drivers/net/ethernet/ibm/ibmvnic.c

   1 // SPDX-License-Identifier: GPL-2.0-or-later
   2 /**************************************************************************/
   3 /*                                                                        */
   4 /*  IBM System i and System p Virtual NIC Device Driver                   */
   5 /*  Copyright (C) 2014 IBM Corp.                                          */
   6 /*  Santiago Leon (santi_leon@yahoo.com)                                  */
   7 /*  Thomas Falcon (tlfalcon@linux.vnet.ibm.com)                           */
   8 /*  John Allen (jallen@linux.vnet.ibm.com)                                */
   9 /*                                                                        */
  10 /*                                                                        */
  11 /* This module contains the implementation of a virtual ethernet device   */
  12 /* for use with IBM i/p Series LPAR Linux. It utilizes the logical LAN    */
  13 /* option of the RS/6000 Platform Architecture to interface with virtual  */
  14 /* ethernet NICs that are presented to the partition by the hypervisor.   */
  15 /*                                                                         */
  16 /* Messages are passed between the VNIC driver and the VNIC server using  */
  17 /* Command/Response Queues (CRQs) and sub CRQs (sCRQs). CRQs are used to  */
  18 /* issue and receive commands that initiate communication with the server */
  19 /* on driver initialization. Sub CRQs (sCRQs) are similar to CRQs, but    */
  20 /* are used by the driver to notify the server that a packet is           */
  21 /* ready for transmission or that a buffer has been added to receive a    */
  22 /* packet. Subsequently, sCRQs are used by the server to notify the       */
  23 /* driver that a packet transmission has been completed or that a packet  */
  24 /* has been received and placed in a waiting buffer.                      */
  25 /*                                                                        */
  26 /* In lieu of a more conventional "on-the-fly" DMA mapping strategy in    */
  27 /* which skbs are DMA mapped and immediately unmapped when the transmit   */
  28 /* or receive has been completed, the VNIC driver is required to use      */
  29 /* "long term mapping". This entails that large, continuous DMA mapped    */
  30 /* buffers are allocated on driver initialization and these buffers are   */
  31 /* then continuously reused to pass skbs to and from the VNIC server.     */
  32 /*                                                                        */
  33 /**************************************************************************/
  34
  35 #include <linux/module.h>
  36 #include <linux/moduleparam.h>
  37 #include <linux/types.h>
  38 #include <linux/errno.h>
  39 #include <linux/completion.h>
  40 #include <linux/ioport.h>
  41 #include <linux/dma-mapping.h>
  42 #include <linux/kernel.h>
  43 #include <linux/netdevice.h>
  44 #include <linux/etherdevice.h>
  45 #include <linux/skbuff.h>
  46 #include <linux/init.h>
  47 #include <linux/delay.h>
  48 #include <linux/mm.h>
  49 #include <linux/ethtool.h>
  50 #include <linux/proc_fs.h>
  51 #include <linux/if_arp.h>
  52 #include <linux/in.h>
  53 #include <linux/ip.h>
  54 #include <linux/ipv6.h>
  55 #include <linux/irq.h>
  56 #include <linux/irqdomain.h>
  57 #include <linux/kthread.h>
  58 #include <linux/seq_file.h>
  59 #include <linux/interrupt.h>
  60 #include <net/net_namespace.h>
  61 #include <asm/hvcall.h>
  62 #include <linux/atomic.h>
  63 #include <asm/vio.h>
  64 #include <asm/xive.h>
  65 #include <asm/iommu.h>
  66 #include <linux/uaccess.h>
  67 #include <asm/firmware.h>
  68 #include <linux/workqueue.h>
  69 #include <linux/if_vlan.h>
  70 #include <linux/utsname.h>
  71
  72 #include "ibmvnic.h"
  73
  74 static const char ibmvnic_driver_name[] = "ibmvnic";
  75 static const char ibmvnic_driver_string[] = "IBM System i/p Virtual NIC Driver";
  76
  77 MODULE_AUTHOR("Santiago Leon");
  78 MODULE_DESCRIPTION("IBM System i/p Virtual NIC Driver");
  79 MODULE_LICENSE("GPL");
  80 MODULE_VERSION(IBMVNIC_DRIVER_VERSION);
  81
  82 static int ibmvnic_version = IBMVNIC_INITIAL_VERSION;
  83 static void release_sub_crqs(struct ibmvnic_adapter *, bool);
  84 static int ibmvnic_reset_crq(struct ibmvnic_adapter *);
  85 static int ibmvnic_send_crq_init(struct ibmvnic_adapter *);
  86 static int ibmvnic_reenable_crq_queue(struct ibmvnic_adapter *);
  87 static int ibmvnic_send_crq(struct ibmvnic_adapter *, union ibmvnic_crq *);
  88 static int send_subcrq_indirect(struct ibmvnic_adapter *, u64, u64, u64);
  89 static irqreturn_t ibmvnic_interrupt_rx(int irq, void *instance);
  90 static int enable_scrq_irq(struct ibmvnic_adapter *,
  91                            struct ibmvnic_sub_crq_queue *);
  92 static int disable_scrq_irq(struct ibmvnic_adapter *,
  93                             struct ibmvnic_sub_crq_queue *);
  94 static int pending_scrq(struct ibmvnic_adapter *,
  95                         struct ibmvnic_sub_crq_queue *);
  96 static union sub_crq *ibmvnic_next_scrq(struct ibmvnic_adapter *,
  97                                         struct ibmvnic_sub_crq_queue *);
  98 static int ibmvnic_poll(struct napi_struct *napi, int data);
  99 static void send_query_map(struct ibmvnic_adapter *adapter);
 100 static int send_request_map(struct ibmvnic_adapter *, dma_addr_t, u32, u8);
 101 static int send_request_unmap(struct ibmvnic_adapter *, u8);
 102 static int send_login(struct ibmvnic_adapter *adapter);
 103 static void send_query_cap(struct ibmvnic_adapter *adapter);
 104 static int init_sub_crqs(struct ibmvnic_adapter *);
 105 static int init_sub_crq_irqs(struct ibmvnic_adapter *adapter);
 106 static int ibmvnic_reset_init(struct ibmvnic_adapter *, bool reset);
 107 static void release_crq_queue(struct ibmvnic_adapter *);
 108 static int __ibmvnic_set_mac(struct net_device *, u8 *);
 109 static int init_crq_queue(struct ibmvnic_adapter *adapter);
 110 static int send_query_phys_parms(struct ibmvnic_adapter *adapter);
 111 static void ibmvnic_tx_scrq_clean_buffer(struct ibmvnic_adapter *adapter,
 112                                          struct ibmvnic_sub_crq_queue *tx_scrq);
 113 static void free_long_term_buff(struct ibmvnic_adapter *adapter,
 114                                 struct ibmvnic_long_term_buff *ltb);
 115 static void ibmvnic_disable_irqs(struct ibmvnic_adapter *adapter);
 116
 117 struct ibmvnic_stat {
 118         char name[ETH_GSTRING_LEN];
 119         int offset;
 120 };
 121
 122 #define IBMVNIC_STAT_OFF(stat) (offsetof(struct ibmvnic_adapter, stats) + \
 123                              offsetof(struct ibmvnic_statistics, stat))
 124 #define IBMVNIC_GET_STAT(a, off) (*((u64 *)(((unsigned long)(a)) + (off))))
 125
 126 static const struct ibmvnic_stat ibmvnic_stats[] = {
 127         {"rx_packets", IBMVNIC_STAT_OFF(rx_packets)},
 128         {"rx_bytes", IBMVNIC_STAT_OFF(rx_bytes)},
 129         {"tx_packets", IBMVNIC_STAT_OFF(tx_packets)},
 130         {"tx_bytes", IBMVNIC_STAT_OFF(tx_bytes)},
 131         {"ucast_tx_packets", IBMVNIC_STAT_OFF(ucast_tx_packets)},
 132         {"ucast_rx_packets", IBMVNIC_STAT_OFF(ucast_rx_packets)},
 133         {"mcast_tx_packets", IBMVNIC_STAT_OFF(mcast_tx_packets)},
 134         {"mcast_rx_packets", IBMVNIC_STAT_OFF(mcast_rx_packets)},
 135         {"bcast_tx_packets", IBMVNIC_STAT_OFF(bcast_tx_packets)},
 136         {"bcast_rx_packets", IBMVNIC_STAT_OFF(bcast_rx_packets)},
 137         {"align_errors", IBMVNIC_STAT_OFF(align_errors)},
 138         {"fcs_errors", IBMVNIC_STAT_OFF(fcs_errors)},
 139         {"single_collision_frames", IBMVNIC_STAT_OFF(single_collision_frames)},
 140         {"multi_collision_frames", IBMVNIC_STAT_OFF(multi_collision_frames)},
 141         {"sqe_test_errors", IBMVNIC_STAT_OFF(sqe_test_errors)},
 142         {"deferred_tx", IBMVNIC_STAT_OFF(deferred_tx)},
 143         {"late_collisions", IBMVNIC_STAT_OFF(late_collisions)},
 144         {"excess_collisions", IBMVNIC_STAT_OFF(excess_collisions)},
 145         {"internal_mac_tx_errors", IBMVNIC_STAT_OFF(internal_mac_tx_errors)},
 146         {"carrier_sense", IBMVNIC_STAT_OFF(carrier_sense)},
 147         {"too_long_frames", IBMVNIC_STAT_OFF(too_long_frames)},
 148         {"internal_mac_rx_errors", IBMVNIC_STAT_OFF(internal_mac_rx_errors)},
 149 };
 150
 151 static int send_crq_init_complete(struct ibmvnic_adapter *adapter)
 152 {
 153         union ibmvnic_crq crq;
 154
 155         memset(&crq, 0, sizeof(crq));
 156         crq.generic.first = IBMVNIC_CRQ_INIT_CMD;
 157         crq.generic.cmd = IBMVNIC_CRQ_INIT_COMPLETE;
 158
 159         return ibmvnic_send_crq(adapter, &crq);
 160 }
 161
 162 static int send_version_xchg(struct ibmvnic_adapter *adapter)
 163 {
 164         union ibmvnic_crq crq;
 165
 166         memset(&crq, 0, sizeof(crq));
 167         crq.version_exchange.first = IBMVNIC_CRQ_CMD;
 168         crq.version_exchange.cmd = VERSION_EXCHANGE;
 169         crq.version_exchange.version = cpu_to_be16(ibmvnic_version);
 170
 171         return ibmvnic_send_crq(adapter, &crq);
 172 }
 173
 174 static long h_reg_sub_crq(unsigned long unit_address, unsigned long token,
 175                           unsigned long length, unsigned long *number,
 176                           unsigned long *irq)
 177 {
 178         unsigned long retbuf[PLPAR_HCALL_BUFSIZE];
 179         long rc;
 180
 181         rc = plpar_hcall(H_REG_SUB_CRQ, retbuf, unit_address, token, length);
 182         *number = retbuf[0];
 183         *irq = retbuf[1];
 184
 185         return rc;
 186 }
 187
 188 /**
 189  * ibmvnic_wait_for_completion - Check device state and wait for completion
 190  * @adapter: private device data
 191  * @comp_done: completion structure to wait for
 192  * @timeout: time to wait in milliseconds
 193  *
 194  * Wait for a completion signal or until the timeout limit is reached
 195  * while checking that the device is still active.
 196  */
 197 static int ibmvnic_wait_for_completion(struct ibmvnic_adapter *adapter,
 198                                        struct completion *comp_done,
 199                                        unsigned long timeout)
 200 {
 201         struct net_device *netdev;
 202         unsigned long div_timeout;
 203         u8 retry;
 204
 205         netdev = adapter->netdev;
 206         retry = 5;
 207         div_timeout = msecs_to_jiffies(timeout / retry);
 208         while (true) {
 209                 if (!adapter->crq.active) {
 210                         netdev_err(netdev, "Device down!\n");
 211                         return -ENODEV;
 212                 }
 213                 if (!retry--)
 214                         break;
 215                 if (wait_for_completion_timeout(comp_done, div_timeout))
 216                         return 0;
 217         }
 218         netdev_err(netdev, "Operation timed out.\n");
 219         return -ETIMEDOUT;
 220 }
 221
 222 /**
 223  * reuse_ltb() - Check if a long term buffer can be reused
 224  * @ltb:  The long term buffer to be checked
 225  * @size: The size of the long term buffer.
 226  *
 227  * An LTB can be reused unless its size has changed.
 228  *
 229  * Return: Return true if the LTB can be reused, false otherwise.
 230  */
 231 static bool reuse_ltb(struct ibmvnic_long_term_buff *ltb, int size)
 232 {
 233         return (ltb->buff && ltb->size == size);
 234 }
 235
 236 /**
 237  * alloc_long_term_buff() - Allocate a long term buffer (LTB)
 238  *
 239  * @adapter: ibmvnic adapter associated to the LTB
 240  * @ltb:     container object for the LTB
 241  * @size:    size of the LTB
 242  *
 243  * Allocate an LTB of the specified size and notify VIOS.
 244  *
 245  * If the given @ltb already has the correct size, reuse it. Otherwise if
 246  * its non-NULL, free it. Then allocate a new one of the correct size.
 247  * Notify the VIOS either way since we may now be working with a new VIOS.
 248  *
 249  * Allocating larger chunks of memory during resets, specially LPM or under
 250  * low memory situations can cause resets to fail/timeout and for LPAR to
 251  * lose connectivity. So hold onto the LTB even if we fail to communicate
 252  * with the VIOS and reuse it on next open. Free LTB when adapter is closed.
