355                               struct ibmvnic_rx_pool *pool)
 356 {
 357         int count = pool->size - atomic_read(&pool->available);
 358         u64 handle = adapter->rx_scrq[pool->index]->handle;
 359         struct device *dev = &adapter->vdev->dev;
 360         struct ibmvnic_ind_xmit_queue *ind_bufp;
 361         struct ibmvnic_sub_crq_queue *rx_scrq;
 362         union sub_crq *sub_crq;
 363         int buffers_added = 0;
 364         unsigned long lpar_rc;
 365         struct sk_buff *skb;
 366         unsigned int offset;
 367         dma_addr_t dma_addr;
 368         unsigned char *dst;
 369         int shift = 0;
 370         int index;
 371         int i;
 372
 373         if (!pool->active)
 374                 return;
 375
 376         rx_scrq = adapter->rx_scrq[pool->index];
 377         ind_bufp = &rx_scrq->ind_buf;
 378
 379         /* netdev_skb_alloc() could have failed after we saved a few skbs
 380          * in the indir_buf and we would not have sent them to VIOS yet.
 381          * To account for them, start the loop at ind_bufp->index rather
 382          * than 0. If we pushed all the skbs to VIOS, ind_bufp->index will
 383          * be 0.
 384          */
 385         for (i = ind_bufp->index; i < count; ++i) {
 386                 index = pool->free_map[pool->next_free];
 387
 388                 /* We maybe reusing the skb from earlier resets. Allocate
 389                  * only if necessary. But since the LTB may have changed
 390                  * during reset (see init_rx_pools()), update LTB below
 391                  * even if reusing skb.
 392                  */
 393                 skb = pool->rx_buff[index].skb;
 394                 if (!skb) {
 395                         skb = netdev_alloc_skb(adapter->netdev,
 396                                                pool->buff_size);
 397                         if (!skb) {
 398                                 dev_err(dev, "Couldn't replenish rx buff\n");
 399                                 adapter->replenish_no_mem++;
 400                                 break;
 401                         }
 402                 }
 403
 404                 pool->free_map[pool->next_free] = IBMVNIC_INVALID_MAP;
 405                 pool->next_free = (pool->next_free + 1) % pool->size;
 406
 407                 /* Copy the skb to the long term mapped DMA buffer */
 408                 offset = index * pool->buff_size;
 409                 dst = pool->long_term_buff.buff + offset;
 410                 memset(dst, 0, pool->buff_size);
 411                 dma_addr = pool->long_term_buff.addr + offset;
 412
 413                 /* add the skb to an rx_buff in the pool */
 414                 pool->rx_buff[index].data = dst;
 415                 pool->rx_buff[index].dma = dma_addr;
 416                 pool->rx_buff[index].skb = skb;
 417                 pool->rx_buff[index].pool_index = pool->index;
 418                 pool->rx_buff[index].size = pool->buff_size;
 419
 420                 /* queue the rx_buff for the next send_subcrq_indirect */
 421                 sub_crq = &ind_bufp->indir_arr[ind_bufp->index++];
 422                 memset(sub_crq, 0, sizeof(*sub_crq));
 423                 sub_crq->rx_add.first = IBMVNIC_CRQ_CMD;
 424                 sub_crq->rx_add.correlator =
 425                     cpu_to_be64((u64)&pool->rx_buff[index]);
 426                 sub_crq->rx_add.ioba = cpu_to_be32(dma_addr);
 427                 sub_crq->rx_add.map_id = pool->long_term_buff.map_id;
 428
 429                 /* The length field of the sCRQ is defined to be 24 bits so the
 430                  * buffer size needs to be left shifted by a byte before it is
 431                  * converted to big endian to prevent the last byte from being
 432                  * truncated.
 433                  */
 434 #ifdef __LITTLE_ENDIAN__
 435                 shift = 8;
 436 #endif
 437                 sub_crq->rx_add.len = cpu_to_be32(pool->buff_size << shift);
 438
 439                 /* if send_subcrq_indirect queue is full, flush to VIOS */
 440                 if (ind_bufp->index == IBMVNIC_MAX_IND_DESCS ||
 441                     i == count - 1) {
 442                         lpar_rc =
 443                                 send_subcrq_indirect(adapter, handle,
 444                                                      (u64)ind_bufp->indir_dma,
 445                                                      (u64)ind_bufp->index);
 446                         if (lpar_rc != H_SUCCESS)
 447                                 goto failure;
 448                         buffers_added += ind_bufp->index;
 449                         adapter->replenish_add_buff_success += ind_bufp->index;
 450                         ind_bufp->index = 0;
 451                 }
 452         }
 453         atomic_add(buffers_added, &pool->available);
 454         return;
 455
 456 failure:
 457         if (lpar_rc != H_PARAMETER && lpar_rc != H_CLOSED)
 458                 dev_err_ratelimited(dev, "rx: replenish packet buffer failed\n");
 459         for (i = ind_bufp->index - 1; i >= 0; --i) {
 460                 struct ibmvnic_rx_buff *rx_buff;
 461
 462                 pool->next_free = pool->next_free == 0 ?
 463                                   pool->size - 1 : pool->next_free - 1;
 464                 sub_crq = &ind_bufp->indir_arr[i];
 465                 rx_buff = (struct ibmvnic_rx_buff *)
 466                                 be64_to_cpu(sub_crq->rx_add.correlator);
 467                 index = (int)(rx_buff - pool->rx_buff);
 468                 pool->free_map[pool->next_free] = index;
 469                 dev_kfree_skb_any(pool->rx_buff[index].skb);
 470                 pool->rx_buff[index].skb = NULL;
 471         }
 472         adapter->replenish_add_buff_failure += ind_bufp->index;
 473         atomic_add(buffers_added, &pool->available);
 474         ind_bufp->index = 0;
 475         if (lpar_rc == H_CLOSED || adapter->failover_pending) {
 476                 /* Disable buffer pool replenishment and report carrier off if
 477                  * queue is closed or pending failover.
 478                  * Firmware guarantees that a signal will be sent to the
 479                  * driver, triggering a reset.
 480                  */
 481                 deactivate_rx_pools(adapter);
 482                 netif_carrier_off(adapter->netdev);
 483         }
 484 }
 485
 486 static void replenish_pools(struct ibmvnic_adapter *adapter)
 487 {
 488         int i;
 489
 490         adapter->replenish_task_cycles++;
 491         for (i = 0; i < adapter->num_active_rx_pools; i++) {
 492                 if (adapter->rx_pool[i].active)
 493                         replenish_rx_pool(adapter, &adapter->rx_pool[i]);
 494         }
 495
 496         netdev_dbg(adapter->netdev, "Replenished %d pools\n", i);
 497 }
 498
 499 static void release_stats_buffers(struct ibmvnic_adapter *adapter)
 500 {
 501         kfree(adapter->tx_stats_buffers);
 502         kfree(adapter->rx_stats_buffers);
 503         adapter->tx_stats_buffers = NULL;
 504         adapter->rx_stats_buffers = NULL;
 505 }
 506
 507 static int init_stats_buffers(struct ibmvnic_adapter *adapter)
 508 {
 509         adapter->tx_stats_buffers =
 510                                 kcalloc(IBMVNIC_MAX_QUEUES,
 511                                         sizeof(struct ibmvnic_tx_queue_stats),
 512                                         GFP_KERNEL);
 513         if (!adapter->tx_stats_buffers)
 514                 return -ENOMEM;
 515
 516         adapter->rx_stats_buffers =
 517                                 kcalloc(IBMVNIC_MAX_QUEUES,
 518                                         sizeof(struct ibmvnic_rx_queue_stats),
 519                                         GFP_KERNEL);
 520         if (!adapter->rx_stats_buffers)
 521                 return -ENOMEM;
 522
 523         return 0;
 524 }
 525
 526 static void release_stats_token(struct ibmvnic_adapter *adapter)
 527 {
 528         struct device *dev = &adapter->vdev->dev;
 529
 530         if (!adapter->stats_token)
 531                 return;
 532
 533         dma_unmap_single(dev, adapter->stats_token,
 534                          sizeof(struct ibmvnic_statistics),
 535                          DMA_FROM_DEVICE);
 536         adapter->stats_token = 0;
 537 }
 538
 539 static int init_stats_token(struct ibmvnic_adapter *adapter)
 540 {
 541         struct device *dev = &adapter->vdev->dev;
 542         dma_addr_t stok;
 543
 544         stok = dma_map_single(dev, &adapter->stats,
 545                               sizeof(struct ibmvnic_statistics),
 546                               DMA_FROM_DEVICE);
 547         if (dma_mapping_error(dev, stok)) {
 548                 dev_err(dev, "Couldn't map stats buffer\n");
 549                 return -1;
 550         }
 551
 552         adapter->stats_token = stok;
 553         netdev_dbg(adapter->netdev, "Stats token initialized (%llx)\n", stok);
 554         return 0;
 555 }
 556
 557 /**
 558  * release_rx_pools() - Release any rx pools attached to @adapter.
 559  * @adapter: ibmvnic adapter
 560  *
 561  * Safe to call this multiple times - even if no pools are attached.
 562  */
 563 static void release_rx_pools(struct ibmvnic_adapter *adapter)
 564 {
 565         struct ibmvnic_rx_pool *rx_pool;
 566         int i, j;
 567
 568         if (!adapter->rx_pool)
 569                 return;
 570
 571         for (i = 0; i < adapter->num_active_rx_pools; i++) {
 572                 rx_pool = &adapter->rx_pool[i];
 573
 574                 netdev_dbg(adapter->netdev, "Releasing rx_pool[%d]\n", i);
 575
 576                 kfree(rx_pool->free_map);
 577
 578                 free_long_term_buff(adapter, &rx_pool->long_term_buff);
 579
 580                 if (!rx_pool->rx_buff)
 581                         continue;
 582
 583                 for (j = 0; j < rx_pool->size; j++) {
 584                         if (rx_pool->rx_buff[j].skb) {
 585                                 dev_kfree_skb_any(rx_pool->rx_buff[j].skb);
 586                                 rx_pool->rx_buff[j].skb = NULL;
 587                         }
 588                 }
 589
 590                 kfree(rx_pool->rx_buff);
 591         }
 592
 593         kfree(adapter->rx_pool);
 594         adapter->rx_pool = NULL;
 595         adapter->num_active_rx_pools = 0;
 596         adapter->prev_rx_pool_size = 0;
 597 }
 598
 599 /**
 600  * reuse_rx_pools() - Check if the existing rx pools can be reused.
 601  * @adapter: ibmvnic adapter
 602  *
 603  * Check if the existing rx pools in the adapter can be reused. The
 604  * pools can be reused if the pool parameters (number of pools,
 605  * number of buffers in the pool and size of each buffer) have not
 606  * changed.
 607  *
 608  * NOTE: This assumes that all pools have the same number of buffers
 609  *       which is the case currently. If that changes, we must fix this.
 610  *
 611  * Return: true if the rx pools can be reused, false otherwise.
 612  */
 613 static bool reuse_rx_pools(struct ibmvnic_adapter *adapter)
 614 {
 615         u64 old_num_pools, new_num_pools;
 616         u64 old_pool_size, new_pool_size;
 617         u64 old_buff_size, new_buff_size;
 618
 619         if (!adapter->rx_pool)
 620                 return false;
 621
 622         old_num_pools = adapter->num_active_rx_pools;
 623         new_num_pools = adapter->req_rx_queues;
 624
 625         old_pool_size = adapter->prev_rx_pool_size;
 626         new_pool_size = adapter->req_rx_add_entries_per_subcrq;
 627
 628         old_buff_size = adapter->prev_rx_buf_sz;
 629         new_buff_size = adapter->cur_rx_buf_sz;
 630
 631         /* Require buff size to be exactly same for now */
 632         if (old_buff_size != new_buff_size)
 633                 return false;
 634
 635         if (old_num_pools == new_num_pools && old_pool_size == new_pool_size)
 636                 return true;
 637
 638         if (old_num_pools < adapter->min_rx_queues ||
 639             old_num_pools > adapter->max_rx_queues ||
 640             old_pool_size < adapter->min_rx_add_entries_per_subcrq ||
 641             old_pool_size > adapter->max_rx_add_entries_per_subcrq)
 642                 return false;
 643
 644         return true;
 645 }
 646
 647 /**
 648  * init_rx_pools(): Initialize the set of receiver pools in the adapter.
 649  * @netdev: net device associated with the vnic interface
 650  *
 651  * Initialize the set of receiver pools in the ibmvnic adapter associated
 652  * with the net_device @netdev. If possible, reuse the existing rx pools.
 653  * Otherwise free any existing pools and  allocate a new set of pools
 654  * before initializing them.
 655  *
 656  * Return: 0 on success and negative value on error.
 657  */
 658 static int init_rx_pools(struct net_device *netdev)
 659 {
 660         struct ibmvnic_adapter *adapter = netdev_priv(netdev);
 661         struct device *dev = &adapter->vdev->dev;
 662         struct ibmvnic_rx_pool *rx_pool;
 663         u64 num_pools;
 664         u64 pool_size;          /* # of buffers in one pool */
 665         u64 buff_size;
 666         int i, j;
 667
 668         pool_size = adapter->req_rx_add_entries_per_subcrq;
 669         num_pools = adapter->req_rx_queues;
 670         buff_size = adapter->cur_rx_buf_sz;
 671
 672         if (reuse_rx_pools(adapter)) {
 673                 dev_dbg(dev, "Reusing rx pools\n");
 674                 goto update_ltb;
 675         }
 676
 677         /* Allocate/populate the pools. */
 678         release_rx_pools(adapter);
 679
 680         adapter->rx_pool = kcalloc(num_pools,
 681                                    sizeof(struct ibmvnic_rx_pool),
 682                                    GFP_KERNEL);
 683         if (!adapter->rx_pool) {
 684                 dev_err(dev, "Failed to allocate rx pools\n");
 685                 return -1;
 686         }
 687
 688         /* Set num_active_rx_pools early. If we fail below after partial
 689          * allocation, release_rx_pools() will know how many to look for.
 690          */
 691         adapter->num_active_rx_pools = num_pools;
 692
 693         for (i = 0; i < num_pools; i++) {
 694                 rx_pool = &adapter->rx_pool[i];
 695
 696                 netdev_dbg(adapter->netdev,
 697                            "Initializing rx_pool[%d], %lld buffs, %lld bytes each\n",
 698                            i, pool_size, buff_size);
 699
 700                 rx_pool->size = pool_size;
 701                 rx_pool->index = i;
 702                 rx_pool->buff_size = ALIGN(buff_size, L1_CACHE_BYTES);
 703
 704                 rx_pool->free_map = kcalloc(rx_pool->size, sizeof(int),
 705                                             GFP_KERNEL);
 706                 if (!rx_pool->free_map) {
 707                         dev_err(dev, "Couldn't alloc free_map %d\n", i);
 708                         goto out_release;
 709                 }
 710
 711                 rx_pool->rx_buff = kcalloc(rx_pool->size,
 712                                            sizeof(struct ibmvnic_rx_buff),
 713                                            GFP_KERNEL);
 714                 if (!rx_pool->rx_buff) {
 715                         dev_err(dev, "Couldn't alloc rx buffers\n");
 716                         goto out_release;
 717                 }
 718         }
 719
 720         adapter->prev_rx_pool_size = pool_size;
 721         adapter->prev_rx_buf_sz = adapter->cur_rx_buf_sz;
 722
 723 update_ltb:
 724         for (i = 0; i < num_pools; i++) {
 725                 rx_pool = &adapter->rx_pool[i];
 726                 dev_dbg(dev, "Updating LTB for rx pool %d [%d, %d]\n",
 727                         i, rx_pool->size, rx_pool->buff_size);
 728
 729                 if (alloc_long_term_buff(adapter, &rx_pool->long_term_buff,
 730                                          rx_pool->size * rx_pool->buff_size))
 731                         goto out;
 732
 733                 for (j = 0; j < rx_pool->size; ++j) {
 734                         struct ibmvnic_rx_buff *rx_buff;
 735
 736                         rx_pool->free_map[j] = j;
 737
 738                         /* NOTE: Don't clear rx_buff->skb here - will leak
 739                          * memory! replenish_rx_pool() will reuse skbs or
 740                          * allocate as necessary.
 741                          */
 742                         rx_buff = &rx_pool->rx_buff[j];
 743                         rx_buff->dma = 0;
 744                         rx_buff->data = 0;
 745                         rx_buff->size = 0;
 746                         rx_buff->pool_index = 0;
 747                 }
 748
 749                 /* Mark pool "empty" so replenish_rx_pools() will
 750                  * update the LTB info for each buffer
 751                  */
 752                 atomic_set(&rx_pool->available, 0);
 753                 rx_pool->next_alloc = 0;
 754                 rx_pool->next_free = 0;
 755                 /* replenish_rx_pool() may have called deactivate_rx_pools()
 756                  * on failover. Ensure pool is active now.
 757                  */
 758                 rx_pool->active = 1;
 759         }
 760         return 0;
 761 out_release:
 762         release_rx_pools(adapter);
 763 out:
 764         /* We failed to allocate one or more LTBs or map them on the VIOS.
 765          * Hold onto the pools and any LTBs that we did allocate/map.
 766          */
 767         return -1;
 768 }
 769
 770 static void release_vpd_data(struct ibmvnic_adapter *adapter)
 771 {
 772         if (!adapter->vpd)
 773                 return;
 774
 775         kfree(adapter->vpd->buff);
 776         kfree(adapter->vpd);
 777
 778         adapter->vpd = NULL;
 779 }
 780
 781 static void release_one_tx_pool(struct ibmvnic_adapter *adapter,
 782                                 struct ibmvnic_tx_pool *tx_pool)
 783 {
 784         kfree(tx_pool->tx_buff);
 785         kfree(tx_pool->free_map);
 786         free_long_term_buff(adapter, &tx_pool->long_term_buff);
 787 }
 788
 789 /**
 790  * release_tx_pools() - Release any tx pools attached to @adapter.
 791  * @adapter: ibmvnic adapter
 792  *
 793  * Safe to call this multiple times - even if no pools are attached.
 794  */
 795 static void release_tx_pools(struct ibmvnic_adapter *adapter)
 796 {
 797         int i;
 798
 799         /* init_tx_pools() ensures that ->tx_pool and ->tso_pool are
 800          * both NULL or both non-NULL. So we only need to check one.
 801          */
 802         if (!adapter->tx_pool)
 803                 return;
 804
 805         for (i = 0; i < adapter->num_active_tx_pools; i++) {
 806                 release_one_tx_pool(adapter, &adapter->tx_pool[i]);
 807                 release_one_tx_pool(adapter, &adapter->tso_pool[i]);
 808         }
 809
 810         kfree(adapter->tx_pool);
 811         adapter->tx_pool = NULL;
 812         kfree(adapter->tso_pool);
 813         adapter->tso_pool = NULL;
 814         adapter->num_active_tx_pools = 0;
 815         adapter->prev_tx_pool_size = 0;
 816 }
 817
 818 static int init_one_tx_pool(struct net_device *netdev,
 819                             struct ibmvnic_tx_pool *tx_pool,
 820                             int pool_size, int buf_size)
 821 {
 822         int i;
 823
 824         tx_pool->tx_buff = kcalloc(pool_size,
 825                                    sizeof(struct ibmvnic_tx_buff),
 826                                    GFP_KERNEL);
 827         if (!tx_pool->tx_buff)
 828                 return -1;
 829
 830         tx_pool->free_map = kcalloc(pool_size, sizeof(int), GFP_KERNEL);
 831         if (!tx_pool->free_map) {
 832                 kfree(tx_pool->tx_buff);
 833                 tx_pool->tx_buff = NULL;
 834                 return -1;
 835         }
 836
 837         for (i = 0; i < pool_size; i++)
 838                 tx_pool->free_map[i] = i;
 839
 840         tx_pool->consumer_index = 0;
 841         tx_pool->producer_index = 0;
 842         tx_pool->num_buffers = pool_size;
 843         tx_pool->buf_size = buf_size;
 844
 845         return 0;
 846 }
 847
 848 /**
 849  * reuse_tx_pools() - Check if the existing tx pools can be reused.
 850  * @adapter: ibmvnic adapter
 851  *
 852  * Check if the existing tx pools in the adapter can be reused. The
 853  * pools can be reused if the pool parameters (number of pools,
 854  * number of buffers in the pool and mtu) have not changed.
 855  *
 856  * NOTE: This assumes that all pools have the same number of buffers
 857  *       which is the case currently. If that changes, we must fix this.
 858  *
 859  * Return: true if the tx pools can be reused, false otherwise.
 860  */
 861 static bool reuse_tx_pools(struct ibmvnic_adapter *adapter)
 862 {
 863         u64 old_num_pools, new_num_pools;
 864         u64 old_pool_size, new_pool_size;
 865         u64 old_mtu, new_mtu;
 866
 867         if (!adapter->tx_pool)
 868                 return false;
 869
 870         old_num_pools = adapter->num_active_tx_pools;
 871         new_num_pools = adapter->num_active_tx_scrqs;
 872         old_pool_size = adapter->prev_tx_pool_size;
 873         new_pool_size = adapter->req_tx_entries_per_subcrq;
 874         old_mtu = adapter->prev_mtu;
 875         new_mtu = adapter->req_mtu;
 876
 877         /* Require MTU to be exactly same to reuse pools for now */
 878         if (old_mtu != new_mtu)
 879                 return false;
 880
 881         if (old_num_pools == new_num_pools && old_pool_size == new_pool_size)
 882                 return true;
 883
 884         if (old_num_pools < adapter->min_tx_queues ||
 885             old_num_pools > adapter->max_tx_queues ||
 886             old_pool_size < adapter->min_tx_entries_per_subcrq ||
 887             old_pool_size > adapter->max_tx_entries_per_subcrq)
 888                 return false;
 889
 890         return true;
 891 }
 892
 893 /**
 894  * init_tx_pools(): Initialize the set of transmit pools in the adapter.
 895  * @netdev: net device associated with the vnic interface
 896  *
 897  * Initialize the set of transmit pools in the ibmvnic adapter associated
 898  * with the net_device @netdev. If possible, reuse the existing tx pools.
 899  * Otherwise free any existing pools and  allocate a new set of pools
 900  * before initializing them.
 901  *
 902  * Return: 0 on success and negative value on error.
 903  */
 904 static int init_tx_pools(struct net_device *netdev)
 905 {
 906         struct ibmvnic_adapter *adapter = netdev_priv(netdev);
 907         struct device *dev = &adapter->vdev->dev;
 908         int num_pools;
 909         u64 pool_size;          /* # of buffers in pool */
 910         u64 buff_size;
 911         int i, j, rc;
 912
 913         num_pools = adapter->req_tx_queues;
 914
 915         /* We must notify the VIOS about the LTB on all resets - but we only
 916          * need to alloc/populate pools if either the number of buffers or
 917          * size of each buffer in the pool has changed.
 918          */
 919         if (reuse_tx_pools(adapter)) {
 920                 netdev_dbg(netdev, "Reusing tx pools\n");
 921                 goto update_ltb;
 922         }
 923
 924         /* Allocate/populate the pools. */
 925         release_tx_pools(adapter);
 926
 927         pool_size = adapter->req_tx_entries_per_subcrq;
 928         num_pools = adapter->num_active_tx_scrqs;
 929
 930         adapter->tx_pool = kcalloc(num_pools,
 931                                    sizeof(struct ibmvnic_tx_pool), GFP_KERNEL);
 932         if (!adapter->tx_pool)
 933                 return -1;
 934
 935         adapter->tso_pool = kcalloc(num_pools,
 936                                     sizeof(struct ibmvnic_tx_pool), GFP_KERNEL);
 937         /* To simplify release_tx_pools() ensure that ->tx_pool and
 938          * ->tso_pool are either both NULL or both non-NULL.
 939          */
 940         if (!adapter->tso_pool) {
 941                 kfree(adapter->tx_pool);
 942                 adapter->tx_pool = NULL;
 943                 return -1;
 944         }
 945
 946         /* Set num_active_tx_pools early. If we fail below after partial
 947          * allocation, release_tx_pools() will know how many to look for.
 948          */
 949         adapter->num_active_tx_pools = num_pools;
 950
 951         buff_size = adapter->req_mtu + VLAN_HLEN;
 952         buff_size = ALIGN(buff_size, L1_CACHE_BYTES);
 953
 954         for (i = 0; i < num_pools; i++) {
 955                 dev_dbg(dev, "Init tx pool %d [%llu, %llu]\n",
 956                         i, adapter->req_tx_entries_per_subcrq, buff_size);
 957
 958                 rc = init_one_tx_pool(netdev, &adapter->tx_pool[i],
 959                                       pool_size, buff_size);
 960                 if (rc)
 961                         goto out_release;
 962
 963                 rc = init_one_tx_pool(netdev, &adapter->tso_pool[i],
 964                                       IBMVNIC_TSO_BUFS,
 965                                       IBMVNIC_TSO_BUF_SZ);
 966                 if (rc)
 967                         goto out_release;
 968         }
 969
 970         adapter->prev_tx_pool_size = pool_size;
 971         adapter->prev_mtu = adapter->req_mtu;
 972
 973 update_ltb:
 974         /* NOTE: All tx_pools have the same number of buffers (which is
 975          *       same as pool_size). All tso_pools have IBMVNIC_TSO_BUFS
 976          *       buffers (see calls init_one_tx_pool() for these).
 977          *       For consistency, we use tx_pool->num_buffers and
 978          *       tso_pool->num_buffers below.
 979          */
 980         rc = -1;
 981         for (i = 0; i < num_pools; i++) {
 982                 struct ibmvnic_tx_pool *tso_pool;
 983                 struct ibmvnic_tx_pool *tx_pool;
 984                 u32 ltb_size;
 985
 986                 tx_pool = &adapter->tx_pool[i];
 987                 ltb_size = tx_pool->num_buffers * tx_pool->buf_size;
 988                 if (alloc_long_term_buff(adapter, &tx_pool->long_term_buff,
 989                                          ltb_size))
 990                         goto out;
 991
 992                 dev_dbg(dev, "Updated LTB for tx pool %d [%p, %d, %d]\n",
 993                         i, tx_pool->long_term_buff.buff,
 994                         tx_pool->num_buffers, tx_pool->buf_size);
 995
 996                 tx_pool->consumer_index = 0;
 997                 tx_pool->producer_index = 0;
 998
 999                 for (j = 0; j < tx_pool->num_buffers; j++)
1000                         tx_pool->free_map[j] = j;
1001
1002                 tso_pool = &adapter->tso_pool[i];
1003                 ltb_size = tso_pool->num_buffers * tso_pool->buf_size;
1004                 if (alloc_long_term_buff(adapter, &tso_pool->long_term_buff,
1005                                          ltb_size))
1006                         goto out;
1007
1008                 dev_dbg(dev, "Updated LTB for tso pool %d [%p, %d, %d]\n",
1009                         i, tso_pool->long_term_buff.buff,
1010                         tso_pool->num_buffers, tso_pool->buf_size);
1011
1012                 tso_pool->consumer_index = 0;
1013                 tso_pool->producer_index = 0;
1014
1015                 for (j = 0; j < tso_pool->num_buffers; j++)
1016                         tso_pool->free_map[j] = j;
1017         }
1018
1019         return 0;
1020 out_release:
1021         release_tx_pools(adapter);
1022 out:
1023         /* We failed to allocate one or more LTBs or map them on the VIOS.
1024          * Hold onto the pools and any LTBs that we did allocate/map.