 253  *
 254  * Return: 0 if we were able to allocate the LTB and notify the VIOS and
 255  *         a negative value otherwise.
 256  */
 257 static int alloc_long_term_buff(struct ibmvnic_adapter *adapter,
 258                                 struct ibmvnic_long_term_buff *ltb, int size)
 259 {
 260         struct device *dev = &adapter->vdev->dev;
 261         u64 prev = 0;
 262         int rc;
 263
 264         if (!reuse_ltb(ltb, size)) {
 265                 dev_dbg(dev,
 266                         "LTB size changed from 0x%llx to 0x%x, reallocating\n",
 267                          ltb->size, size);
 268                 prev = ltb->size;
 269                 free_long_term_buff(adapter, ltb);
 270         }
 271
 272         if (ltb->buff) {
 273                 dev_dbg(dev, "Reusing LTB [map %d, size 0x%llx]\n",
 274                         ltb->map_id, ltb->size);
 275         } else {
 276                 ltb->buff = dma_alloc_coherent(dev, size, &ltb->addr,
 277                                                GFP_KERNEL);
 278                 if (!ltb->buff) {
 279                         dev_err(dev, "Couldn't alloc long term buffer\n");
 280                         return -ENOMEM;
 281                 }
 282                 ltb->size = size;
 283
 284                 ltb->map_id = find_first_zero_bit(adapter->map_ids,
 285                                                   MAX_MAP_ID);
 286                 bitmap_set(adapter->map_ids, ltb->map_id, 1);
 287
 288                 dev_dbg(dev,
 289                         "Allocated new LTB [map %d, size 0x%llx was 0x%llx]\n",
 290                          ltb->map_id, ltb->size, prev);
 291         }
 292
 293         /* Ensure ltb is zeroed - specially when reusing it. */
 294         memset(ltb->buff, 0, ltb->size);
 295
 296         mutex_lock(&adapter->fw_lock);
 297         adapter->fw_done_rc = 0;
 298         reinit_completion(&adapter->fw_done);
 299
 300         rc = send_request_map(adapter, ltb->addr, ltb->size, ltb->map_id);
 301         if (rc) {
 302                 dev_err(dev, "send_request_map failed, rc = %d\n", rc);
 303                 goto out;
 304         }
 305
 306         rc = ibmvnic_wait_for_completion(adapter, &adapter->fw_done, 10000);
 307         if (rc) {
 308                 dev_err(dev, "LTB map request aborted or timed out, rc = %d\n",
 309                         rc);
 310                 goto out;
 311         }
 312
 313         if (adapter->fw_done_rc) {
 314                 dev_err(dev, "Couldn't map LTB, rc = %d\n",
 315                         adapter->fw_done_rc);
 316                 rc = -EIO;
 317                 goto out;
 318         }
 319         rc = 0;
 320 out:
 321         /* don't free LTB on communication error - see function header */
 322         mutex_unlock(&adapter->fw_lock);
 323         return rc;
 324 }
 325
 326 static void free_long_term_buff(struct ibmvnic_adapter *adapter,
 327                                 struct ibmvnic_long_term_buff *ltb)
 328 {
 329         struct device *dev = &adapter->vdev->dev;
 330
 331         if (!ltb->buff)
 332                 return;
 333
 334         /* VIOS automatically unmaps the long term buffer at remote
 335          * end for the following resets:
 336          * FAILOVER, MOBILITY, TIMEOUT.
 337          */
 338         if (adapter->reset_reason != VNIC_RESET_FAILOVER &&
 339             adapter->reset_reason != VNIC_RESET_MOBILITY &&
 340             adapter->reset_reason != VNIC_RESET_TIMEOUT)
 341                 send_request_unmap(adapter, ltb->map_id);
 342
 343         dma_free_coherent(dev, ltb->size, ltb->buff, ltb->addr);
 344
 345         ltb->buff = NULL;
 346         /* mark this map_id free */
 347         bitmap_clear(adapter->map_ids, ltb->map_id, 1);
 348         ltb->map_id = 0;
 349 }
 350
 351 /**
 352  * free_ltb_set - free the given set of long term buffers (LTBS)
 353  * @adapter: The ibmvnic adapter containing this ltb set
 354  * @ltb_set: The ltb_set to be freed
 355  *
 356  * Free the set of LTBs in the given set.
 357  */
 358
 359 static void free_ltb_set(struct ibmvnic_adapter *adapter,
 360                          struct ibmvnic_ltb_set *ltb_set)
 361 {
 362         int i;
 363
 364         for (i = 0; i < ltb_set->num_ltbs; i++)
 365                 free_long_term_buff(adapter, &ltb_set->ltbs[i]);
 366
 367         kfree(ltb_set->ltbs);
 368         ltb_set->ltbs = NULL;
 369         ltb_set->num_ltbs = 0;
 370 }
 371
 372 /**
 373  * alloc_ltb_set() - Allocate a set of long term buffers (LTBs)
 374  *
 375  * @adapter: ibmvnic adapter associated to the LTB
 376  * @ltb_set: container object for the set of LTBs
 377  * @num_buffs: Number of buffers in the LTB
 378  * @buff_size: Size of each buffer in the LTB
 379  *
 380  * Allocate a set of LTBs to accommodate @num_buffs buffers of @buff_size
 381  * each. We currently cap size each LTB to IBMVNIC_ONE_LTB_SIZE. If the
 382  * new set of LTBs have fewer LTBs than the old set, free the excess LTBs.
 383  * If new set needs more than in old set, allocate the remaining ones.
 384  * Try and reuse as many LTBs as possible and avoid reallocation.
 385  *
 386  * Any changes to this allocation strategy must be reflected in
 387  * map_rxpool_buff_to_ltb() and map_txpool_buff_to_ltb().
 388  */
 389 static int alloc_ltb_set(struct ibmvnic_adapter *adapter,
 390                          struct ibmvnic_ltb_set *ltb_set, int num_buffs,
 391                          int buff_size)
 392 {
 393         struct device *dev = &adapter->vdev->dev;
 394         struct ibmvnic_ltb_set old_set;
 395         struct ibmvnic_ltb_set new_set;
 396         int rem_size;
 397         int tot_size;           /* size of all ltbs */
 398         int ltb_size;           /* size of one ltb */
 399         int nltbs;
 400         int rc;
 401         int n;
 402         int i;
 403
 404         dev_dbg(dev, "%s() num_buffs %d, buff_size %d\n", __func__, num_buffs,
 405                 buff_size);
 406
 407         ltb_size = rounddown(IBMVNIC_ONE_LTB_SIZE, buff_size);
 408         tot_size = num_buffs * buff_size;
 409
 410         if (ltb_size > tot_size)
 411                 ltb_size = tot_size;
 412
 413         nltbs = tot_size / ltb_size;
 414         if (tot_size % ltb_size)
 415                 nltbs++;
 416
 417         old_set = *ltb_set;
 418
 419         if (old_set.num_ltbs == nltbs) {
 420                 new_set = old_set;
 421         } else {
 422                 int tmp = nltbs * sizeof(struct ibmvnic_long_term_buff);
 423
 424                 new_set.ltbs = kzalloc(tmp, GFP_KERNEL);
 425                 if (!new_set.ltbs)
 426                         return -ENOMEM;
 427
 428                 new_set.num_ltbs = nltbs;
 429
 430                 /* Free any excess ltbs in old set */
 431                 for (i = new_set.num_ltbs; i < old_set.num_ltbs; i++)
 432                         free_long_term_buff(adapter, &old_set.ltbs[i]);
 433
 434                 /* Copy remaining ltbs to new set. All LTBs except the
 435                  * last one are of the same size. alloc_long_term_buff()
 436                  * will realloc if the size changes.
 437                  */
 438                 n = min(old_set.num_ltbs, new_set.num_ltbs);
 439                 for (i = 0; i < n; i++)
 440                         new_set.ltbs[i] = old_set.ltbs[i];
 441
 442                 /* Any additional ltbs in new set will have NULL ltbs for
 443                  * now and will be allocated in alloc_long_term_buff().
 444                  */
 445
 446                 /* We no longer need the old_set so free it. Note that we
 447                  * may have reused some ltbs from old set and freed excess
 448                  * ltbs above. So we only need to free the container now
 449                  * not the LTBs themselves. (i.e. dont free_ltb_set()!)
 450                  */
 451                 kfree(old_set.ltbs);
 452                 old_set.ltbs = NULL;
 453                 old_set.num_ltbs = 0;
 454
 455                 /* Install the new set. If allocations fail below, we will
 456                  * retry later and know what size LTBs we need.
 457                  */
 458                 *ltb_set = new_set;
 459         }
 460
 461         i = 0;
 462         rem_size = tot_size;
 463         while (rem_size) {
 464                 if (ltb_size > rem_size)
 465                         ltb_size = rem_size;
 466
 467                 rem_size -= ltb_size;
 468
 469                 rc = alloc_long_term_buff(adapter, &new_set.ltbs[i], ltb_size);
 470                 if (rc)
 471                         goto out;
 472                 i++;
 473         }
 474
 475         WARN_ON(i != new_set.num_ltbs);
 476
 477         return 0;
 478 out:
 479         /* We may have allocated one/more LTBs before failing and we
 480          * want to try and reuse on next reset. So don't free ltb set.
 481          */
 482         return rc;
 483 }
 484
 485 /**
 486  * map_rxpool_buf_to_ltb - Map given rxpool buffer to offset in an LTB.
 487  * @rxpool: The receive buffer pool containing buffer
 488  * @bufidx: Index of buffer in rxpool
 489  * @ltbp: (Output) pointer to the long term buffer containing the buffer
 490  * @offset: (Output) offset of buffer in the LTB from @ltbp
 491  *
 492  * Map the given buffer identified by [rxpool, bufidx] to an LTB in the
 493  * pool and its corresponding offset. Assume for now that each LTB is of
 494  * different size but could possibly be optimized based on the allocation
 495  * strategy in alloc_ltb_set().
 496  */
 497 static void map_rxpool_buf_to_ltb(struct ibmvnic_rx_pool *rxpool,
 498                                   unsigned int bufidx,
 499                                   struct ibmvnic_long_term_buff **ltbp,
 500                                   unsigned int *offset)
 501 {
 502         struct ibmvnic_long_term_buff *ltb;
 503         int nbufs;      /* # of buffers in one ltb */
 504         int i;
 505
 506         WARN_ON(bufidx >= rxpool->size);
 507
 508         for (i = 0; i < rxpool->ltb_set.num_ltbs; i++) {
 509                 ltb = &rxpool->ltb_set.ltbs[i];
 510                 nbufs = ltb->size / rxpool->buff_size;
 511                 if (bufidx < nbufs)
 512                         break;
 513                 bufidx -= nbufs;
 514         }
 515
 516         *ltbp = ltb;
 517         *offset = bufidx * rxpool->buff_size;
 518 }
 519
 520 /**
 521  * map_txpool_buf_to_ltb - Map given txpool buffer to offset in an LTB.
 522  * @txpool: The transmit buffer pool containing buffer
 523  * @bufidx: Index of buffer in txpool
 524  * @ltbp: (Output) pointer to the long term buffer (LTB) containing the buffer
 525  * @offset: (Output) offset of buffer in the LTB from @ltbp
 526  *
 527  * Map the given buffer identified by [txpool, bufidx] to an LTB in the
 528  * pool and its corresponding offset.
 529  */
 530 static void map_txpool_buf_to_ltb(struct ibmvnic_tx_pool *txpool,
 531                                   unsigned int bufidx,
 532                                   struct ibmvnic_long_term_buff **ltbp,
 533                                   unsigned int *offset)
 534 {
 535         struct ibmvnic_long_term_buff *ltb;
 536         int nbufs;      /* # of buffers in one ltb */
 537         int i;
 538
 539         WARN_ON_ONCE(bufidx >= txpool->num_buffers);
 540
 541         for (i = 0; i < txpool->ltb_set.num_ltbs; i++) {
 542                 ltb = &txpool->ltb_set.ltbs[i];
 543                 nbufs = ltb->size / txpool->buf_size;
 544                 if (bufidx < nbufs)
 545                         break;
 546                 bufidx -= nbufs;
 547         }
 548
 549         *ltbp = ltb;
 550         *offset = bufidx * txpool->buf_size;
 551 }
 552
 553 static void deactivate_rx_pools(struct ibmvnic_adapter *adapter)
 554 {
 555         int i;
 556
 557         for (i = 0; i < adapter->num_active_rx_pools; i++)
 558                 adapter->rx_pool[i].active = 0;
 559 }
 560
 561 static void replenish_rx_pool(struct ibmvnic_adapter *adapter,
 562                               struct ibmvnic_rx_pool *pool)
 563 {
 564         int count = pool->size - atomic_read(&pool->available);
 565         u64 handle = adapter->rx_scrq[pool->index]->handle;
 566         struct device *dev = &adapter->vdev->dev;
 567         struct ibmvnic_ind_xmit_queue *ind_bufp;
 568         struct ibmvnic_sub_crq_queue *rx_scrq;
 569         struct ibmvnic_long_term_buff *ltb;
 570         union sub_crq *sub_crq;
 571         int buffers_added = 0;
 572         unsigned long lpar_rc;
 573         struct sk_buff *skb;
 574         unsigned int offset;
 575         dma_addr_t dma_addr;
 576         unsigned char *dst;
 577         int shift = 0;
 578         int bufidx;
 579         int i;
 580
 581         if (!pool->active)
 582                 return;
 583
 584         rx_scrq = adapter->rx_scrq[pool->index];
 585         ind_bufp = &rx_scrq->ind_buf;
 586
 587         /* netdev_skb_alloc() could have failed after we saved a few skbs
 588          * in the indir_buf and we would not have sent them to VIOS yet.
 589          * To account for them, start the loop at ind_bufp->index rather
 590          * than 0. If we pushed all the skbs to VIOS, ind_bufp->index will
 591          * be 0.
 592          */
 593         for (i = ind_bufp->index; i < count; ++i) {
 594                 bufidx = pool->free_map[pool->next_free];
 595
 596                 /* We maybe reusing the skb from earlier resets. Allocate
 597                  * only if necessary. But since the LTB may have changed
 598                  * during reset (see init_rx_pools()), update LTB below
 599                  * even if reusing skb.
 600                  */
 601                 skb = pool->rx_buff[bufidx].skb;
 602                 if (!skb) {
 603                         skb = netdev_alloc_skb(adapter->netdev,
 604                                                pool->buff_size);
 605                         if (!skb) {
 606                                 dev_err(dev, "Couldn't replenish rx buff\n");
 607                                 adapter->replenish_no_mem++;
 608                                 break;
 609                         }
 610                 }
 611
 612                 pool->free_map[pool->next_free] = IBMVNIC_INVALID_MAP;
 613                 pool->next_free = (pool->next_free + 1) % pool->size;
 614
 615                 /* Copy the skb to the long term mapped DMA buffer */
 616                 map_rxpool_buf_to_ltb(pool, bufidx, &ltb, &offset);
 617                 dst = ltb->buff + offset;
 618                 memset(dst, 0, pool->buff_size);
 619                 dma_addr = ltb->addr + offset;
 620
 621                 /* add the skb to an rx_buff in the pool */
 622                 pool->rx_buff[bufidx].data = dst;
 623                 pool->rx_buff[bufidx].dma = dma_addr;
 624                 pool->rx_buff[bufidx].skb = skb;
 625                 pool->rx_buff[bufidx].pool_index = pool->index;
 626                 pool->rx_buff[bufidx].size = pool->buff_size;
 627
 628                 /* queue the rx_buff for the next send_subcrq_indirect */
 629                 sub_crq = &ind_bufp->indir_arr[ind_bufp->index++];
 630                 memset(sub_crq, 0, sizeof(*sub_crq));
 631                 sub_crq->rx_add.first = IBMVNIC_CRQ_CMD;
 632                 sub_crq->rx_add.correlator =
 633                     cpu_to_be64((u64)&pool->rx_buff[bufidx]);
 634                 sub_crq->rx_add.ioba = cpu_to_be32(dma_addr);
 635                 sub_crq->rx_add.map_id = ltb->map_id;
 636
 637                 /* The length field of the sCRQ is defined to be 24 bits so the
 638                  * buffer size needs to be left shifted by a byte before it is
 639                  * converted to big endian to prevent the last byte from being
 640                  * truncated.
 641                  */
 642 #ifdef __LITTLE_ENDIAN__
 643                 shift = 8;
 644 #endif
 645                 sub_crq->rx_add.len = cpu_to_be32(pool->buff_size << shift);
 646
 647                 /* if send_subcrq_indirect queue is full, flush to VIOS */
 648                 if (ind_bufp->index == IBMVNIC_MAX_IND_DESCS ||
 649                     i == count - 1) {
 650                         lpar_rc =
 651                                 send_subcrq_indirect(adapter, handle,
 652                                                      (u64)ind_bufp->indir_dma,
 653                                                      (u64)ind_bufp->index);
 654                         if (lpar_rc != H_SUCCESS)
 655                                 goto failure;
 656                         buffers_added += ind_bufp->index;
 657                         adapter->replenish_add_buff_success += ind_bufp->index;
 658                         ind_bufp->index = 0;
 659                 }
 660         }
 661         atomic_add(buffers_added, &pool->available);
 662         return;
 663
 664 failure:
 665         if (lpar_rc != H_PARAMETER && lpar_rc != H_CLOSED)
 666                 dev_err_ratelimited(dev, "rx: replenish packet buffer failed\n");
 667         for (i = ind_bufp->index - 1; i >= 0; --i) {
 668                 struct ibmvnic_rx_buff *rx_buff;
 669
 670                 pool->next_free = pool->next_free == 0 ?
 671                                   pool->size - 1 : pool->next_free - 1;
 672                 sub_crq = &ind_bufp->indir_arr[i];
 673                 rx_buff = (struct ibmvnic_rx_buff *)
 674                                 be64_to_cpu(sub_crq->rx_add.correlator);
 675                 bufidx = (int)(rx_buff - pool->rx_buff);
 676                 pool->free_map[pool->next_free] = bufidx;
 677                 dev_kfree_skb_any(pool->rx_buff[bufidx].skb);
 678                 pool->rx_buff[bufidx].skb = NULL;
 679         }
 680         adapter->replenish_add_buff_failure += ind_bufp->index;
 681         atomic_add(buffers_added, &pool->available);
 682         ind_bufp->index = 0;
 683         if (lpar_rc == H_CLOSED || adapter->failover_pending) {
 684                 /* Disable buffer pool replenishment and report carrier off if
 685                  * queue is closed or pending failover.
 686                  * Firmware guarantees that a signal will be sent to the
 687                  * driver, triggering a reset.
 688                  */
 689                 deactivate_rx_pools(adapter);
 690                 netif_carrier_off(adapter->netdev);
 691         }
 692 }
 693
 694 static void replenish_pools(struct ibmvnic_adapter *adapter)
 695 {
 696         int i;
 697
 698         adapter->replenish_task_cycles++;
 699         for (i = 0; i < adapter->num_active_rx_pools; i++) {
 700                 if (adapter->rx_pool[i].active)
 701                         replenish_rx_pool(adapter, &adapter->rx_pool[i]);
 702         }
 703
 704         netdev_dbg(adapter->netdev, "Replenished %d pools\n", i);
 705 }
 706
 707 static void release_stats_buffers(struct ibmvnic_adapter *adapter)
 708 {
 709         kfree(adapter->tx_stats_buffers);
 710         kfree(adapter->rx_stats_buffers);
 711         adapter->tx_stats_buffers = NULL;
 712         adapter->rx_stats_buffers = NULL;
 713 }
 714
 715 static int init_stats_buffers(struct ibmvnic_adapter *adapter)
 716 {
 717         adapter->tx_stats_buffers =
 718                                 kcalloc(IBMVNIC_MAX_QUEUES,
 719                                         sizeof(struct ibmvnic_tx_queue_stats),
 720                                         GFP_KERNEL);
 721         if (!adapter->tx_stats_buffers)
 722                 return -ENOMEM;
 723
 724         adapter->rx_stats_buffers =
 725                                 kcalloc(IBMVNIC_MAX_QUEUES,
 726                                         sizeof(struct ibmvnic_rx_queue_stats),
 727                                         GFP_KERNEL);
 728         if (!adapter->rx_stats_buffers)
 729                 return -ENOMEM;
 730
 731         return 0;
 732 }
 733
 734 static void release_stats_token(struct ibmvnic_adapter *adapter)
 735 {
 736         struct device *dev = &adapter->vdev->dev;
 737
 738         if (!adapter->stats_token)
 739                 return;
 740
 741         dma_unmap_single(dev, adapter->stats_token,
 742                          sizeof(struct ibmvnic_statistics),
 743                          DMA_FROM_DEVICE);
 744         adapter->stats_token = 0;
 745 }
 746
 747 static int init_stats_token(struct ibmvnic_adapter *adapter)
 748 {
 749         struct device *dev = &adapter->vdev->dev;
 750         dma_addr_t stok;
 751         int rc;
 752
 753         stok = dma_map_single(dev, &adapter->stats,
 754                               sizeof(struct ibmvnic_statistics),
 755                               DMA_FROM_DEVICE);
 756         rc = dma_mapping_error(dev, stok);
 757         if (rc) {
 758                 dev_err(dev, "Couldn't map stats buffer, rc = %d\n", rc);
 759                 return rc;
 760         }
 761
 762         adapter->stats_token = stok;
 763         netdev_dbg(adapter->netdev, "Stats token initialized (%llx)\n", stok);
 764         return 0;
 765 }
 766
 767 /**
 768  * release_rx_pools() - Release any rx pools attached to @adapter.
 769  * @adapter: ibmvnic adapter
 770  *
 771  * Safe to call this multiple times - even if no pools are attached.
 772  */
 773 static void release_rx_pools(struct ibmvnic_adapter *adapter)
 774 {
 775         struct ibmvnic_rx_pool *rx_pool;
 776         int i, j;
 777
 778         if (!adapter->rx_pool)
 779                 return;
 780
 781         for (i = 0; i < adapter->num_active_rx_pools; i++) {
 782                 rx_pool = &adapter->rx_pool[i];
 783
 784                 netdev_dbg(adapter->netdev, "Releasing rx_pool[%d]\n", i);
 785
 786                 kfree(rx_pool->free_map);
 787
 788                 free_ltb_set(adapter, &rx_pool->ltb_set);
 789
 790                 if (!rx_pool->rx_buff)
 791                         continue;
 792
 793                 for (j = 0; j < rx_pool->size; j++) {
 794                         if (rx_pool->rx_buff[j].skb) {
 795                                 dev_kfree_skb_any(rx_pool->rx_buff[j].skb);
 796                                 rx_pool->rx_buff[j].skb = NULL;
 797                         }
 798                 }
 799
 800                 kfree(rx_pool->rx_buff);
 801         }
 802
 803         kfree(adapter->rx_pool);
 804         adapter->rx_pool = NULL;
 805         adapter->num_active_rx_pools = 0;
 806         adapter->prev_rx_pool_size = 0;
 807 }
 808
 809 /**
 810  * reuse_rx_pools() - Check if the existing rx pools can be reused.
 811  * @adapter: ibmvnic adapter
 812  *
 813  * Check if the existing rx pools in the adapter can be reused. The
 814  * pools can be reused if the pool parameters (number of pools,
 815  * number of buffers in the pool and size of each buffer) have not
 816  * changed.
 817  *
 818  * NOTE: This assumes that all pools have the same number of buffers
 819  *       which is the case currently. If that changes, we must fix this.
 820  *
 821  * Return: true if the rx pools can be reused, false otherwise.
 822  */
 823 static bool reuse_rx_pools(struct ibmvnic_adapter *adapter)
 824 {
 825         u64 old_num_pools, new_num_pools;
 826         u64 old_pool_size, new_pool_size;
 827         u64 old_buff_size, new_buff_size;
 828
 829         if (!adapter->rx_pool)
 830                 return false;
 831
 832         old_num_pools = adapter->num_active_rx_pools;
 833         new_num_pools = adapter->req_rx_queues;
 834
 835         old_pool_size = adapter->prev_rx_pool_size;
 836         new_pool_size = adapter->req_rx_add_entries_per_subcrq;
 837
 838         old_buff_size = adapter->prev_rx_buf_sz;
 839         new_buff_size = adapter->cur_rx_buf_sz;
 840
 841         if (old_buff_size != new_buff_size ||
 842             old_num_pools != new_num_pools ||
 843             old_pool_size != new_pool_size)
 844                 return false;
 845
 846         return true;
 847 }
 848
 849 /**
 850  * init_rx_pools(): Initialize the set of receiver pools in the adapter.
 851  * @netdev: net device associated with the vnic interface
 852  *
 853  * Initialize the set of receiver pools in the ibmvnic adapter associated
 854  * with the net_device @netdev. If possible, reuse the existing rx pools.
 855  * Otherwise free any existing pools and  allocate a new set of pools
 856  * before initializing them.
 857  *
 858  * Return: 0 on success and negative value on error.
 859  */
 860 static int init_rx_pools(struct net_device *netdev)
 861 {
 862         struct ibmvnic_adapter *adapter = netdev_priv(netdev);
 863         struct device *dev = &adapter->vdev->dev;
 864         struct ibmvnic_rx_pool *rx_pool;
 865         u64 num_pools;
 866         u64 pool_size;          /* # of buffers in one pool */
 867         u64 buff_size;
 868         int i, j, rc;
 869
 870         pool_size = adapter->req_rx_add_entries_per_subcrq;
 871         num_pools = adapter->req_rx_queues;
 872         buff_size = adapter->cur_rx_buf_sz;
 873
 874         if (reuse_rx_pools(adapter)) {
 875                 dev_dbg(dev, "Reusing rx pools\n");
 876                 goto update_ltb;
 877         }
 878
 879         /* Allocate/populate the pools. */
 880         release_rx_pools(adapter);
 881
 882         adapter->rx_pool = kcalloc(num_pools,
 883                                    sizeof(struct ibmvnic_rx_pool),
 884                                    GFP_KERNEL);
 885         if (!adapter->rx_pool) {
 886                 dev_err(dev, "Failed to allocate rx pools\n");
 887                 return -ENOMEM;
 888         }
 889
 890         /* Set num_active_rx_pools early. If we fail below after partial
 891          * allocation, release_rx_pools() will know how many to look for.