1025          */
1026         return rc;
1027 }
1028
1029 static void ibmvnic_napi_enable(struct ibmvnic_adapter *adapter)
1030 {
1031         int i;
1032
1033         if (adapter->napi_enabled)
1034                 return;
1035
1036         for (i = 0; i < adapter->req_rx_queues; i++)
1037                 napi_enable(&adapter->napi[i]);
1038
1039         adapter->napi_enabled = true;
1040 }
1041
1042 static void ibmvnic_napi_disable(struct ibmvnic_adapter *adapter)
1043 {
1044         int i;
1045
1046         if (!adapter->napi_enabled)
1047                 return;
1048
1049         for (i = 0; i < adapter->req_rx_queues; i++) {
1050                 netdev_dbg(adapter->netdev, "Disabling napi[%d]\n", i);
1051                 napi_disable(&adapter->napi[i]);
1052         }
1053
1054         adapter->napi_enabled = false;
1055 }
1056
1057 static int init_napi(struct ibmvnic_adapter *adapter)
1058 {
1059         int i;
1060
1061         adapter->napi = kcalloc(adapter->req_rx_queues,
1062                                 sizeof(struct napi_struct), GFP_KERNEL);
1063         if (!adapter->napi)
1064                 return -ENOMEM;
1065
1066         for (i = 0; i < adapter->req_rx_queues; i++) {
1067                 netdev_dbg(adapter->netdev, "Adding napi[%d]\n", i);
1068                 netif_napi_add(adapter->netdev, &adapter->napi[i],
1069                                ibmvnic_poll, NAPI_POLL_WEIGHT);
1070         }
1071
1072         adapter->num_active_rx_napi = adapter->req_rx_queues;
1073         return 0;
1074 }
1075
1076 static void release_napi(struct ibmvnic_adapter *adapter)
1077 {
1078         int i;
1079
1080         if (!adapter->napi)
1081                 return;
1082
1083         for (i = 0; i < adapter->num_active_rx_napi; i++) {
1084                 netdev_dbg(adapter->netdev, "Releasing napi[%d]\n", i);
1085                 netif_napi_del(&adapter->napi[i]);
1086         }
1087
1088         kfree(adapter->napi);
1089         adapter->napi = NULL;
1090         adapter->num_active_rx_napi = 0;
1091         adapter->napi_enabled = false;
1092 }
1093
1094 static const char *adapter_state_to_string(enum vnic_state state)
1095 {
1096         switch (state) {
1097         case VNIC_PROBING:
1098                 return "PROBING";
1099         case VNIC_PROBED:
1100                 return "PROBED";
1101         case VNIC_OPENING:
1102                 return "OPENING";
1103         case VNIC_OPEN:
1104                 return "OPEN";
1105         case VNIC_CLOSING:
1106                 return "CLOSING";
1107         case VNIC_CLOSED:
1108                 return "CLOSED";
1109         case VNIC_REMOVING:
1110                 return "REMOVING";
1111         case VNIC_REMOVED:
1112                 return "REMOVED";
1113         case VNIC_DOWN:
1114                 return "DOWN";
1115         }
1116         return "UNKNOWN";
1117 }
1118
1119 static int ibmvnic_login(struct net_device *netdev)
1120 {
1121         struct ibmvnic_adapter *adapter = netdev_priv(netdev);
1122         unsigned long timeout = msecs_to_jiffies(20000);
1123         int retry_count = 0;
1124         int retries = 10;
1125         bool retry;
1126         int rc;
1127
1128         do {
1129                 retry = false;
1130                 if (retry_count > retries) {
1131                         netdev_warn(netdev, "Login attempts exceeded\n");
1132                         return -1;
1133                 }
1134
1135                 adapter->init_done_rc = 0;
1136                 reinit_completion(&adapter->init_done);
1137                 rc = send_login(adapter);
1138                 if (rc)
1139                         return rc;
1140
1141                 if (!wait_for_completion_timeout(&adapter->init_done,
1142                                                  timeout)) {
1143                         netdev_warn(netdev, "Login timed out, retrying...\n");
1144                         retry = true;
1145                         adapter->init_done_rc = 0;
1146                         retry_count++;
1147                         continue;
1148                 }
1149
1150                 if (adapter->init_done_rc == ABORTED) {
1151                         netdev_warn(netdev, "Login aborted, retrying...\n");
1152                         retry = true;
1153                         adapter->init_done_rc = 0;
1154                         retry_count++;
1155                         /* FW or device may be busy, so
1156                          * wait a bit before retrying login
1157                          */
1158                         msleep(500);
1159                 } else if (adapter->init_done_rc == PARTIALSUCCESS) {
1160                         retry_count++;
1161                         release_sub_crqs(adapter, 1);
1162
1163                         retry = true;
1164                         netdev_dbg(netdev,
1165                                    "Received partial success, retrying...\n");
1166                         adapter->init_done_rc = 0;
1167                         reinit_completion(&adapter->init_done);
1168                         send_query_cap(adapter);
1169                         if (!wait_for_completion_timeout(&adapter->init_done,
1170                                                          timeout)) {
1171                                 netdev_warn(netdev,
1172                                             "Capabilities query timed out\n");
1173                                 return -1;
1174                         }
1175
1176                         rc = init_sub_crqs(adapter);
1177                         if (rc) {
1178                                 netdev_warn(netdev,
1179                                             "SCRQ initialization failed\n");
1180                                 return -1;
1181                         }
1182
1183                         rc = init_sub_crq_irqs(adapter);
1184                         if (rc) {
1185                                 netdev_warn(netdev,
1186                                             "SCRQ irq initialization failed\n");
1187                                 return -1;
1188                         }
1189                 } else if (adapter->init_done_rc) {
1190                         netdev_warn(netdev, "Adapter login failed\n");
1191                         return -1;
1192                 }
1193         } while (retry);
1194
1195         __ibmvnic_set_mac(netdev, adapter->mac_addr);
1196
1197         netdev_dbg(netdev, "[S:%s] Login succeeded\n", adapter_state_to_string(adapter->state));
1198         return 0;
1199 }
1200
1201 static void release_login_buffer(struct ibmvnic_adapter *adapter)
1202 {
1203         kfree(adapter->login_buf);
1204         adapter->login_buf = NULL;
1205 }
1206
1207 static void release_login_rsp_buffer(struct ibmvnic_adapter *adapter)
1208 {
1209         kfree(adapter->login_rsp_buf);
1210         adapter->login_rsp_buf = NULL;
1211 }
1212
1213 static void release_resources(struct ibmvnic_adapter *adapter)
1214 {
1215         release_vpd_data(adapter);
1216
1217         release_napi(adapter);
1218         release_login_buffer(adapter);
1219         release_login_rsp_buffer(adapter);
1220 }
1221
1222 static int set_link_state(struct ibmvnic_adapter *adapter, u8 link_state)
1223 {
1224         struct net_device *netdev = adapter->netdev;
1225         unsigned long timeout = msecs_to_jiffies(20000);
1226         union ibmvnic_crq crq;
1227         bool resend;
1228         int rc;
1229
1230         netdev_dbg(netdev, "setting link state %d\n", link_state);
1231
1232         memset(&crq, 0, sizeof(crq));
1233         crq.logical_link_state.first = IBMVNIC_CRQ_CMD;
1234         crq.logical_link_state.cmd = LOGICAL_LINK_STATE;
1235         crq.logical_link_state.link_state = link_state;
1236
1237         do {
1238                 resend = false;
1239
1240                 reinit_completion(&adapter->init_done);
1241                 rc = ibmvnic_send_crq(adapter, &crq);
1242                 if (rc) {
1243                         netdev_err(netdev, "Failed to set link state\n");
1244                         return rc;
1245                 }
1246
1247                 if (!wait_for_completion_timeout(&adapter->init_done,
1248                                                  timeout)) {
1249                         netdev_err(netdev, "timeout setting link state\n");
1250                         return -1;
1251                 }
1252
1253                 if (adapter->init_done_rc == PARTIALSUCCESS) {
1254                         /* Partuial success, delay and re-send */
1255                         mdelay(1000);
1256                         resend = true;
1257                 } else if (adapter->init_done_rc) {
1258                         netdev_warn(netdev, "Unable to set link state, rc=%d\n",
1259                                     adapter->init_done_rc);
1260                         return adapter->init_done_rc;
1261                 }
1262         } while (resend);
1263
1264         return 0;
1265 }
1266
1267 static int set_real_num_queues(struct net_device *netdev)
1268 {
1269         struct ibmvnic_adapter *adapter = netdev_priv(netdev);
1270         int rc;
1271
1272         netdev_dbg(netdev, "Setting real tx/rx queues (%llx/%llx)\n",
1273                    adapter->req_tx_queues, adapter->req_rx_queues);
1274
1275         rc = netif_set_real_num_tx_queues(netdev, adapter->req_tx_queues);
1276         if (rc) {
1277                 netdev_err(netdev, "failed to set the number of tx queues\n");
1278                 return rc;
1279         }
1280
1281         rc = netif_set_real_num_rx_queues(netdev, adapter->req_rx_queues);
1282         if (rc)
1283                 netdev_err(netdev, "failed to set the number of rx queues\n");
1284
1285         return rc;
1286 }
1287
1288 static int ibmvnic_get_vpd(struct ibmvnic_adapter *adapter)
1289 {
1290         struct device *dev = &adapter->vdev->dev;
1291         union ibmvnic_crq crq;
1292         int len = 0;
1293         int rc;
1294
1295         if (adapter->vpd->buff)
1296                 len = adapter->vpd->len;
1297
1298         mutex_lock(&adapter->fw_lock);
1299         adapter->fw_done_rc = 0;
1300         reinit_completion(&adapter->fw_done);
1301
1302         crq.get_vpd_size.first = IBMVNIC_CRQ_CMD;
1303         crq.get_vpd_size.cmd = GET_VPD_SIZE;
1304         rc = ibmvnic_send_crq(adapter, &crq);
1305         if (rc) {
1306                 mutex_unlock(&adapter->fw_lock);
1307                 return rc;
1308         }
1309
1310         rc = ibmvnic_wait_for_completion(adapter, &adapter->fw_done, 10000);
1311         if (rc) {
1312                 dev_err(dev, "Could not retrieve VPD size, rc = %d\n", rc);
1313                 mutex_unlock(&adapter->fw_lock);
1314                 return rc;
1315         }
1316         mutex_unlock(&adapter->fw_lock);
1317
1318         if (!adapter->vpd->len)
1319                 return -ENODATA;
1320
1321         if (!adapter->vpd->buff)
1322                 adapter->vpd->buff = kzalloc(adapter->vpd->len, GFP_KERNEL);
1323         else if (adapter->vpd->len != len)
1324                 adapter->vpd->buff =
1325                         krealloc(adapter->vpd->buff,
1326                                  adapter->vpd->len, GFP_KERNEL);
1327
1328         if (!adapter->vpd->buff) {
1329                 dev_err(dev, "Could allocate VPD buffer\n");
1330                 return -ENOMEM;
1331         }
1332
1333         adapter->vpd->dma_addr =
1334                 dma_map_single(dev, adapter->vpd->buff, adapter->vpd->len,
1335                                DMA_FROM_DEVICE);
1336         if (dma_mapping_error(dev, adapter->vpd->dma_addr)) {
1337                 dev_err(dev, "Could not map VPD buffer\n");
1338                 kfree(adapter->vpd->buff);
1339                 adapter->vpd->buff = NULL;
1340                 return -ENOMEM;
1341         }
1342
1343         mutex_lock(&adapter->fw_lock);
1344         adapter->fw_done_rc = 0;
1345         reinit_completion(&adapter->fw_done);
1346
1347         crq.get_vpd.first = IBMVNIC_CRQ_CMD;
1348         crq.get_vpd.cmd = GET_VPD;
1349         crq.get_vpd.ioba = cpu_to_be32(adapter->vpd->dma_addr);
1350         crq.get_vpd.len = cpu_to_be32((u32)adapter->vpd->len);
1351         rc = ibmvnic_send_crq(adapter, &crq);
1352         if (rc) {
1353                 kfree(adapter->vpd->buff);
1354                 adapter->vpd->buff = NULL;
1355                 mutex_unlock(&adapter->fw_lock);
1356                 return rc;
1357         }
1358
1359         rc = ibmvnic_wait_for_completion(adapter, &adapter->fw_done, 10000);
1360         if (rc) {
1361                 dev_err(dev, "Unable to retrieve VPD, rc = %d\n", rc);
1362                 kfree(adapter->vpd->buff);
1363                 adapter->vpd->buff = NULL;
1364                 mutex_unlock(&adapter->fw_lock);
1365                 return rc;
1366         }
1367
1368         mutex_unlock(&adapter->fw_lock);
1369         return 0;
1370 }
1371
1372 static int init_resources(struct ibmvnic_adapter *adapter)
1373 {
1374         struct net_device *netdev = adapter->netdev;
1375         int rc;
1376
1377         rc = set_real_num_queues(netdev);
1378         if (rc)
1379                 return rc;
1380
1381         adapter->vpd = kzalloc(sizeof(*adapter->vpd), GFP_KERNEL);
1382         if (!adapter->vpd)
1383                 return -ENOMEM;
1384
1385         /* Vital Product Data (VPD) */
1386         rc = ibmvnic_get_vpd(adapter);
1387         if (rc) {
1388                 netdev_err(netdev, "failed to initialize Vital Product Data (VPD)\n");
1389                 return rc;
1390         }
1391
1392         rc = init_napi(adapter);
1393         if (rc)
1394                 return rc;
1395
1396         send_query_map(adapter);
1397
1398         rc = init_rx_pools(netdev);
1399         if (rc)
1400                 return rc;
1401
1402         rc = init_tx_pools(netdev);
1403         return rc;
1404 }
1405
1406 static int __ibmvnic_open(struct net_device *netdev)
1407 {
1408         struct ibmvnic_adapter *adapter = netdev_priv(netdev);
1409         enum vnic_state prev_state = adapter->state;
1410         int i, rc;
1411
1412         adapter->state = VNIC_OPENING;
1413         replenish_pools(adapter);
1414         ibmvnic_napi_enable(adapter);
1415
1416         /* We're ready to receive frames, enable the sub-crq interrupts and
1417          * set the logical link state to up
1418          */
1419         for (i = 0; i < adapter->req_rx_queues; i++) {
1420                 netdev_dbg(netdev, "Enabling rx_scrq[%d] irq\n", i);
1421                 if (prev_state == VNIC_CLOSED)
1422                         enable_irq(adapter->rx_scrq[i]->irq);
1423                 enable_scrq_irq(adapter, adapter->rx_scrq[i]);
1424         }
1425
1426         for (i = 0; i < adapter->req_tx_queues; i++) {
1427                 netdev_dbg(netdev, "Enabling tx_scrq[%d] irq\n", i);
1428                 if (prev_state == VNIC_CLOSED)
1429                         enable_irq(adapter->tx_scrq[i]->irq);
1430                 enable_scrq_irq(adapter, adapter->tx_scrq[i]);
1431                 netdev_tx_reset_queue(netdev_get_tx_queue(netdev, i));
1432         }
1433
1434         rc = set_link_state(adapter, IBMVNIC_LOGICAL_LNK_UP);
1435         if (rc) {
1436                 ibmvnic_napi_disable(adapter);
1437                 release_resources(adapter);
1438                 return rc;
1439         }
1440
1441         netif_tx_start_all_queues(netdev);
1442
1443         if (prev_state == VNIC_CLOSED) {
1444                 for (i = 0; i < adapter->req_rx_queues; i++)
1445                         napi_schedule(&adapter->napi[i]);
1446         }
1447
1448         adapter->state = VNIC_OPEN;
1449         return rc;
1450 }
1451
1452 static int ibmvnic_open(struct net_device *netdev)
1453 {
1454         struct ibmvnic_adapter *adapter = netdev_priv(netdev);
1455         int rc;
1456
1457         ASSERT_RTNL();
1458
1459         /* If device failover is pending or we are about to reset, just set
1460          * device state and return. Device operation will be handled by reset
1461          * routine.
1462          *
1463          * It should be safe to overwrite the adapter->state here. Since
1464          * we hold the rtnl, either the reset has not actually started or
1465          * the rtnl got dropped during the set_link_state() in do_reset().
1466          * In the former case, no one else is changing the state (again we
1467          * have the rtnl) and in the latter case, do_reset() will detect and
1468          * honor our setting below.
1469          */
1470         if (adapter->failover_pending || (test_bit(0, &adapter->resetting))) {
1471                 netdev_dbg(netdev, "[S:%s FOP:%d] Resetting, deferring open\n",
1472                            adapter_state_to_string(adapter->state),
1473                            adapter->failover_pending);
1474                 adapter->state = VNIC_OPEN;
1475                 rc = 0;
1476                 goto out;
1477         }
1478
1479         if (adapter->state != VNIC_CLOSED) {
1480                 rc = ibmvnic_login(netdev);
1481                 if (rc)
1482                         goto out;
1483
1484                 rc = init_resources(adapter);
1485                 if (rc) {
1486                         netdev_err(netdev, "failed to initialize resources\n");
1487                         release_resources(adapter);
1488                         release_rx_pools(adapter);
1489                         release_tx_pools(adapter);
1490                         goto out;
1491                 }
1492         }
1493
1494         rc = __ibmvnic_open(netdev);
1495
1496 out:
1497         /* If open failed and there is a pending failover or in-progress reset,
1498          * set device state and return. Device operation will be handled by
1499          * reset routine. See also comments above regarding rtnl.
1500          */
1501         if (rc &&
1502             (adapter->failover_pending || (test_bit(0, &adapter->resetting)))) {
1503                 adapter->state = VNIC_OPEN;
1504                 rc = 0;
1505         }
1506         return rc;
1507 }
1508
1509 static void clean_rx_pools(struct ibmvnic_adapter *adapter)
1510 {
1511         struct ibmvnic_rx_pool *rx_pool;
1512         struct ibmvnic_rx_buff *rx_buff;
1513         u64 rx_entries;
1514         int rx_scrqs;
1515         int i, j;
1516
1517         if (!adapter->rx_pool)
1518                 return;
1519
1520         rx_scrqs = adapter->num_active_rx_pools;
1521         rx_entries = adapter->req_rx_add_entries_per_subcrq;
1522
1523         /* Free any remaining skbs in the rx buffer pools */
1524         for (i = 0; i < rx_scrqs; i++) {
1525                 rx_pool = &adapter->rx_pool[i];
1526                 if (!rx_pool || !rx_pool->rx_buff)
1527                         continue;
1528
1529                 netdev_dbg(adapter->netdev, "Cleaning rx_pool[%d]\n", i);
1530                 for (j = 0; j < rx_entries; j++) {
1531                         rx_buff = &rx_pool->rx_buff[j];
1532                         if (rx_buff && rx_buff->skb) {
1533                                 dev_kfree_skb_any(rx_buff->skb);
1534                                 rx_buff->skb = NULL;
1535                         }
1536                 }
1537         }
1538 }
1539
1540 static void clean_one_tx_pool(struct ibmvnic_adapter *adapter,
1541                               struct ibmvnic_tx_pool *tx_pool)
1542 {
1543         struct ibmvnic_tx_buff *tx_buff;
1544         u64 tx_entries;
1545         int i;
1546
1547         if (!tx_pool || !tx_pool->tx_buff)
1548                 return;
1549
1550         tx_entries = tx_pool->num_buffers;
1551
1552         for (i = 0; i < tx_entries; i++) {
1553                 tx_buff = &tx_pool->tx_buff[i];
1554                 if (tx_buff && tx_buff->skb) {
1555                         dev_kfree_skb_any(tx_buff->skb);
1556                         tx_buff->skb = NULL;
1557                 }
1558         }
1559 }
1560
1561 static void clean_tx_pools(struct ibmvnic_adapter *adapter)
1562 {
1563         int tx_scrqs;
1564         int i;
1565
1566         if (!adapter->tx_pool || !adapter->tso_pool)
1567                 return;
1568
1569         tx_scrqs = adapter->num_active_tx_pools;
1570
1571         /* Free any remaining skbs in the tx buffer pools */
1572         for (i = 0; i < tx_scrqs; i++) {
1573                 netdev_dbg(adapter->netdev, "Cleaning tx_pool[%d]\n", i);
1574                 clean_one_tx_pool(adapter, &adapter->tx_pool[i]);
1575                 clean_one_tx_pool(adapter, &adapter->tso_pool[i]);
1576         }
1577 }
1578
1579 static void ibmvnic_disable_irqs(struct ibmvnic_adapter *adapter)
1580 {
1581         struct net_device *netdev = adapter->netdev;
1582         int i;
1583
1584         if (adapter->tx_scrq) {
1585                 for (i = 0; i < adapter->req_tx_queues; i++)
1586                         if (adapter->tx_scrq[i]->irq) {
1587                                 netdev_dbg(netdev,
1588                                            "Disabling tx_scrq[%d] irq\n", i);
1589                                 disable_scrq_irq(adapter, adapter->tx_scrq[i]);
1590                                 disable_irq(adapter->tx_scrq[i]->irq);
1591                         }
1592         }
1593
1594         if (adapter->rx_scrq) {
1595                 for (i = 0; i < adapter->req_rx_queues; i++) {
1596                         if (adapter->rx_scrq[i]->irq) {
1597                                 netdev_dbg(netdev,
1598                                            "Disabling rx_scrq[%d] irq\n", i);
1599                                 disable_scrq_irq(adapter, adapter->rx_scrq[i]);
1600                                 disable_irq(adapter->rx_scrq[i]->irq);
1601                         }
1602                 }
1603         }
1604 }
1605
1606 static void ibmvnic_cleanup(struct net_device *netdev)
1607 {
1608         struct ibmvnic_adapter *adapter = netdev_priv(netdev);
1609
1610         /* ensure that transmissions are stopped if called by do_reset */
1611         if (test_bit(0, &adapter->resetting))
1612                 netif_tx_disable(netdev);
1613         else
1614                 netif_tx_stop_all_queues(netdev);
1615
1616         ibmvnic_napi_disable(adapter);
1617         ibmvnic_disable_irqs(adapter);
1618 }
1619
1620 static int __ibmvnic_close(struct net_device *netdev)
1621 {
1622         struct ibmvnic_adapter *adapter = netdev_priv(netdev);
1623         int rc = 0;
1624
1625         adapter->state = VNIC_CLOSING;
1626         rc = set_link_state(adapter, IBMVNIC_LOGICAL_LNK_DN);
1627         adapter->state = VNIC_CLOSED;
1628         return rc;
1629 }
1630
1631 static int ibmvnic_close(struct net_device *netdev)
1632 {
1633         struct ibmvnic_adapter *adapter = netdev_priv(netdev);
1634         int rc;
1635
1636         netdev_dbg(netdev, "[S:%s FOP:%d FRR:%d] Closing\n",
1637                    adapter_state_to_string(adapter->state),
1638                    adapter->failover_pending,
1639                    adapter->force_reset_recovery);
1640
1641         /* If device failover is pending, just set device state and return.
1642          * Device operation will be handled by reset routine.
1643          */
1644         if (adapter->failover_pending) {
1645                 adapter->state = VNIC_CLOSED;
1646                 return 0;
1647         }
1648
1649         rc = __ibmvnic_close(netdev);
1650         ibmvnic_cleanup(netdev);
1651         clean_rx_pools(adapter);
1652         clean_tx_pools(adapter);
1653
1654         return rc;
1655 }
1656
1657 /**
1658  * build_hdr_data - creates L2/L3/L4 header data buffer
1659  * @hdr_field: bitfield determining needed headers
1660  * @skb: socket buffer
1661  * @hdr_len: array of header lengths
1662  * @hdr_data: buffer to write the header to
1663  *
1664  * Reads hdr_field to determine which headers are needed by firmware.
1665  * Builds a buffer containing these headers.  Saves individual header
1666  * lengths and total buffer length to be used to build descriptors.
1667  */
1668 static int build_hdr_data(u8 hdr_field, struct sk_buff *skb,
1669                           int *hdr_len, u8 *hdr_data)
1670 {
1671         int len = 0;
1672         u8 *hdr;
1673
1674         if (skb_vlan_tagged(skb) && !skb_vlan_tag_present(skb))
1675                 hdr_len[0] = sizeof(struct vlan_ethhdr);
1676         else
1677                 hdr_len[0] = sizeof(struct ethhdr);
1678
1679         if (skb->protocol == htons(ETH_P_IP)) {
1680                 hdr_len[1] = ip_hdr(skb)->ihl * 4;
1681                 if (ip_hdr(skb)->protocol == IPPROTO_TCP)
1682                         hdr_len[2] = tcp_hdrlen(skb);
1683                 else if (ip_hdr(skb)->protocol == IPPROTO_UDP)
1684                         hdr_len[2] = sizeof(struct udphdr);
1685         } else if (skb->protocol == htons(ETH_P_IPV6)) {
1686                 hdr_len[1] = sizeof(struct ipv6hdr);
1687                 if (ipv6_hdr(skb)->nexthdr == IPPROTO_TCP)
1688                         hdr_len[2] = tcp_hdrlen(skb);
1689                 else if (ipv6_hdr(skb)->nexthdr == IPPROTO_UDP)
1690                         hdr_len[2] = sizeof(struct udphdr);
1691         } else if (skb->protocol == htons(ETH_P_ARP)) {
1692                 hdr_len[1] = arp_hdr_len(skb->dev);
1693                 hdr_len[2] = 0;
1694         }
1695
1696         memset(hdr_data, 0, 120);
1697         if ((hdr_field >> 6) & 1) {
1698                 hdr = skb_mac_header(skb);
1699                 memcpy(hdr_data, hdr, hdr_len[0]);
1700                 len += hdr_len[0];
1701         }
1702
1703         if ((hdr_field >> 5) & 1) {
1704                 hdr = skb_network_header(skb);
1705                 memcpy(hdr_data + len, hdr, hdr_len[1]);
1706                 len += hdr_len[1];
1707         }
1708
1709         if ((hdr_field >> 4) & 1) {
1710                 hdr = skb_transport_header(skb);
1711                 memcpy(hdr_data + len, hdr, hdr_len[2]);
1712                 len += hdr_len[2];
1713         }
1714         return len;
1715 }
1716
1717 /**
1718  * create_hdr_descs - create header and header extension descriptors
1719  * @hdr_field: bitfield determining needed headers
1720  * @hdr_data: buffer containing header data
1721  * @len: length of data buffer
1722  * @hdr_len: array of individual header lengths
1723  * @scrq_arr: descriptor array
1724  *
1725  * Creates header and, if needed, header extension descriptors and
1726  * places them in a descriptor array, scrq_arr
1727  */
1728
1729 static int create_hdr_descs(u8 hdr_field, u8 *hdr_data, int len, int *hdr_len,
1730                             union sub_crq *scrq_arr)
1731 {
1732         union sub_crq hdr_desc;
1733         int tmp_len = len;
1734         int num_descs = 0;
1735         u8 *data, *cur;
1736         int tmp;
1737
1738         while (tmp_len > 0) {
1739                 cur = hdr_data + len - tmp_len;
1740
1741                 memset(&hdr_desc, 0, sizeof(hdr_desc));
1742                 if (cur != hdr_data) {
1743                         data = hdr_desc.hdr_ext.data;
1744                         tmp = tmp_len > 29 ? 29 : tmp_len;
1745                         hdr_desc.hdr_ext.first = IBMVNIC_CRQ_CMD;
1746                         hdr_desc.hdr_ext.type = IBMVNIC_HDR_EXT_DESC;
1747                         hdr_desc.hdr_ext.len = tmp;
1748                 } else {
1749                         data = hdr_desc.hdr.data;
1750                         tmp = tmp_len > 24 ? 24 : tmp_len;
1751                         hdr_desc.hdr.first = IBMVNIC_CRQ_CMD;
1752                         hdr_desc.hdr.type = IBMVNIC_HDR_DESC;
1753                         hdr_desc.hdr.len = tmp;
1754                         hdr_desc.hdr.l2_len = (u8)hdr_len[0];
1755                         hdr_desc.hdr.l3_len = cpu_to_be16((u16)hdr_len[1]);
1756                         hdr_desc.hdr.l4_len = (u8)hdr_len[2];
1757                         hdr_desc.hdr.flag = hdr_field << 1;
1758                 }
1759                 memcpy(data, cur, tmp);
1760                 tmp_len -= tmp;
1761                 *scrq_arr = hdr_desc;
1762                 scrq_arr++;
1763                 num_descs++;
1764         }
1765
1766         return num_descs;
1767 }
1768
1769 /**
1770  * build_hdr_descs_arr - build a header descriptor array
1771  * @skb: tx socket buffer
1772  * @indir_arr: indirect array
1773  * @num_entries: number of descriptors to be sent
1774  * @hdr_field: bit field determining which headers will be sent
1775  *
1776  * This function will build a TX descriptor array with applicable
1777  * L2/L3/L4 packet header descriptors to be sent by send_subcrq_indirect.
1778  */
1779
1780 static void build_hdr_descs_arr(struct sk_buff *skb,
1781                                 union sub_crq *indir_arr,
1782                                 int *num_entries, u8 hdr_field)
1783 {
1784         int hdr_len[3] = {0, 0, 0};
1785         u8 hdr_data[140] = {0};
1786         int tot_len;
1787
1788         tot_len = build_hdr_data(hdr_field, skb, hdr_len,
1789                                  hdr_data);
1790         *num_entries += create_hdr_descs(hdr_field, hdr_data, tot_len, hdr_len,
1791                                          indir_arr + 1);
1792 }
1793
1794 static int ibmvnic_xmit_workarounds(struct sk_buff *skb,
1795                                     struct net_device *netdev)
1796 {
1797         /* For some backing devices, mishandling of small packets
1798          * can result in a loss of connection or TX stall. Device
1799          * architects recommend that no packet should be smaller
1800          * than the minimum MTU value provided to the driver, so
1801          * pad any packets to that length
1802          */
1803         if (skb->len < netdev->min_mtu)
1804                 return skb_put_padto(skb, netdev->min_mtu);
1805
1806         return 0;
1807 }
1808
1809 static void ibmvnic_tx_scrq_clean_buffer(struct ibmvnic_adapter *adapter,
1810                                          struct ibmvnic_sub_crq_queue *tx_scrq)
1811 {
1812         struct ibmvnic_ind_xmit_queue *ind_bufp;
1813         struct ibmvnic_tx_buff *tx_buff;
1814         struct ibmvnic_tx_pool *tx_pool;
1815         union sub_crq tx_scrq_entry;
1816         int queue_num;
1817         int entries;
1818         int index;
1819         int i;
1820
1821         ind_bufp = &tx_scrq->ind_buf;
1822         entries = (u64)ind_bufp->index;
1823         queue_num = tx_scrq->pool_index;
1824
1825         for (i = entries - 1; i >= 0; --i) {
1826                 tx_scrq_entry = ind_bufp->indir_arr[i];
1827                 if (tx_scrq_entry.v1.type != IBMVNIC_TX_DESC)
1828                         continue;
1829                 index = be32_to_cpu(tx_scrq_entry.v1.correlator);
1830                 if (index & IBMVNIC_TSO_POOL_MASK) {
1831                         tx_pool = &adapter->tso_pool[queue_num];
1832                         index &= ~IBMVNIC_TSO_POOL_MASK;
1833                 } else {
1834                         tx_pool = &adapter->tx_pool[queue_num];
1835                 }
1836                 tx_pool->free_map[tx_pool->consumer_index] = index;
1837                 tx_pool->consumer_index = tx_pool->consumer_index == 0 ?