 892          */
 893         adapter->num_active_rx_pools = num_pools;
 894
 895         for (i = 0; i < num_pools; i++) {
 896                 rx_pool = &adapter->rx_pool[i];
 897
 898                 netdev_dbg(adapter->netdev,
 899                            "Initializing rx_pool[%d], %lld buffs, %lld bytes each\n",
 900                            i, pool_size, buff_size);
 901
 902                 rx_pool->size = pool_size;
 903                 rx_pool->index = i;
 904                 rx_pool->buff_size = ALIGN(buff_size, L1_CACHE_BYTES);
 905
 906                 rx_pool->free_map = kcalloc(rx_pool->size, sizeof(int),
 907                                             GFP_KERNEL);
 908                 if (!rx_pool->free_map) {
 909                         dev_err(dev, "Couldn't alloc free_map %d\n", i);
 910                         rc = -ENOMEM;
 911                         goto out_release;
 912                 }
 913
 914                 rx_pool->rx_buff = kcalloc(rx_pool->size,
 915                                            sizeof(struct ibmvnic_rx_buff),
 916                                            GFP_KERNEL);
 917                 if (!rx_pool->rx_buff) {
 918                         dev_err(dev, "Couldn't alloc rx buffers\n");
 919                         rc = -ENOMEM;
 920                         goto out_release;
 921                 }
 922         }
 923
 924         adapter->prev_rx_pool_size = pool_size;
 925         adapter->prev_rx_buf_sz = adapter->cur_rx_buf_sz;
 926
 927 update_ltb:
 928         for (i = 0; i < num_pools; i++) {
 929                 rx_pool = &adapter->rx_pool[i];
 930                 dev_dbg(dev, "Updating LTB for rx pool %d [%d, %d]\n",
 931                         i, rx_pool->size, rx_pool->buff_size);
 932
 933                 rc = alloc_ltb_set(adapter, &rx_pool->ltb_set,
 934                                    rx_pool->size, rx_pool->buff_size);
 935                 if (rc)
 936                         goto out;
 937
 938                 for (j = 0; j < rx_pool->size; ++j) {
 939                         struct ibmvnic_rx_buff *rx_buff;
 940
 941                         rx_pool->free_map[j] = j;
 942
 943                         /* NOTE: Don't clear rx_buff->skb here - will leak
 944                          * memory! replenish_rx_pool() will reuse skbs or
 945                          * allocate as necessary.
 946                          */
 947                         rx_buff = &rx_pool->rx_buff[j];
 948                         rx_buff->dma = 0;
 949                         rx_buff->data = 0;
 950                         rx_buff->size = 0;
 951                         rx_buff->pool_index = 0;
 952                 }
 953
 954                 /* Mark pool "empty" so replenish_rx_pools() will
 955                  * update the LTB info for each buffer
 956                  */
 957                 atomic_set(&rx_pool->available, 0);
 958                 rx_pool->next_alloc = 0;
 959                 rx_pool->next_free = 0;
 960                 /* replenish_rx_pool() may have called deactivate_rx_pools()
 961                  * on failover. Ensure pool is active now.
 962                  */
 963                 rx_pool->active = 1;
 964         }
 965         return 0;
 966 out_release:
 967         release_rx_pools(adapter);
 968 out:
 969         /* We failed to allocate one or more LTBs or map them on the VIOS.
 970          * Hold onto the pools and any LTBs that we did allocate/map.
 971          */
 972         return rc;
 973 }
 974
 975 static void release_vpd_data(struct ibmvnic_adapter *adapter)
 976 {
 977         if (!adapter->vpd)
 978                 return;
 979
 980         kfree(adapter->vpd->buff);
 981         kfree(adapter->vpd);
 982
 983         adapter->vpd = NULL;
 984 }
 985
 986 static void release_one_tx_pool(struct ibmvnic_adapter *adapter,
 987                                 struct ibmvnic_tx_pool *tx_pool)
 988 {
 989         kfree(tx_pool->tx_buff);
 990         kfree(tx_pool->free_map);
 991         free_ltb_set(adapter, &tx_pool->ltb_set);
 992 }
 993
 994 /**
 995  * release_tx_pools() - Release any tx pools attached to @adapter.
 996  * @adapter: ibmvnic adapter
 997  *
 998  * Safe to call this multiple times - even if no pools are attached.
 999  */
1000 static void release_tx_pools(struct ibmvnic_adapter *adapter)
1001 {
1002         int i;
1003
1004         /* init_tx_pools() ensures that ->tx_pool and ->tso_pool are
1005          * both NULL or both non-NULL. So we only need to check one.
1006          */
1007         if (!adapter->tx_pool)
1008                 return;
1009
1010         for (i = 0; i < adapter->num_active_tx_pools; i++) {
1011                 release_one_tx_pool(adapter, &adapter->tx_pool[i]);
1012                 release_one_tx_pool(adapter, &adapter->tso_pool[i]);
1013         }
1014
1015         kfree(adapter->tx_pool);
1016         adapter->tx_pool = NULL;
1017         kfree(adapter->tso_pool);
1018         adapter->tso_pool = NULL;
1019         adapter->num_active_tx_pools = 0;
1020         adapter->prev_tx_pool_size = 0;
1021 }
1022
1023 static int init_one_tx_pool(struct net_device *netdev,
1024                             struct ibmvnic_tx_pool *tx_pool,
1025                             int pool_size, int buf_size)
1026 {
1027         int i;
1028
1029         tx_pool->tx_buff = kcalloc(pool_size,
1030                                    sizeof(struct ibmvnic_tx_buff),
1031                                    GFP_KERNEL);
1032         if (!tx_pool->tx_buff)
1033                 return -ENOMEM;
1034
1035         tx_pool->free_map = kcalloc(pool_size, sizeof(int), GFP_KERNEL);
1036         if (!tx_pool->free_map) {
1037                 kfree(tx_pool->tx_buff);
1038                 tx_pool->tx_buff = NULL;
1039                 return -ENOMEM;
1040         }
1041
1042         for (i = 0; i < pool_size; i++)
1043                 tx_pool->free_map[i] = i;
1044
1045         tx_pool->consumer_index = 0;
1046         tx_pool->producer_index = 0;
1047         tx_pool->num_buffers = pool_size;
1048         tx_pool->buf_size = buf_size;
1049
1050         return 0;
1051 }
1052
1053 /**
1054  * reuse_tx_pools() - Check if the existing tx pools can be reused.
1055  * @adapter: ibmvnic adapter
1056  *
1057  * Check if the existing tx pools in the adapter can be reused. The
1058  * pools can be reused if the pool parameters (number of pools,
1059  * number of buffers in the pool and mtu) have not changed.
1060  *
1061  * NOTE: This assumes that all pools have the same number of buffers
1062  *       which is the case currently. If that changes, we must fix this.
1063  *
1064  * Return: true if the tx pools can be reused, false otherwise.
1065  */
1066 static bool reuse_tx_pools(struct ibmvnic_adapter *adapter)
1067 {
1068         u64 old_num_pools, new_num_pools;
1069         u64 old_pool_size, new_pool_size;
1070         u64 old_mtu, new_mtu;
1071
1072         if (!adapter->tx_pool)
1073                 return false;
1074
1075         old_num_pools = adapter->num_active_tx_pools;
1076         new_num_pools = adapter->num_active_tx_scrqs;
1077         old_pool_size = adapter->prev_tx_pool_size;
1078         new_pool_size = adapter->req_tx_entries_per_subcrq;
1079         old_mtu = adapter->prev_mtu;
1080         new_mtu = adapter->req_mtu;
1081
1082         if (old_mtu != new_mtu ||
1083             old_num_pools != new_num_pools ||
1084             old_pool_size != new_pool_size)
1085                 return false;
1086
1087         return true;
1088 }
1089
1090 /**
1091  * init_tx_pools(): Initialize the set of transmit pools in the adapter.
1092  * @netdev: net device associated with the vnic interface
1093  *
1094  * Initialize the set of transmit pools in the ibmvnic adapter associated
1095  * with the net_device @netdev. If possible, reuse the existing tx pools.
1096  * Otherwise free any existing pools and  allocate a new set of pools
1097  * before initializing them.
1098  *
1099  * Return: 0 on success and negative value on error.
1100  */
1101 static int init_tx_pools(struct net_device *netdev)
1102 {
1103         struct ibmvnic_adapter *adapter = netdev_priv(netdev);
1104         struct device *dev = &adapter->vdev->dev;
1105         int num_pools;
1106         u64 pool_size;          /* # of buffers in pool */
1107         u64 buff_size;
1108         int i, j, rc;
1109
1110         num_pools = adapter->req_tx_queues;
1111
1112         /* We must notify the VIOS about the LTB on all resets - but we only
1113          * need to alloc/populate pools if either the number of buffers or
1114          * size of each buffer in the pool has changed.
1115          */
1116         if (reuse_tx_pools(adapter)) {
1117                 netdev_dbg(netdev, "Reusing tx pools\n");
1118                 goto update_ltb;
1119         }
1120
1121         /* Allocate/populate the pools. */
1122         release_tx_pools(adapter);
1123
1124         pool_size = adapter->req_tx_entries_per_subcrq;
1125         num_pools = adapter->num_active_tx_scrqs;
1126
1127         adapter->tx_pool = kcalloc(num_pools,
1128                                    sizeof(struct ibmvnic_tx_pool), GFP_KERNEL);
1129         if (!adapter->tx_pool)
1130                 return -ENOMEM;
1131
1132         adapter->tso_pool = kcalloc(num_pools,
1133                                     sizeof(struct ibmvnic_tx_pool), GFP_KERNEL);
1134         /* To simplify release_tx_pools() ensure that ->tx_pool and
1135          * ->tso_pool are either both NULL or both non-NULL.
1136          */
1137         if (!adapter->tso_pool) {
1138                 kfree(adapter->tx_pool);
1139                 adapter->tx_pool = NULL;
1140                 return -ENOMEM;
1141         }
1142
1143         /* Set num_active_tx_pools early. If we fail below after partial
1144          * allocation, release_tx_pools() will know how many to look for.
1145          */
1146         adapter->num_active_tx_pools = num_pools;
1147
1148         buff_size = adapter->req_mtu + VLAN_HLEN;
1149         buff_size = ALIGN(buff_size, L1_CACHE_BYTES);
1150
1151         for (i = 0; i < num_pools; i++) {
1152                 dev_dbg(dev, "Init tx pool %d [%llu, %llu]\n",
1153                         i, adapter->req_tx_entries_per_subcrq, buff_size);
1154
1155                 rc = init_one_tx_pool(netdev, &adapter->tx_pool[i],
1156                                       pool_size, buff_size);
1157                 if (rc)
1158                         goto out_release;
1159
1160                 rc = init_one_tx_pool(netdev, &adapter->tso_pool[i],
1161                                       IBMVNIC_TSO_BUFS,
1162                                       IBMVNIC_TSO_BUF_SZ);
1163                 if (rc)
1164                         goto out_release;
1165         }
1166
1167         adapter->prev_tx_pool_size = pool_size;
1168         adapter->prev_mtu = adapter->req_mtu;
1169
1170 update_ltb:
1171         /* NOTE: All tx_pools have the same number of buffers (which is
1172          *       same as pool_size). All tso_pools have IBMVNIC_TSO_BUFS
1173          *       buffers (see calls init_one_tx_pool() for these).
1174          *       For consistency, we use tx_pool->num_buffers and
1175          *       tso_pool->num_buffers below.
1176          */
1177         rc = -1;
1178         for (i = 0; i < num_pools; i++) {
1179                 struct ibmvnic_tx_pool *tso_pool;
1180                 struct ibmvnic_tx_pool *tx_pool;
1181
1182                 tx_pool = &adapter->tx_pool[i];
1183
1184                 dev_dbg(dev, "Updating LTB for tx pool %d [%d, %d]\n",
1185                         i, tx_pool->num_buffers, tx_pool->buf_size);
1186
1187                 rc = alloc_ltb_set(adapter, &tx_pool->ltb_set,
1188                                    tx_pool->num_buffers, tx_pool->buf_size);
1189                 if (rc)
1190                         goto out;
1191
1192                 tx_pool->consumer_index = 0;
1193                 tx_pool->producer_index = 0;
1194
1195                 for (j = 0; j < tx_pool->num_buffers; j++)
1196                         tx_pool->free_map[j] = j;
1197
1198                 tso_pool = &adapter->tso_pool[i];
1199
1200                 dev_dbg(dev, "Updating LTB for tso pool %d [%d, %d]\n",
1201                         i, tso_pool->num_buffers, tso_pool->buf_size);
1202
1203                 rc = alloc_ltb_set(adapter, &tso_pool->ltb_set,
1204                                    tso_pool->num_buffers, tso_pool->buf_size);
1205                 if (rc)
1206                         goto out;
1207
1208                 tso_pool->consumer_index = 0;
1209                 tso_pool->producer_index = 0;
1210
1211                 for (j = 0; j < tso_pool->num_buffers; j++)
1212                         tso_pool->free_map[j] = j;
1213         }
1214
1215         return 0;
1216 out_release:
1217         release_tx_pools(adapter);
1218 out:
1219         /* We failed to allocate one or more LTBs or map them on the VIOS.
1220          * Hold onto the pools and any LTBs that we did allocate/map.