1838                                           tx_pool->num_buffers - 1 :
1839                                           tx_pool->consumer_index - 1;
1840                 tx_buff = &tx_pool->tx_buff[index];
1841                 adapter->netdev->stats.tx_packets--;
1842                 adapter->netdev->stats.tx_bytes -= tx_buff->skb->len;
1843                 adapter->tx_stats_buffers[queue_num].packets--;
1844                 adapter->tx_stats_buffers[queue_num].bytes -=
1845                                                 tx_buff->skb->len;
1846                 dev_kfree_skb_any(tx_buff->skb);
1847                 tx_buff->skb = NULL;
1848                 adapter->netdev->stats.tx_dropped++;
1849         }
1850         ind_bufp->index = 0;
1851         if (atomic_sub_return(entries, &tx_scrq->used) <=
1852             (adapter->req_tx_entries_per_subcrq / 2) &&
1853             __netif_subqueue_stopped(adapter->netdev, queue_num) &&
1854             !test_bit(0, &adapter->resetting)) {
1855                 netif_wake_subqueue(adapter->netdev, queue_num);
1856                 netdev_dbg(adapter->netdev, "Started queue %d\n",
1857                            queue_num);
1858         }
1859 }
1860
1861 static int ibmvnic_tx_scrq_flush(struct ibmvnic_adapter *adapter,
1862                                  struct ibmvnic_sub_crq_queue *tx_scrq)
1863 {
1864         struct ibmvnic_ind_xmit_queue *ind_bufp;
1865         u64 dma_addr;
1866         u64 entries;
1867         u64 handle;
1868         int rc;
1869
1870         ind_bufp = &tx_scrq->ind_buf;
1871         dma_addr = (u64)ind_bufp->indir_dma;
1872         entries = (u64)ind_bufp->index;
1873         handle = tx_scrq->handle;
1874
1875         if (!entries)
1876                 return 0;
1877         rc = send_subcrq_indirect(adapter, handle, dma_addr, entries);
1878         if (rc)
1879                 ibmvnic_tx_scrq_clean_buffer(adapter, tx_scrq);
1880         else
1881                 ind_bufp->index = 0;
1882         return 0;
1883 }
1884
1885 static netdev_tx_t ibmvnic_xmit(struct sk_buff *skb, struct net_device *netdev)
1886 {
1887         struct ibmvnic_adapter *adapter = netdev_priv(netdev);
1888         int queue_num = skb_get_queue_mapping(skb);
1889         u8 *hdrs = (u8 *)&adapter->tx_rx_desc_req;
1890         struct device *dev = &adapter->vdev->dev;
1891         struct ibmvnic_ind_xmit_queue *ind_bufp;
1892         struct ibmvnic_tx_buff *tx_buff = NULL;
1893         struct ibmvnic_sub_crq_queue *tx_scrq;
1894         struct ibmvnic_tx_pool *tx_pool;
1895         unsigned int tx_send_failed = 0;
1896         netdev_tx_t ret = NETDEV_TX_OK;
1897         unsigned int tx_map_failed = 0;
1898         union sub_crq indir_arr[16];
1899         unsigned int tx_dropped = 0;
1900         unsigned int tx_packets = 0;
1901         unsigned int tx_bytes = 0;
1902         dma_addr_t data_dma_addr;
1903         struct netdev_queue *txq;
1904         unsigned long lpar_rc;
1905         union sub_crq tx_crq;
1906         unsigned int offset;
1907         int num_entries = 1;
1908         unsigned char *dst;
1909         int index = 0;
1910         u8 proto = 0;
1911
1912         tx_scrq = adapter->tx_scrq[queue_num];
1913         txq = netdev_get_tx_queue(netdev, queue_num);
1914         ind_bufp = &tx_scrq->ind_buf;
1915
1916         if (test_bit(0, &adapter->resetting)) {
1917                 dev_kfree_skb_any(skb);
1918
1919                 tx_send_failed++;
1920                 tx_dropped++;
1921                 ret = NETDEV_TX_OK;
1922                 goto out;
1923         }
1924
1925         if (ibmvnic_xmit_workarounds(skb, netdev)) {
1926                 tx_dropped++;
1927                 tx_send_failed++;
1928                 ret = NETDEV_TX_OK;
1929                 ibmvnic_tx_scrq_flush(adapter, tx_scrq);
1930                 goto out;
1931         }
1932         if (skb_is_gso(skb))
1933                 tx_pool = &adapter->tso_pool[queue_num];
1934         else
1935                 tx_pool = &adapter->tx_pool[queue_num];
1936
1937         index = tx_pool->free_map[tx_pool->consumer_index];
1938
1939         if (index == IBMVNIC_INVALID_MAP) {
1940                 dev_kfree_skb_any(skb);
1941                 tx_send_failed++;
1942                 tx_dropped++;
1943                 ibmvnic_tx_scrq_flush(adapter, tx_scrq);
1944                 ret = NETDEV_TX_OK;
1945                 goto out;
1946         }
1947
1948         tx_pool->free_map[tx_pool->consumer_index] = IBMVNIC_INVALID_MAP;
1949
1950         offset = index * tx_pool->buf_size;
1951         dst = tx_pool->long_term_buff.buff + offset;
1952         memset(dst, 0, tx_pool->buf_size);
1953         data_dma_addr = tx_pool->long_term_buff.addr + offset;
1954
1955         if (skb_shinfo(skb)->nr_frags) {
1956                 int cur, i;
1957
1958                 /* Copy the head */
1959                 skb_copy_from_linear_data(skb, dst, skb_headlen(skb));
1960                 cur = skb_headlen(skb);
1961
1962                 /* Copy the frags */
1963                 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
1964                         const skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
1965
1966                         memcpy(dst + cur, skb_frag_address(frag),
1967                                skb_frag_size(frag));
1968                         cur += skb_frag_size(frag);
1969                 }
1970         } else {
1971                 skb_copy_from_linear_data(skb, dst, skb->len);
1972         }
1973
1974         /* post changes to long_term_buff *dst before VIOS accessing it */
1975         dma_wmb();
1976
1977         tx_pool->consumer_index =
1978             (tx_pool->consumer_index + 1) % tx_pool->num_buffers;
1979
1980         tx_buff = &tx_pool->tx_buff[index];
1981         tx_buff->skb = skb;
1982         tx_buff->index = index;
1983         tx_buff->pool_index = queue_num;
1984
1985         memset(&tx_crq, 0, sizeof(tx_crq));
1986         tx_crq.v1.first = IBMVNIC_CRQ_CMD;
1987         tx_crq.v1.type = IBMVNIC_TX_DESC;
1988         tx_crq.v1.n_crq_elem = 1;
1989         tx_crq.v1.n_sge = 1;
1990         tx_crq.v1.flags1 = IBMVNIC_TX_COMP_NEEDED;
1991
1992         if (skb_is_gso(skb))
1993                 tx_crq.v1.correlator =
1994                         cpu_to_be32(index | IBMVNIC_TSO_POOL_MASK);
1995         else
1996                 tx_crq.v1.correlator = cpu_to_be32(index);
1997         tx_crq.v1.dma_reg = cpu_to_be16(tx_pool->long_term_buff.map_id);
1998         tx_crq.v1.sge_len = cpu_to_be32(skb->len);
1999         tx_crq.v1.ioba = cpu_to_be64(data_dma_addr);
2000
2001         if (adapter->vlan_header_insertion && skb_vlan_tag_present(skb)) {
2002                 tx_crq.v1.flags2 |= IBMVNIC_TX_VLAN_INSERT;
2003                 tx_crq.v1.vlan_id = cpu_to_be16(skb->vlan_tci);
2004         }
2005
2006         if (skb->protocol == htons(ETH_P_IP)) {
2007                 tx_crq.v1.flags1 |= IBMVNIC_TX_PROT_IPV4;
2008                 proto = ip_hdr(skb)->protocol;
2009         } else if (skb->protocol == htons(ETH_P_IPV6)) {
2010                 tx_crq.v1.flags1 |= IBMVNIC_TX_PROT_IPV6;
2011                 proto = ipv6_hdr(skb)->nexthdr;
2012         }
2013
2014         if (proto == IPPROTO_TCP)
2015                 tx_crq.v1.flags1 |= IBMVNIC_TX_PROT_TCP;
2016         else if (proto == IPPROTO_UDP)
2017                 tx_crq.v1.flags1 |= IBMVNIC_TX_PROT_UDP;
2018
2019         if (skb->ip_summed == CHECKSUM_PARTIAL) {
2020                 tx_crq.v1.flags1 |= IBMVNIC_TX_CHKSUM_OFFLOAD;
2021                 hdrs += 2;
2022         }
2023         if (skb_is_gso(skb)) {
2024                 tx_crq.v1.flags1 |= IBMVNIC_TX_LSO;
2025                 tx_crq.v1.mss = cpu_to_be16(skb_shinfo(skb)->gso_size);
2026                 hdrs += 2;
2027         }
2028
2029         if ((*hdrs >> 7) & 1)
2030                 build_hdr_descs_arr(skb, indir_arr, &num_entries, *hdrs);
2031
2032         tx_crq.v1.n_crq_elem = num_entries;
2033         tx_buff->num_entries = num_entries;
2034         /* flush buffer if current entry can not fit */
2035         if (num_entries + ind_bufp->index > IBMVNIC_MAX_IND_DESCS) {
2036                 lpar_rc = ibmvnic_tx_scrq_flush(adapter, tx_scrq);
2037                 if (lpar_rc != H_SUCCESS)
2038                         goto tx_flush_err;
2039         }
2040
2041         indir_arr[0] = tx_crq;
2042         memcpy(&ind_bufp->indir_arr[ind_bufp->index], &indir_arr[0],
2043                num_entries * sizeof(struct ibmvnic_generic_scrq));
2044         ind_bufp->index += num_entries;
2045         if (__netdev_tx_sent_queue(txq, skb->len,
2046                                    netdev_xmit_more() &&
2047                                    ind_bufp->index < IBMVNIC_MAX_IND_DESCS)) {
2048                 lpar_rc = ibmvnic_tx_scrq_flush(adapter, tx_scrq);
2049                 if (lpar_rc != H_SUCCESS)
2050                         goto tx_err;
2051         }
2052
2053         if (atomic_add_return(num_entries, &tx_scrq->used)
2054                                         >= adapter->req_tx_entries_per_subcrq) {
2055                 netdev_dbg(netdev, "Stopping queue %d\n", queue_num);
2056                 netif_stop_subqueue(netdev, queue_num);
2057         }
2058
2059         tx_packets++;
2060         tx_bytes += skb->len;
2061         txq->trans_start = jiffies;
2062         ret = NETDEV_TX_OK;
2063         goto out;
2064
2065 tx_flush_err:
2066         dev_kfree_skb_any(skb);
2067         tx_buff->skb = NULL;
2068         tx_pool->consumer_index = tx_pool->consumer_index == 0 ?
2069                                   tx_pool->num_buffers - 1 :
2070                                   tx_pool->consumer_index - 1;
2071         tx_dropped++;
2072 tx_err:
2073         if (lpar_rc != H_CLOSED && lpar_rc != H_PARAMETER)
2074                 dev_err_ratelimited(dev, "tx: send failed\n");
2075
2076         if (lpar_rc == H_CLOSED || adapter->failover_pending) {
2077                 /* Disable TX and report carrier off if queue is closed
2078                  * or pending failover.
2079                  * Firmware guarantees that a signal will be sent to the
2080                  * driver, triggering a reset or some other action.
2081                  */
2082                 netif_tx_stop_all_queues(netdev);
2083                 netif_carrier_off(netdev);
2084         }
2085 out:
2086         netdev->stats.tx_dropped += tx_dropped;
2087         netdev->stats.tx_bytes += tx_bytes;
2088         netdev->stats.tx_packets += tx_packets;
2089         adapter->tx_send_failed += tx_send_failed;
2090         adapter->tx_map_failed += tx_map_failed;
2091         adapter->tx_stats_buffers[queue_num].packets += tx_packets;
2092         adapter->tx_stats_buffers[queue_num].bytes += tx_bytes;
2093         adapter->tx_stats_buffers[queue_num].dropped_packets += tx_dropped;
2094
2095         return ret;
2096 }
2097
2098 static void ibmvnic_set_multi(struct net_device *netdev)
2099 {
2100         struct ibmvnic_adapter *adapter = netdev_priv(netdev);
2101         struct netdev_hw_addr *ha;
2102         union ibmvnic_crq crq;
2103
2104         memset(&crq, 0, sizeof(crq));
2105         crq.request_capability.first = IBMVNIC_CRQ_CMD;
2106         crq.request_capability.cmd = REQUEST_CAPABILITY;
2107
2108         if (netdev->flags & IFF_PROMISC) {
2109                 if (!adapter->promisc_supported)
2110                         return;
2111         } else {
2112                 if (netdev->flags & IFF_ALLMULTI) {
2113                         /* Accept all multicast */
2114                         memset(&crq, 0, sizeof(crq));
2115                         crq.multicast_ctrl.first = IBMVNIC_CRQ_CMD;
2116                         crq.multicast_ctrl.cmd = MULTICAST_CTRL;
2117                         crq.multicast_ctrl.flags = IBMVNIC_ENABLE_ALL;
2118                         ibmvnic_send_crq(adapter, &crq);
2119                 } else if (netdev_mc_empty(netdev)) {
2120                         /* Reject all multicast */
2121                         memset(&crq, 0, sizeof(crq));
2122                         crq.multicast_ctrl.first = IBMVNIC_CRQ_CMD;
2123                         crq.multicast_ctrl.cmd = MULTICAST_CTRL;
2124                         crq.multicast_ctrl.flags = IBMVNIC_DISABLE_ALL;
2125                         ibmvnic_send_crq(adapter, &crq);
2126                 } else {
2127                         /* Accept one or more multicast(s) */
2128                         netdev_for_each_mc_addr(ha, netdev) {
2129                                 memset(&crq, 0, sizeof(crq));
2130                                 crq.multicast_ctrl.first = IBMVNIC_CRQ_CMD;
2131                                 crq.multicast_ctrl.cmd = MULTICAST_CTRL;
2132                                 crq.multicast_ctrl.flags = IBMVNIC_ENABLE_MC;
2133                                 ether_addr_copy(&crq.multicast_ctrl.mac_addr[0],
2134                                                 ha->addr);
2135                                 ibmvnic_send_crq(adapter, &crq);
2136                         }
2137                 }
2138         }
2139 }
2140
2141 static int __ibmvnic_set_mac(struct net_device *netdev, u8 *dev_addr)
2142 {
2143         struct ibmvnic_adapter *adapter = netdev_priv(netdev);
2144         union ibmvnic_crq crq;
2145         int rc;
2146
2147         if (!is_valid_ether_addr(dev_addr)) {
2148                 rc = -EADDRNOTAVAIL;
2149                 goto err;
2150         }
2151
2152         memset(&crq, 0, sizeof(crq));
2153         crq.change_mac_addr.first = IBMVNIC_CRQ_CMD;
2154         crq.change_mac_addr.cmd = CHANGE_MAC_ADDR;
2155         ether_addr_copy(&crq.change_mac_addr.mac_addr[0], dev_addr);
2156
2157         mutex_lock(&adapter->fw_lock);
2158         adapter->fw_done_rc = 0;
2159         reinit_completion(&adapter->fw_done);
2160
2161         rc = ibmvnic_send_crq(adapter, &crq);
2162         if (rc) {
2163                 rc = -EIO;
2164                 mutex_unlock(&adapter->fw_lock);
2165                 goto err;
2166         }
2167
2168         rc = ibmvnic_wait_for_completion(adapter, &adapter->fw_done, 10000);
2169         /* netdev->dev_addr is changed in handle_change_mac_rsp function */
2170         if (rc || adapter->fw_done_rc) {
2171                 rc = -EIO;
2172                 mutex_unlock(&adapter->fw_lock);
2173                 goto err;
2174         }
2175         mutex_unlock(&adapter->fw_lock);
2176         return 0;
2177 err:
2178         ether_addr_copy(adapter->mac_addr, netdev->dev_addr);
2179         return rc;
2180 }
2181
2182 static int ibmvnic_set_mac(struct net_device *netdev, void *p)
2183 {
2184         struct ibmvnic_adapter *adapter = netdev_priv(netdev);
2185         struct sockaddr *addr = p;
2186         int rc;
2187
2188         rc = 0;
2189         if (!is_valid_ether_addr(addr->sa_data))
2190                 return -EADDRNOTAVAIL;
2191
2192         ether_addr_copy(adapter->mac_addr, addr->sa_data);
2193         if (adapter->state != VNIC_PROBED)
2194                 rc = __ibmvnic_set_mac(netdev, addr->sa_data);
2195
2196         return rc;
2197 }
2198
2199 static const char *reset_reason_to_string(enum ibmvnic_reset_reason reason)
2200 {
2201         switch (reason) {
2202         case VNIC_RESET_FAILOVER:
2203                 return "FAILOVER";
2204         case VNIC_RESET_MOBILITY:
2205                 return "MOBILITY";
2206         case VNIC_RESET_FATAL:
2207                 return "FATAL";
2208         case VNIC_RESET_NON_FATAL:
2209                 return "NON_FATAL";
2210         case VNIC_RESET_TIMEOUT:
2211                 return "TIMEOUT";
2212         case VNIC_RESET_CHANGE_PARAM:
2213                 return "CHANGE_PARAM";
2214         case VNIC_RESET_PASSIVE_INIT:
2215                 return "PASSIVE_INIT";
2216         }
2217         return "UNKNOWN";
2218 }
2219
2220 /*
2221  * do_reset returns zero if we are able to keep processing reset events, or
2222  * non-zero if we hit a fatal error and must halt.
2223  */
2224 static int do_reset(struct ibmvnic_adapter *adapter,
2225                     struct ibmvnic_rwi *rwi, u32 reset_state)
2226 {
2227         struct net_device *netdev = adapter->netdev;
2228         u64 old_num_rx_queues, old_num_tx_queues;
2229         u64 old_num_rx_slots, old_num_tx_slots;
2230         int rc;
2231
2232         netdev_dbg(adapter->netdev,
2233                    "[S:%s FOP:%d] Reset reason: %s, reset_state: %s\n",
2234                    adapter_state_to_string(adapter->state),
2235                    adapter->failover_pending,
2236                    reset_reason_to_string(rwi->reset_reason),
2237                    adapter_state_to_string(reset_state));
2238
2239         adapter->reset_reason = rwi->reset_reason;
2240         /* requestor of VNIC_RESET_CHANGE_PARAM already has the rtnl lock */
2241         if (!(adapter->reset_reason == VNIC_RESET_CHANGE_PARAM))
2242                 rtnl_lock();
2243
2244         /* Now that we have the rtnl lock, clear any pending failover.
2245          * This will ensure ibmvnic_open() has either completed or will
2246          * block until failover is complete.
2247          */
2248         if (rwi->reset_reason == VNIC_RESET_FAILOVER)
2249                 adapter->failover_pending = false;
2250
2251         /* read the state and check (again) after getting rtnl */
2252         reset_state = adapter->state;
2253
2254         if (reset_state == VNIC_REMOVING || reset_state == VNIC_REMOVED) {
2255                 rc = -EBUSY;
2256                 goto out;
2257         }
2258
2259         netif_carrier_off(netdev);
2260
2261         old_num_rx_queues = adapter->req_rx_queues;
2262         old_num_tx_queues = adapter->req_tx_queues;
2263         old_num_rx_slots = adapter->req_rx_add_entries_per_subcrq;
2264         old_num_tx_slots = adapter->req_tx_entries_per_subcrq;
2265
2266         ibmvnic_cleanup(netdev);
2267
2268         if (reset_state == VNIC_OPEN &&
2269             adapter->reset_reason != VNIC_RESET_MOBILITY &&
2270             adapter->reset_reason != VNIC_RESET_FAILOVER) {
2271                 if (adapter->reset_reason == VNIC_RESET_CHANGE_PARAM) {
2272                         rc = __ibmvnic_close(netdev);
2273                         if (rc)
2274                                 goto out;
2275                 } else {
2276                         adapter->state = VNIC_CLOSING;
2277
2278                         /* Release the RTNL lock before link state change and
2279                          * re-acquire after the link state change to allow
2280                          * linkwatch_event to grab the RTNL lock and run during
2281                          * a reset.
2282                          */
2283                         rtnl_unlock();
2284                         rc = set_link_state(adapter, IBMVNIC_LOGICAL_LNK_DN);
2285                         rtnl_lock();
2286                         if (rc)
2287                                 goto out;
2288
2289                         if (adapter->state == VNIC_OPEN) {
2290                                 /* When we dropped rtnl, ibmvnic_open() got
2291                                  * it and noticed that we are resetting and
2292                                  * set the adapter state to OPEN. Update our
2293                                  * new "target" state, and resume the reset
2294                                  * from VNIC_CLOSING state.
2295                                  */
2296                                 netdev_dbg(netdev,
2297                                            "Open changed state from %s, updating.\n",
2298                                            adapter_state_to_string(reset_state));
2299                                 reset_state = VNIC_OPEN;
2300                                 adapter->state = VNIC_CLOSING;
2301                         }
2302
2303                         if (adapter->state != VNIC_CLOSING) {
2304                                 /* If someone else changed the adapter state
2305                                  * when we dropped the rtnl, fail the reset
2306                                  */
2307                                 rc = -1;
2308                                 goto out;
2309                         }
2310                         adapter->state = VNIC_CLOSED;
2311                 }
2312         }
2313
2314         if (adapter->reset_reason == VNIC_RESET_CHANGE_PARAM) {
2315                 release_resources(adapter);
2316                 release_sub_crqs(adapter, 1);
2317                 release_crq_queue(adapter);
2318         }
2319
2320         if (adapter->reset_reason != VNIC_RESET_NON_FATAL) {
2321                 /* remove the closed state so when we call open it appears
2322                  * we are coming from the probed state.
2323                  */
2324                 adapter->state = VNIC_PROBED;
2325
2326                 if (adapter->reset_reason == VNIC_RESET_CHANGE_PARAM) {
2327                         rc = init_crq_queue(adapter);
2328                 } else if (adapter->reset_reason == VNIC_RESET_MOBILITY) {
2329                         rc = ibmvnic_reenable_crq_queue(adapter);
2330                         release_sub_crqs(adapter, 1);
2331                 } else {
2332                         rc = ibmvnic_reset_crq(adapter);
2333                         if (rc == H_CLOSED || rc == H_SUCCESS) {
2334                                 rc = vio_enable_interrupts(adapter->vdev);
2335                                 if (rc)
2336                                         netdev_err(adapter->netdev,
2337                                                    "Reset failed to enable interrupts. rc=%d\n",
2338                                                    rc);
2339                         }
2340                 }
2341
2342                 if (rc) {
2343                         netdev_err(adapter->netdev,
2344                                    "Reset couldn't initialize crq. rc=%d\n", rc);
2345                         goto out;
2346                 }
2347
2348                 rc = ibmvnic_reset_init(adapter, true);
2349                 if (rc) {
2350                         rc = IBMVNIC_INIT_FAILED;
2351                         goto out;
2352                 }
2353
2354                 /* If the adapter was in PROBE or DOWN state prior to the reset,
2355                  * exit here.
2356                  */
2357                 if (reset_state == VNIC_PROBED || reset_state == VNIC_DOWN) {
2358                         rc = 0;
2359                         goto out;
2360                 }
2361
2362                 rc = ibmvnic_login(netdev);
2363                 if (rc)
2364                         goto out;
2365
2366                 if (adapter->reset_reason == VNIC_RESET_CHANGE_PARAM) {
2367                         rc = init_resources(adapter);
2368                         if (rc)
2369                                 goto out;
2370                 } else if (adapter->req_rx_queues != old_num_rx_queues ||
2371                     adapter->req_tx_queues != old_num_tx_queues ||
2372                     adapter->req_rx_add_entries_per_subcrq !=
2373                     old_num_rx_slots ||
2374                     adapter->req_tx_entries_per_subcrq !=
2375                     old_num_tx_slots ||
2376                     !adapter->rx_pool ||
2377                     !adapter->tso_pool ||
2378                     !adapter->tx_pool) {
2379                         release_napi(adapter);
2380                         release_vpd_data(adapter);
2381
2382                         rc = init_resources(adapter);
2383                         if (rc)
2384                                 goto out;
2385
2386                 } else {
2387                         rc = init_tx_pools(netdev);
2388                         if (rc) {
2389                                 netdev_dbg(netdev,
2390                                            "init tx pools failed (%d)\n",
2391                                            rc);
2392                                 goto out;
2393                         }
2394
2395                         rc = init_rx_pools(netdev);
2396                         if (rc) {
2397                                 netdev_dbg(netdev,
2398                                            "init rx pools failed (%d)\n",
2399                                            rc);
2400                                 goto out;
2401                         }
2402                 }
2403                 ibmvnic_disable_irqs(adapter);
2404         }
2405         adapter->state = VNIC_CLOSED;
2406
2407         if (reset_state == VNIC_CLOSED) {
2408                 rc = 0;
2409                 goto out;
2410         }
2411
2412         rc = __ibmvnic_open(netdev);
2413         if (rc) {
2414                 rc = IBMVNIC_OPEN_FAILED;
2415                 goto out;
2416         }
2417
2418         /* refresh device's multicast list */
2419         ibmvnic_set_multi(netdev);
2420
2421         if (adapter->reset_reason == VNIC_RESET_FAILOVER ||
2422             adapter->reset_reason == VNIC_RESET_MOBILITY)
2423                 __netdev_notify_peers(netdev);
2424
2425         rc = 0;
2426
2427 out:
2428         /* restore the adapter state if reset failed */
2429         if (rc)
2430                 adapter->state = reset_state;
2431         /* requestor of VNIC_RESET_CHANGE_PARAM should still hold the rtnl lock */
2432         if (!(adapter->reset_reason == VNIC_RESET_CHANGE_PARAM))
2433                 rtnl_unlock();
2434
2435         netdev_dbg(adapter->netdev, "[S:%s FOP:%d] Reset done, rc %d\n",
2436                    adapter_state_to_string(adapter->state),
2437                    adapter->failover_pending, rc);
2438         return rc;
2439 }
2440
2441 static int do_hard_reset(struct ibmvnic_adapter *adapter,
2442                          struct ibmvnic_rwi *rwi, u32 reset_state)
2443 {
2444         struct net_device *netdev = adapter->netdev;
2445         int rc;
2446
2447         netdev_dbg(adapter->netdev, "Hard resetting driver (%s)\n",
2448                    reset_reason_to_string(rwi->reset_reason));
2449
2450         /* read the state and check (again) after getting rtnl */
2451         reset_state = adapter->state;
2452
2453         if (reset_state == VNIC_REMOVING || reset_state == VNIC_REMOVED) {
2454                 rc = -EBUSY;
2455                 goto out;
2456         }
2457
2458         netif_carrier_off(netdev);
2459         adapter->reset_reason = rwi->reset_reason;
2460
2461         ibmvnic_cleanup(netdev);
2462         release_resources(adapter);
2463         release_sub_crqs(adapter, 0);
2464         release_crq_queue(adapter);
2465
2466         /* remove the closed state so when we call open it appears
2467          * we are coming from the probed state.
2468          */
2469         adapter->state = VNIC_PROBED;
2470
2471         reinit_completion(&adapter->init_done);
2472         rc = init_crq_queue(adapter);
2473         if (rc) {
2474                 netdev_err(adapter->netdev,
2475                            "Couldn't initialize crq. rc=%d\n", rc);
2476                 goto out;
2477         }
2478
2479         rc = ibmvnic_reset_init(adapter, false);
2480         if (rc)
2481                 goto out;
2482
2483         /* If the adapter was in PROBE or DOWN state prior to the reset,
2484          * exit here.
2485          */
2486         if (reset_state == VNIC_PROBED || reset_state == VNIC_DOWN)
2487                 goto out;
2488
2489         rc = ibmvnic_login(netdev);
2490         if (rc)
2491                 goto out;
2492
2493         rc = init_resources(adapter);
2494         if (rc)
2495                 goto out;
2496
2497         ibmvnic_disable_irqs(adapter);
2498         adapter->state = VNIC_CLOSED;
2499
2500         if (reset_state == VNIC_CLOSED)
2501                 goto out;
2502
2503         rc = __ibmvnic_open(netdev);
2504         if (rc) {
2505                 rc = IBMVNIC_OPEN_FAILED;
2506                 goto out;
2507         }
2508
2509         __netdev_notify_peers(netdev);
2510 out:
2511         /* restore adapter state if reset failed */
2512         if (rc)
2513                 adapter->state = reset_state;
2514         netdev_dbg(adapter->netdev, "[S:%s FOP:%d] Hard reset done, rc %d\n",
2515                    adapter_state_to_string(adapter->state),
2516                    adapter->failover_pending, rc);
2517         return rc;
2518 }
2519
2520 static struct ibmvnic_rwi *get_next_rwi(struct ibmvnic_adapter *adapter)
2521 {
2522         struct ibmvnic_rwi *rwi;
2523         unsigned long flags;
2524
2525         spin_lock_irqsave(&adapter->rwi_lock, flags);
2526
2527         if (!list_empty(&adapter->rwi_list)) {
2528                 rwi = list_first_entry(&adapter->rwi_list, struct ibmvnic_rwi,
2529                                        list);
2530                 list_del(&rwi->list);
2531         } else {
2532                 rwi = NULL;
2533         }
2534
2535         spin_unlock_irqrestore(&adapter->rwi_lock, flags);
2536         return rwi;
2537 }
2538
2539 /**
2540  * do_passive_init - complete probing when partner device is detected.
2541  * @adapter: ibmvnic_adapter struct
2542  *
2543  * If the ibmvnic device does not have a partner device to communicate with at boot
2544  * and that partner device comes online at a later time, this function is called
2545  * to complete the initialization process of ibmvnic device.
2546  * Caller is expected to hold rtnl_lock().
2547  *
2548  * Returns non-zero if sub-CRQs are not initialized properly leaving the device
2549  * in the down state.
2550  * Returns 0 upon success and the device is in PROBED state.
2551  */
2552
2553 static int do_passive_init(struct ibmvnic_adapter *adapter)
2554 {
2555         unsigned long timeout = msecs_to_jiffies(30000);
2556         struct net_device *netdev = adapter->netdev;
2557         struct device *dev = &adapter->vdev->dev;
2558         int rc;
2559
2560         netdev_dbg(netdev, "Partner device found, probing.\n");
2561
2562         adapter->state = VNIC_PROBING;
2563         reinit_completion(&adapter->init_done);
2564         adapter->init_done_rc = 0;
2565         adapter->crq.active = true;
2566
2567         rc = send_crq_init_complete(adapter);
2568         if (rc)
2569                 goto out;
2570
2571         rc = send_version_xchg(adapter);
2572         if (rc)
2573                 netdev_dbg(adapter->netdev, "send_version_xchg failed, rc=%d\n", rc);
2574
2575         if (!wait_for_completion_timeout(&adapter->init_done, timeout)) {
2576                 dev_err(dev, "Initialization sequence timed out\n");
2577                 rc = -ETIMEDOUT;
2578                 goto out;
2579         }
2580
2581         rc = init_sub_crqs(adapter);
2582         if (rc) {
2583                 dev_err(dev, "Initialization of sub crqs failed, rc=%d\n", rc);
2584                 goto out;
2585         }
2586
2587         rc = init_sub_crq_irqs(adapter);
2588         if (rc) {
2589                 dev_err(dev, "Failed to initialize sub crq irqs\n, rc=%d", rc);
2590                 goto init_failed;
2591         }
2592
2593         netdev->mtu = adapter->req_mtu - ETH_HLEN;
2594         netdev->min_mtu = adapter->min_mtu - ETH_HLEN;
2595         netdev->max_mtu = adapter->max_mtu - ETH_HLEN;
2596
2597         adapter->state = VNIC_PROBED;
2598         netdev_dbg(netdev, "Probed successfully. Waiting for signal from partner device.\n");
2599
2600         return 0;
2601
2602 init_failed:
2603         release_sub_crqs(adapter, 1);
2604 out:
2605         adapter->state = VNIC_DOWN;
2606         return rc;
2607 }
2608
2609 static void __ibmvnic_reset(struct work_struct *work)
2610 {
2611         struct ibmvnic_adapter *adapter;
2612         bool saved_state = false;
2613         struct ibmvnic_rwi *tmprwi;
2614         struct ibmvnic_rwi *rwi;
2615         unsigned long flags;
2616         u32 reset_state;
2617         int rc = 0;
2618
2619         adapter = container_of(work, struct ibmvnic_adapter, ibmvnic_reset);
2620
2621         if (test_and_set_bit_lock(0, &adapter->resetting)) {
2622                 queue_delayed_work(system_long_wq,
2623                                    &adapter->ibmvnic_delayed_reset,
2624                                    IBMVNIC_RESET_DELAY);
2625                 return;
2626         }
2627
2628         rwi = get_next_rwi(adapter);
2629         while (rwi) {
2630                 spin_lock_irqsave(&adapter->state_lock, flags);
2631
2632                 if (adapter->state == VNIC_REMOVING ||
2633                     adapter->state == VNIC_REMOVED) {
2634                         spin_unlock_irqrestore(&adapter->state_lock, flags);
2635                         kfree(rwi);
2636                         rc = EBUSY;
2637                         break;
2638                 }
2639
2640                 if (!saved_state) {
2641                         reset_state = adapter->state;
2642                         saved_state = true;
2643                 }
2644                 spin_unlock_irqrestore(&adapter->state_lock, flags);
2645
2646                 if (rwi->reset_reason == VNIC_RESET_PASSIVE_INIT) {
2647                         rtnl_lock();
2648                         rc = do_passive_init(adapter);
2649                         rtnl_unlock();
2650                         if (!rc)
2651                                 netif_carrier_on(adapter->netdev);
2652                 } else if (adapter->force_reset_recovery) {
2653                         /* Since we are doing a hard reset now, clear the
2654                          * failover_pending flag so we don't ignore any
2655                          * future MOBILITY or other resets.
2656                          */
2657                         adapter->failover_pending = false;
2658
2659                         /* Transport event occurred during previous reset */
2660                         if (adapter->wait_for_reset) {
2661                                 /* Previous was CHANGE_PARAM; caller locked */
2662                                 adapter->force_reset_recovery = false;
2663                                 rc = do_hard_reset(adapter, rwi, reset_state);
2664                         } else {
2665                                 rtnl_lock();
2666                                 adapter->force_reset_recovery = false;
2667                                 rc = do_hard_reset(adapter, rwi, reset_state);
2668                                 rtnl_unlock();
2669                         }
2670                         if (rc) {
2671                                 /* give backing device time to settle down */
2672                                 netdev_dbg(adapter->netdev,
2673                                            "[S:%s] Hard reset failed, waiting 60 secs\n",
2674                                            adapter_state_to_string(adapter->state));
2675                                 set_current_state(TASK_UNINTERRUPTIBLE);
2676                                 schedule_timeout(60 * HZ);
2677                         }
2678                 } else {
2679                         rc = do_reset(adapter, rwi, reset_state);
2680                 }
2681                 tmprwi = rwi;
2682                 adapter->last_reset_time = jiffies;
2683
2684                 if (rc)
2685                         netdev_dbg(adapter->netdev, "Reset failed, rc=%d\n", rc);
2686
2687                 rwi = get_next_rwi(adapter);
2688
2689                 /*
2690                  * If there is another reset queued, free the previous rwi
2691                  * and process the new reset even if previous reset failed
2692                  * (the previous reset could have failed because of a fail
2693                  * over for instance, so process the fail over).
2694                  *
2695                  * If there are no resets queued and the previous reset failed,
2696                  * the adapter would be in an undefined state. So retry the
2697                  * previous reset as a hard reset.
2698                  */
2699                 if (rwi)
2700                         kfree(tmprwi);
2701                 else if (rc)
2702                         rwi = tmprwi;
2703
2704                 if (rwi && (rwi->reset_reason == VNIC_RESET_FAILOVER ||
2705                             rwi->reset_reason == VNIC_RESET_MOBILITY || rc))
2706                         adapter->force_reset_recovery = true;
2707         }
2708
2709         if (adapter->wait_for_reset) {
2710                 adapter->reset_done_rc = rc;
2711                 complete(&adapter->reset_done);
2712         }
2713
2714         clear_bit_unlock(0, &adapter->resetting);
2715
2716         netdev_dbg(adapter->netdev,
2717                    "[S:%s FRR:%d WFR:%d] Done processing resets\n",
2718                    adapter_state_to_string(adapter->state),
2719                    adapter->force_reset_recovery,
2720                    adapter->wait_for_reset);
2721 }
2722
2723 static void __ibmvnic_delayed_reset(struct work_struct *work)
2724 {
2725         struct ibmvnic_adapter *adapter;
2726
2727         adapter = container_of(work, struct ibmvnic_adapter,
2728                                ibmvnic_delayed_reset.work);
2729         __ibmvnic_reset(&adapter->ibmvnic_reset);
2730 }
2731
2732 static int ibmvnic_reset(struct ibmvnic_adapter *adapter,
2733                          enum ibmvnic_reset_reason reason)
2734 {
2735         struct list_head *entry, *tmp_entry;
2736         struct ibmvnic_rwi *rwi, *tmp;
2737         struct net_device *netdev = adapter->netdev;
2738         unsigned long flags;
2739         int ret;
2740
2741         spin_lock_irqsave(&adapter->rwi_lock, flags);
2742
2743         /* If failover is pending don't schedule any other reset.
2744          * Instead let the failover complete. If there is already a
2745          * a failover reset scheduled, we will detect and drop the
2746          * duplicate reset when walking the ->rwi_list below.
2747          */
2748         if (adapter->state == VNIC_REMOVING ||
2749             adapter->state == VNIC_REMOVED ||
2750             (adapter->failover_pending && reason != VNIC_RESET_FAILOVER)) {
2751                 ret = EBUSY;
2752                 netdev_dbg(netdev, "Adapter removing or pending failover, skipping reset\n");
2753                 goto err;
2754         }
2755
2756         if (adapter->state == VNIC_PROBING) {
2757                 netdev_warn(netdev, "Adapter reset during probe\n");
2758                 adapter->init_done_rc = -EAGAIN;
2759                 ret = EAGAIN;
2760                 goto err;
2761         }
2762
2763         list_for_each_entry(tmp, &adapter->rwi_list, list) {
2764                 if (tmp->reset_reason == reason) {
2765                         netdev_dbg(netdev, "Skipping matching reset, reason=%s\n",
2766                                    reset_reason_to_string(reason));
2767                         ret = EBUSY;
2768                         goto err;
2769                 }
2770         }
2771
2772         rwi = kzalloc(sizeof(*rwi), GFP_ATOMIC);
2773         if (!rwi) {
2774                 ret = ENOMEM;
2775                 goto err;
2776         }
2777         /* if we just received a transport event,
2778          * flush reset queue and process this reset
2779          */
2780         if (adapter->force_reset_recovery && !list_empty(&adapter->rwi_list)) {
2781                 list_for_each_safe(entry, tmp_entry, &adapter->rwi_list)
2782                         list_del(entry);
2783         }
2784         rwi->reset_reason = reason;
2785         list_add_tail(&rwi->list, &adapter->rwi_list);
2786         netdev_dbg(adapter->netdev, "Scheduling reset (reason %s)\n",
2787                    reset_reason_to_string(reason));
2788         queue_work(system_long_wq, &adapter->ibmvnic_reset);
2789
2790         ret = 0;
2791 err:
2792         /* ibmvnic_close() below can block, so drop the lock first */
2793         spin_unlock_irqrestore(&adapter->rwi_lock, flags);
2794
2795         if (ret == ENOMEM)
2796                 ibmvnic_close(netdev);
2797
2798         return -ret;
2799 }
2800
2801 static void ibmvnic_tx_timeout(struct net_device *dev, unsigned int txqueue)
2802 {
2803         struct ibmvnic_adapter *adapter = netdev_priv(dev);
2804
2805         if (test_bit(0, &adapter->resetting)) {
2806                 netdev_err(adapter->netdev,
2807                            "Adapter is resetting, skip timeout reset\n");
2808                 return;
2809         }
2810         /* No queuing up reset until at least 5 seconds (default watchdog val)
2811          * after last reset
2812          */
2813         if (time_before(jiffies, (adapter->last_reset_time + dev->watchdog_timeo))) {
2814                 netdev_dbg(dev, "Not yet time to tx timeout.\n");
2815                 return;
2816         }
2817         ibmvnic_reset(adapter, VNIC_RESET_TIMEOUT);
2818 }
2819
2820 static void remove_buff_from_pool(struct ibmvnic_adapter *adapter,
2821                                   struct ibmvnic_rx_buff *rx_buff)
2822 {
2823         struct ibmvnic_rx_pool *pool = &adapter->rx_pool[rx_buff->pool_index];
2824
2825         rx_buff->skb = NULL;
2826
2827         pool->free_map[pool->next_alloc] = (int)(rx_buff - pool->rx_buff);
2828         pool->next_alloc = (pool->next_alloc + 1) % pool->size;
2829
2830         atomic_dec(&pool->available);
2831 }
2832
2833 static int ibmvnic_poll(struct napi_struct *napi, int budget)
2834 {
2835         struct ibmvnic_sub_crq_queue *rx_scrq;
2836         struct ibmvnic_adapter *adapter;
2837         struct net_device *netdev;
2838         int frames_processed;
2839         int scrq_num;
2840
2841         netdev = napi->dev;
2842         adapter = netdev_priv(netdev);
2843         scrq_num = (int)(napi - adapter->napi);
2844         frames_processed = 0;
2845         rx_scrq = adapter->rx_scrq[scrq_num];
2846
2847 restart_poll:
2848         while (frames_processed < budget) {
2849                 struct sk_buff *skb;
2850                 struct ibmvnic_rx_buff *rx_buff;
2851                 union sub_crq *next;
2852                 u32 length;
2853                 u16 offset;
2854                 u8 flags = 0;
2855
2856                 if (unlikely(test_bit(0, &adapter->resetting) &&
2857                              adapter->reset_reason != VNIC_RESET_NON_FATAL)) {
2858                         enable_scrq_irq(adapter, rx_scrq);
2859                         napi_complete_done(napi, frames_processed);
2860                         return frames_processed;
2861                 }
2862
2863                 if (!pending_scrq(adapter, rx_scrq))
2864                         break;
2865                 next = ibmvnic_next_scrq(adapter, rx_scrq);
2866                 rx_buff = (struct ibmvnic_rx_buff *)
2867                           be64_to_cpu(next->rx_comp.correlator);
2868                 /* do error checking */
2869                 if (next->rx_comp.rc) {
2870                         netdev_dbg(netdev, "rx buffer returned with rc %x\n",
2871                                    be16_to_cpu(next->rx_comp.rc));
2872                         /* free the entry */
2873                         next->rx_comp.first = 0;
2874                         dev_kfree_skb_any(rx_buff->skb);
2875                         remove_buff_from_pool(adapter, rx_buff);
2876                         continue;
2877                 } else if (!rx_buff->skb) {
2878                         /* free the entry */
2879                         next->rx_comp.first = 0;
2880                         remove_buff_from_pool(adapter, rx_buff);
2881                         continue;
2882                 }
2883
2884                 length = be32_to_cpu(next->rx_comp.len);
2885                 offset = be16_to_cpu(next->rx_comp.off_frame_data);
2886                 flags = next->rx_comp.flags;
2887                 skb = rx_buff->skb;
2888                 /* load long_term_buff before copying to skb */
2889                 dma_rmb();
2890                 skb_copy_to_linear_data(skb, rx_buff->data + offset,
2891                                         length);
2892
2893                 /* VLAN Header has been stripped by the system firmware and
2894                  * needs to be inserted by the driver
2895                  */
2896                 if (adapter->rx_vlan_header_insertion &&
2897                     (flags & IBMVNIC_VLAN_STRIPPED))
2898                         __vlan_hwaccel_put_tag(skb, htons(ETH_P_8021Q),
2899                                                ntohs(next->rx_comp.vlan_tci));
2900
2901                 /* free the entry */
2902                 next->rx_comp.first = 0;
2903                 remove_buff_from_pool(adapter, rx_buff);
2904
2905                 skb_put(skb, length);
2906                 skb->protocol = eth_type_trans(skb, netdev);
2907                 skb_record_rx_queue(skb, scrq_num);
2908
2909                 if (flags & IBMVNIC_IP_CHKSUM_GOOD &&
2910                     flags & IBMVNIC_TCP_UDP_CHKSUM_GOOD) {
2911                         skb->ip_summed = CHECKSUM_UNNECESSARY;
2912                 }
2913
2914                 length = skb->len;
2915                 napi_gro_receive(napi, skb); /* send it up */
2916                 netdev->stats.rx_packets++;
2917                 netdev->stats.rx_bytes += length;
2918                 adapter->rx_stats_buffers[scrq_num].packets++;
2919                 adapter->rx_stats_buffers[scrq_num].bytes += length;
2920                 frames_processed++;
2921         }
2922
2923         if (adapter->state != VNIC_CLOSING &&
2924             ((atomic_read(&adapter->rx_pool[scrq_num].available) <
2925               adapter->req_rx_add_entries_per_subcrq / 2) ||
2926               frames_processed < budget))
2927                 replenish_rx_pool(adapter, &adapter->rx_pool[scrq_num]);
2928         if (frames_processed < budget) {
2929                 if (napi_complete_done(napi, frames_processed)) {
2930                         enable_scrq_irq(adapter, rx_scrq);
2931                         if (pending_scrq(adapter, rx_scrq)) {
2932                                 if (napi_reschedule(napi)) {
2933                                         disable_scrq_irq(adapter, rx_scrq);
2934                                         goto restart_poll;
2935                                 }
2936                         }
2937                 }
2938         }
2939         return frames_processed;
2940 }
2941
2942 static int wait_for_reset(struct ibmvnic_adapter *adapter)
2943 {
2944         int rc, ret;
2945
2946         adapter->fallback.mtu = adapter->req_mtu;
2947         adapter->fallback.rx_queues = adapter->req_rx_queues;
2948         adapter->fallback.tx_queues = adapter->req_tx_queues;
2949         adapter->fallback.rx_entries = adapter->req_rx_add_entries_per_subcrq;
2950         adapter->fallback.tx_entries = adapter->req_tx_entries_per_subcrq;
2951
2952         reinit_completion(&adapter->reset_done);
2953         adapter->wait_for_reset = true;
2954         rc = ibmvnic_reset(adapter, VNIC_RESET_CHANGE_PARAM);
2955
2956         if (rc) {
2957                 ret = rc;
2958                 goto out;
2959         }
2960         rc = ibmvnic_wait_for_completion(adapter, &adapter->reset_done, 60000);
2961         if (rc) {
2962                 ret = -ENODEV;
2963                 goto out;
2964         }
2965
2966         ret = 0;
2967         if (adapter->reset_done_rc) {
2968                 ret = -EIO;
2969                 adapter->desired.mtu = adapter->fallback.mtu;
2970                 adapter->desired.rx_queues = adapter->fallback.rx_queues;
2971                 adapter->desired.tx_queues = adapter->fallback.tx_queues;
2972                 adapter->desired.rx_entries = adapter->fallback.rx_entries;
2973                 adapter->desired.tx_entries = adapter->fallback.tx_entries;
2974
2975                 reinit_completion(&adapter->reset_done);
2976                 adapter->wait_for_reset = true;
2977                 rc = ibmvnic_reset(adapter, VNIC_RESET_CHANGE_PARAM);
2978                 if (rc) {
2979                         ret = rc;
2980                         goto out;
2981                 }
2982                 rc = ibmvnic_wait_for_completion(adapter, &adapter->reset_done,
2983                                                  60000);
2984                 if (rc) {
2985                         ret = -ENODEV;
2986                         goto out;
2987                 }
2988         }
2989 out:
2990         adapter->wait_for_reset = false;
2991
2992         return ret;
2993 }
2994
2995 static int ibmvnic_change_mtu(struct net_device *netdev, int new_mtu)
2996 {
2997         struct ibmvnic_adapter *adapter = netdev_priv(netdev);
2998
2999         adapter->desired.mtu = new_mtu + ETH_HLEN;
3000
3001         return wait_for_reset(adapter);
3002 }
3003
3004 static netdev_features_t ibmvnic_features_check(struct sk_buff *skb,
3005                                                 struct net_device *dev,
3006                                                 netdev_features_t features)
3007 {
3008         /* Some backing hardware adapters can not
3009          * handle packets with a MSS less than 224
3010          * or with only one segment.
3011          */
3012         if (skb_is_gso(skb)) {
3013                 if (skb_shinfo(skb)->gso_size < 224 ||
3014                     skb_shinfo(skb)->gso_segs == 1)
3015                         features &= ~NETIF_F_GSO_MASK;
3016         }
3017
3018         return features;
3019 }
3020
3021 static const struct net_device_ops ibmvnic_netdev_ops = {
3022         .ndo_open               = ibmvnic_open,
3023         .ndo_stop               = ibmvnic_close,
3024         .ndo_start_xmit         = ibmvnic_xmit,
3025         .ndo_set_rx_mode        = ibmvnic_set_multi,
3026         .ndo_set_mac_address    = ibmvnic_set_mac,
3027         .ndo_validate_addr      = eth_validate_addr,
3028         .ndo_tx_timeout         = ibmvnic_tx_timeout,
3029         .ndo_change_mtu         = ibmvnic_change_mtu,
3030         .ndo_features_check     = ibmvnic_features_check,
3031 };
3032
3033 /* ethtool functions */
3034
3035 static int ibmvnic_get_link_ksettings(struct net_device *netdev,
3036                                       struct ethtool_link_ksettings *cmd)
3037 {
3038         struct ibmvnic_adapter *adapter = netdev_priv(netdev);
3039         int rc;
3040
3041         rc = send_query_phys_parms(adapter);
3042         if (rc) {
3043                 adapter->speed = SPEED_UNKNOWN;
3044                 adapter->duplex = DUPLEX_UNKNOWN;
3045         }
3046         cmd->base.speed = adapter->speed;
3047         cmd->base.duplex = adapter->duplex;
3048         cmd->base.port = PORT_FIBRE;
3049         cmd->base.phy_address = 0;
3050         cmd->base.autoneg = AUTONEG_ENABLE;
3051
3052         return 0;
3053 }
3054
3055 static void ibmvnic_get_drvinfo(struct net_device *netdev,
3056                                 struct ethtool_drvinfo *info)
3057 {
3058         struct ibmvnic_adapter *adapter = netdev_priv(netdev);
3059
3060         strscpy(info->driver, ibmvnic_driver_name, sizeof(info->driver));
3061         strscpy(info->version, IBMVNIC_DRIVER_VERSION, sizeof(info->version));
3062         strscpy(info->fw_version, adapter->fw_version,
3063                 sizeof(info->fw_version));
3064 }
3065
3066 static u32 ibmvnic_get_msglevel(struct net_device *netdev)
3067 {
3068         struct ibmvnic_adapter *adapter = netdev_priv(netdev);
3069
3070         return adapter->msg_enable;
3071 }
3072
3073 static void ibmvnic_set_msglevel(struct net_device *netdev, u32 data)
3074 {
3075         struct ibmvnic_adapter *adapter = netdev_priv(netdev);
3076
3077         adapter->msg_enable = data;
3078 }
3079
3080 static u32 ibmvnic_get_link(struct net_device *netdev)
3081 {
3082         struct ibmvnic_adapter *adapter = netdev_priv(netdev);
3083
3084         /* Don't need to send a query because we request a logical link up at
3085          * init and then we wait for link state indications
3086          */
3087         return adapter->logical_link_state;
3088 }
3089
3090 static void ibmvnic_get_ringparam(struct net_device *netdev,
3091                                   struct ethtool_ringparam *ring)
3092 {
3093         struct ibmvnic_adapter *adapter = netdev_priv(netdev);
3094
3095         if (adapter->priv_flags & IBMVNIC_USE_SERVER_MAXES) {
3096                 ring->rx_max_pending = adapter->max_rx_add_entries_per_subcrq;
3097                 ring->tx_max_pending = adapter->max_tx_entries_per_subcrq;
3098         } else {
3099                 ring->rx_max_pending = IBMVNIC_MAX_QUEUE_SZ;
3100                 ring->tx_max_pending = IBMVNIC_MAX_QUEUE_SZ;
3101         }
3102         ring->rx_mini_max_pending = 0;
3103         ring->rx_jumbo_max_pending = 0;
3104         ring->rx_pending = adapter->req_rx_add_entries_per_subcrq;
3105         ring->tx_pending = adapter->req_tx_entries_per_subcrq;
3106         ring->rx_mini_pending = 0;
3107         ring->rx_jumbo_pending = 0;
3108 }
3109
3110 static int ibmvnic_set_ringparam(struct net_device *netdev,
3111                                  struct ethtool_ringparam *ring)
3112 {
3113         struct ibmvnic_adapter *adapter = netdev_priv(netdev);
3114         int ret;
3115
3116         ret = 0;
3117         adapter->desired.rx_entries = ring->rx_pending;
3118         adapter->desired.tx_entries = ring->tx_pending;
3119
3120         ret = wait_for_reset(adapter);
3121
3122         if (!ret &&
3123             (adapter->req_rx_add_entries_per_subcrq != ring->rx_pending ||
3124              adapter->req_tx_entries_per_subcrq != ring->tx_pending))
3125                 netdev_info(netdev,
3126                             "Could not match full ringsize request. Requested: RX %d, TX %d; Allowed: RX %llu, TX %llu\n",
3127                             ring->rx_pending, ring->tx_pending,
3128                             adapter->req_rx_add_entries_per_subcrq,
3129                             adapter->req_tx_entries_per_subcrq);
3130         return ret;
3131 }
3132
3133 static void ibmvnic_get_channels(struct net_device *netdev,
3134                                  struct ethtool_channels *channels)
3135 {
3136         struct ibmvnic_adapter *adapter = netdev_priv(netdev);
3137
3138         if (adapter->priv_flags & IBMVNIC_USE_SERVER_MAXES) {
3139                 channels->max_rx = adapter->max_rx_queues;
3140                 channels->max_tx = adapter->max_tx_queues;
3141         } else {
3142                 channels->max_rx = IBMVNIC_MAX_QUEUES;
3143                 channels->max_tx = IBMVNIC_MAX_QUEUES;
3144         }
3145
3146         channels->max_other = 0;
3147         channels->max_combined = 0;
3148         channels->rx_count = adapter->req_rx_queues;
3149         channels->tx_count = adapter->req_tx_queues;
3150         channels->other_count = 0;
3151         channels->combined_count = 0;
3152 }
3153
3154 static int ibmvnic_set_channels(struct net_device *netdev,
3155                                 struct ethtool_channels *channels)
3156 {
3157         struct ibmvnic_adapter *adapter = netdev_priv(netdev);
3158         int ret;
3159
3160         ret = 0;
3161         adapter->desired.rx_queues = channels->rx_count;
3162         adapter->desired.tx_queues = channels->tx_count;
3163
3164         ret = wait_for_reset(adapter);
3165
3166         if (!ret &&
3167             (adapter->req_rx_queues != channels->rx_count ||
3168              adapter->req_tx_queues != channels->tx_count))
3169                 netdev_info(netdev,
3170                             "Could not match full channels request. Requested: RX %d, TX %d; Allowed: RX %llu, TX %llu\n",
3171                             channels->rx_count, channels->tx_count,
3172                             adapter->req_rx_queues, adapter->req_tx_queues);
3173         return ret;
3174 }
3175
3176 static void ibmvnic_get_strings(struct net_device *dev, u32 stringset, u8 *data)
3177 {
3178         struct ibmvnic_adapter *adapter = netdev_priv(dev);
3179         int i;
3180
3181         switch (stringset) {
3182         case ETH_SS_STATS:
3183                 for (i = 0; i < ARRAY_SIZE(ibmvnic_stats);
3184                                 i++, data += ETH_GSTRING_LEN)
3185                         memcpy(data, ibmvnic_stats[i].name, ETH_GSTRING_LEN);
3186
3187                 for (i = 0; i < adapter->req_tx_queues; i++) {
3188                         snprintf(data, ETH_GSTRING_LEN, "tx%d_packets", i);
3189                         data += ETH_GSTRING_LEN;
3190
3191                         snprintf(data, ETH_GSTRING_LEN, "tx%d_bytes", i);
3192                         data += ETH_GSTRING_LEN;
3193
3194                         snprintf(data, ETH_GSTRING_LEN,
3195                                  "tx%d_dropped_packets", i);
3196                         data += ETH_GSTRING_LEN;
3197                 }
3198
3199                 for (i = 0; i < adapter->req_rx_queues; i++) {
3200                         snprintf(data, ETH_GSTRING_LEN, "rx%d_packets", i);
3201                         data += ETH_GSTRING_LEN;
3202
3203                         snprintf(data, ETH_GSTRING_LEN, "rx%d_bytes", i);
3204                         data += ETH_GSTRING_LEN;
3205
3206                         snprintf(data, ETH_GSTRING_LEN, "rx%d_interrupts", i);
3207                         data += ETH_GSTRING_LEN;
3208                 }
3209                 break;
3210
3211         case ETH_SS_PRIV_FLAGS:
3212                 for (i = 0; i < ARRAY_SIZE(ibmvnic_priv_flags); i++)
3213                         strcpy(data + i * ETH_GSTRING_LEN,
3214                                ibmvnic_priv_flags[i]);
3215                 break;
3216         default:
3217                 return;
3218         }
3219 }
3220
3221 static int ibmvnic_get_sset_count(struct net_device *dev, int sset)
3222 {
3223         struct ibmvnic_adapter *adapter = netdev_priv(dev);
3224
3225         switch (sset) {
3226         case ETH_SS_STATS:
3227                 return ARRAY_SIZE(ibmvnic_stats) +
3228                        adapter->req_tx_queues * NUM_TX_STATS +
3229                        adapter->req_rx_queues * NUM_RX_STATS;
3230         case ETH_SS_PRIV_FLAGS:
3231                 return ARRAY_SIZE(ibmvnic_priv_flags);
3232         default:
3233                 return -EOPNOTSUPP;
3234         }
3235 }
3236
3237 static void ibmvnic_get_ethtool_stats(struct net_device *dev,
3238                                       struct ethtool_stats *stats, u64 *data)
3239 {
3240         struct ibmvnic_adapter *adapter = netdev_priv(dev);
3241         union ibmvnic_crq crq;
3242         int i, j;
3243         int rc;
3244
3245         memset(&crq, 0, sizeof(crq));
3246         crq.request_statistics.first = IBMVNIC_CRQ_CMD;
3247         crq.request_statistics.cmd = REQUEST_STATISTICS;
3248         crq.request_statistics.ioba = cpu_to_be32(adapter->stats_token);
3249         crq.request_statistics.len =
3250             cpu_to_be32(sizeof(struct ibmvnic_statistics));
3251
3252         /* Wait for data to be written */
3253         reinit_completion(&adapter->stats_done);
3254         rc = ibmvnic_send_crq(adapter, &crq);
3255         if (rc)
3256                 return;
3257         rc = ibmvnic_wait_for_completion(adapter, &adapter->stats_done, 10000);
3258         if (rc)
3259                 return;
3260
3261         for (i = 0; i < ARRAY_SIZE(ibmvnic_stats); i++)
3262                 data[i] = be64_to_cpu(IBMVNIC_GET_STAT
3263                                       (adapter, ibmvnic_stats[i].offset));
3264
3265         for (j = 0; j < adapter->req_tx_queues; j++) {
3266                 data[i] = adapter->tx_stats_buffers[j].packets;
3267                 i++;
3268                 data[i] = adapter->tx_stats_buffers[j].bytes;
3269                 i++;
3270                 data[i] = adapter->tx_stats_buffers[j].dropped_packets;
3271                 i++;
3272         }
3273
3274         for (j = 0; j < adapter->req_rx_queues; j++) {
3275                 data[i] = adapter->rx_stats_buffers[j].packets;
3276                 i++;
3277                 data[i] = adapter->rx_stats_buffers[j].bytes;
3278                 i++;
3279                 data[i] = adapter->rx_stats_buffers[j].interrupts;
3280                 i++;
3281         }
3282 }
3283
3284 static u32 ibmvnic_get_priv_flags(struct net_device *netdev)
3285 {
3286         struct ibmvnic_adapter *adapter = netdev_priv(netdev);
3287
3288         return adapter->priv_flags;
3289 }
3290
3291 static int ibmvnic_set_priv_flags(struct net_device *netdev, u32 flags)
3292 {
3293         struct ibmvnic_adapter *adapter = netdev_priv(netdev);
3294         bool which_maxes = !!(flags & IBMVNIC_USE_SERVER_MAXES);
3295
3296         if (which_maxes)
3297                 adapter->priv_flags |= IBMVNIC_USE_SERVER_MAXES;
3298         else
3299                 adapter->priv_flags &= ~IBMVNIC_USE_SERVER_MAXES;
3300
3301         return 0;
3302 }
3303
3304 static const struct ethtool_ops ibmvnic_ethtool_ops = {
3305         .get_drvinfo            = ibmvnic_get_drvinfo,
3306         .get_msglevel           = ibmvnic_get_msglevel,
3307         .set_msglevel           = ibmvnic_set_msglevel,
3308         .get_link               = ibmvnic_get_link,
3309         .get_ringparam          = ibmvnic_get_ringparam,
3310         .set_ringparam          = ibmvnic_set_ringparam,
3311         .get_channels           = ibmvnic_get_channels,
3312         .set_channels           = ibmvnic_set_channels,
3313         .get_strings            = ibmvnic_get_strings,
3314         .get_sset_count         = ibmvnic_get_sset_count,
3315         .get_ethtool_stats      = ibmvnic_get_ethtool_stats,
3316         .get_link_ksettings     = ibmvnic_get_link_ksettings,
3317         .get_priv_flags         = ibmvnic_get_priv_flags,
3318         .set_priv_flags         = ibmvnic_set_priv_flags,
3319 };
3320
3321 /* Routines for managing CRQs/sCRQs  */
3322
3323 static int reset_one_sub_crq_queue(struct ibmvnic_adapter *adapter,
3324                                    struct ibmvnic_sub_crq_queue *scrq)
3325 {
3326         int rc;
3327
3328         if (!scrq) {
3329                 netdev_dbg(adapter->netdev, "Invalid scrq reset.\n");
3330                 return -EINVAL;
3331         }
3332
3333         if (scrq->irq) {
3334                 free_irq(scrq->irq, scrq);
3335                 irq_dispose_mapping(scrq->irq);
3336                 scrq->irq = 0;
3337         }
3338
3339         if (scrq->msgs) {
3340                 memset(scrq->msgs, 0, 4 * PAGE_SIZE);
3341                 atomic_set(&scrq->used, 0);
3342                 scrq->cur = 0;
3343                 scrq->ind_buf.index = 0;
3344         } else {
3345                 netdev_dbg(adapter->netdev, "Invalid scrq reset\n");
3346                 return -EINVAL;
3347         }
3348