1221          */
1222         return rc;
1223 }
1224
1225 static void ibmvnic_napi_enable(struct ibmvnic_adapter *adapter)
1226 {
1227         int i;
1228
1229         if (adapter->napi_enabled)
1230                 return;
1231
1232         for (i = 0; i < adapter->req_rx_queues; i++)
1233                 napi_enable(&adapter->napi[i]);
1234
1235         adapter->napi_enabled = true;
1236 }
1237
1238 static void ibmvnic_napi_disable(struct ibmvnic_adapter *adapter)
1239 {
1240         int i;
1241
1242         if (!adapter->napi_enabled)
1243                 return;
1244
1245         for (i = 0; i < adapter->req_rx_queues; i++) {
1246                 netdev_dbg(adapter->netdev, "Disabling napi[%d]\n", i);
1247                 napi_disable(&adapter->napi[i]);
1248         }
1249
1250         adapter->napi_enabled = false;
1251 }
1252
1253 static int init_napi(struct ibmvnic_adapter *adapter)
1254 {
1255         int i;
1256
1257         adapter->napi = kcalloc(adapter->req_rx_queues,
1258                                 sizeof(struct napi_struct), GFP_KERNEL);
1259         if (!adapter->napi)
1260                 return -ENOMEM;
1261
1262         for (i = 0; i < adapter->req_rx_queues; i++) {
1263                 netdev_dbg(adapter->netdev, "Adding napi[%d]\n", i);
1264                 netif_napi_add(adapter->netdev, &adapter->napi[i],
1265                                ibmvnic_poll, NAPI_POLL_WEIGHT);
1266         }
1267
1268         adapter->num_active_rx_napi = adapter->req_rx_queues;
1269         return 0;
1270 }
1271
1272 static void release_napi(struct ibmvnic_adapter *adapter)
1273 {
1274         int i;
1275
1276         if (!adapter->napi)
1277                 return;
1278
1279         for (i = 0; i < adapter->num_active_rx_napi; i++) {
1280                 netdev_dbg(adapter->netdev, "Releasing napi[%d]\n", i);
1281                 netif_napi_del(&adapter->napi[i]);
1282         }
1283
1284         kfree(adapter->napi);
1285         adapter->napi = NULL;
1286         adapter->num_active_rx_napi = 0;
1287         adapter->napi_enabled = false;
1288 }
1289
1290 static const char *adapter_state_to_string(enum vnic_state state)
1291 {
1292         switch (state) {
1293         case VNIC_PROBING:
1294                 return "PROBING";
1295         case VNIC_PROBED:
1296                 return "PROBED";
1297         case VNIC_OPENING:
1298                 return "OPENING";
1299         case VNIC_OPEN:
1300                 return "OPEN";
1301         case VNIC_CLOSING:
1302                 return "CLOSING";
1303         case VNIC_CLOSED:
1304                 return "CLOSED";
1305         case VNIC_REMOVING:
1306                 return "REMOVING";
1307         case VNIC_REMOVED:
1308                 return "REMOVED";
1309         case VNIC_DOWN:
1310                 return "DOWN";
1311         }
1312         return "UNKNOWN";
1313 }
1314
1315 static int ibmvnic_login(struct net_device *netdev)
1316 {
1317         struct ibmvnic_adapter *adapter = netdev_priv(netdev);
1318         unsigned long timeout = msecs_to_jiffies(20000);
1319         int retry_count = 0;
1320         int retries = 10;
1321         bool retry;
1322         int rc;
1323
1324         do {
1325                 retry = false;
1326                 if (retry_count > retries) {
1327                         netdev_warn(netdev, "Login attempts exceeded\n");
1328                         return -EACCES;
1329                 }
1330
1331                 adapter->init_done_rc = 0;
1332                 reinit_completion(&adapter->init_done);
1333                 rc = send_login(adapter);
1334                 if (rc)
1335                         return rc;
1336
1337                 if (!wait_for_completion_timeout(&adapter->init_done,
1338                                                  timeout)) {
1339                         netdev_warn(netdev, "Login timed out, retrying...\n");
1340                         retry = true;
1341                         adapter->init_done_rc = 0;
1342                         retry_count++;
1343                         continue;
1344                 }
1345
1346                 if (adapter->init_done_rc == ABORTED) {
1347                         netdev_warn(netdev, "Login aborted, retrying...\n");
1348                         retry = true;
1349                         adapter->init_done_rc = 0;
1350                         retry_count++;
1351                         /* FW or device may be busy, so
1352                          * wait a bit before retrying login
1353                          */
1354                         msleep(500);
1355                 } else if (adapter->init_done_rc == PARTIALSUCCESS) {
1356                         retry_count++;
1357                         release_sub_crqs(adapter, 1);
1358
1359                         retry = true;
1360                         netdev_dbg(netdev,
1361                                    "Received partial success, retrying...\n");
1362                         adapter->init_done_rc = 0;
1363                         reinit_completion(&adapter->init_done);
1364                         send_query_cap(adapter);
1365                         if (!wait_for_completion_timeout(&adapter->init_done,
1366                                                          timeout)) {
1367                                 netdev_warn(netdev,
1368                                             "Capabilities query timed out\n");
1369                                 return -ETIMEDOUT;
1370                         }
1371
1372                         rc = init_sub_crqs(adapter);
1373                         if (rc) {
1374                                 netdev_warn(netdev,
1375                                             "SCRQ initialization failed\n");
1376                                 return rc;
1377                         }
1378
1379                         rc = init_sub_crq_irqs(adapter);
1380                         if (rc) {
1381                                 netdev_warn(netdev,
1382                                             "SCRQ irq initialization failed\n");
1383                                 return rc;
1384                         }
1385                 } else if (adapter->init_done_rc) {
1386                         netdev_warn(netdev, "Adapter login failed, init_done_rc = %d\n",
1387                                     adapter->init_done_rc);
1388                         return -EIO;
1389                 }
1390         } while (retry);
1391
1392         __ibmvnic_set_mac(netdev, adapter->mac_addr);
1393
1394         netdev_dbg(netdev, "[S:%s] Login succeeded\n", adapter_state_to_string(adapter->state));
1395         return 0;
1396 }
1397
1398 static void release_login_buffer(struct ibmvnic_adapter *adapter)
1399 {
1400         kfree(adapter->login_buf);
1401         adapter->login_buf = NULL;
1402 }
1403
1404 static void release_login_rsp_buffer(struct ibmvnic_adapter *adapter)
1405 {
1406         kfree(adapter->login_rsp_buf);
1407         adapter->login_rsp_buf = NULL;
1408 }
1409
1410 static void release_resources(struct ibmvnic_adapter *adapter)
1411 {
1412         release_vpd_data(adapter);
1413
1414         release_napi(adapter);
1415         release_login_buffer(adapter);
1416         release_login_rsp_buffer(adapter);
1417 }
1418
1419 static int set_link_state(struct ibmvnic_adapter *adapter, u8 link_state)
1420 {
1421         struct net_device *netdev = adapter->netdev;
1422         unsigned long timeout = msecs_to_jiffies(20000);
1423         union ibmvnic_crq crq;
1424         bool resend;
1425         int rc;
1426
1427         netdev_dbg(netdev, "setting link state %d\n", link_state);
1428
1429         memset(&crq, 0, sizeof(crq));
1430         crq.logical_link_state.first = IBMVNIC_CRQ_CMD;
1431         crq.logical_link_state.cmd = LOGICAL_LINK_STATE;
1432         crq.logical_link_state.link_state = link_state;
1433
1434         do {
1435                 resend = false;
1436
1437                 reinit_completion(&adapter->init_done);
1438                 rc = ibmvnic_send_crq(adapter, &crq);
1439                 if (rc) {
1440                         netdev_err(netdev, "Failed to set link state\n");
1441                         return rc;
1442                 }
1443
1444                 if (!wait_for_completion_timeout(&adapter->init_done,
1445                                                  timeout)) {
1446                         netdev_err(netdev, "timeout setting link state\n");
1447                         return -ETIMEDOUT;
1448                 }
1449
1450                 if (adapter->init_done_rc == PARTIALSUCCESS) {
1451                         /* Partuial success, delay and re-send */
1452                         mdelay(1000);
1453                         resend = true;
1454                 } else if (adapter->init_done_rc) {
1455                         netdev_warn(netdev, "Unable to set link state, rc=%d\n",
1456                                     adapter->init_done_rc);
1457                         return adapter->init_done_rc;
1458                 }
1459         } while (resend);
1460
1461         return 0;
1462 }
1463
1464 static int set_real_num_queues(struct net_device *netdev)
1465 {
1466         struct ibmvnic_adapter *adapter = netdev_priv(netdev);
1467         int rc;
1468
1469         netdev_dbg(netdev, "Setting real tx/rx queues (%llx/%llx)\n",
1470                    adapter->req_tx_queues, adapter->req_rx_queues);
1471
1472         rc = netif_set_real_num_tx_queues(netdev, adapter->req_tx_queues);
1473         if (rc) {
1474                 netdev_err(netdev, "failed to set the number of tx queues\n");
1475                 return rc;
1476         }
1477
1478         rc = netif_set_real_num_rx_queues(netdev, adapter->req_rx_queues);
1479         if (rc)
1480                 netdev_err(netdev, "failed to set the number of rx queues\n");
1481
1482         return rc;
1483 }
1484
1485 static int ibmvnic_get_vpd(struct ibmvnic_adapter *adapter)
1486 {
1487         struct device *dev = &adapter->vdev->dev;
1488         union ibmvnic_crq crq;
1489         int len = 0;
1490         int rc;
1491
1492         if (adapter->vpd->buff)
1493                 len = adapter->vpd->len;
1494
1495         mutex_lock(&adapter->fw_lock);
1496         adapter->fw_done_rc = 0;
1497         reinit_completion(&adapter->fw_done);
1498
1499         crq.get_vpd_size.first = IBMVNIC_CRQ_CMD;
1500         crq.get_vpd_size.cmd = GET_VPD_SIZE;
1501         rc = ibmvnic_send_crq(adapter, &crq);
1502         if (rc) {
1503                 mutex_unlock(&adapter->fw_lock);
1504                 return rc;
1505         }
1506
1507         rc = ibmvnic_wait_for_completion(adapter, &adapter->fw_done, 10000);
1508         if (rc) {
1509                 dev_err(dev, "Could not retrieve VPD size, rc = %d\n", rc);
1510                 mutex_unlock(&adapter->fw_lock);
1511                 return rc;
1512         }
1513         mutex_unlock(&adapter->fw_lock);
1514
1515         if (!adapter->vpd->len)
1516                 return -ENODATA;
1517
1518         if (!adapter->vpd->buff)
1519                 adapter->vpd->buff = kzalloc(adapter->vpd->len, GFP_KERNEL);
1520         else if (adapter->vpd->len != len)
1521                 adapter->vpd->buff =
1522                         krealloc(adapter->vpd->buff,
1523                                  adapter->vpd->len, GFP_KERNEL);
1524
1525         if (!adapter->vpd->buff) {
1526                 dev_err(dev, "Could allocate VPD buffer\n");
1527                 return -ENOMEM;
1528         }
1529
1530         adapter->vpd->dma_addr =
1531                 dma_map_single(dev, adapter->vpd->buff, adapter->vpd->len,
1532                                DMA_FROM_DEVICE);
1533         if (dma_mapping_error(dev, adapter->vpd->dma_addr)) {
1534                 dev_err(dev, "Could not map VPD buffer\n");
1535                 kfree(adapter->vpd->buff);
1536                 adapter->vpd->buff = NULL;
1537                 return -ENOMEM;
1538         }
1539
1540         mutex_lock(&adapter->fw_lock);
1541         adapter->fw_done_rc = 0;
1542         reinit_completion(&adapter->fw_done);
1543
1544         crq.get_vpd.first = IBMVNIC_CRQ_CMD;
1545         crq.get_vpd.cmd = GET_VPD;
1546         crq.get_vpd.ioba = cpu_to_be32(adapter->vpd->dma_addr);
1547         crq.get_vpd.len = cpu_to_be32((u32)adapter->vpd->len);
1548         rc = ibmvnic_send_crq(adapter, &crq);
1549         if (rc) {
1550                 kfree(adapter->vpd->buff);
1551                 adapter->vpd->buff = NULL;
1552                 mutex_unlock(&adapter->fw_lock);
1553                 return rc;
1554         }
1555
1556         rc = ibmvnic_wait_for_completion(adapter, &adapter->fw_done, 10000);
1557         if (rc) {
1558                 dev_err(dev, "Unable to retrieve VPD, rc = %d\n", rc);
1559                 kfree(adapter->vpd->buff);
1560                 adapter->vpd->buff = NULL;
1561                 mutex_unlock(&adapter->fw_lock);
1562                 return rc;
1563         }
1564
1565         mutex_unlock(&adapter->fw_lock);
1566         return 0;
1567 }
1568
1569 static int init_resources(struct ibmvnic_adapter *adapter)
1570 {
1571         struct net_device *netdev = adapter->netdev;
1572         int rc;
1573
1574         rc = set_real_num_queues(netdev);
1575         if (rc)
1576                 return rc;
1577
1578         adapter->vpd = kzalloc(sizeof(*adapter->vpd), GFP_KERNEL);
1579         if (!adapter->vpd)
1580                 return -ENOMEM;
1581
1582         /* Vital Product Data (VPD) */
1583         rc = ibmvnic_get_vpd(adapter);
1584         if (rc) {
1585                 netdev_err(netdev, "failed to initialize Vital Product Data (VPD)\n");
1586                 return rc;
1587         }
1588
1589         rc = init_napi(adapter);
1590         if (rc)
1591                 return rc;
1592
1593         send_query_map(adapter);
1594
1595         rc = init_rx_pools(netdev);
1596         if (rc)
1597                 return rc;
1598
1599         rc = init_tx_pools(netdev);
1600         return rc;
1601 }
1602
1603 static int __ibmvnic_open(struct net_device *netdev)
1604 {
1605         struct ibmvnic_adapter *adapter = netdev_priv(netdev);
1606         enum vnic_state prev_state = adapter->state;
1607         int i, rc;
1608
1609         adapter->state = VNIC_OPENING;
1610         replenish_pools(adapter);
1611         ibmvnic_napi_enable(adapter);
1612
1613         /* We're ready to receive frames, enable the sub-crq interrupts and
1614          * set the logical link state to up
1615          */
1616         for (i = 0; i < adapter->req_rx_queues; i++) {
1617                 netdev_dbg(netdev, "Enabling rx_scrq[%d] irq\n", i);
1618                 if (prev_state == VNIC_CLOSED)
1619                         enable_irq(adapter->rx_scrq[i]->irq);
1620                 enable_scrq_irq(adapter, adapter->rx_scrq[i]);
1621         }
1622
1623         for (i = 0; i < adapter->req_tx_queues; i++) {
1624                 netdev_dbg(netdev, "Enabling tx_scrq[%d] irq\n", i);
1625                 if (prev_state == VNIC_CLOSED)
1626                         enable_irq(adapter->tx_scrq[i]->irq);
1627                 enable_scrq_irq(adapter, adapter->tx_scrq[i]);
1628                 netdev_tx_reset_queue(netdev_get_tx_queue(netdev, i));
1629         }
1630
1631         rc = set_link_state(adapter, IBMVNIC_LOGICAL_LNK_UP);
1632         if (rc) {
1633                 ibmvnic_napi_disable(adapter);
1634                 ibmvnic_disable_irqs(adapter);
1635                 return rc;
1636         }
1637
1638         adapter->tx_queues_active = true;
1639
1640         /* Since queues were stopped until now, there shouldn't be any
1641          * one in ibmvnic_complete_tx() or ibmvnic_xmit() so maybe we
1642          * don't need the synchronize_rcu()? Leaving it for consistency
1643          * with setting ->tx_queues_active = false.