3349         rc = h_reg_sub_crq(adapter->vdev->unit_address, scrq->msg_token,
3350                            4 * PAGE_SIZE, &scrq->crq_num, &scrq->hw_irq);
3351         return rc;
3352 }
3353
3354 static int reset_sub_crq_queues(struct ibmvnic_adapter *adapter)
3355 {
3356         int i, rc;
3357
3358         if (!adapter->tx_scrq || !adapter->rx_scrq)
3359                 return -EINVAL;
3360
3361         for (i = 0; i < adapter->req_tx_queues; i++) {
3362                 netdev_dbg(adapter->netdev, "Re-setting tx_scrq[%d]\n", i);
3363                 rc = reset_one_sub_crq_queue(adapter, adapter->tx_scrq[i]);
3364                 if (rc)
3365                         return rc;
3366         }
3367
3368         for (i = 0; i < adapter->req_rx_queues; i++) {
3369                 netdev_dbg(adapter->netdev, "Re-setting rx_scrq[%d]\n", i);
3370                 rc = reset_one_sub_crq_queue(adapter, adapter->rx_scrq[i]);
3371                 if (rc)
3372                         return rc;
3373         }
3374
3375         return rc;
3376 }
3377
3378 static void release_sub_crq_queue(struct ibmvnic_adapter *adapter,
3379                                   struct ibmvnic_sub_crq_queue *scrq,
3380                                   bool do_h_free)
3381 {
3382         struct device *dev = &adapter->vdev->dev;
3383         long rc;
3384
3385         netdev_dbg(adapter->netdev, "Releasing sub-CRQ\n");
3386
3387         if (do_h_free) {
3388                 /* Close the sub-crqs */
3389                 do {
3390                         rc = plpar_hcall_norets(H_FREE_SUB_CRQ,
3391                                                 adapter->vdev->unit_address,
3392                                                 scrq->crq_num);
3393                 } while (rc == H_BUSY || H_IS_LONG_BUSY(rc));
3394
3395                 if (rc) {
3396                         netdev_err(adapter->netdev,
3397                                    "Failed to release sub-CRQ %16lx, rc = %ld\n",
3398                                    scrq->crq_num, rc);
3399                 }
3400         }
3401
3402         dma_free_coherent(dev,
3403                           IBMVNIC_IND_ARR_SZ,
3404                           scrq->ind_buf.indir_arr,
3405                           scrq->ind_buf.indir_dma);
3406
3407         dma_unmap_single(dev, scrq->msg_token, 4 * PAGE_SIZE,
3408                          DMA_BIDIRECTIONAL);
3409         free_pages((unsigned long)scrq->msgs, 2);
3410         kfree(scrq);
3411 }
3412
3413 static struct ibmvnic_sub_crq_queue *init_sub_crq_queue(struct ibmvnic_adapter
3414                                                         *adapter)
3415 {
3416         struct device *dev = &adapter->vdev->dev;
3417         struct ibmvnic_sub_crq_queue *scrq;
3418         int rc;
3419
3420         scrq = kzalloc(sizeof(*scrq), GFP_KERNEL);
3421         if (!scrq)
3422                 return NULL;
3423
3424         scrq->msgs =
3425                 (union sub_crq *)__get_free_pages(GFP_KERNEL | __GFP_ZERO, 2);
3426         if (!scrq->msgs) {
3427                 dev_warn(dev, "Couldn't allocate crq queue messages page\n");
3428                 goto zero_page_failed;
3429         }
3430
3431         scrq->msg_token = dma_map_single(dev, scrq->msgs, 4 * PAGE_SIZE,
3432                                          DMA_BIDIRECTIONAL);
3433         if (dma_mapping_error(dev, scrq->msg_token)) {
3434                 dev_warn(dev, "Couldn't map crq queue messages page\n");
3435                 goto map_failed;
3436         }
3437
3438         rc = h_reg_sub_crq(adapter->vdev->unit_address, scrq->msg_token,
3439                            4 * PAGE_SIZE, &scrq->crq_num, &scrq->hw_irq);
3440
3441         if (rc == H_RESOURCE)
3442                 rc = ibmvnic_reset_crq(adapter);
3443
3444         if (rc == H_CLOSED) {
3445                 dev_warn(dev, "Partner adapter not ready, waiting.\n");
3446         } else if (rc) {
3447                 dev_warn(dev, "Error %d registering sub-crq\n", rc);
3448                 goto reg_failed;
3449         }
3450
3451         scrq->adapter = adapter;
3452         scrq->size = 4 * PAGE_SIZE / sizeof(*scrq->msgs);
3453         scrq->ind_buf.index = 0;
3454
3455         scrq->ind_buf.indir_arr =
3456                 dma_alloc_coherent(dev,
3457                                    IBMVNIC_IND_ARR_SZ,
3458                                    &scrq->ind_buf.indir_dma,
3459                                    GFP_KERNEL);
3460
3461         if (!scrq->ind_buf.indir_arr)
3462                 goto indir_failed;
3463
3464         spin_lock_init(&scrq->lock);
3465
3466         netdev_dbg(adapter->netdev,
3467                    "sub-crq initialized, num %lx, hw_irq=%lx, irq=%x\n",
3468                    scrq->crq_num, scrq->hw_irq, scrq->irq);
3469
3470         return scrq;
3471
3472 indir_failed:
3473         do {
3474                 rc = plpar_hcall_norets(H_FREE_SUB_CRQ,
3475                                         adapter->vdev->unit_address,
3476                                         scrq->crq_num);
3477         } while (rc == H_BUSY || rc == H_IS_LONG_BUSY(rc));
3478 reg_failed:
3479         dma_unmap_single(dev, scrq->msg_token, 4 * PAGE_SIZE,
3480                          DMA_BIDIRECTIONAL);
3481 map_failed:
3482         free_pages((unsigned long)scrq->msgs, 2);
3483 zero_page_failed:
3484         kfree(scrq);
3485
3486         return NULL;
3487 }
3488
3489 static void release_sub_crqs(struct ibmvnic_adapter *adapter, bool do_h_free)
3490 {
3491         int i;
3492
3493         if (adapter->tx_scrq) {
3494                 for (i = 0; i < adapter->num_active_tx_scrqs; i++) {
3495                         if (!adapter->tx_scrq[i])
3496                                 continue;
3497
3498                         netdev_dbg(adapter->netdev, "Releasing tx_scrq[%d]\n",
3499                                    i);
3500                         ibmvnic_tx_scrq_clean_buffer(adapter, adapter->tx_scrq[i]);
3501                         if (adapter->tx_scrq[i]->irq) {
3502                                 free_irq(adapter->tx_scrq[i]->irq,
3503                                          adapter->tx_scrq[i]);
3504                                 irq_dispose_mapping(adapter->tx_scrq[i]->irq);
3505                                 adapter->tx_scrq[i]->irq = 0;
3506                         }
3507
3508                         release_sub_crq_queue(adapter, adapter->tx_scrq[i],
3509                                               do_h_free);
3510                 }
3511
3512                 kfree(adapter->tx_scrq);
3513                 adapter->tx_scrq = NULL;
3514                 adapter->num_active_tx_scrqs = 0;
3515         }
3516
3517         if (adapter->rx_scrq) {
3518                 for (i = 0; i < adapter->num_active_rx_scrqs; i++) {
3519                         if (!adapter->rx_scrq[i])
3520                                 continue;
3521
3522                         netdev_dbg(adapter->netdev, "Releasing rx_scrq[%d]\n",
3523                                    i);
3524                         if (adapter->rx_scrq[i]->irq) {
3525                                 free_irq(adapter->rx_scrq[i]->irq,
3526                                          adapter->rx_scrq[i]);
3527                                 irq_dispose_mapping(adapter->rx_scrq[i]->irq);
3528                                 adapter->rx_scrq[i]->irq = 0;
3529                         }
3530
3531                         release_sub_crq_queue(adapter, adapter->rx_scrq[i],
3532                                               do_h_free);
3533                 }
3534
3535                 kfree(adapter->rx_scrq);
3536                 adapter->rx_scrq = NULL;
3537                 adapter->num_active_rx_scrqs = 0;
3538         }
3539 }
3540
3541 static int disable_scrq_irq(struct ibmvnic_adapter *adapter,
3542                             struct ibmvnic_sub_crq_queue *scrq)
3543 {
3544         struct device *dev = &adapter->vdev->dev;
3545         unsigned long rc;
3546
3547         rc = plpar_hcall_norets(H_VIOCTL, adapter->vdev->unit_address,
3548                                 H_DISABLE_VIO_INTERRUPT, scrq->hw_irq, 0, 0);
3549         if (rc)
3550                 dev_err(dev, "Couldn't disable scrq irq 0x%lx. rc=%ld\n",
3551                         scrq->hw_irq, rc);
3552         return rc;
3553 }
3554
3555 static int enable_scrq_irq(struct ibmvnic_adapter *adapter,
3556                            struct ibmvnic_sub_crq_queue *scrq)
3557 {
3558         struct device *dev = &adapter->vdev->dev;
3559         unsigned long rc;
3560
3561         if (scrq->hw_irq > 0x100000000ULL) {
3562                 dev_err(dev, "bad hw_irq = %lx\n", scrq->hw_irq);
3563                 return 1;
3564         }
3565
3566         if (test_bit(0, &adapter->resetting) &&
3567             adapter->reset_reason == VNIC_RESET_MOBILITY) {
3568                 u64 val = (0xff000000) | scrq->hw_irq;
3569
3570                 rc = plpar_hcall_norets(H_EOI, val);
3571                 /* H_EOI would fail with rc = H_FUNCTION when running
3572                  * in XIVE mode which is expected, but not an error.
3573                  */
3574                 if (rc && (rc != H_FUNCTION))
3575                         dev_err(dev, "H_EOI FAILED irq 0x%llx. rc=%ld\n",
3576                                 val, rc);
3577         }
3578
3579         rc = plpar_hcall_norets(H_VIOCTL, adapter->vdev->unit_address,
3580                                 H_ENABLE_VIO_INTERRUPT, scrq->hw_irq, 0, 0);
3581         if (rc)
3582                 dev_err(dev, "Couldn't enable scrq irq 0x%lx. rc=%ld\n",
3583                         scrq->hw_irq, rc);
3584         return rc;
3585 }
3586
3587 static int ibmvnic_complete_tx(struct ibmvnic_adapter *adapter,
3588                                struct ibmvnic_sub_crq_queue *scrq)
3589 {
3590         struct device *dev = &adapter->vdev->dev;
3591         struct ibmvnic_tx_pool *tx_pool;
3592         struct ibmvnic_tx_buff *txbuff;
3593         struct netdev_queue *txq;
3594         union sub_crq *next;
3595         int index;
3596         int i;
3597
3598 restart_loop:
3599         while (pending_scrq(adapter, scrq)) {
3600                 unsigned int pool = scrq->pool_index;
3601                 int num_entries = 0;
3602                 int total_bytes = 0;
3603                 int num_packets = 0;
3604
3605                 next = ibmvnic_next_scrq(adapter, scrq);
3606                 for (i = 0; i < next->tx_comp.num_comps; i++) {
3607                         index = be32_to_cpu(next->tx_comp.correlators[i]);
3608                         if (index & IBMVNIC_TSO_POOL_MASK) {
3609                                 tx_pool = &adapter->tso_pool[pool];
3610                                 index &= ~IBMVNIC_TSO_POOL_MASK;
3611                         } else {
3612                                 tx_pool = &adapter->tx_pool[pool];
3613                         }
3614
3615                         txbuff = &tx_pool->tx_buff[index];
3616                         num_packets++;
3617                         num_entries += txbuff->num_entries;
3618                         if (txbuff->skb) {
3619                                 total_bytes += txbuff->skb->len;
3620                                 if (next->tx_comp.rcs[i]) {
3621                                         dev_err(dev, "tx error %x\n",
3622                                                 next->tx_comp.rcs[i]);
3623                                         dev_kfree_skb_irq(txbuff->skb);
3624                                 } else {
3625                                         dev_consume_skb_irq(txbuff->skb);
3626                                 }
3627                                 txbuff->skb = NULL;
3628                         } else {
3629                                 netdev_warn(adapter->netdev,
3630                                             "TX completion received with NULL socket buffer\n");
3631                         }
3632                         tx_pool->free_map[tx_pool->producer_index] = index;
3633                         tx_pool->producer_index =
3634                                 (tx_pool->producer_index + 1) %
3635                                         tx_pool->num_buffers;
3636                 }
3637                 /* remove tx_comp scrq*/
3638                 next->tx_comp.first = 0;
3639
3640                 txq = netdev_get_tx_queue(adapter->netdev, scrq->pool_index);
3641                 netdev_tx_completed_queue(txq, num_packets, total_bytes);
3642
3643                 if (atomic_sub_return(num_entries, &scrq->used) <=
3644                     (adapter->req_tx_entries_per_subcrq / 2) &&
3645                     __netif_subqueue_stopped(adapter->netdev,
3646                                              scrq->pool_index)) {
3647                         netif_wake_subqueue(adapter->netdev, scrq->pool_index);
3648                         netdev_dbg(adapter->netdev, "Started queue %d\n",
3649                                    scrq->pool_index);
3650                 }
3651         }
3652
3653         enable_scrq_irq(adapter, scrq);
3654
3655         if (pending_scrq(adapter, scrq)) {
3656                 disable_scrq_irq(adapter, scrq);
3657                 goto restart_loop;
3658         }
3659
3660         return 0;
3661 }
3662
3663 static irqreturn_t ibmvnic_interrupt_tx(int irq, void *instance)
3664 {
3665         struct ibmvnic_sub_crq_queue *scrq = instance;
3666         struct ibmvnic_adapter *adapter = scrq->adapter;
3667
3668         disable_scrq_irq(adapter, scrq);
3669         ibmvnic_complete_tx(adapter, scrq);
3670
3671         return IRQ_HANDLED;
3672 }
3673
3674 static irqreturn_t ibmvnic_interrupt_rx(int irq, void *instance)
3675 {
3676         struct ibmvnic_sub_crq_queue *scrq = instance;
3677         struct ibmvnic_adapter *adapter = scrq->adapter;
3678
3679         /* When booting a kdump kernel we can hit pending interrupts
3680          * prior to completing driver initialization.
3681          */
3682         if (unlikely(adapter->state != VNIC_OPEN))
3683                 return IRQ_NONE;
3684
3685         adapter->rx_stats_buffers[scrq->scrq_num].interrupts++;
3686
3687         if (napi_schedule_prep(&adapter->napi[scrq->scrq_num])) {
3688                 disable_scrq_irq(adapter, scrq);
3689                 __napi_schedule(&adapter->napi[scrq->scrq_num]);
3690         }
3691
3692         return IRQ_HANDLED;
3693 }
3694
3695 static int init_sub_crq_irqs(struct ibmvnic_adapter *adapter)
3696 {
3697         struct device *dev = &adapter->vdev->dev;
3698         struct ibmvnic_sub_crq_queue *scrq;
3699         int i = 0, j = 0;
3700         int rc = 0;
3701
3702         for (i = 0; i < adapter->req_tx_queues; i++) {
3703                 netdev_dbg(adapter->netdev, "Initializing tx_scrq[%d] irq\n",
3704                            i);
3705                 scrq = adapter->tx_scrq[i];
3706                 scrq->irq = irq_create_mapping(NULL, scrq->hw_irq);
3707
3708                 if (!scrq->irq) {
3709                         rc = -EINVAL;
3710                         dev_err(dev, "Error mapping irq\n");
3711                         goto req_tx_irq_failed;
3712                 }
3713
3714                 snprintf(scrq->name, sizeof(scrq->name), "ibmvnic-%x-tx%d",
3715                          adapter->vdev->unit_address, i);
3716                 rc = request_irq(scrq->irq, ibmvnic_interrupt_tx,
3717                                  0, scrq->name, scrq);
3718
3719                 if (rc) {
3720                         dev_err(dev, "Couldn't register tx irq 0x%x. rc=%d\n",
3721                                 scrq->irq, rc);
3722                         irq_dispose_mapping(scrq->irq);
3723                         goto req_tx_irq_failed;
3724                 }
3725         }
3726
3727         for (i = 0; i < adapter->req_rx_queues; i++) {
3728                 netdev_dbg(adapter->netdev, "Initializing rx_scrq[%d] irq\n",
3729                            i);
3730                 scrq = adapter->rx_scrq[i];
3731                 scrq->irq = irq_create_mapping(NULL, scrq->hw_irq);
3732                 if (!scrq->irq) {
3733                         rc = -EINVAL;
3734                         dev_err(dev, "Error mapping irq\n");
3735                         goto req_rx_irq_failed;
3736                 }
3737                 snprintf(scrq->name, sizeof(scrq->name), "ibmvnic-%x-rx%d",
3738                          adapter->vdev->unit_address, i);
3739                 rc = request_irq(scrq->irq, ibmvnic_interrupt_rx,
3740                                  0, scrq->name, scrq);
3741                 if (rc) {
3742                         dev_err(dev, "Couldn't register rx irq 0x%x. rc=%d\n",
3743                                 scrq->irq, rc);
3744                         irq_dispose_mapping(scrq->irq);
3745                         goto req_rx_irq_failed;
3746                 }
3747         }
3748         return rc;
3749
3750 req_rx_irq_failed:
3751         for (j = 0; j < i; j++) {
3752                 free_irq(adapter->rx_scrq[j]->irq, adapter->rx_scrq[j]);
3753                 irq_dispose_mapping(adapter->rx_scrq[j]->irq);
3754         }
3755         i = adapter->req_tx_queues;
3756 req_tx_irq_failed:
3757         for (j = 0; j < i; j++) {
3758                 free_irq(adapter->tx_scrq[j]->irq, adapter->tx_scrq[j]);
3759                 irq_dispose_mapping(adapter->tx_scrq[j]->irq);
3760         }
3761         release_sub_crqs(adapter, 1);
3762         return rc;
3763 }
3764
3765 static int init_sub_crqs(struct ibmvnic_adapter *adapter)
3766 {
3767         struct device *dev = &adapter->vdev->dev;
3768         struct ibmvnic_sub_crq_queue **allqueues;
3769         int registered_queues = 0;
3770         int total_queues;
3771         int more = 0;
3772         int i;
3773
3774         total_queues = adapter->req_tx_queues + adapter->req_rx_queues;
3775
3776         allqueues = kcalloc(total_queues, sizeof(*allqueues), GFP_KERNEL);
3777         if (!allqueues)
3778                 return -1;
3779
3780         for (i = 0; i < total_queues; i++) {
3781                 allqueues[i] = init_sub_crq_queue(adapter);
3782                 if (!allqueues[i]) {
3783                         dev_warn(dev, "Couldn't allocate all sub-crqs\n");
3784                         break;
3785                 }
3786                 registered_queues++;
3787         }
3788
3789         /* Make sure we were able to register the minimum number of queues */
3790         if (registered_queues <
3791             adapter->min_tx_queues + adapter->min_rx_queues) {
3792                 dev_err(dev, "Fatal: Couldn't init  min number of sub-crqs\n");
3793                 goto tx_failed;
3794         }
3795
3796         /* Distribute the failed allocated queues*/
3797         for (i = 0; i < total_queues - registered_queues + more ; i++) {
3798                 netdev_dbg(adapter->netdev, "Reducing number of queues\n");
3799                 switch (i % 3) {
3800                 case 0:
3801                         if (adapter->req_rx_queues > adapter->min_rx_queues)
3802                                 adapter->req_rx_queues--;
3803                         else
3804                                 more++;
3805                         break;
3806                 case 1:
3807                         if (adapter->req_tx_queues > adapter->min_tx_queues)
3808                                 adapter->req_tx_queues--;
3809                         else
3810                                 more++;
3811                         break;
3812                 }
3813         }
3814
3815         adapter->tx_scrq = kcalloc(adapter->req_tx_queues,
3816                                    sizeof(*adapter->tx_scrq), GFP_KERNEL);
3817         if (!adapter->tx_scrq)
3818                 goto tx_failed;
3819
3820         for (i = 0; i < adapter->req_tx_queues; i++) {
3821                 adapter->tx_scrq[i] = allqueues[i];
3822                 adapter->tx_scrq[i]->pool_index = i;
3823                 adapter->num_active_tx_scrqs++;
3824         }
3825
3826         adapter->rx_scrq = kcalloc(adapter->req_rx_queues,
3827                                    sizeof(*adapter->rx_scrq), GFP_KERNEL);
3828         if (!adapter->rx_scrq)
3829                 goto rx_failed;
3830
3831         for (i = 0; i < adapter->req_rx_queues; i++) {
3832                 adapter->rx_scrq[i] = allqueues[i + adapter->req_tx_queues];
3833                 adapter->rx_scrq[i]->scrq_num = i;
3834                 adapter->num_active_rx_scrqs++;
3835         }
3836
3837         kfree(allqueues);
3838         return 0;
3839
3840 rx_failed:
3841         kfree(adapter->tx_scrq);
3842         adapter->tx_scrq = NULL;
3843 tx_failed:
3844         for (i = 0; i < registered_queues; i++)
3845                 release_sub_crq_queue(adapter, allqueues[i], 1);
3846         kfree(allqueues);
3847         return -1;
3848 }
3849
3850 static void send_request_cap(struct ibmvnic_adapter *adapter, int retry)
3851 {
3852         struct device *dev = &adapter->vdev->dev;
3853         union ibmvnic_crq crq;
3854         int max_entries;
3855
3856         if (!retry) {
3857                 /* Sub-CRQ entries are 32 byte long */
3858                 int entries_page = 4 * PAGE_SIZE / (sizeof(u64) * 4);
3859
3860                 if (adapter->min_tx_entries_per_subcrq > entries_page ||
3861                     adapter->min_rx_add_entries_per_subcrq > entries_page) {
3862                         dev_err(dev, "Fatal, invalid entries per sub-crq\n");
3863                         return;
3864                 }
3865
3866                 if (adapter->desired.mtu)
3867                         adapter->req_mtu = adapter->desired.mtu;
3868                 else
3869                         adapter->req_mtu = adapter->netdev->mtu + ETH_HLEN;
3870
3871                 if (!adapter->desired.tx_entries)
3872                         adapter->desired.tx_entries =
3873                                         adapter->max_tx_entries_per_subcrq;
3874                 if (!adapter->desired.rx_entries)
3875                         adapter->desired.rx_entries =
3876                                         adapter->max_rx_add_entries_per_subcrq;
3877
3878                 max_entries = IBMVNIC_MAX_LTB_SIZE /
3879                               (adapter->req_mtu + IBMVNIC_BUFFER_HLEN);
3880
3881                 if ((adapter->req_mtu + IBMVNIC_BUFFER_HLEN) *
3882                         adapter->desired.tx_entries > IBMVNIC_MAX_LTB_SIZE) {
3883                         adapter->desired.tx_entries = max_entries;
3884                 }
3885
3886                 if ((adapter->req_mtu + IBMVNIC_BUFFER_HLEN) *
3887                         adapter->desired.rx_entries > IBMVNIC_MAX_LTB_SIZE) {
3888                         adapter->desired.rx_entries = max_entries;
3889                 }
3890
3891                 if (adapter->desired.tx_entries)
3892                         adapter->req_tx_entries_per_subcrq =
3893                                         adapter->desired.tx_entries;
3894                 else
3895                         adapter->req_tx_entries_per_subcrq =
3896                                         adapter->max_tx_entries_per_subcrq;
3897
3898                 if (adapter->desired.rx_entries)
3899                         adapter->req_rx_add_entries_per_subcrq =
3900                                         adapter->desired.rx_entries;
3901                 else
3902                         adapter->req_rx_add_entries_per_subcrq =
3903                                         adapter->max_rx_add_entries_per_subcrq;
3904
3905                 if (adapter->desired.tx_queues)
3906                         adapter->req_tx_queues =
3907                                         adapter->desired.tx_queues;
3908                 else
3909                         adapter->req_tx_queues =
3910                                         adapter->opt_tx_comp_sub_queues;
3911
3912                 if (adapter->desired.rx_queues)
3913                         adapter->req_rx_queues =
3914                                         adapter->desired.rx_queues;
3915                 else
3916                         adapter->req_rx_queues =
3917                                         adapter->opt_rx_comp_queues;
3918
3919                 adapter->req_rx_add_queues = adapter->max_rx_add_queues;
3920         }
3921
3922         memset(&crq, 0, sizeof(crq));
3923         crq.request_capability.first = IBMVNIC_CRQ_CMD;
3924         crq.request_capability.cmd = REQUEST_CAPABILITY;
3925
3926         crq.request_capability.capability = cpu_to_be16(REQ_TX_QUEUES);
3927         crq.request_capability.number = cpu_to_be64(adapter->req_tx_queues);
3928         atomic_inc(&adapter->running_cap_crqs);
3929         ibmvnic_send_crq(adapter, &crq);
3930
3931         crq.request_capability.capability = cpu_to_be16(REQ_RX_QUEUES);
3932         crq.request_capability.number = cpu_to_be64(adapter->req_rx_queues);
3933         atomic_inc(&adapter->running_cap_crqs);
3934         ibmvnic_send_crq(adapter, &crq);
3935
3936         crq.request_capability.capability = cpu_to_be16(REQ_RX_ADD_QUEUES);
3937         crq.request_capability.number = cpu_to_be64(adapter->req_rx_add_queues);
3938         atomic_inc(&adapter->running_cap_crqs);
3939         ibmvnic_send_crq(adapter, &crq);
3940
3941         crq.request_capability.capability =
3942             cpu_to_be16(REQ_TX_ENTRIES_PER_SUBCRQ);
3943         crq.request_capability.number =
3944             cpu_to_be64(adapter->req_tx_entries_per_subcrq);
3945         atomic_inc(&adapter->running_cap_crqs);
3946         ibmvnic_send_crq(adapter, &crq);
3947
3948         crq.request_capability.capability =
3949             cpu_to_be16(REQ_RX_ADD_ENTRIES_PER_SUBCRQ);
3950         crq.request_capability.number =
3951             cpu_to_be64(adapter->req_rx_add_entries_per_subcrq);
3952         atomic_inc(&adapter->running_cap_crqs);
3953         ibmvnic_send_crq(adapter, &crq);
3954
3955         crq.request_capability.capability = cpu_to_be16(REQ_MTU);
3956         crq.request_capability.number = cpu_to_be64(adapter->req_mtu);
3957         atomic_inc(&adapter->running_cap_crqs);
3958         ibmvnic_send_crq(adapter, &crq);
3959
3960         if (adapter->netdev->flags & IFF_PROMISC) {
3961                 if (adapter->promisc_supported) {
3962                         crq.request_capability.capability =
3963                             cpu_to_be16(PROMISC_REQUESTED);
3964                         crq.request_capability.number = cpu_to_be64(1);
3965                         atomic_inc(&adapter->running_cap_crqs);
3966                         ibmvnic_send_crq(adapter, &crq);
3967                 }
3968         } else {
3969                 crq.request_capability.capability =
3970                     cpu_to_be16(PROMISC_REQUESTED);
3971                 crq.request_capability.number = cpu_to_be64(0);
3972                 atomic_inc(&adapter->running_cap_crqs);
3973                 ibmvnic_send_crq(adapter, &crq);
3974         }
3975 }
3976
3977 static int pending_scrq(struct ibmvnic_adapter *adapter,
3978                         struct ibmvnic_sub_crq_queue *scrq)
3979 {
3980         union sub_crq *entry = &scrq->msgs[scrq->cur];
3981         int rc;
3982
3983         rc = !!(entry->generic.first & IBMVNIC_CRQ_CMD_RSP);
3984
3985         /* Ensure that the SCRQ valid flag is loaded prior to loading the
3986          * contents of the SCRQ descriptor
3987          */
3988         dma_rmb();
3989
3990         return rc;
3991 }
3992
3993 static union sub_crq *ibmvnic_next_scrq(struct ibmvnic_adapter *adapter,
3994                                         struct ibmvnic_sub_crq_queue *scrq)
3995 {
3996         union sub_crq *entry;
3997         unsigned long flags;
3998
3999         spin_lock_irqsave(&scrq->lock, flags);
4000         entry = &scrq->msgs[scrq->cur];
4001         if (entry->generic.first & IBMVNIC_CRQ_CMD_RSP) {
4002                 if (++scrq->cur == scrq->size)
4003                         scrq->cur = 0;
4004         } else {
4005                 entry = NULL;
4006         }
4007         spin_unlock_irqrestore(&scrq->lock, flags);
4008
4009         /* Ensure that the SCRQ valid flag is loaded prior to loading the
4010          * contents of the SCRQ descriptor
4011          */
4012         dma_rmb();
4013
4014         return entry;
4015 }
4016
4017 static union ibmvnic_crq *ibmvnic_next_crq(struct ibmvnic_adapter *adapter)
4018 {
4019         struct ibmvnic_crq_queue *queue = &adapter->crq;
4020         union ibmvnic_crq *crq;
4021
4022         crq = &queue->msgs[queue->cur];
4023         if (crq->generic.first & IBMVNIC_CRQ_CMD_RSP) {
4024                 if (++queue->cur == queue->size)
4025                         queue->cur = 0;
4026         } else {
4027                 crq = NULL;
4028         }
4029
4030         return crq;
4031 }
4032
4033 static void print_subcrq_error(struct device *dev, int rc, const char *func)
4034 {
4035         switch (rc) {
4036         case H_PARAMETER:
4037                 dev_warn_ratelimited(dev,
4038                                      "%s failed: Send request is malformed or adapter failover pending. (rc=%d)\n",
4039                                      func, rc);
4040                 break;
4041         case H_CLOSED:
4042                 dev_warn_ratelimited(dev,
4043                                      "%s failed: Backing queue closed. Adapter is down or failover pending. (rc=%d)\n",
4044                                      func, rc);
4045                 break;
4046         default:
4047                 dev_err_ratelimited(dev, "%s failed: (rc=%d)\n", func, rc);
4048                 break;
4049         }
4050 }
4051
4052 static int send_subcrq_indirect(struct ibmvnic_adapter *adapter,
4053                                 u64 remote_handle, u64 ioba, u64 num_entries)
4054 {
4055         unsigned int ua = adapter->vdev->unit_address;
4056         struct device *dev = &adapter->vdev->dev;
4057         int rc;
4058
4059         /* Make sure the hypervisor sees the complete request */
4060         dma_wmb();
4061         rc = plpar_hcall_norets(H_SEND_SUB_CRQ_INDIRECT, ua,
4062                                 cpu_to_be64(remote_handle),
4063                                 ioba, num_entries);
4064
4065         if (rc)
4066                 print_subcrq_error(dev, rc, __func__);
4067
4068         return rc;
4069 }
4070
4071 static int ibmvnic_send_crq(struct ibmvnic_adapter *adapter,
4072                             union ibmvnic_crq *crq)
4073 {
4074         unsigned int ua = adapter->vdev->unit_address;
4075         struct device *dev = &adapter->vdev->dev;
4076         u64 *u64_crq = (u64 *)crq;
4077         int rc;
4078
4079         netdev_dbg(adapter->netdev, "Sending CRQ: %016lx %016lx\n",
4080                    (unsigned long)cpu_to_be64(u64_crq[0]),
4081                    (unsigned long)cpu_to_be64(u64_crq[1]));
4082
4083         if (!adapter->crq.active &&
4084             crq->generic.first != IBMVNIC_CRQ_INIT_CMD) {
4085                 dev_warn(dev, "Invalid request detected while CRQ is inactive, possible device state change during reset\n");
4086                 return -EINVAL;
4087         }
4088
4089         /* Make sure the hypervisor sees the complete request */
4090         dma_wmb();
4091
4092         rc = plpar_hcall_norets(H_SEND_CRQ, ua,
4093                                 cpu_to_be64(u64_crq[0]),
4094                                 cpu_to_be64(u64_crq[1]));
4095
4096         if (rc) {
4097                 if (rc == H_CLOSED) {
4098                         dev_warn(dev, "CRQ Queue closed\n");
4099                         /* do not reset, report the fail, wait for passive init from server */
4100                 }
4101
4102                 dev_warn(dev, "Send error (rc=%d)\n", rc);
4103         }
4104
4105         return rc;
4106 }
4107
4108 static int ibmvnic_send_crq_init(struct ibmvnic_adapter *adapter)
4109 {
4110         struct device *dev = &adapter->vdev->dev;
4111         union ibmvnic_crq crq;
4112         int retries = 100;
4113         int rc;
4114
4115         memset(&crq, 0, sizeof(crq));
4116         crq.generic.first = IBMVNIC_CRQ_INIT_CMD;
4117         crq.generic.cmd = IBMVNIC_CRQ_INIT;
4118         netdev_dbg(adapter->netdev, "Sending CRQ init\n");
4119
4120         do {
4121                 rc = ibmvnic_send_crq(adapter, &crq);
4122                 if (rc != H_CLOSED)
4123                         break;
4124                 retries--;
4125                 msleep(50);
4126
4127         } while (retries > 0);
4128
4129         if (rc) {
4130                 dev_err(dev, "Failed to send init request, rc = %d\n", rc);
4131                 return rc;
4132         }
4133
4134         return 0;
4135 }
4136
4137 struct vnic_login_client_data {
4138         u8      type;
4139         __be16  len;
4140         char    name[];
4141 } __packed;
4142
4143 static int vnic_client_data_len(struct ibmvnic_adapter *adapter)
4144 {
4145         int len;
4146
4147         /* Calculate the amount of buffer space needed for the
4148          * vnic client data in the login buffer. There are four entries,
4149          * OS name, LPAR name, device name, and a null last entry.