1644          */
1645         synchronize_rcu();
1646
1647         netif_tx_start_all_queues(netdev);
1648
1649         if (prev_state == VNIC_CLOSED) {
1650                 for (i = 0; i < adapter->req_rx_queues; i++)
1651                         napi_schedule(&adapter->napi[i]);
1652         }
1653
1654         adapter->state = VNIC_OPEN;
1655         return rc;
1656 }
1657
1658 static int ibmvnic_open(struct net_device *netdev)
1659 {
1660         struct ibmvnic_adapter *adapter = netdev_priv(netdev);
1661         int rc;
1662
1663         ASSERT_RTNL();
1664
1665         /* If device failover is pending or we are about to reset, just set
1666          * device state and return. Device operation will be handled by reset
1667          * routine.
1668          *
1669          * It should be safe to overwrite the adapter->state here. Since
1670          * we hold the rtnl, either the reset has not actually started or
1671          * the rtnl got dropped during the set_link_state() in do_reset().
1672          * In the former case, no one else is changing the state (again we
1673          * have the rtnl) and in the latter case, do_reset() will detect and
1674          * honor our setting below.
1675          */
1676         if (adapter->failover_pending || (test_bit(0, &adapter->resetting))) {
1677                 netdev_dbg(netdev, "[S:%s FOP:%d] Resetting, deferring open\n",
1678                            adapter_state_to_string(adapter->state),
1679                            adapter->failover_pending);
1680                 adapter->state = VNIC_OPEN;
1681                 rc = 0;
1682                 goto out;
1683         }
1684
1685         if (adapter->state != VNIC_CLOSED) {
1686                 rc = ibmvnic_login(netdev);
1687                 if (rc)
1688                         goto out;
1689
1690                 rc = init_resources(adapter);
1691                 if (rc) {
1692                         netdev_err(netdev, "failed to initialize resources\n");
1693                         goto out;
1694                 }
1695         }
1696
1697         rc = __ibmvnic_open(netdev);
1698
1699 out:
1700         /* If open failed and there is a pending failover or in-progress reset,
1701          * set device state and return. Device operation will be handled by
1702          * reset routine. See also comments above regarding rtnl.
1703          */
1704         if (rc &&
1705             (adapter->failover_pending || (test_bit(0, &adapter->resetting)))) {
1706                 adapter->state = VNIC_OPEN;
1707                 rc = 0;
1708         }
1709
1710         if (rc) {
1711                 release_resources(adapter);
1712                 release_rx_pools(adapter);
1713                 release_tx_pools(adapter);
1714         }
1715
1716         return rc;
1717 }
1718
1719 static void clean_rx_pools(struct ibmvnic_adapter *adapter)
1720 {
1721         struct ibmvnic_rx_pool *rx_pool;
1722         struct ibmvnic_rx_buff *rx_buff;
1723         u64 rx_entries;
1724         int rx_scrqs;
1725         int i, j;
1726
1727         if (!adapter->rx_pool)
1728                 return;
1729
1730         rx_scrqs = adapter->num_active_rx_pools;
1731         rx_entries = adapter->req_rx_add_entries_per_subcrq;
1732
1733         /* Free any remaining skbs in the rx buffer pools */
1734         for (i = 0; i < rx_scrqs; i++) {
1735                 rx_pool = &adapter->rx_pool[i];
1736                 if (!rx_pool || !rx_pool->rx_buff)
1737                         continue;
1738
1739                 netdev_dbg(adapter->netdev, "Cleaning rx_pool[%d]\n", i);
1740                 for (j = 0; j < rx_entries; j++) {
1741                         rx_buff = &rx_pool->rx_buff[j];
1742                         if (rx_buff && rx_buff->skb) {
1743                                 dev_kfree_skb_any(rx_buff->skb);
1744                                 rx_buff->skb = NULL;
1745                         }
1746                 }
1747         }
1748 }
1749
1750 static void clean_one_tx_pool(struct ibmvnic_adapter *adapter,
1751                               struct ibmvnic_tx_pool *tx_pool)
1752 {
1753         struct ibmvnic_tx_buff *tx_buff;
1754         u64 tx_entries;
1755         int i;
1756
1757         if (!tx_pool || !tx_pool->tx_buff)
1758                 return;
1759
1760         tx_entries = tx_pool->num_buffers;
1761
1762         for (i = 0; i < tx_entries; i++) {
1763                 tx_buff = &tx_pool->tx_buff[i];
1764                 if (tx_buff && tx_buff->skb) {
1765                         dev_kfree_skb_any(tx_buff->skb);
1766                         tx_buff->skb = NULL;
1767                 }
1768         }
1769 }
1770
1771 static void clean_tx_pools(struct ibmvnic_adapter *adapter)
1772 {
1773         int tx_scrqs;
1774         int i;
1775
1776         if (!adapter->tx_pool || !adapter->tso_pool)
1777                 return;
1778
1779         tx_scrqs = adapter->num_active_tx_pools;
1780
1781         /* Free any remaining skbs in the tx buffer pools */
1782         for (i = 0; i < tx_scrqs; i++) {
1783                 netdev_dbg(adapter->netdev, "Cleaning tx_pool[%d]\n", i);
1784                 clean_one_tx_pool(adapter, &adapter->tx_pool[i]);
1785                 clean_one_tx_pool(adapter, &adapter->tso_pool[i]);
1786         }
1787 }
1788
1789 static void ibmvnic_disable_irqs(struct ibmvnic_adapter *adapter)
1790 {
1791         struct net_device *netdev = adapter->netdev;
1792         int i;
1793
1794         if (adapter->tx_scrq) {
1795                 for (i = 0; i < adapter->req_tx_queues; i++)
1796                         if (adapter->tx_scrq[i]->irq) {
1797                                 netdev_dbg(netdev,
1798                                            "Disabling tx_scrq[%d] irq\n", i);
1799                                 disable_scrq_irq(adapter, adapter->tx_scrq[i]);
1800                                 disable_irq(adapter->tx_scrq[i]->irq);
1801                         }
1802         }
1803
1804         if (adapter->rx_scrq) {
1805                 for (i = 0; i < adapter->req_rx_queues; i++) {
1806                         if (adapter->rx_scrq[i]->irq) {
1807                                 netdev_dbg(netdev,
1808                                            "Disabling rx_scrq[%d] irq\n", i);
1809                                 disable_scrq_irq(adapter, adapter->rx_scrq[i]);
1810                                 disable_irq(adapter->rx_scrq[i]->irq);
1811                         }
1812                 }
1813         }
1814 }
1815
1816 static void ibmvnic_cleanup(struct net_device *netdev)
1817 {
1818         struct ibmvnic_adapter *adapter = netdev_priv(netdev);
1819
1820         /* ensure that transmissions are stopped if called by do_reset */
1821
1822         adapter->tx_queues_active = false;
1823
1824         /* Ensure complete_tx() and ibmvnic_xmit() see ->tx_queues_active
1825          * update so they don't restart a queue after we stop it below.
1826          */
1827         synchronize_rcu();
1828
1829         if (test_bit(0, &adapter->resetting))
1830                 netif_tx_disable(netdev);
1831         else
1832                 netif_tx_stop_all_queues(netdev);
1833
1834         ibmvnic_napi_disable(adapter);
1835         ibmvnic_disable_irqs(adapter);
1836 }
1837
1838 static int __ibmvnic_close(struct net_device *netdev)
1839 {
1840         struct ibmvnic_adapter *adapter = netdev_priv(netdev);
1841         int rc = 0;
1842
1843         adapter->state = VNIC_CLOSING;
1844         rc = set_link_state(adapter, IBMVNIC_LOGICAL_LNK_DN);
1845         adapter->state = VNIC_CLOSED;
1846         return rc;
1847 }
1848
1849 static int ibmvnic_close(struct net_device *netdev)
1850 {
1851         struct ibmvnic_adapter *adapter = netdev_priv(netdev);
1852         int rc;
1853
1854         netdev_dbg(netdev, "[S:%s FOP:%d FRR:%d] Closing\n",
1855                    adapter_state_to_string(adapter->state),
1856                    adapter->failover_pending,
1857                    adapter->force_reset_recovery);
1858
1859         /* If device failover is pending, just set device state and return.
1860          * Device operation will be handled by reset routine.
1861          */
1862         if (adapter->failover_pending) {
1863                 adapter->state = VNIC_CLOSED;
1864                 return 0;
1865         }
1866
1867         rc = __ibmvnic_close(netdev);
1868         ibmvnic_cleanup(netdev);
1869         clean_rx_pools(adapter);
1870         clean_tx_pools(adapter);
1871
1872         return rc;
1873 }
1874
1875 /**
1876  * build_hdr_data - creates L2/L3/L4 header data buffer
1877  * @hdr_field: bitfield determining needed headers
1878  * @skb: socket buffer
1879  * @hdr_len: array of header lengths
1880  * @hdr_data: buffer to write the header to
1881  *
1882  * Reads hdr_field to determine which headers are needed by firmware.
1883  * Builds a buffer containing these headers.  Saves individual header
1884  * lengths and total buffer length to be used to build descriptors.
1885  */
1886 static int build_hdr_data(u8 hdr_field, struct sk_buff *skb,
1887                           int *hdr_len, u8 *hdr_data)
1888 {
1889         int len = 0;
1890         u8 *hdr;
1891
1892         if (skb_vlan_tagged(skb) && !skb_vlan_tag_present(skb))
1893                 hdr_len[0] = sizeof(struct vlan_ethhdr);
1894         else
1895                 hdr_len[0] = sizeof(struct ethhdr);
1896
1897         if (skb->protocol == htons(ETH_P_IP)) {
1898                 hdr_len[1] = ip_hdr(skb)->ihl * 4;
1899                 if (ip_hdr(skb)->protocol == IPPROTO_TCP)
1900                         hdr_len[2] = tcp_hdrlen(skb);
1901                 else if (ip_hdr(skb)->protocol == IPPROTO_UDP)
1902                         hdr_len[2] = sizeof(struct udphdr);
1903         } else if (skb->protocol == htons(ETH_P_IPV6)) {
1904                 hdr_len[1] = sizeof(struct ipv6hdr);
1905                 if (ipv6_hdr(skb)->nexthdr == IPPROTO_TCP)
1906                         hdr_len[2] = tcp_hdrlen(skb);
1907                 else if (ipv6_hdr(skb)->nexthdr == IPPROTO_UDP)
1908                         hdr_len[2] = sizeof(struct udphdr);
1909         } else if (skb->protocol == htons(ETH_P_ARP)) {
1910                 hdr_len[1] = arp_hdr_len(skb->dev);
1911                 hdr_len[2] = 0;
1912         }
1913
1914         memset(hdr_data, 0, 120);
1915         if ((hdr_field >> 6) & 1) {
1916                 hdr = skb_mac_header(skb);
1917                 memcpy(hdr_data, hdr, hdr_len[0]);
1918                 len += hdr_len[0];
1919         }
1920
1921         if ((hdr_field >> 5) & 1) {
1922                 hdr = skb_network_header(skb);
1923                 memcpy(hdr_data + len, hdr, hdr_len[1]);
1924                 len += hdr_len[1];
1925         }
1926
1927         if ((hdr_field >> 4) & 1) {
1928                 hdr = skb_transport_header(skb);
1929                 memcpy(hdr_data + len, hdr, hdr_len[2]);
1930                 len += hdr_len[2];
1931         }
1932         return len;
1933 }
1934
1935 /**
1936  * create_hdr_descs - create header and header extension descriptors
1937  * @hdr_field: bitfield determining needed headers
1938  * @hdr_data: buffer containing header data
1939  * @len: length of data buffer
1940  * @hdr_len: array of individual header lengths
1941  * @scrq_arr: descriptor array
1942  *
1943  * Creates header and, if needed, header extension descriptors and
1944  * places them in a descriptor array, scrq_arr
1945  */
1946
1947 static int create_hdr_descs(u8 hdr_field, u8 *hdr_data, int len, int *hdr_len,
1948                             union sub_crq *scrq_arr)
1949 {
1950         union sub_crq hdr_desc;
1951         int tmp_len = len;
1952         int num_descs = 0;
1953         u8 *data, *cur;
1954         int tmp;
1955
1956         while (tmp_len > 0) {
1957                 cur = hdr_data + len - tmp_len;
1958
1959                 memset(&hdr_desc, 0, sizeof(hdr_desc));
1960                 if (cur != hdr_data) {
1961                         data = hdr_desc.hdr_ext.data;
1962                         tmp = tmp_len > 29 ? 29 : tmp_len;
1963                         hdr_desc.hdr_ext.first = IBMVNIC_CRQ_CMD;
1964                         hdr_desc.hdr_ext.type = IBMVNIC_HDR_EXT_DESC;
1965                         hdr_desc.hdr_ext.len = tmp;
1966                 } else {
1967                         data = hdr_desc.hdr.data;
1968                         tmp = tmp_len > 24 ? 24 : tmp_len;
1969                         hdr_desc.hdr.first = IBMVNIC_CRQ_CMD;
1970                         hdr_desc.hdr.type = IBMVNIC_HDR_DESC;
1971                         hdr_desc.hdr.len = tmp;
1972                         hdr_desc.hdr.l2_len = (u8)hdr_len[0];
1973                         hdr_desc.hdr.l3_len = cpu_to_be16((u16)hdr_len[1]);
1974                         hdr_desc.hdr.l4_len = (u8)hdr_len[2];
1975                         hdr_desc.hdr.flag = hdr_field << 1;
1976                 }
1977                 memcpy(data, cur, tmp);
1978                 tmp_len -= tmp;
1979                 *scrq_arr = hdr_desc;
1980                 scrq_arr++;
1981                 num_descs++;
1982         }
1983
1984         return num_descs;
1985 }
1986
1987 /**
1988  * build_hdr_descs_arr - build a header descriptor array
1989  * @skb: tx socket buffer
1990  * @indir_arr: indirect array
1991  * @num_entries: number of descriptors to be sent
1992  * @hdr_field: bit field determining which headers will be sent
1993  *
1994  * This function will build a TX descriptor array with applicable
1995  * L2/L3/L4 packet header descriptors to be sent by send_subcrq_indirect.