4150          */
4151         len = 4 * sizeof(struct vnic_login_client_data);
4152         len += 6; /* "Linux" plus NULL */
4153         len += strlen(utsname()->nodename) + 1;
4154         len += strlen(adapter->netdev->name) + 1;
4155
4156         return len;
4157 }
4158
4159 static void vnic_add_client_data(struct ibmvnic_adapter *adapter,
4160                                  struct vnic_login_client_data *vlcd)
4161 {
4162         const char *os_name = "Linux";
4163         int len;
4164
4165         /* Type 1 - LPAR OS */
4166         vlcd->type = 1;
4167         len = strlen(os_name) + 1;
4168         vlcd->len = cpu_to_be16(len);
4169         strscpy(vlcd->name, os_name, len);
4170         vlcd = (struct vnic_login_client_data *)(vlcd->name + len);
4171
4172         /* Type 2 - LPAR name */
4173         vlcd->type = 2;
4174         len = strlen(utsname()->nodename) + 1;
4175         vlcd->len = cpu_to_be16(len);
4176         strscpy(vlcd->name, utsname()->nodename, len);
4177         vlcd = (struct vnic_login_client_data *)(vlcd->name + len);
4178
4179         /* Type 3 - device name */
4180         vlcd->type = 3;
4181         len = strlen(adapter->netdev->name) + 1;
4182         vlcd->len = cpu_to_be16(len);
4183         strscpy(vlcd->name, adapter->netdev->name, len);
4184 }
4185
4186 static int send_login(struct ibmvnic_adapter *adapter)
4187 {
4188         struct ibmvnic_login_rsp_buffer *login_rsp_buffer;
4189         struct ibmvnic_login_buffer *login_buffer;
4190         struct device *dev = &adapter->vdev->dev;
4191         struct vnic_login_client_data *vlcd;
4192         dma_addr_t rsp_buffer_token;
4193         dma_addr_t buffer_token;
4194         size_t rsp_buffer_size;
4195         union ibmvnic_crq crq;
4196         int client_data_len;
4197         size_t buffer_size;
4198         __be64 *tx_list_p;
4199         __be64 *rx_list_p;
4200         int rc;
4201         int i;
4202
4203         if (!adapter->tx_scrq || !adapter->rx_scrq) {
4204                 netdev_err(adapter->netdev,
4205                            "RX or TX queues are not allocated, device login failed\n");
4206                 return -1;
4207         }
4208
4209         release_login_buffer(adapter);
4210         release_login_rsp_buffer(adapter);
4211
4212         client_data_len = vnic_client_data_len(adapter);
4213
4214         buffer_size =
4215             sizeof(struct ibmvnic_login_buffer) +
4216             sizeof(u64) * (adapter->req_tx_queues + adapter->req_rx_queues) +
4217             client_data_len;
4218
4219         login_buffer = kzalloc(buffer_size, GFP_ATOMIC);
4220         if (!login_buffer)
4221                 goto buf_alloc_failed;
4222
4223         buffer_token = dma_map_single(dev, login_buffer, buffer_size,
4224                                       DMA_TO_DEVICE);
4225         if (dma_mapping_error(dev, buffer_token)) {
4226                 dev_err(dev, "Couldn't map login buffer\n");
4227                 goto buf_map_failed;
4228         }
4229
4230         rsp_buffer_size = sizeof(struct ibmvnic_login_rsp_buffer) +
4231                           sizeof(u64) * adapter->req_tx_queues +
4232                           sizeof(u64) * adapter->req_rx_queues +
4233                           sizeof(u64) * adapter->req_rx_queues +
4234                           sizeof(u8) * IBMVNIC_TX_DESC_VERSIONS;
4235
4236         login_rsp_buffer = kmalloc(rsp_buffer_size, GFP_ATOMIC);
4237         if (!login_rsp_buffer)
4238                 goto buf_rsp_alloc_failed;
4239
4240         rsp_buffer_token = dma_map_single(dev, login_rsp_buffer,
4241                                           rsp_buffer_size, DMA_FROM_DEVICE);
4242         if (dma_mapping_error(dev, rsp_buffer_token)) {
4243                 dev_err(dev, "Couldn't map login rsp buffer\n");
4244                 goto buf_rsp_map_failed;
4245         }
4246
4247         adapter->login_buf = login_buffer;
4248         adapter->login_buf_token = buffer_token;
4249         adapter->login_buf_sz = buffer_size;
4250         adapter->login_rsp_buf = login_rsp_buffer;
4251         adapter->login_rsp_buf_token = rsp_buffer_token;
4252         adapter->login_rsp_buf_sz = rsp_buffer_size;
4253
4254         login_buffer->len = cpu_to_be32(buffer_size);
4255         login_buffer->version = cpu_to_be32(INITIAL_VERSION_LB);
4256         login_buffer->num_txcomp_subcrqs = cpu_to_be32(adapter->req_tx_queues);
4257         login_buffer->off_txcomp_subcrqs =
4258             cpu_to_be32(sizeof(struct ibmvnic_login_buffer));
4259         login_buffer->num_rxcomp_subcrqs = cpu_to_be32(adapter->req_rx_queues);
4260         login_buffer->off_rxcomp_subcrqs =
4261             cpu_to_be32(sizeof(struct ibmvnic_login_buffer) +
4262                         sizeof(u64) * adapter->req_tx_queues);
4263         login_buffer->login_rsp_ioba = cpu_to_be32(rsp_buffer_token);
4264         login_buffer->login_rsp_len = cpu_to_be32(rsp_buffer_size);
4265
4266         tx_list_p = (__be64 *)((char *)login_buffer +
4267                                       sizeof(struct ibmvnic_login_buffer));
4268         rx_list_p = (__be64 *)((char *)login_buffer +
4269                                       sizeof(struct ibmvnic_login_buffer) +
4270                                       sizeof(u64) * adapter->req_tx_queues);
4271
4272         for (i = 0; i < adapter->req_tx_queues; i++) {
4273                 if (adapter->tx_scrq[i]) {
4274                         tx_list_p[i] =
4275                                 cpu_to_be64(adapter->tx_scrq[i]->crq_num);
4276                 }
4277         }
4278
4279         for (i = 0; i < adapter->req_rx_queues; i++) {
4280                 if (adapter->rx_scrq[i]) {
4281                         rx_list_p[i] =
4282                                 cpu_to_be64(adapter->rx_scrq[i]->crq_num);
4283                 }
4284         }
4285
4286         /* Insert vNIC login client data */
4287         vlcd = (struct vnic_login_client_data *)
4288                 ((char *)rx_list_p + (sizeof(u64) * adapter->req_rx_queues));
4289         login_buffer->client_data_offset =
4290                         cpu_to_be32((char *)vlcd - (char *)login_buffer);
4291         login_buffer->client_data_len = cpu_to_be32(client_data_len);
4292
4293         vnic_add_client_data(adapter, vlcd);
4294
4295         netdev_dbg(adapter->netdev, "Login Buffer:\n");
4296         for (i = 0; i < (adapter->login_buf_sz - 1) / 8 + 1; i++) {
4297                 netdev_dbg(adapter->netdev, "%016lx\n",
4298                            ((unsigned long *)(adapter->login_buf))[i]);
4299         }
4300
4301         memset(&crq, 0, sizeof(crq));
4302         crq.login.first = IBMVNIC_CRQ_CMD;
4303         crq.login.cmd = LOGIN;
4304         crq.login.ioba = cpu_to_be32(buffer_token);
4305         crq.login.len = cpu_to_be32(buffer_size);
4306
4307         adapter->login_pending = true;
4308         rc = ibmvnic_send_crq(adapter, &crq);
4309         if (rc) {
4310                 adapter->login_pending = false;
4311                 netdev_err(adapter->netdev, "Failed to send login, rc=%d\n", rc);
4312                 goto buf_rsp_map_failed;
4313         }
4314
4315         return 0;
4316
4317 buf_rsp_map_failed:
4318         kfree(login_rsp_buffer);
4319         adapter->login_rsp_buf = NULL;
4320 buf_rsp_alloc_failed:
4321         dma_unmap_single(dev, buffer_token, buffer_size, DMA_TO_DEVICE);
4322 buf_map_failed:
4323         kfree(login_buffer);
4324         adapter->login_buf = NULL;
4325 buf_alloc_failed:
4326         return -1;
4327 }
4328
4329 static int send_request_map(struct ibmvnic_adapter *adapter, dma_addr_t addr,
4330                             u32 len, u8 map_id)
4331 {
4332         union ibmvnic_crq crq;
4333
4334         memset(&crq, 0, sizeof(crq));
4335         crq.request_map.first = IBMVNIC_CRQ_CMD;
4336         crq.request_map.cmd = REQUEST_MAP;
4337         crq.request_map.map_id = map_id;
4338         crq.request_map.ioba = cpu_to_be32(addr);
4339         crq.request_map.len = cpu_to_be32(len);
4340         return ibmvnic_send_crq(adapter, &crq);
4341 }
4342
4343 static int send_request_unmap(struct ibmvnic_adapter *adapter, u8 map_id)
4344 {
4345         union ibmvnic_crq crq;
4346
4347         memset(&crq, 0, sizeof(crq));
4348         crq.request_unmap.first = IBMVNIC_CRQ_CMD;
4349         crq.request_unmap.cmd = REQUEST_UNMAP;
4350         crq.request_unmap.map_id = map_id;
4351         return ibmvnic_send_crq(adapter, &crq);
4352 }
4353
4354 static void send_query_map(struct ibmvnic_adapter *adapter)
4355 {
4356         union ibmvnic_crq crq;
4357
4358         memset(&crq, 0, sizeof(crq));
4359         crq.query_map.first = IBMVNIC_CRQ_CMD;
4360         crq.query_map.cmd = QUERY_MAP;
4361         ibmvnic_send_crq(adapter, &crq);
4362 }
4363
4364 /* Send a series of CRQs requesting various capabilities of the VNIC server */
4365 static void send_query_cap(struct ibmvnic_adapter *adapter)
4366 {
4367         union ibmvnic_crq crq;
4368
4369         atomic_set(&adapter->running_cap_crqs, 0);
4370         memset(&crq, 0, sizeof(crq));
4371         crq.query_capability.first = IBMVNIC_CRQ_CMD;
4372         crq.query_capability.cmd = QUERY_CAPABILITY;
4373
4374         crq.query_capability.capability = cpu_to_be16(MIN_TX_QUEUES);
4375         atomic_inc(&adapter->running_cap_crqs);
4376         ibmvnic_send_crq(adapter, &crq);
4377
4378         crq.query_capability.capability = cpu_to_be16(MIN_RX_QUEUES);
4379         atomic_inc(&adapter->running_cap_crqs);
4380         ibmvnic_send_crq(adapter, &crq);
4381
4382         crq.query_capability.capability = cpu_to_be16(MIN_RX_ADD_QUEUES);
4383         atomic_inc(&adapter->running_cap_crqs);
4384         ibmvnic_send_crq(adapter, &crq);
4385
4386         crq.query_capability.capability = cpu_to_be16(MAX_TX_QUEUES);
4387         atomic_inc(&adapter->running_cap_crqs);
4388         ibmvnic_send_crq(adapter, &crq);
4389
4390         crq.query_capability.capability = cpu_to_be16(MAX_RX_QUEUES);
4391         atomic_inc(&adapter->running_cap_crqs);
4392         ibmvnic_send_crq(adapter, &crq);
4393
4394         crq.query_capability.capability = cpu_to_be16(MAX_RX_ADD_QUEUES);
4395         atomic_inc(&adapter->running_cap_crqs);
4396         ibmvnic_send_crq(adapter, &crq);
4397
4398         crq.query_capability.capability =
4399             cpu_to_be16(MIN_TX_ENTRIES_PER_SUBCRQ);
4400         atomic_inc(&adapter->running_cap_crqs);
4401         ibmvnic_send_crq(adapter, &crq);
4402
4403         crq.query_capability.capability =
4404             cpu_to_be16(MIN_RX_ADD_ENTRIES_PER_SUBCRQ);
4405         atomic_inc(&adapter->running_cap_crqs);
4406         ibmvnic_send_crq(adapter, &crq);
4407
4408         crq.query_capability.capability =
4409             cpu_to_be16(MAX_TX_ENTRIES_PER_SUBCRQ);
4410         atomic_inc(&adapter->running_cap_crqs);
4411         ibmvnic_send_crq(adapter, &crq);
4412
4413         crq.query_capability.capability =
4414             cpu_to_be16(MAX_RX_ADD_ENTRIES_PER_SUBCRQ);
4415         atomic_inc(&adapter->running_cap_crqs);
4416         ibmvnic_send_crq(adapter, &crq);
4417
4418         crq.query_capability.capability = cpu_to_be16(TCP_IP_OFFLOAD);
4419         atomic_inc(&adapter->running_cap_crqs);
4420         ibmvnic_send_crq(adapter, &crq);
4421
4422         crq.query_capability.capability = cpu_to_be16(PROMISC_SUPPORTED);
4423         atomic_inc(&adapter->running_cap_crqs);
4424         ibmvnic_send_crq(adapter, &crq);
4425
4426         crq.query_capability.capability = cpu_to_be16(MIN_MTU);
4427         atomic_inc(&adapter->running_cap_crqs);
4428         ibmvnic_send_crq(adapter, &crq);
4429
4430         crq.query_capability.capability = cpu_to_be16(MAX_MTU);
4431         atomic_inc(&adapter->running_cap_crqs);
4432         ibmvnic_send_crq(adapter, &crq);
4433
4434         crq.query_capability.capability = cpu_to_be16(MAX_MULTICAST_FILTERS);
4435         atomic_inc(&adapter->running_cap_crqs);
4436         ibmvnic_send_crq(adapter, &crq);
4437
4438         crq.query_capability.capability = cpu_to_be16(VLAN_HEADER_INSERTION);
4439         atomic_inc(&adapter->running_cap_crqs);
4440         ibmvnic_send_crq(adapter, &crq);
4441
4442         crq.query_capability.capability = cpu_to_be16(RX_VLAN_HEADER_INSERTION);
4443         atomic_inc(&adapter->running_cap_crqs);
4444         ibmvnic_send_crq(adapter, &crq);
4445
4446         crq.query_capability.capability = cpu_to_be16(MAX_TX_SG_ENTRIES);
4447         atomic_inc(&adapter->running_cap_crqs);
4448         ibmvnic_send_crq(adapter, &crq);
4449
4450         crq.query_capability.capability = cpu_to_be16(RX_SG_SUPPORTED);
4451         atomic_inc(&adapter->running_cap_crqs);
4452         ibmvnic_send_crq(adapter, &crq);
4453
4454         crq.query_capability.capability = cpu_to_be16(OPT_TX_COMP_SUB_QUEUES);
4455         atomic_inc(&adapter->running_cap_crqs);
4456         ibmvnic_send_crq(adapter, &crq);
4457
4458         crq.query_capability.capability = cpu_to_be16(OPT_RX_COMP_QUEUES);
4459         atomic_inc(&adapter->running_cap_crqs);
4460         ibmvnic_send_crq(adapter, &crq);
4461
4462         crq.query_capability.capability =
4463                         cpu_to_be16(OPT_RX_BUFADD_Q_PER_RX_COMP_Q);
4464         atomic_inc(&adapter->running_cap_crqs);
4465         ibmvnic_send_crq(adapter, &crq);
4466
4467         crq.query_capability.capability =
4468                         cpu_to_be16(OPT_TX_ENTRIES_PER_SUBCRQ);
4469         atomic_inc(&adapter->running_cap_crqs);
4470         ibmvnic_send_crq(adapter, &crq);
4471
4472         crq.query_capability.capability =
4473                         cpu_to_be16(OPT_RXBA_ENTRIES_PER_SUBCRQ);
4474         atomic_inc(&adapter->running_cap_crqs);
4475         ibmvnic_send_crq(adapter, &crq);
4476
4477         crq.query_capability.capability = cpu_to_be16(TX_RX_DESC_REQ);
4478         atomic_inc(&adapter->running_cap_crqs);
4479         ibmvnic_send_crq(adapter, &crq);
4480 }
4481
4482 static void send_query_ip_offload(struct ibmvnic_adapter *adapter)
4483 {
4484         int buf_sz = sizeof(struct ibmvnic_query_ip_offload_buffer);
4485         struct device *dev = &adapter->vdev->dev;
4486         union ibmvnic_crq crq;
4487
4488         adapter->ip_offload_tok =
4489                 dma_map_single(dev,
4490                                &adapter->ip_offload_buf,
4491                                buf_sz,
4492                                DMA_FROM_DEVICE);
4493
4494         if (dma_mapping_error(dev, adapter->ip_offload_tok)) {
4495                 if (!firmware_has_feature(FW_FEATURE_CMO))
4496                         dev_err(dev, "Couldn't map offload buffer\n");
4497                 return;
4498         }
4499
4500         memset(&crq, 0, sizeof(crq));
4501         crq.query_ip_offload.first = IBMVNIC_CRQ_CMD;
4502         crq.query_ip_offload.cmd = QUERY_IP_OFFLOAD;
4503         crq.query_ip_offload.len = cpu_to_be32(buf_sz);
4504         crq.query_ip_offload.ioba =
4505             cpu_to_be32(adapter->ip_offload_tok);
4506
4507         ibmvnic_send_crq(adapter, &crq);
4508 }
4509
4510 static void send_control_ip_offload(struct ibmvnic_adapter *adapter)
4511 {
4512         struct ibmvnic_control_ip_offload_buffer *ctrl_buf = &adapter->ip_offload_ctrl;
4513         struct ibmvnic_query_ip_offload_buffer *buf = &adapter->ip_offload_buf;
4514         struct device *dev = &adapter->vdev->dev;
4515         netdev_features_t old_hw_features = 0;
4516         union ibmvnic_crq crq;
4517
4518         adapter->ip_offload_ctrl_tok =
4519                 dma_map_single(dev,
4520                                ctrl_buf,
4521                                sizeof(adapter->ip_offload_ctrl),
4522                                DMA_TO_DEVICE);
4523
4524         if (dma_mapping_error(dev, adapter->ip_offload_ctrl_tok)) {
4525                 dev_err(dev, "Couldn't map ip offload control buffer\n");
4526                 return;
4527         }
4528
4529         ctrl_buf->len = cpu_to_be32(sizeof(adapter->ip_offload_ctrl));
4530         ctrl_buf->version = cpu_to_be32(INITIAL_VERSION_IOB);
4531         ctrl_buf->ipv4_chksum = buf->ipv4_chksum;
4532         ctrl_buf->ipv6_chksum = buf->ipv6_chksum;
4533         ctrl_buf->tcp_ipv4_chksum = buf->tcp_ipv4_chksum;
4534         ctrl_buf->udp_ipv4_chksum = buf->udp_ipv4_chksum;
4535         ctrl_buf->tcp_ipv6_chksum = buf->tcp_ipv6_chksum;
4536         ctrl_buf->udp_ipv6_chksum = buf->udp_ipv6_chksum;
4537         ctrl_buf->large_tx_ipv4 = buf->large_tx_ipv4;
4538         ctrl_buf->large_tx_ipv6 = buf->large_tx_ipv6;
4539
4540         /* large_rx disabled for now, additional features needed */
4541         ctrl_buf->large_rx_ipv4 = 0;
4542         ctrl_buf->large_rx_ipv6 = 0;
4543
4544         if (adapter->state != VNIC_PROBING) {
4545                 old_hw_features = adapter->netdev->hw_features;
4546                 adapter->netdev->hw_features = 0;
4547         }
4548
4549         adapter->netdev->hw_features = NETIF_F_SG | NETIF_F_GSO | NETIF_F_GRO;
4550
4551         if (buf->tcp_ipv4_chksum || buf->udp_ipv4_chksum)
4552                 adapter->netdev->hw_features |= NETIF_F_IP_CSUM;
4553
4554         if (buf->tcp_ipv6_chksum || buf->udp_ipv6_chksum)
4555                 adapter->netdev->hw_features |= NETIF_F_IPV6_CSUM;
4556
4557         if ((adapter->netdev->features &
4558             (NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM)))
4559                 adapter->netdev->hw_features |= NETIF_F_RXCSUM;
4560
4561         if (buf->large_tx_ipv4)
4562                 adapter->netdev->hw_features |= NETIF_F_TSO;
4563         if (buf->large_tx_ipv6)
4564                 adapter->netdev->hw_features |= NETIF_F_TSO6;
4565
4566         if (adapter->state == VNIC_PROBING) {
4567                 adapter->netdev->features |= adapter->netdev->hw_features;
4568         } else if (old_hw_features != adapter->netdev->hw_features) {
4569                 netdev_features_t tmp = 0;
4570
4571                 /* disable features no longer supported */
4572                 adapter->netdev->features &= adapter->netdev->hw_features;
4573                 /* turn on features now supported if previously enabled */
4574                 tmp = (old_hw_features ^ adapter->netdev->hw_features) &
4575                         adapter->netdev->hw_features;
4576                 adapter->netdev->features |=
4577                                 tmp & adapter->netdev->wanted_features;
4578         }
4579
4580         memset(&crq, 0, sizeof(crq));
4581         crq.control_ip_offload.first = IBMVNIC_CRQ_CMD;
4582         crq.control_ip_offload.cmd = CONTROL_IP_OFFLOAD;
4583         crq.control_ip_offload.len =
4584             cpu_to_be32(sizeof(adapter->ip_offload_ctrl));
4585         crq.control_ip_offload.ioba = cpu_to_be32(adapter->ip_offload_ctrl_tok);
4586         ibmvnic_send_crq(adapter, &crq);
4587 }
4588
4589 static void handle_vpd_size_rsp(union ibmvnic_crq *crq,
4590                                 struct ibmvnic_adapter *adapter)
4591 {
4592         struct device *dev = &adapter->vdev->dev;
4593
4594         if (crq->get_vpd_size_rsp.rc.code) {
4595                 dev_err(dev, "Error retrieving VPD size, rc=%x\n",
4596                         crq->get_vpd_size_rsp.rc.code);
4597                 complete(&adapter->fw_done);
4598                 return;
4599         }
4600
4601         adapter->vpd->len = be64_to_cpu(crq->get_vpd_size_rsp.len);
4602         complete(&adapter->fw_done);
4603 }
4604
4605 static void handle_vpd_rsp(union ibmvnic_crq *crq,
4606                            struct ibmvnic_adapter *adapter)
4607 {
4608         struct device *dev = &adapter->vdev->dev;
4609         unsigned char *substr = NULL;
4610         u8 fw_level_len = 0;
4611
4612         memset(adapter->fw_version, 0, 32);
4613
4614         dma_unmap_single(dev, adapter->vpd->dma_addr, adapter->vpd->len,
4615                          DMA_FROM_DEVICE);
4616
4617         if (crq->get_vpd_rsp.rc.code) {
4618                 dev_err(dev, "Error retrieving VPD from device, rc=%x\n",
4619                         crq->get_vpd_rsp.rc.code);
4620                 goto complete;
4621         }
4622
4623         /* get the position of the firmware version info
4624          * located after the ASCII 'RM' substring in the buffer
4625          */
4626         substr = strnstr(adapter->vpd->buff, "RM", adapter->vpd->len);
4627         if (!substr) {
4628                 dev_info(dev, "Warning - No FW level has been provided in the VPD buffer by the VIOS Server\n");
4629                 goto complete;
4630         }
4631
4632         /* get length of firmware level ASCII substring */
4633         if ((substr + 2) < (adapter->vpd->buff + adapter->vpd->len)) {
4634                 fw_level_len = *(substr + 2);
4635         } else {
4636                 dev_info(dev, "Length of FW substr extrapolated VDP buff\n");
4637                 goto complete;
4638         }
4639
4640         /* copy firmware version string from vpd into adapter */
4641         if ((substr + 3 + fw_level_len) <
4642             (adapter->vpd->buff + adapter->vpd->len)) {
4643                 strncpy((char *)adapter->fw_version, substr + 3, fw_level_len);
4644         } else {
4645                 dev_info(dev, "FW substr extrapolated VPD buff\n");
4646         }
4647
4648 complete:
4649         if (adapter->fw_version[0] == '\0')
4650                 strscpy((char *)adapter->fw_version, "N/A", sizeof(adapter->fw_version));
4651         complete(&adapter->fw_done);
4652 }
4653
4654 static void handle_query_ip_offload_rsp(struct ibmvnic_adapter *adapter)
4655 {
4656         struct device *dev = &adapter->vdev->dev;
4657         struct ibmvnic_query_ip_offload_buffer *buf = &adapter->ip_offload_buf;
4658         int i;
4659
4660         dma_unmap_single(dev, adapter->ip_offload_tok,
4661                          sizeof(adapter->ip_offload_buf), DMA_FROM_DEVICE);
4662
4663         netdev_dbg(adapter->netdev, "Query IP Offload Buffer:\n");
4664         for (i = 0; i < (sizeof(adapter->ip_offload_buf) - 1) / 8 + 1; i++)
4665                 netdev_dbg(adapter->netdev, "%016lx\n",
4666                            ((unsigned long *)(buf))[i]);
4667
4668         netdev_dbg(adapter->netdev, "ipv4_chksum = %d\n", buf->ipv4_chksum);
4669         netdev_dbg(adapter->netdev, "ipv6_chksum = %d\n", buf->ipv6_chksum);
4670         netdev_dbg(adapter->netdev, "tcp_ipv4_chksum = %d\n",
4671                    buf->tcp_ipv4_chksum);
4672         netdev_dbg(adapter->netdev, "tcp_ipv6_chksum = %d\n",
4673                    buf->tcp_ipv6_chksum);
4674         netdev_dbg(adapter->netdev, "udp_ipv4_chksum = %d\n",
4675                    buf->udp_ipv4_chksum);
4676         netdev_dbg(adapter->netdev, "udp_ipv6_chksum = %d\n",
4677                    buf->udp_ipv6_chksum);
4678         netdev_dbg(adapter->netdev, "large_tx_ipv4 = %d\n",
4679                    buf->large_tx_ipv4);
4680         netdev_dbg(adapter->netdev, "large_tx_ipv6 = %d\n",
4681                    buf->large_tx_ipv6);
4682         netdev_dbg(adapter->netdev, "large_rx_ipv4 = %d\n",
4683                    buf->large_rx_ipv4);
4684         netdev_dbg(adapter->netdev, "large_rx_ipv6 = %d\n",
4685                    buf->large_rx_ipv6);
4686         netdev_dbg(adapter->netdev, "max_ipv4_hdr_sz = %d\n",
4687                    buf->max_ipv4_header_size);
4688         netdev_dbg(adapter->netdev, "max_ipv6_hdr_sz = %d\n",
4689                    buf->max_ipv6_header_size);
4690         netdev_dbg(adapter->netdev, "max_tcp_hdr_size = %d\n",
4691                    buf->max_tcp_header_size);
4692         netdev_dbg(adapter->netdev, "max_udp_hdr_size = %d\n",
4693                    buf->max_udp_header_size);
4694         netdev_dbg(adapter->netdev, "max_large_tx_size = %d\n",
4695                    buf->max_large_tx_size);
4696         netdev_dbg(adapter->netdev, "max_large_rx_size = %d\n",
4697                    buf->max_large_rx_size);
4698         netdev_dbg(adapter->netdev, "ipv6_ext_hdr = %d\n",
4699                    buf->ipv6_extension_header);
4700         netdev_dbg(adapter->netdev, "tcp_pseudosum_req = %d\n",
4701                    buf->tcp_pseudosum_req);
4702         netdev_dbg(adapter->netdev, "num_ipv6_ext_hd = %d\n",
4703                    buf->num_ipv6_ext_headers);
4704         netdev_dbg(adapter->netdev, "off_ipv6_ext_hd = %d\n",
4705                    buf->off_ipv6_ext_headers);
4706
4707         send_control_ip_offload(adapter);
4708 }
4709
4710 static const char *ibmvnic_fw_err_cause(u16 cause)
4711 {
4712         switch (cause) {
4713         case ADAPTER_PROBLEM:
4714                 return "adapter problem";
4715         case BUS_PROBLEM:
4716                 return "bus problem";
4717         case FW_PROBLEM:
4718                 return "firmware problem";
4719         case DD_PROBLEM:
4720                 return "device driver problem";
4721         case EEH_RECOVERY:
4722                 return "EEH recovery";
4723         case FW_UPDATED:
4724                 return "firmware updated";
4725         case LOW_MEMORY:
4726                 return "low Memory";
4727         default:
4728                 return "unknown";
4729         }
4730 }
4731
4732 static void handle_error_indication(union ibmvnic_crq *crq,
4733                                     struct ibmvnic_adapter *adapter)
4734 {
4735         struct device *dev = &adapter->vdev->dev;
4736         u16 cause;
4737
4738         cause = be16_to_cpu(crq->error_indication.error_cause);
4739
4740         dev_warn_ratelimited(dev,
4741                              "Firmware reports %serror, cause: %s. Starting recovery...\n",
4742                              crq->error_indication.flags
4743                                 & IBMVNIC_FATAL_ERROR ? "FATAL " : "",
4744                              ibmvnic_fw_err_cause(cause));
4745
4746         if (crq->error_indication.flags & IBMVNIC_FATAL_ERROR)
4747                 ibmvnic_reset(adapter, VNIC_RESET_FATAL);
4748         else
4749                 ibmvnic_reset(adapter, VNIC_RESET_NON_FATAL);
4750 }
4751
4752 static int handle_change_mac_rsp(union ibmvnic_crq *crq,
4753                                  struct ibmvnic_adapter *adapter)
4754 {
4755         struct net_device *netdev = adapter->netdev;
4756         struct device *dev = &adapter->vdev->dev;
4757         long rc;
4758
4759         rc = crq->change_mac_addr_rsp.rc.code;
4760         if (rc) {
4761                 dev_err(dev, "Error %ld in CHANGE_MAC_ADDR_RSP\n", rc);
4762                 goto out;
4763         }
4764         /* crq->change_mac_addr.mac_addr is the requested one
4765          * crq->change_mac_addr_rsp.mac_addr is the returned valid one.