1996  */
1997
1998 static void build_hdr_descs_arr(struct sk_buff *skb,
1999                                 union sub_crq *indir_arr,
2000                                 int *num_entries, u8 hdr_field)
2001 {
2002         int hdr_len[3] = {0, 0, 0};
2003         u8 hdr_data[140] = {0};
2004         int tot_len;
2005
2006         tot_len = build_hdr_data(hdr_field, skb, hdr_len,
2007                                  hdr_data);
2008         *num_entries += create_hdr_descs(hdr_field, hdr_data, tot_len, hdr_len,
2009                                          indir_arr + 1);
2010 }
2011
2012 static int ibmvnic_xmit_workarounds(struct sk_buff *skb,
2013                                     struct net_device *netdev)
2014 {
2015         /* For some backing devices, mishandling of small packets
2016          * can result in a loss of connection or TX stall. Device
2017          * architects recommend that no packet should be smaller
2018          * than the minimum MTU value provided to the driver, so
2019          * pad any packets to that length
2020          */
2021         if (skb->len < netdev->min_mtu)
2022                 return skb_put_padto(skb, netdev->min_mtu);
2023
2024         return 0;
2025 }
2026
2027 static void ibmvnic_tx_scrq_clean_buffer(struct ibmvnic_adapter *adapter,
2028                                          struct ibmvnic_sub_crq_queue *tx_scrq)
2029 {
2030         struct ibmvnic_ind_xmit_queue *ind_bufp;
2031         struct ibmvnic_tx_buff *tx_buff;
2032         struct ibmvnic_tx_pool *tx_pool;
2033         union sub_crq tx_scrq_entry;
2034         int queue_num;
2035         int entries;
2036         int index;
2037         int i;
2038
2039         ind_bufp = &tx_scrq->ind_buf;
2040         entries = (u64)ind_bufp->index;
2041         queue_num = tx_scrq->pool_index;
2042
2043         for (i = entries - 1; i >= 0; --i) {
2044                 tx_scrq_entry = ind_bufp->indir_arr[i];
2045                 if (tx_scrq_entry.v1.type != IBMVNIC_TX_DESC)
2046                         continue;
2047                 index = be32_to_cpu(tx_scrq_entry.v1.correlator);
2048                 if (index & IBMVNIC_TSO_POOL_MASK) {
2049                         tx_pool = &adapter->tso_pool[queue_num];
2050                         index &= ~IBMVNIC_TSO_POOL_MASK;
2051                 } else {
2052                         tx_pool = &adapter->tx_pool[queue_num];
2053                 }
2054                 tx_pool->free_map[tx_pool->consumer_index] = index;
2055                 tx_pool->consumer_index = tx_pool->consumer_index == 0 ?
2056                                           tx_pool->num_buffers - 1 :
2057                                           tx_pool->consumer_index - 1;
2058                 tx_buff = &tx_pool->tx_buff[index];
2059                 adapter->netdev->stats.tx_packets--;
2060                 adapter->netdev->stats.tx_bytes -= tx_buff->skb->len;
2061                 adapter->tx_stats_buffers[queue_num].packets--;
2062                 adapter->tx_stats_buffers[queue_num].bytes -=
2063                                                 tx_buff->skb->len;
2064                 dev_kfree_skb_any(tx_buff->skb);
2065                 tx_buff->skb = NULL;
2066                 adapter->netdev->stats.tx_dropped++;
2067         }
2068
2069         ind_bufp->index = 0;
2070
2071         if (atomic_sub_return(entries, &tx_scrq->used) <=
2072             (adapter->req_tx_entries_per_subcrq / 2) &&
2073             __netif_subqueue_stopped(adapter->netdev, queue_num)) {
2074                 rcu_read_lock();
2075
2076                 if (adapter->tx_queues_active) {
2077                         netif_wake_subqueue(adapter->netdev, queue_num);
2078                         netdev_dbg(adapter->netdev, "Started queue %d\n",
2079                                    queue_num);
2080                 }
2081
2082                 rcu_read_unlock();
2083         }
2084 }
2085
2086 static int ibmvnic_tx_scrq_flush(struct ibmvnic_adapter *adapter,
2087                                  struct ibmvnic_sub_crq_queue *tx_scrq)
2088 {
2089         struct ibmvnic_ind_xmit_queue *ind_bufp;
2090         u64 dma_addr;
2091         u64 entries;
2092         u64 handle;
2093         int rc;
2094
2095         ind_bufp = &tx_scrq->ind_buf;
2096         dma_addr = (u64)ind_bufp->indir_dma;
2097         entries = (u64)ind_bufp->index;
2098         handle = tx_scrq->handle;
2099
2100         if (!entries)
2101                 return 0;
2102         rc = send_subcrq_indirect(adapter, handle, dma_addr, entries);
2103         if (rc)
2104                 ibmvnic_tx_scrq_clean_buffer(adapter, tx_scrq);
2105         else
2106                 ind_bufp->index = 0;
2107         return 0;
2108 }
2109
2110 static netdev_tx_t ibmvnic_xmit(struct sk_buff *skb, struct net_device *netdev)
2111 {
2112         struct ibmvnic_adapter *adapter = netdev_priv(netdev);
2113         int queue_num = skb_get_queue_mapping(skb);
2114         u8 *hdrs = (u8 *)&adapter->tx_rx_desc_req;
2115         struct device *dev = &adapter->vdev->dev;
2116         struct ibmvnic_ind_xmit_queue *ind_bufp;
2117         struct ibmvnic_tx_buff *tx_buff = NULL;
2118         struct ibmvnic_sub_crq_queue *tx_scrq;
2119         struct ibmvnic_long_term_buff *ltb;
2120         struct ibmvnic_tx_pool *tx_pool;
2121         unsigned int tx_send_failed = 0;
2122         netdev_tx_t ret = NETDEV_TX_OK;
2123         unsigned int tx_map_failed = 0;
2124         union sub_crq indir_arr[16];
2125         unsigned int tx_dropped = 0;
2126         unsigned int tx_packets = 0;
2127         unsigned int tx_bytes = 0;
2128         dma_addr_t data_dma_addr;
2129         struct netdev_queue *txq;
2130         unsigned long lpar_rc;
2131         union sub_crq tx_crq;
2132         unsigned int offset;
2133         int num_entries = 1;
2134         unsigned char *dst;
2135         int bufidx = 0;
2136         u8 proto = 0;
2137
2138         /* If a reset is in progress, drop the packet since
2139          * the scrqs may get torn down. Otherwise use the
2140          * rcu to ensure reset waits for us to complete.
2141          */
2142         rcu_read_lock();
2143         if (!adapter->tx_queues_active) {
2144                 dev_kfree_skb_any(skb);
2145
2146                 tx_send_failed++;
2147                 tx_dropped++;
2148                 ret = NETDEV_TX_OK;
2149                 goto out;
2150         }
2151
2152         tx_scrq = adapter->tx_scrq[queue_num];
2153         txq = netdev_get_tx_queue(netdev, queue_num);
2154         ind_bufp = &tx_scrq->ind_buf;
2155
2156         if (ibmvnic_xmit_workarounds(skb, netdev)) {
2157                 tx_dropped++;
2158                 tx_send_failed++;
2159                 ret = NETDEV_TX_OK;
2160                 ibmvnic_tx_scrq_flush(adapter, tx_scrq);
2161                 goto out;
2162         }
2163
2164         if (skb_is_gso(skb))
2165                 tx_pool = &adapter->tso_pool[queue_num];
2166         else
2167                 tx_pool = &adapter->tx_pool[queue_num];
2168
2169         bufidx = tx_pool->free_map[tx_pool->consumer_index];
2170
2171         if (bufidx == IBMVNIC_INVALID_MAP) {
2172                 dev_kfree_skb_any(skb);
2173                 tx_send_failed++;
2174                 tx_dropped++;
2175                 ibmvnic_tx_scrq_flush(adapter, tx_scrq);
2176                 ret = NETDEV_TX_OK;
2177                 goto out;
2178         }
2179
2180         tx_pool->free_map[tx_pool->consumer_index] = IBMVNIC_INVALID_MAP;
2181
2182         map_txpool_buf_to_ltb(tx_pool, bufidx, &ltb, &offset);
2183
2184         dst = ltb->buff + offset;
2185         memset(dst, 0, tx_pool->buf_size);
2186         data_dma_addr = ltb->addr + offset;
2187
2188         if (skb_shinfo(skb)->nr_frags) {
2189                 int cur, i;
2190
2191                 /* Copy the head */
2192                 skb_copy_from_linear_data(skb, dst, skb_headlen(skb));
2193                 cur = skb_headlen(skb);
2194
2195                 /* Copy the frags */
2196                 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
2197                         const skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
2198
2199                         memcpy(dst + cur, skb_frag_address(frag),
2200                                skb_frag_size(frag));
2201                         cur += skb_frag_size(frag);
2202                 }
2203         } else {
2204                 skb_copy_from_linear_data(skb, dst, skb->len);
2205         }
2206
2207         /* post changes to long_term_buff *dst before VIOS accessing it */
2208         dma_wmb();
2209
2210         tx_pool->consumer_index =
2211             (tx_pool->consumer_index + 1) % tx_pool->num_buffers;
2212
2213         tx_buff = &tx_pool->tx_buff[bufidx];
2214         tx_buff->skb = skb;
2215         tx_buff->index = bufidx;
2216         tx_buff->pool_index = queue_num;
2217
2218         memset(&tx_crq, 0, sizeof(tx_crq));
2219         tx_crq.v1.first = IBMVNIC_CRQ_CMD;
2220         tx_crq.v1.type = IBMVNIC_TX_DESC;
2221         tx_crq.v1.n_crq_elem = 1;
2222         tx_crq.v1.n_sge = 1;
2223         tx_crq.v1.flags1 = IBMVNIC_TX_COMP_NEEDED;
2224
2225         if (skb_is_gso(skb))
2226                 tx_crq.v1.correlator =
2227                         cpu_to_be32(bufidx | IBMVNIC_TSO_POOL_MASK);
2228         else
2229                 tx_crq.v1.correlator = cpu_to_be32(bufidx);
2230         tx_crq.v1.dma_reg = cpu_to_be16(ltb->map_id);
2231         tx_crq.v1.sge_len = cpu_to_be32(skb->len);
2232         tx_crq.v1.ioba = cpu_to_be64(data_dma_addr);
2233
2234         if (adapter->vlan_header_insertion && skb_vlan_tag_present(skb)) {
2235                 tx_crq.v1.flags2 |= IBMVNIC_TX_VLAN_INSERT;
2236                 tx_crq.v1.vlan_id = cpu_to_be16(skb->vlan_tci);
2237         }
2238
2239         if (skb->protocol == htons(ETH_P_IP)) {
2240                 tx_crq.v1.flags1 |= IBMVNIC_TX_PROT_IPV4;
2241                 proto = ip_hdr(skb)->protocol;
2242         } else if (skb->protocol == htons(ETH_P_IPV6)) {
2243                 tx_crq.v1.flags1 |= IBMVNIC_TX_PROT_IPV6;
2244                 proto = ipv6_hdr(skb)->nexthdr;
2245         }
2246
2247         if (proto == IPPROTO_TCP)
2248                 tx_crq.v1.flags1 |= IBMVNIC_TX_PROT_TCP;
2249         else if (proto == IPPROTO_UDP)
2250                 tx_crq.v1.flags1 |= IBMVNIC_TX_PROT_UDP;
2251
2252         if (skb->ip_summed == CHECKSUM_PARTIAL) {
2253                 tx_crq.v1.flags1 |= IBMVNIC_TX_CHKSUM_OFFLOAD;
2254                 hdrs += 2;
2255         }
2256         if (skb_is_gso(skb)) {
2257                 tx_crq.v1.flags1 |= IBMVNIC_TX_LSO;
2258                 tx_crq.v1.mss = cpu_to_be16(skb_shinfo(skb)->gso_size);
2259                 hdrs += 2;
2260         }
2261
2262         if ((*hdrs >> 7) & 1)
2263                 build_hdr_descs_arr(skb, indir_arr, &num_entries, *hdrs);
2264
2265         tx_crq.v1.n_crq_elem = num_entries;
2266         tx_buff->num_entries = num_entries;
2267         /* flush buffer if current entry can not fit */
2268         if (num_entries + ind_bufp->index > IBMVNIC_MAX_IND_DESCS) {
2269                 lpar_rc = ibmvnic_tx_scrq_flush(adapter, tx_scrq);
2270                 if (lpar_rc != H_SUCCESS)
2271                         goto tx_flush_err;
2272         }
2273
2274         indir_arr[0] = tx_crq;
2275         memcpy(&ind_bufp->indir_arr[ind_bufp->index], &indir_arr[0],
2276                num_entries * sizeof(struct ibmvnic_generic_scrq));
2277         ind_bufp->index += num_entries;
2278         if (__netdev_tx_sent_queue(txq, skb->len,
2279                                    netdev_xmit_more() &&
2280                                    ind_bufp->index < IBMVNIC_MAX_IND_DESCS)) {
2281                 lpar_rc = ibmvnic_tx_scrq_flush(adapter, tx_scrq);
2282                 if (lpar_rc != H_SUCCESS)
2283                         goto tx_err;
2284         }
2285
2286         if (atomic_add_return(num_entries, &tx_scrq->used)
2287                                         >= adapter->req_tx_entries_per_subcrq) {
2288                 netdev_dbg(netdev, "Stopping queue %d\n", queue_num);
2289                 netif_stop_subqueue(netdev, queue_num);
2290         }
2291
2292         tx_packets++;
2293         tx_bytes += skb->len;
2294         txq_trans_cond_update(txq);
2295         ret = NETDEV_TX_OK;
2296         goto out;
2297
2298 tx_flush_err:
2299         dev_kfree_skb_any(skb);
2300         tx_buff->skb = NULL;
2301         tx_pool->consumer_index = tx_pool->consumer_index == 0 ?