4766          */
4767         eth_hw_addr_set(netdev, &crq->change_mac_addr_rsp.mac_addr[0]);
4768         ether_addr_copy(adapter->mac_addr,
4769                         &crq->change_mac_addr_rsp.mac_addr[0]);
4770 out:
4771         complete(&adapter->fw_done);
4772         return rc;
4773 }
4774
4775 static void handle_request_cap_rsp(union ibmvnic_crq *crq,
4776                                    struct ibmvnic_adapter *adapter)
4777 {
4778         struct device *dev = &adapter->vdev->dev;
4779         u64 *req_value;
4780         char *name;
4781
4782         atomic_dec(&adapter->running_cap_crqs);
4783         switch (be16_to_cpu(crq->request_capability_rsp.capability)) {
4784         case REQ_TX_QUEUES:
4785                 req_value = &adapter->req_tx_queues;
4786                 name = "tx";
4787                 break;
4788         case REQ_RX_QUEUES:
4789                 req_value = &adapter->req_rx_queues;
4790                 name = "rx";
4791                 break;
4792         case REQ_RX_ADD_QUEUES:
4793                 req_value = &adapter->req_rx_add_queues;
4794                 name = "rx_add";
4795                 break;
4796         case REQ_TX_ENTRIES_PER_SUBCRQ:
4797                 req_value = &adapter->req_tx_entries_per_subcrq;
4798                 name = "tx_entries_per_subcrq";
4799                 break;
4800         case REQ_RX_ADD_ENTRIES_PER_SUBCRQ:
4801                 req_value = &adapter->req_rx_add_entries_per_subcrq;
4802                 name = "rx_add_entries_per_subcrq";
4803                 break;
4804         case REQ_MTU:
4805                 req_value = &adapter->req_mtu;
4806                 name = "mtu";
4807                 break;
4808         case PROMISC_REQUESTED:
4809                 req_value = &adapter->promisc;
4810                 name = "promisc";
4811                 break;
4812         default:
4813                 dev_err(dev, "Got invalid cap request rsp %d\n",
4814                         crq->request_capability.capability);
4815                 return;
4816         }
4817
4818         switch (crq->request_capability_rsp.rc.code) {
4819         case SUCCESS:
4820                 break;
4821         case PARTIALSUCCESS:
4822                 dev_info(dev, "req=%lld, rsp=%ld in %s queue, retrying.\n",
4823                          *req_value,
4824                          (long)be64_to_cpu(crq->request_capability_rsp.number),
4825                          name);
4826
4827                 if (be16_to_cpu(crq->request_capability_rsp.capability) ==
4828                     REQ_MTU) {
4829                         pr_err("mtu of %llu is not supported. Reverting.\n",
4830                                *req_value);
4831                         *req_value = adapter->fallback.mtu;
4832                 } else {
4833                         *req_value =
4834                                 be64_to_cpu(crq->request_capability_rsp.number);
4835                 }
4836
4837                 send_request_cap(adapter, 1);
4838                 return;
4839         default:
4840                 dev_err(dev, "Error %d in request cap rsp\n",
4841                         crq->request_capability_rsp.rc.code);
4842                 return;
4843         }
4844
4845         /* Done receiving requested capabilities, query IP offload support */
4846         if (atomic_read(&adapter->running_cap_crqs) == 0) {
4847                 adapter->wait_capability = false;
4848                 send_query_ip_offload(adapter);
4849         }
4850 }
4851
4852 static int handle_login_rsp(union ibmvnic_crq *login_rsp_crq,
4853                             struct ibmvnic_adapter *adapter)
4854 {
4855         struct device *dev = &adapter->vdev->dev;
4856         struct net_device *netdev = adapter->netdev;
4857         struct ibmvnic_login_rsp_buffer *login_rsp = adapter->login_rsp_buf;
4858         struct ibmvnic_login_buffer *login = adapter->login_buf;
4859         u64 *tx_handle_array;
4860         u64 *rx_handle_array;
4861         int num_tx_pools;
4862         int num_rx_pools;
4863         u64 *size_array;
4864         int i;
4865
4866         /* CHECK: Test/set of login_pending does not need to be atomic
4867          * because only ibmvnic_tasklet tests/clears this.
4868          */
4869         if (!adapter->login_pending) {
4870                 netdev_warn(netdev, "Ignoring unexpected login response\n");
4871                 return 0;
4872         }
4873         adapter->login_pending = false;
4874
4875         dma_unmap_single(dev, adapter->login_buf_token, adapter->login_buf_sz,
4876                          DMA_TO_DEVICE);
4877         dma_unmap_single(dev, adapter->login_rsp_buf_token,
4878                          adapter->login_rsp_buf_sz, DMA_FROM_DEVICE);
4879
4880         /* If the number of queues requested can't be allocated by the
4881          * server, the login response will return with code 1. We will need
4882          * to resend the login buffer with fewer queues requested.
4883          */
4884         if (login_rsp_crq->generic.rc.code) {
4885                 adapter->init_done_rc = login_rsp_crq->generic.rc.code;
4886                 complete(&adapter->init_done);
4887                 return 0;
4888         }
4889
4890         if (adapter->failover_pending) {
4891                 adapter->init_done_rc = -EAGAIN;
4892                 netdev_dbg(netdev, "Failover pending, ignoring login response\n");
4893                 complete(&adapter->init_done);
4894                 /* login response buffer will be released on reset */
4895                 return 0;
4896         }
4897
4898         netdev->mtu = adapter->req_mtu - ETH_HLEN;
4899
4900         netdev_dbg(adapter->netdev, "Login Response Buffer:\n");
4901         for (i = 0; i < (adapter->login_rsp_buf_sz - 1) / 8 + 1; i++) {
4902                 netdev_dbg(adapter->netdev, "%016lx\n",
4903                            ((unsigned long *)(adapter->login_rsp_buf))[i]);
4904         }
4905
4906         /* Sanity checks */
4907         if (login->num_txcomp_subcrqs != login_rsp->num_txsubm_subcrqs ||
4908             (be32_to_cpu(login->num_rxcomp_subcrqs) *
4909              adapter->req_rx_add_queues !=
4910              be32_to_cpu(login_rsp->num_rxadd_subcrqs))) {
4911                 dev_err(dev, "FATAL: Inconsistent login and login rsp\n");
4912                 ibmvnic_reset(adapter, VNIC_RESET_FATAL);
4913                 return -EIO;
4914         }
4915         size_array = (u64 *)((u8 *)(adapter->login_rsp_buf) +
4916                 be32_to_cpu(adapter->login_rsp_buf->off_rxadd_buff_size));
4917         /* variable buffer sizes are not supported, so just read the
4918          * first entry.
4919          */
4920         adapter->cur_rx_buf_sz = be64_to_cpu(size_array[0]);
4921
4922         num_tx_pools = be32_to_cpu(adapter->login_rsp_buf->num_txsubm_subcrqs);
4923         num_rx_pools = be32_to_cpu(adapter->login_rsp_buf->num_rxadd_subcrqs);
4924
4925         tx_handle_array = (u64 *)((u8 *)(adapter->login_rsp_buf) +
4926                                   be32_to_cpu(adapter->login_rsp_buf->off_txsubm_subcrqs));
4927         rx_handle_array = (u64 *)((u8 *)(adapter->login_rsp_buf) +
4928                                   be32_to_cpu(adapter->login_rsp_buf->off_rxadd_subcrqs));
4929
4930         for (i = 0; i < num_tx_pools; i++)
4931                 adapter->tx_scrq[i]->handle = tx_handle_array[i];
4932
4933         for (i = 0; i < num_rx_pools; i++)
4934                 adapter->rx_scrq[i]->handle = rx_handle_array[i];
4935
4936         adapter->num_active_tx_scrqs = num_tx_pools;
4937         adapter->num_active_rx_scrqs = num_rx_pools;
4938         release_login_rsp_buffer(adapter);
4939         release_login_buffer(adapter);
4940         complete(&adapter->init_done);
4941
4942         return 0;
4943 }
4944
4945 static void handle_request_unmap_rsp(union ibmvnic_crq *crq,
4946                                      struct ibmvnic_adapter *adapter)
4947 {
4948         struct device *dev = &adapter->vdev->dev;
4949         long rc;
4950
4951         rc = crq->request_unmap_rsp.rc.code;
4952         if (rc)
4953                 dev_err(dev, "Error %ld in REQUEST_UNMAP_RSP\n", rc);
4954 }
4955
4956 static void handle_query_map_rsp(union ibmvnic_crq *crq,
4957                                  struct ibmvnic_adapter *adapter)
4958 {
4959         struct net_device *netdev = adapter->netdev;
4960         struct device *dev = &adapter->vdev->dev;
4961         long rc;
4962
4963         rc = crq->query_map_rsp.rc.code;
4964         if (rc) {
4965                 dev_err(dev, "Error %ld in QUERY_MAP_RSP\n", rc);
4966                 return;
4967         }
4968         netdev_dbg(netdev, "page_size = %d\ntot_pages = %u\nfree_pages = %u\n",
4969                    crq->query_map_rsp.page_size,
4970                    __be32_to_cpu(crq->query_map_rsp.tot_pages),
4971                    __be32_to_cpu(crq->query_map_rsp.free_pages));
4972 }
4973
4974 static void handle_query_cap_rsp(union ibmvnic_crq *crq,
4975                                  struct ibmvnic_adapter *adapter)
4976 {
4977         struct net_device *netdev = adapter->netdev;
4978         struct device *dev = &adapter->vdev->dev;
4979         long rc;
4980
4981         atomic_dec(&adapter->running_cap_crqs);
4982         netdev_dbg(netdev, "Outstanding queries: %d\n",
4983                    atomic_read(&adapter->running_cap_crqs));
4984         rc = crq->query_capability.rc.code;
4985         if (rc) {
4986                 dev_err(dev, "Error %ld in QUERY_CAP_RSP\n", rc);
4987                 goto out;
4988         }
4989
4990         switch (be16_to_cpu(crq->query_capability.capability)) {
4991         case MIN_TX_QUEUES:
4992                 adapter->min_tx_queues =
4993                     be64_to_cpu(crq->query_capability.number);
4994                 netdev_dbg(netdev, "min_tx_queues = %lld\n",
4995                            adapter->min_tx_queues);
4996                 break;
4997         case MIN_RX_QUEUES:
4998                 adapter->min_rx_queues =
4999                     be64_to_cpu(crq->query_capability.number);
5000                 netdev_dbg(netdev, "min_rx_queues = %lld\n",
5001                            adapter->min_rx_queues);
5002                 break;
5003         case MIN_RX_ADD_QUEUES:
5004                 adapter->min_rx_add_queues =
5005                     be64_to_cpu(crq->query_capability.number);
5006                 netdev_dbg(netdev, "min_rx_add_queues = %lld\n",
5007                            adapter->min_rx_add_queues);
5008                 break;
5009         case MAX_TX_QUEUES:
5010                 adapter->max_tx_queues =
5011                     be64_to_cpu(crq->query_capability.number);
5012                 netdev_dbg(netdev, "max_tx_queues = %lld\n",
5013                            adapter->max_tx_queues);
5014                 break;
5015         case MAX_RX_QUEUES:
5016                 adapter->max_rx_queues =
5017                     be64_to_cpu(crq->query_capability.number);
5018                 netdev_dbg(netdev, "max_rx_queues = %lld\n",
5019                            adapter->max_rx_queues);
5020                 break;
5021         case MAX_RX_ADD_QUEUES:
5022                 adapter->max_rx_add_queues =
5023                     be64_to_cpu(crq->query_capability.number);
5024                 netdev_dbg(netdev, "max_rx_add_queues = %lld\n",
5025                            adapter->max_rx_add_queues);
5026                 break;
5027         case MIN_TX_ENTRIES_PER_SUBCRQ:
5028                 adapter->min_tx_entries_per_subcrq =
5029                     be64_to_cpu(crq->query_capability.number);
5030                 netdev_dbg(netdev, "min_tx_entries_per_subcrq = %lld\n",
5031                            adapter->min_tx_entries_per_subcrq);
5032                 break;
5033         case MIN_RX_ADD_ENTRIES_PER_SUBCRQ:
5034                 adapter->min_rx_add_entries_per_subcrq =
5035                     be64_to_cpu(crq->query_capability.number);
5036                 netdev_dbg(netdev, "min_rx_add_entrs_per_subcrq = %lld\n",
5037                            adapter->min_rx_add_entries_per_subcrq);
5038                 break;
5039         case MAX_TX_ENTRIES_PER_SUBCRQ:
5040                 adapter->max_tx_entries_per_subcrq =
5041                     be64_to_cpu(crq->query_capability.number);
5042                 netdev_dbg(netdev, "max_tx_entries_per_subcrq = %lld\n",
5043                            adapter->max_tx_entries_per_subcrq);
5044                 break;
5045         case MAX_RX_ADD_ENTRIES_PER_SUBCRQ:
5046                 adapter->max_rx_add_entries_per_subcrq =
5047                     be64_to_cpu(crq->query_capability.number);
5048                 netdev_dbg(netdev, "max_rx_add_entrs_per_subcrq = %lld\n",
5049                            adapter->max_rx_add_entries_per_subcrq);
5050                 break;
5051         case TCP_IP_OFFLOAD:
5052                 adapter->tcp_ip_offload =
5053                     be64_to_cpu(crq->query_capability.number);
5054                 netdev_dbg(netdev, "tcp_ip_offload = %lld\n",
5055                            adapter->tcp_ip_offload);
5056                 break;
5057         case PROMISC_SUPPORTED:
5058                 adapter->promisc_supported =
5059                     be64_to_cpu(crq->query_capability.number);
5060                 netdev_dbg(netdev, "promisc_supported = %lld\n",
5061                            adapter->promisc_supported);
5062                 break;
5063         case MIN_MTU:
5064                 adapter->min_mtu = be64_to_cpu(crq->query_capability.number);
5065                 netdev->min_mtu = adapter->min_mtu - ETH_HLEN;
5066                 netdev_dbg(netdev, "min_mtu = %lld\n", adapter->min_mtu);
5067                 break;
5068         case MAX_MTU:
5069                 adapter->max_mtu = be64_to_cpu(crq->query_capability.number);
5070                 netdev->max_mtu = adapter->max_mtu - ETH_HLEN;
5071                 netdev_dbg(netdev, "max_mtu = %lld\n", adapter->max_mtu);
5072                 break;
5073         case MAX_MULTICAST_FILTERS:
5074                 adapter->max_multicast_filters =
5075                     be64_to_cpu(crq->query_capability.number);
5076                 netdev_dbg(netdev, "max_multicast_filters = %lld\n",
5077                            adapter->max_multicast_filters);
5078                 break;
5079         case VLAN_HEADER_INSERTION:
5080                 adapter->vlan_header_insertion =
5081                     be64_to_cpu(crq->query_capability.number);
5082                 if (adapter->vlan_header_insertion)
5083                         netdev->features |= NETIF_F_HW_VLAN_STAG_TX;
5084                 netdev_dbg(netdev, "vlan_header_insertion = %lld\n",
5085                            adapter->vlan_header_insertion);
5086                 break;
5087         case RX_VLAN_HEADER_INSERTION:
5088                 adapter->rx_vlan_header_insertion =
5089                     be64_to_cpu(crq->query_capability.number);
5090                 netdev_dbg(netdev, "rx_vlan_header_insertion = %lld\n",
5091                            adapter->rx_vlan_header_insertion);
5092                 break;
5093         case MAX_TX_SG_ENTRIES:
5094                 adapter->max_tx_sg_entries =
5095                     be64_to_cpu(crq->query_capability.number);
5096                 netdev_dbg(netdev, "max_tx_sg_entries = %lld\n",
5097                            adapter->max_tx_sg_entries);
5098                 break;
5099         case RX_SG_SUPPORTED:
5100                 adapter->rx_sg_supported =
5101                     be64_to_cpu(crq->query_capability.number);
5102                 netdev_dbg(netdev, "rx_sg_supported = %lld\n",
5103                            adapter->rx_sg_supported);
5104                 break;
5105         case OPT_TX_COMP_SUB_QUEUES:
5106                 adapter->opt_tx_comp_sub_queues =
5107                     be64_to_cpu(crq->query_capability.number);
5108                 netdev_dbg(netdev, "opt_tx_comp_sub_queues = %lld\n",
5109                            adapter->opt_tx_comp_sub_queues);
5110                 break;
5111         case OPT_RX_COMP_QUEUES:
5112                 adapter->opt_rx_comp_queues =
5113                     be64_to_cpu(crq->query_capability.number);
5114                 netdev_dbg(netdev, "opt_rx_comp_queues = %lld\n",
5115                            adapter->opt_rx_comp_queues);
5116                 break;
5117         case OPT_RX_BUFADD_Q_PER_RX_COMP_Q:
5118                 adapter->opt_rx_bufadd_q_per_rx_comp_q =
5119                     be64_to_cpu(crq->query_capability.number);
5120                 netdev_dbg(netdev, "opt_rx_bufadd_q_per_rx_comp_q = %lld\n",
5121                            adapter->opt_rx_bufadd_q_per_rx_comp_q);
5122                 break;
5123         case OPT_TX_ENTRIES_PER_SUBCRQ:
5124                 adapter->opt_tx_entries_per_subcrq =
5125                     be64_to_cpu(crq->query_capability.number);
5126                 netdev_dbg(netdev, "opt_tx_entries_per_subcrq = %lld\n",
5127                            adapter->opt_tx_entries_per_subcrq);
5128                 break;
5129         case OPT_RXBA_ENTRIES_PER_SUBCRQ:
5130                 adapter->opt_rxba_entries_per_subcrq =
5131                     be64_to_cpu(crq->query_capability.number);
5132                 netdev_dbg(netdev, "opt_rxba_entries_per_subcrq = %lld\n",
5133                            adapter->opt_rxba_entries_per_subcrq);
5134                 break;
5135         case TX_RX_DESC_REQ:
5136                 adapter->tx_rx_desc_req = crq->query_capability.number;
5137                 netdev_dbg(netdev, "tx_rx_desc_req = %llx\n",
5138                            adapter->tx_rx_desc_req);
5139                 break;
5140
5141         default:
5142                 netdev_err(netdev, "Got invalid cap rsp %d\n",
5143                            crq->query_capability.capability);
5144         }
5145
5146 out:
5147         if (atomic_read(&adapter->running_cap_crqs) == 0) {
5148                 adapter->wait_capability = false;
5149                 send_request_cap(adapter, 0);
5150         }
5151 }
5152
5153 static int send_query_phys_parms(struct ibmvnic_adapter *adapter)
5154 {
5155         union ibmvnic_crq crq;
5156         int rc;
5157
5158         memset(&crq, 0, sizeof(crq));
5159         crq.query_phys_parms.first = IBMVNIC_CRQ_CMD;
5160         crq.query_phys_parms.cmd = QUERY_PHYS_PARMS;
5161
5162         mutex_lock(&adapter->fw_lock);
5163         adapter->fw_done_rc = 0;
5164         reinit_completion(&adapter->fw_done);
5165
5166         rc = ibmvnic_send_crq(adapter, &crq);
5167         if (rc) {
5168                 mutex_unlock(&adapter->fw_lock);
5169                 return rc;
5170         }
5171
5172         rc = ibmvnic_wait_for_completion(adapter, &adapter->fw_done, 10000);
5173         if (rc) {
5174                 mutex_unlock(&adapter->fw_lock);
5175                 return rc;
5176         }
5177
5178         mutex_unlock(&adapter->fw_lock);
5179         return adapter->fw_done_rc ? -EIO : 0;
5180 }
5181
5182 static int handle_query_phys_parms_rsp(union ibmvnic_crq *crq,
5183                                        struct ibmvnic_adapter *adapter)
5184 {
5185         struct net_device *netdev = adapter->netdev;
5186         int rc;
5187         __be32 rspeed = cpu_to_be32(crq->query_phys_parms_rsp.speed);
5188
5189         rc = crq->query_phys_parms_rsp.rc.code;
5190         if (rc) {
5191                 netdev_err(netdev, "Error %d in QUERY_PHYS_PARMS\n", rc);
5192                 return rc;
5193         }
5194         switch (rspeed) {
5195         case IBMVNIC_10MBPS:
5196                 adapter->speed = SPEED_10;
5197                 break;
5198         case IBMVNIC_100MBPS:
5199                 adapter->speed = SPEED_100;
5200                 break;
5201         case IBMVNIC_1GBPS:
5202                 adapter->speed = SPEED_1000;
5203                 break;
5204         case IBMVNIC_10GBPS:
5205                 adapter->speed = SPEED_10000;
5206                 break;
5207         case IBMVNIC_25GBPS:
5208                 adapter->speed = SPEED_25000;
5209                 break;
5210         case IBMVNIC_40GBPS:
5211                 adapter->speed = SPEED_40000;
5212                 break;
5213         case IBMVNIC_50GBPS:
5214                 adapter->speed = SPEED_50000;
5215                 break;
5216         case IBMVNIC_100GBPS:
5217                 adapter->speed = SPEED_100000;
5218                 break;
5219         case IBMVNIC_200GBPS:
5220                 adapter->speed = SPEED_200000;
5221                 break;
5222         default:
5223                 if (netif_carrier_ok(netdev))
5224                         netdev_warn(netdev, "Unknown speed 0x%08x\n", rspeed);
5225                 adapter->speed = SPEED_UNKNOWN;
5226         }
5227         if (crq->query_phys_parms_rsp.flags1 & IBMVNIC_FULL_DUPLEX)
5228                 adapter->duplex = DUPLEX_FULL;
5229         else if (crq->query_phys_parms_rsp.flags1 & IBMVNIC_HALF_DUPLEX)
5230                 adapter->duplex = DUPLEX_HALF;
5231         else
5232                 adapter->duplex = DUPLEX_UNKNOWN;
5233
5234         return rc;
5235 }
5236
5237 static void ibmvnic_handle_crq(union ibmvnic_crq *crq,
5238                                struct ibmvnic_adapter *adapter)
5239 {
5240         struct ibmvnic_generic_crq *gen_crq = &crq->generic;
5241         struct net_device *netdev = adapter->netdev;
5242         struct device *dev = &adapter->vdev->dev;
5243         u64 *u64_crq = (u64 *)crq;
5244         long rc;
5245
5246         netdev_dbg(netdev, "Handling CRQ: %016lx %016lx\n",
5247                    (unsigned long)cpu_to_be64(u64_crq[0]),
5248                    (unsigned long)cpu_to_be64(u64_crq[1]));
5249         switch (gen_crq->first) {
5250         case IBMVNIC_CRQ_INIT_RSP:
5251                 switch (gen_crq->cmd) {
5252                 case IBMVNIC_CRQ_INIT:
5253                         dev_info(dev, "Partner initialized\n");
5254                         adapter->from_passive_init = true;
5255                         /* Discard any stale login responses from prev reset.
5256                          * CHECK: should we clear even on INIT_COMPLETE?
5257                          */
5258                         adapter->login_pending = false;
5259
5260                         if (adapter->state == VNIC_DOWN)
5261                                 rc = ibmvnic_reset(adapter, VNIC_RESET_PASSIVE_INIT);
5262                         else
5263                                 rc = ibmvnic_reset(adapter, VNIC_RESET_FAILOVER);
5264
5265                         if (rc && rc != -EBUSY) {
5266                                 /* We were unable to schedule the failover
5267                                  * reset either because the adapter was still
5268                                  * probing (eg: during kexec) or we could not
5269                                  * allocate memory. Clear the failover_pending
5270                                  * flag since no one else will. We ignore
5271                                  * EBUSY because it means either FAILOVER reset
5272                                  * is already scheduled or the adapter is
5273                                  * being removed.
5274                                  */
5275                                 netdev_err(netdev,
5276                                            "Error %ld scheduling failover reset\n",
5277                                            rc);
5278                                 adapter->failover_pending = false;
5279                         }
5280
5281                         if (!completion_done(&adapter->init_done)) {
5282                                 complete(&adapter->init_done);
5283                                 if (!adapter->init_done_rc)
5284                                         adapter->init_done_rc = -EAGAIN;
5285                         }
5286
5287                         break;
5288                 case IBMVNIC_CRQ_INIT_COMPLETE:
5289                         dev_info(dev, "Partner initialization complete\n");
5290                         adapter->crq.active = true;
5291                         send_version_xchg(adapter);
5292                         break;
5293                 default:
5294                         dev_err(dev, "Unknown crq cmd: %d\n", gen_crq->cmd);
5295                 }
5296                 return;
5297         case IBMVNIC_CRQ_XPORT_EVENT:
5298                 netif_carrier_off(netdev);
5299                 adapter->crq.active = false;
5300                 /* terminate any thread waiting for a response
5301                  * from the device
5302                  */
5303                 if (!completion_done(&adapter->fw_done)) {
5304                         adapter->fw_done_rc = -EIO;
5305                         complete(&adapter->fw_done);
5306                 }
5307                 if (!completion_done(&adapter->stats_done))
5308                         complete(&adapter->stats_done);
5309                 if (test_bit(0, &adapter->resetting))
5310                         adapter->force_reset_recovery = true;
5311                 if (gen_crq->cmd == IBMVNIC_PARTITION_MIGRATED) {
5312                         dev_info(dev, "Migrated, re-enabling adapter\n");
5313                         ibmvnic_reset(adapter, VNIC_RESET_MOBILITY);
5314                 } else if (gen_crq->cmd == IBMVNIC_DEVICE_FAILOVER) {
5315                         dev_info(dev, "Backing device failover detected\n");
5316                         adapter->failover_pending = true;
5317                 } else {
5318                         /* The adapter lost the connection */
5319                         dev_err(dev, "Virtual Adapter failed (rc=%d)\n",
5320                                 gen_crq->cmd);
5321                         ibmvnic_reset(adapter, VNIC_RESET_FATAL);
5322                 }
5323                 return;
5324         case IBMVNIC_CRQ_CMD_RSP:
5325                 break;
5326         default:
5327                 dev_err(dev, "Got an invalid msg type 0x%02x\n",
5328                         gen_crq->first);
5329                 return;
5330         }
5331
5332         switch (gen_crq->cmd) {
5333         case VERSION_EXCHANGE_RSP:
5334                 rc = crq->version_exchange_rsp.rc.code;
5335                 if (rc) {
5336                         dev_err(dev, "Error %ld in VERSION_EXCHG_RSP\n", rc);
5337                         break;
5338                 }
5339                 ibmvnic_version =
5340                             be16_to_cpu(crq->version_exchange_rsp.version);
5341                 dev_info(dev, "Partner protocol version is %d\n",
5342                          ibmvnic_version);
5343                 send_query_cap(adapter);
5344                 break;
5345         case QUERY_CAPABILITY_RSP:
5346                 handle_query_cap_rsp(crq, adapter);
5347                 break;
5348         case QUERY_MAP_RSP:
5349                 handle_query_map_rsp(crq, adapter);
5350                 break;
5351         case REQUEST_MAP_RSP:
5352                 adapter->fw_done_rc = crq->request_map_rsp.rc.code;
5353                 complete(&adapter->fw_done);
5354                 break;
5355         case REQUEST_UNMAP_RSP:
5356                 handle_request_unmap_rsp(crq, adapter);
5357                 break;
5358         case REQUEST_CAPABILITY_RSP:
5359                 handle_request_cap_rsp(crq, adapter);
5360                 break;
5361         case LOGIN_RSP:
5362                 netdev_dbg(netdev, "Got Login Response\n");
5363                 handle_login_rsp(crq, adapter);
5364                 break;
5365         case LOGICAL_LINK_STATE_RSP:
5366                 netdev_dbg(netdev,
5367                            "Got Logical Link State Response, state: %d rc: %d\n",
5368                            crq->logical_link_state_rsp.link_state,
5369                            crq->logical_link_state_rsp.rc.code);
5370                 adapter->logical_link_state =
5371                     crq->logical_link_state_rsp.link_state;
5372                 adapter->init_done_rc = crq->logical_link_state_rsp.rc.code;
5373                 complete(&adapter->init_done);
5374                 break;
5375         case LINK_STATE_INDICATION:
5376                 netdev_dbg(netdev, "Got Logical Link State Indication\n");
5377                 adapter->phys_link_state =
5378                     crq->link_state_indication.phys_link_state;
5379                 adapter->logical_link_state =
5380                     crq->link_state_indication.logical_link_state;
5381                 if (adapter->phys_link_state && adapter->logical_link_state)
5382                         netif_carrier_on(netdev);
5383                 else
5384                         netif_carrier_off(netdev);
5385                 break;
5386         case CHANGE_MAC_ADDR_RSP:
5387                 netdev_dbg(netdev, "Got MAC address change Response\n");
5388                 adapter->fw_done_rc = handle_change_mac_rsp(crq, adapter);
5389                 break;
5390         case ERROR_INDICATION:
5391                 netdev_dbg(netdev, "Got Error Indication\n");
5392                 handle_error_indication(crq, adapter);
5393                 break;
5394         case REQUEST_STATISTICS_RSP:
5395                 netdev_dbg(netdev, "Got Statistics Response\n");
5396                 complete(&adapter->stats_done);
5397                 break;
5398         case QUERY_IP_OFFLOAD_RSP:
5399                 netdev_dbg(netdev, "Got Query IP offload Response\n");
5400                 handle_query_ip_offload_rsp(adapter);
5401                 break;
5402         case MULTICAST_CTRL_RSP:
5403                 netdev_dbg(netdev, "Got multicast control Response\n");
5404                 break;
5405         case CONTROL_IP_OFFLOAD_RSP:
5406                 netdev_dbg(netdev, "Got Control IP offload Response\n");
5407                 dma_unmap_single(dev, adapter->ip_offload_ctrl_tok,
5408                                  sizeof(adapter->ip_offload_ctrl),
5409                                  DMA_TO_DEVICE);
5410                 complete(&adapter->init_done);
5411                 break;
5412         case COLLECT_FW_TRACE_RSP:
5413                 netdev_dbg(netdev, "Got Collect firmware trace Response\n");
5414                 complete(&adapter->fw_done);
5415                 break;
5416         case GET_VPD_SIZE_RSP:
5417                 handle_vpd_size_rsp(crq, adapter);
5418                 break;
5419         case GET_VPD_RSP:
5420                 handle_vpd_rsp(crq, adapter);
5421                 break;
5422         case QUERY_PHYS_PARMS_RSP:
5423                 adapter->fw_done_rc = handle_query_phys_parms_rsp(crq, adapter);
5424                 complete(&adapter->fw_done);
5425                 break;
5426         default:
5427                 netdev_err(netdev, "Got an invalid cmd type 0x%02x\n",
5428                            gen_crq->cmd);
5429         }
5430 }
5431
5432 static irqreturn_t ibmvnic_interrupt(int irq, void *instance)
5433 {
5434         struct ibmvnic_adapter *adapter = instance;
5435
5436         tasklet_schedule(&adapter->tasklet);
5437         return IRQ_HANDLED;
5438 }
5439
5440 static void ibmvnic_tasklet(struct tasklet_struct *t)
5441 {
5442         struct ibmvnic_adapter *adapter = from_tasklet(adapter, t, tasklet);
5443         struct ibmvnic_crq_queue *queue = &adapter->crq;
5444         union ibmvnic_crq *crq;
5445         unsigned long flags;
5446         bool done = false;
5447
5448         spin_lock_irqsave(&queue->lock, flags);
5449         while (!done) {
5450                 /* Pull all the valid messages off the CRQ */
5451                 while ((crq = ibmvnic_next_crq(adapter)) != NULL) {
5452                         /* This barrier makes sure ibmvnic_next_crq()'s
5453                          * crq->generic.first & IBMVNIC_CRQ_CMD_RSP is loaded
5454                          * before ibmvnic_handle_crq()'s
5455                          * switch(gen_crq->first) and switch(gen_crq->cmd).
5456                          */
5457                         dma_rmb();
5458                         ibmvnic_handle_crq(crq, adapter);
5459                         crq->generic.first = 0;
5460                 }
5461
5462                 /* remain in tasklet until all
5463                  * capabilities responses are received
5464                  */
5465                 if (!adapter->wait_capability)
5466                         done = true;
5467         }
5468         /* if capabilities CRQ's were sent in this tasklet, the following
5469          * tasklet must wait until all responses are received
5470          */
5471         if (atomic_read(&adapter->running_cap_crqs) != 0)
5472                 adapter->wait_capability = true;
5473         spin_unlock_irqrestore(&queue->lock, flags);
5474 }
5475
5476 static int ibmvnic_reenable_crq_queue(struct ibmvnic_adapter *adapter)
5477 {
5478         struct vio_dev *vdev = adapter->vdev;
5479         int rc;
5480
5481         do {
5482                 rc = plpar_hcall_norets(H_ENABLE_CRQ, vdev->unit_address);
5483         } while (rc == H_IN_PROGRESS || rc == H_BUSY || H_IS_LONG_BUSY(rc));
5484
5485         if (rc)
5486                 dev_err(&vdev->dev, "Error enabling adapter (rc=%d)\n", rc);
5487
5488         return rc;
5489 }
5490
5491 static int ibmvnic_reset_crq(struct ibmvnic_adapter *adapter)
5492 {
5493         struct ibmvnic_crq_queue *crq = &adapter->crq;
5494         struct device *dev = &adapter->vdev->dev;
5495         struct vio_dev *vdev = adapter->vdev;
5496         int rc;
5497
5498         /* Close the CRQ */
5499         do {
5500                 rc = plpar_hcall_norets(H_FREE_CRQ, vdev->unit_address);
5501         } while (rc == H_BUSY || H_IS_LONG_BUSY(rc));
5502
5503         /* Clean out the queue */
5504         if (!crq->msgs)
5505                 return -EINVAL;
5506
5507         memset(crq->msgs, 0, PAGE_SIZE);
5508         crq->cur = 0;
5509         crq->active = false;
5510
5511         /* And re-open it again */
5512         rc = plpar_hcall_norets(H_REG_CRQ, vdev->unit_address,
5513                                 crq->msg_token, PAGE_SIZE);
5514
5515         if (rc == H_CLOSED)
5516                 /* Adapter is good, but other end is not ready */
5517                 dev_warn(dev, "Partner adapter not ready\n");
5518         else if (rc != 0)
5519                 dev_warn(dev, "Couldn't register crq (rc=%d)\n", rc);
5520
5521         return rc;
5522 }
5523
5524 static void release_crq_queue(struct ibmvnic_adapter *adapter)
5525 {
5526         struct ibmvnic_crq_queue *crq = &adapter->crq;
5527         struct vio_dev *vdev = adapter->vdev;
5528         long rc;
5529
5530         if (!crq->msgs)
5531                 return;
5532
5533         netdev_dbg(adapter->netdev, "Releasing CRQ\n");
5534         free_irq(vdev->irq, adapter);
5535         tasklet_kill(&adapter->tasklet);
5536         do {
5537                 rc = plpar_hcall_norets(H_FREE_CRQ, vdev->unit_address);
5538         } while (rc == H_BUSY || H_IS_LONG_BUSY(rc));
5539
5540         dma_unmap_single(&vdev->dev, crq->msg_token, PAGE_SIZE,
5541                          DMA_BIDIRECTIONAL);
5542         free_page((unsigned long)crq->msgs);
5543         crq->msgs = NULL;
5544         crq->active = false;
5545 }
5546
5547 static int init_crq_queue(struct ibmvnic_adapter *adapter)
5548 {
5549         struct ibmvnic_crq_queue *crq = &adapter->crq;
5550         struct device *dev = &adapter->vdev->dev;
5551         struct vio_dev *vdev = adapter->vdev;
5552         int rc, retrc = -ENOMEM;
5553
5554         if (crq->msgs)
5555                 return 0;
5556
5557         crq->msgs = (union ibmvnic_crq *)get_zeroed_page(GFP_KERNEL);
5558         /* Should we allocate more than one page? */
5559
5560         if (!crq->msgs)
5561                 return -ENOMEM;
5562
5563         crq->size = PAGE_SIZE / sizeof(*crq->msgs);
5564         crq->msg_token = dma_map_single(dev, crq->msgs, PAGE_SIZE,
5565                                         DMA_BIDIRECTIONAL);
5566         if (dma_mapping_error(dev, crq->msg_token))
5567                 goto map_failed;
5568
5569         rc = plpar_hcall_norets(H_REG_CRQ, vdev->unit_address,
5570                                 crq->msg_token, PAGE_SIZE);
5571
5572         if (rc == H_RESOURCE)
5573                 /* maybe kexecing and resource is busy. try a reset */
5574                 rc = ibmvnic_reset_crq(adapter);
5575         retrc = rc;
5576
5577         if (rc == H_CLOSED) {
5578                 dev_warn(dev, "Partner adapter not ready\n");
5579         } else if (rc) {
5580                 dev_warn(dev, "Error %d opening adapter\n", rc);
5581                 goto reg_crq_failed;
5582         }
5583
5584         retrc = 0;
5585
5586         tasklet_setup(&adapter->tasklet, (void *)ibmvnic_tasklet);
5587
5588         netdev_dbg(adapter->netdev, "registering irq 0x%x\n", vdev->irq);
5589         snprintf(crq->name, sizeof(crq->name), "ibmvnic-%x",
5590                  adapter->vdev->unit_address);
5591         rc = request_irq(vdev->irq, ibmvnic_interrupt, 0, crq->name, adapter);
5592         if (rc) {
5593                 dev_err(dev, "Couldn't register irq 0x%x. rc=%d\n",
5594                         vdev->irq, rc);
5595                 goto req_irq_failed;
5596         }
5597
5598         rc = vio_enable_interrupts(vdev);
5599         if (rc) {
5600                 dev_err(dev, "Error %d enabling interrupts\n", rc);
5601                 goto req_irq_failed;
5602         }
5603
5604         crq->cur = 0;
5605         spin_lock_init(&crq->lock);
5606
5607         /* process any CRQs that were queued before we enabled interrupts */
5608         tasklet_schedule(&adapter->tasklet);
5609
5610         return retrc;
5611
5612 req_irq_failed:
5613         tasklet_kill(&adapter->tasklet);
5614         do {
5615                 rc = plpar_hcall_norets(H_FREE_CRQ, vdev->unit_address);
5616         } while (rc == H_BUSY || H_IS_LONG_BUSY(rc));
5617 reg_crq_failed:
5618         dma_unmap_single(dev, crq->msg_token, PAGE_SIZE, DMA_BIDIRECTIONAL);
5619 map_failed:
5620         free_page((unsigned long)crq->msgs);
5621         crq->msgs = NULL;
5622         return retrc;
5623 }
5624
5625 static int ibmvnic_reset_init(struct ibmvnic_adapter *adapter, bool reset)
5626 {
5627         struct device *dev = &adapter->vdev->dev;
5628         unsigned long timeout = msecs_to_jiffies(20000);
5629         u64 old_num_rx_queues = adapter->req_rx_queues;
5630         u64 old_num_tx_queues = adapter->req_tx_queues;
5631         int rc;
5632
5633         adapter->from_passive_init = false;
5634
5635         if (reset)
5636                 reinit_completion(&adapter->init_done);
5637
5638         adapter->init_done_rc = 0;
5639         rc = ibmvnic_send_crq_init(adapter);
5640         if (rc) {
5641                 dev_err(dev, "Send crq init failed with error %d\n", rc);
5642                 return rc;
5643         }
5644
5645         if (!wait_for_completion_timeout(&adapter->init_done, timeout)) {
5646                 dev_err(dev, "Initialization sequence timed out\n");
5647                 return -1;
5648         }
5649
5650         if (adapter->init_done_rc) {
5651                 release_crq_queue(adapter);
5652                 return adapter->init_done_rc;
5653         }
5654
5655         if (adapter->from_passive_init) {
5656                 adapter->state = VNIC_OPEN;
5657                 adapter->from_passive_init = false;
5658                 return -1;
5659         }
5660
5661         if (reset &&
5662             test_bit(0, &adapter->resetting) && !adapter->wait_for_reset &&
5663             adapter->reset_reason != VNIC_RESET_MOBILITY) {
5664                 if (adapter->req_rx_queues != old_num_rx_queues ||
5665                     adapter->req_tx_queues != old_num_tx_queues) {
5666                         release_sub_crqs(adapter, 0);
5667                         rc = init_sub_crqs(adapter);
5668                 } else {
5669                         rc = reset_sub_crq_queues(adapter);
5670                 }
5671         } else {
5672                 rc = init_sub_crqs(adapter);
5673         }
5674
5675         if (rc) {
5676                 dev_err(dev, "Initialization of sub crqs failed\n");
5677                 release_crq_queue(adapter);
5678                 return rc;
5679         }
5680
5681         rc = init_sub_crq_irqs(adapter);
5682         if (rc) {
5683                 dev_err(dev, "Failed to initialize sub crq irqs\n");
5684                 release_crq_queue(adapter);
5685         }
5686
5687         return rc;
5688 }
5689
5690 static struct device_attribute dev_attr_failover;
5691
5692 static int ibmvnic_probe(struct vio_dev *dev, const struct vio_device_id *id)
5693 {
5694         struct ibmvnic_adapter *adapter;
5695         struct net_device *netdev;
5696         unsigned char *mac_addr_p;
5697         bool init_success;
5698         int rc;
5699
5700         dev_dbg(&dev->dev, "entering ibmvnic_probe for UA 0x%x\n",
5701                 dev->unit_address);
5702
5703         mac_addr_p = (unsigned char *)vio_get_attribute(dev,
5704                                                         VETH_MAC_ADDR, NULL);
5705         if (!mac_addr_p) {
5706                 dev_err(&dev->dev,
5707                         "(%s:%3.3d) ERROR: Can't find MAC_ADDR attribute\n",
5708                         __FILE__, __LINE__);
5709                 return 0;
5710         }
5711
5712         netdev = alloc_etherdev_mq(sizeof(struct ibmvnic_adapter),
5713                                    IBMVNIC_MAX_QUEUES);
5714         if (!netdev)
5715                 return -ENOMEM;
5716
5717         adapter = netdev_priv(netdev);
5718         adapter->state = VNIC_PROBING;
5719         dev_set_drvdata(&dev->dev, netdev);
5720         adapter->vdev = dev;
5721         adapter->netdev = netdev;
5722         adapter->login_pending = false;
5723         memset(&adapter->map_ids, 0, sizeof(adapter->map_ids));
5724         /* map_ids start at 1, so ensure map_id 0 is always "in-use" */
5725         bitmap_set(adapter->map_ids, 0, 1);
5726
5727         ether_addr_copy(adapter->mac_addr, mac_addr_p);
5728         eth_hw_addr_set(netdev, adapter->mac_addr);
5729         netdev->irq = dev->irq;
5730         netdev->netdev_ops = &ibmvnic_netdev_ops;
5731         netdev->ethtool_ops = &ibmvnic_ethtool_ops;
5732         SET_NETDEV_DEV(netdev, &dev->dev);
5733
5734         INIT_WORK(&adapter->ibmvnic_reset, __ibmvnic_reset);
5735         INIT_DELAYED_WORK(&adapter->ibmvnic_delayed_reset,
5736                           __ibmvnic_delayed_reset);
5737         INIT_LIST_HEAD(&adapter->rwi_list);
5738         spin_lock_init(&adapter->rwi_lock);
5739         spin_lock_init(&adapter->state_lock);
5740         mutex_init(&adapter->fw_lock);
5741         init_completion(&adapter->init_done);
5742         init_completion(&adapter->fw_done);
5743         init_completion(&adapter->reset_done);
5744         init_completion(&adapter->stats_done);
5745         clear_bit(0, &adapter->resetting);
5746         adapter->prev_rx_buf_sz = 0;
5747         adapter->prev_mtu = 0;
5748
5749         init_success = false;
5750         do {
5751                 rc = init_crq_queue(adapter);
5752                 if (rc) {
5753                         dev_err(&dev->dev, "Couldn't initialize crq. rc=%d\n",
5754                                 rc);
5755                         goto ibmvnic_init_fail;
5756                 }
5757
5758                 rc = ibmvnic_reset_init(adapter, false);
5759         } while (rc == -EAGAIN);
5760
5761         /* We are ignoring the error from ibmvnic_reset_init() assuming that the
5762          * partner is not ready. CRQ is not active. When the partner becomes
5763          * ready, we will do the passive init reset.
5764          */
5765
5766         if (!rc)
5767                 init_success = true;
5768
5769         rc = init_stats_buffers(adapter);
5770         if (rc)
5771                 goto ibmvnic_init_fail;
5772
5773         rc = init_stats_token(adapter);
5774         if (rc)
5775                 goto ibmvnic_stats_fail;
5776
5777         rc = device_create_file(&dev->dev, &dev_attr_failover);
5778         if (rc)
5779                 goto ibmvnic_dev_file_err;
5780
5781         netif_carrier_off(netdev);
5782         rc = register_netdev(netdev);
5783         if (rc) {
5784                 dev_err(&dev->dev, "failed to register netdev rc=%d\n", rc);
5785                 goto ibmvnic_register_fail;
5786         }
5787         dev_info(&dev->dev, "ibmvnic registered\n");
5788
5789         if (init_success) {
5790                 adapter->state = VNIC_PROBED;
5791                 netdev->mtu = adapter->req_mtu - ETH_HLEN;
5792                 netdev->min_mtu = adapter->min_mtu - ETH_HLEN;
5793                 netdev->max_mtu = adapter->max_mtu - ETH_HLEN;
5794         } else {
5795                 adapter->state = VNIC_DOWN;
5796         }
5797
5798         adapter->wait_for_reset = false;
5799         adapter->last_reset_time = jiffies;
5800         return 0;
5801
5802 ibmvnic_register_fail:
5803         device_remove_file(&dev->dev, &dev_attr_failover);
5804
5805 ibmvnic_dev_file_err:
5806         release_stats_token(adapter);
5807
5808 ibmvnic_stats_fail:
5809         release_stats_buffers(adapter);
5810
5811 ibmvnic_init_fail:
5812         release_sub_crqs(adapter, 1);
5813         release_crq_queue(adapter);
5814         mutex_destroy(&adapter->fw_lock);
5815         free_netdev(netdev);
5816
5817         return rc;
5818 }
5819
5820 static void ibmvnic_remove(struct vio_dev *dev)
5821 {
5822         struct net_device *netdev = dev_get_drvdata(&dev->dev);
5823         struct ibmvnic_adapter *adapter = netdev_priv(netdev);
5824         unsigned long flags;
5825
5826         spin_lock_irqsave(&adapter->state_lock, flags);
5827
5828         /* If ibmvnic_reset() is scheduling a reset, wait for it to
5829          * finish. Then, set the state to REMOVING to prevent it from
5830          * scheduling any more work and to have reset functions ignore
5831          * any resets that have already been scheduled. Drop the lock
5832          * after setting state, so __ibmvnic_reset() which is called
5833          * from the flush_work() below, can make progress.
5834          */
5835         spin_lock(&adapter->rwi_lock);
5836         adapter->state = VNIC_REMOVING;
5837         spin_unlock(&adapter->rwi_lock);
5838
5839         spin_unlock_irqrestore(&adapter->state_lock, flags);
5840
5841         flush_work(&adapter->ibmvnic_reset);
5842         flush_delayed_work(&adapter->ibmvnic_delayed_reset);
5843
5844         rtnl_lock();
5845         unregister_netdevice(netdev);
5846
5847         release_resources(adapter);
5848         release_rx_pools(adapter);
5849         release_tx_pools(adapter);
5850         release_sub_crqs(adapter, 1);
5851         release_crq_queue(adapter);
5852
5853         release_stats_token(adapter);
5854         release_stats_buffers(adapter);
5855
5856         adapter->state = VNIC_REMOVED;
5857
5858         rtnl_unlock();
5859         mutex_destroy(&adapter->fw_lock);
5860         device_remove_file(&dev->dev, &dev_attr_failover);
5861         free_netdev(netdev);
5862         dev_set_drvdata(&dev->dev, NULL);
5863 }
5864
5865 static ssize_t failover_store(struct device *dev, struct device_attribute *attr,
5866                               const char *buf, size_t count)
5867 {
5868         struct net_device *netdev = dev_get_drvdata(dev);
5869         struct ibmvnic_adapter *adapter = netdev_priv(netdev);
5870         unsigned long retbuf[PLPAR_HCALL_BUFSIZE];
5871         __be64 session_token;
5872         long rc;
5873
5874         if (!sysfs_streq(buf, "1"))
5875                 return -EINVAL;
5876
5877         rc = plpar_hcall(H_VIOCTL, retbuf, adapter->vdev->unit_address,
5878                          H_GET_SESSION_TOKEN, 0, 0, 0);
5879         if (rc) {
5880                 netdev_err(netdev, "Couldn't retrieve session token, rc %ld\n",
5881                            rc);
5882                 goto last_resort;
5883         }
5884
5885         session_token = (__be64)retbuf[0];
5886         netdev_dbg(netdev, "Initiating client failover, session id %llx\n",
5887                    be64_to_cpu(session_token));
5888         rc = plpar_hcall_norets(H_VIOCTL, adapter->vdev->unit_address,
5889                                 H_SESSION_ERR_DETECTED, session_token, 0, 0);
5890         if (rc)
5891                 netdev_err(netdev,
5892                            "H_VIOCTL initiated failover failed, rc %ld\n",
5893                            rc);
5894
5895 last_resort:
5896         netdev_dbg(netdev, "Trying to send CRQ_CMD, the last resort\n");
5897         ibmvnic_reset(adapter, VNIC_RESET_FAILOVER);
5898
5899         return count;
5900 }
5901 static DEVICE_ATTR_WO(failover);
5902
5903 static unsigned long ibmvnic_get_desired_dma(struct vio_dev *vdev)
5904 {
5905         struct net_device *netdev = dev_get_drvdata(&vdev->dev);
5906         struct ibmvnic_adapter *adapter;
5907         struct iommu_table *tbl;
5908         unsigned long ret = 0;
5909         int i;
5910
5911         tbl = get_iommu_table_base(&vdev->dev);
5912
5913         /* netdev inits at probe time along with the structures we need below*/
5914         if (!netdev)
5915                 return IOMMU_PAGE_ALIGN(IBMVNIC_IO_ENTITLEMENT_DEFAULT, tbl);
5916
5917         adapter = netdev_priv(netdev);
5918
5919         ret += PAGE_SIZE; /* the crq message queue */
5920         ret += IOMMU_PAGE_ALIGN(sizeof(struct ibmvnic_statistics), tbl);
5921
5922         for (i = 0; i < adapter->req_tx_queues + adapter->req_rx_queues; i++)
5923                 ret += 4 * PAGE_SIZE; /* the scrq message queue */
5924
5925         for (i = 0; i < adapter->num_active_rx_pools; i++)
5926                 ret += adapter->rx_pool[i].size *
5927                     IOMMU_PAGE_ALIGN(adapter->rx_pool[i].buff_size, tbl);
5928
5929         return ret;
5930 }
5931
5932 static int ibmvnic_resume(struct device *dev)
5933 {
5934         struct net_device *netdev = dev_get_drvdata(dev);
5935         struct ibmvnic_adapter *adapter = netdev_priv(netdev);
5936
5937         if (adapter->state != VNIC_OPEN)
5938                 return 0;
5939
5940         tasklet_schedule(&adapter->tasklet);
5941
5942         return 0;
5943 }
5944
5945 static const struct vio_device_id ibmvnic_device_table[] = {
5946         {"network", "IBM,vnic"},
5947         {"", "" }
5948 };
5949 MODULE_DEVICE_TABLE(vio, ibmvnic_device_table);
5950
5951 static const struct dev_pm_ops ibmvnic_pm_ops = {
5952         .resume = ibmvnic_resume
5953 };
5954
5955 static struct vio_driver ibmvnic_driver = {
5956         .id_table       = ibmvnic_device_table,
5957         .probe          = ibmvnic_probe,
5958         .remove         = ibmvnic_remove,
5959         .get_desired_dma = ibmvnic_get_desired_dma,
5960         .name           = ibmvnic_driver_name,
5961         .pm             = &ibmvnic_pm_ops,
5962 };
5963
5964 /* module functions */
5965 static int __init ibmvnic_module_init(void)
5966 {
5967         pr_info("%s: %s %s\n", ibmvnic_driver_name, ibmvnic_driver_string,
5968                 IBMVNIC_DRIVER_VERSION);
5969
5970         return vio_register_driver(&ibmvnic_driver);
5971 }
5972
5973 static void __exit ibmvnic_module_exit(void)
5974 {
5975         vio_unregister_driver(&ibmvnic_driver);
5976 }
5977
5978 module_init(ibmvnic_module_init);
5979 module_exit(ibmvnic_module_exit);