2302                                   tx_pool->num_buffers - 1 :
2303                                   tx_pool->consumer_index - 1;
2304         tx_dropped++;
2305 tx_err:
2306         if (lpar_rc != H_CLOSED && lpar_rc != H_PARAMETER)
2307                 dev_err_ratelimited(dev, "tx: send failed\n");
2308
2309         if (lpar_rc == H_CLOSED || adapter->failover_pending) {
2310                 /* Disable TX and report carrier off if queue is closed
2311                  * or pending failover.
2312                  * Firmware guarantees that a signal will be sent to the
2313                  * driver, triggering a reset or some other action.
2314                  */
2315                 netif_tx_stop_all_queues(netdev);
2316                 netif_carrier_off(netdev);
2317         }
2318 out:
2319         rcu_read_unlock();
2320         netdev->stats.tx_dropped += tx_dropped;
2321         netdev->stats.tx_bytes += tx_bytes;
2322         netdev->stats.tx_packets += tx_packets;
2323         adapter->tx_send_failed += tx_send_failed;
2324         adapter->tx_map_failed += tx_map_failed;
2325         adapter->tx_stats_buffers[queue_num].packets += tx_packets;
2326         adapter->tx_stats_buffers[queue_num].bytes += tx_bytes;
2327         adapter->tx_stats_buffers[queue_num].dropped_packets += tx_dropped;
2328
2329         return ret;
2330 }
2331
2332 static void ibmvnic_set_multi(struct net_device *netdev)
2333 {
2334         struct ibmvnic_adapter *adapter = netdev_priv(netdev);
2335         struct netdev_hw_addr *ha;
2336         union ibmvnic_crq crq;
2337
2338         memset(&crq, 0, sizeof(crq));
2339         crq.request_capability.first = IBMVNIC_CRQ_CMD;
2340         crq.request_capability.cmd = REQUEST_CAPABILITY;
2341
2342         if (netdev->flags & IFF_PROMISC) {
2343                 if (!adapter->promisc_supported)
2344                         return;
2345         } else {
2346                 if (netdev->flags & IFF_ALLMULTI) {
2347                         /* Accept all multicast */
2348                         memset(&crq, 0, sizeof(crq));
2349                         crq.multicast_ctrl.first = IBMVNIC_CRQ_CMD;
2350                         crq.multicast_ctrl.cmd = MULTICAST_CTRL;
2351                         crq.multicast_ctrl.flags = IBMVNIC_ENABLE_ALL;
2352                         ibmvnic_send_crq(adapter, &crq);
2353                 } else if (netdev_mc_empty(netdev)) {
2354                         /* Reject all multicast */
2355                         memset(&crq, 0, sizeof(crq));
2356                         crq.multicast_ctrl.first = IBMVNIC_CRQ_CMD;
2357                         crq.multicast_ctrl.cmd = MULTICAST_CTRL;
2358                         crq.multicast_ctrl.flags = IBMVNIC_DISABLE_ALL;
2359                         ibmvnic_send_crq(adapter, &crq);
2360                 } else {
2361                         /* Accept one or more multicast(s) */
2362                         netdev_for_each_mc_addr(ha, netdev) {
2363                                 memset(&crq, 0, sizeof(crq));
2364                                 crq.multicast_ctrl.first = IBMVNIC_CRQ_CMD;
2365                                 crq.multicast_ctrl.cmd = MULTICAST_CTRL;
2366                                 crq.multicast_ctrl.flags = IBMVNIC_ENABLE_MC;
2367                                 ether_addr_copy(&crq.multicast_ctrl.mac_addr[0],
2368                                                 ha->addr);
2369                                 ibmvnic_send_crq(adapter, &crq);
2370                         }
2371                 }
2372         }
2373 }
2374
2375 static int __ibmvnic_set_mac(struct net_device *netdev, u8 *dev_addr)
2376 {
2377         struct ibmvnic_adapter *adapter = netdev_priv(netdev);
2378         union ibmvnic_crq crq;
2379         int rc;
2380
2381         if (!is_valid_ether_addr(dev_addr)) {
2382                 rc = -EADDRNOTAVAIL;
2383                 goto err;
2384         }
2385
2386         memset(&crq, 0, sizeof(crq));
2387         crq.change_mac_addr.first = IBMVNIC_CRQ_CMD;
2388         crq.change_mac_addr.cmd = CHANGE_MAC_ADDR;
2389         ether_addr_copy(&crq.change_mac_addr.mac_addr[0], dev_addr);
2390
2391         mutex_lock(&adapter->fw_lock);
2392         adapter->fw_done_rc = 0;
2393         reinit_completion(&adapter->fw_done);
2394
2395         rc = ibmvnic_send_crq(adapter, &crq);
2396         if (rc) {
2397                 rc = -EIO;
2398                 mutex_unlock(&adapter->fw_lock);
2399                 goto err;
2400         }
2401
2402         rc = ibmvnic_wait_for_completion(adapter, &adapter->fw_done, 10000);
2403         /* netdev->dev_addr is changed in handle_change_mac_rsp function */
2404         if (rc || adapter->fw_done_rc) {
2405                 rc = -EIO;
2406                 mutex_unlock(&adapter->fw_lock);
2407                 goto err;
2408         }
2409         mutex_unlock(&adapter->fw_lock);
2410         return 0;
2411 err:
2412         ether_addr_copy(adapter->mac_addr, netdev->dev_addr);
2413         return rc;
2414 }
2415
2416 static int ibmvnic_set_mac(struct net_device *netdev, void *p)
2417 {
2418         struct ibmvnic_adapter *adapter = netdev_priv(netdev);
2419         struct sockaddr *addr = p;
2420         int rc;
2421
2422         rc = 0;
2423         if (!is_valid_ether_addr(addr->sa_data))
2424                 return -EADDRNOTAVAIL;
2425
2426         ether_addr_copy(adapter->mac_addr, addr->sa_data);
2427         if (adapter->state != VNIC_PROBED)
2428                 rc = __ibmvnic_set_mac(netdev, addr->sa_data);
2429
2430         return rc;
2431 }
2432
2433 static const char *reset_reason_to_string(enum ibmvnic_reset_reason reason)
2434 {
2435         switch (reason) {
2436         case VNIC_RESET_FAILOVER:
2437                 return "FAILOVER";
2438         case VNIC_RESET_MOBILITY:
2439                 return "MOBILITY";
2440         case VNIC_RESET_FATAL:
2441                 return "FATAL";
2442         case VNIC_RESET_NON_FATAL:
2443                 return "NON_FATAL";
2444         case VNIC_RESET_TIMEOUT:
2445                 return "TIMEOUT";
2446         case VNIC_RESET_CHANGE_PARAM:
2447                 return "CHANGE_PARAM";
2448         case VNIC_RESET_PASSIVE_INIT:
2449                 return "PASSIVE_INIT";
2450         }
2451         return "UNKNOWN";
2452 }
2453
2454 /*
2455  * Initialize the init_done completion and return code values. We
2456  * can get a transport event just after registering the CRQ and the
2457  * tasklet will use this to communicate the transport event. To ensure
2458  * we don't miss the notification/error, initialize these _before_
2459  * regisering the CRQ.
2460  */
2461 static inline void reinit_init_done(struct ibmvnic_adapter *adapter)
2462 {
2463         reinit_completion(&adapter->init_done);
2464         adapter->init_done_rc = 0;
2465 }
2466
2467 /*
2468  * do_reset returns zero if we are able to keep processing reset events, or
2469  * non-zero if we hit a fatal error and must halt.
2470  */
2471 static int do_reset(struct ibmvnic_adapter *adapter,
2472                     struct ibmvnic_rwi *rwi, u32 reset_state)
2473 {
2474         struct net_device *netdev = adapter->netdev;
2475         u64 old_num_rx_queues, old_num_tx_queues;
2476         u64 old_num_rx_slots, old_num_tx_slots;
2477         int rc;
2478
2479         netdev_dbg(adapter->netdev,
2480                    "[S:%s FOP:%d] Reset reason: %s, reset_state: %s\n",
2481                    adapter_state_to_string(adapter->state),
2482                    adapter->failover_pending,
2483                    reset_reason_to_string(rwi->reset_reason),
2484                    adapter_state_to_string(reset_state));
2485
2486         adapter->reset_reason = rwi->reset_reason;
2487         /* requestor of VNIC_RESET_CHANGE_PARAM already has the rtnl lock */
2488         if (!(adapter->reset_reason == VNIC_RESET_CHANGE_PARAM))
2489                 rtnl_lock();
2490
2491         /* Now that we have the rtnl lock, clear any pending failover.
2492          * This will ensure ibmvnic_open() has either completed or will
2493          * block until failover is complete.
2494          */
2495         if (rwi->reset_reason == VNIC_RESET_FAILOVER)
2496                 adapter->failover_pending = false;
2497
2498         /* read the state and check (again) after getting rtnl */
2499         reset_state = adapter->state;
2500
2501         if (reset_state == VNIC_REMOVING || reset_state == VNIC_REMOVED) {
2502                 rc = -EBUSY;
2503                 goto out;
2504         }
2505
2506         netif_carrier_off(netdev);
2507
2508         old_num_rx_queues = adapter->req_rx_queues;
2509         old_num_tx_queues = adapter->req_tx_queues;
2510         old_num_rx_slots = adapter->req_rx_add_entries_per_subcrq;
2511         old_num_tx_slots = adapter->req_tx_entries_per_subcrq;
2512
2513         ibmvnic_cleanup(netdev);
2514
2515         if (reset_state == VNIC_OPEN &&
2516             adapter->reset_reason != VNIC_RESET_MOBILITY &&
2517             adapter->reset_reason != VNIC_RESET_FAILOVER) {
2518                 if (adapter->reset_reason == VNIC_RESET_CHANGE_PARAM) {
2519                         rc = __ibmvnic_close(netdev);
2520                         if (rc)
2521                                 goto out;
2522                 } else {
2523                         adapter->state = VNIC_CLOSING;
2524
2525                         /* Release the RTNL lock before link state change and
2526                          * re-acquire after the link state change to allow
2527                          * linkwatch_event to grab the RTNL lock and run during
2528                          * a reset.
2529                          */
2530                         rtnl_unlock();
2531                         rc = set_link_state(adapter, IBMVNIC_LOGICAL_LNK_DN);
2532                         rtnl_lock();
2533                         if (rc)
2534                                 goto out;
2535
2536                         if (adapter->state == VNIC_OPEN) {
2537                                 /* When we dropped rtnl, ibmvnic_open() got
2538                                  * it and noticed that we are resetting and
2539                                  * set the adapter state to OPEN. Update our
2540                                  * new "target" state, and resume the reset
2541                                  * from VNIC_CLOSING state.
2542                                  */
2543                                 netdev_dbg(netdev,
2544                                            "Open changed state from %s, updating.\n",
2545                                            adapter_state_to_string(reset_state));
2546                                 reset_state = VNIC_OPEN;
2547                                 adapter->state = VNIC_CLOSING;
2548                         }
2549
2550                         if (adapter->state != VNIC_CLOSING) {
2551                                 /* If someone else changed the adapter state
2552                                  * when we dropped the rtnl, fail the reset
2553                                  */
2554                                 rc = -EAGAIN;
2555                                 goto out;
2556                         }
2557                         adapter->state = VNIC_CLOSED;
2558                 }
2559         }
2560
2561         if (adapter->reset_reason == VNIC_RESET_CHANGE_PARAM) {
2562                 release_resources(adapter);
2563                 release_sub_crqs(adapter, 1);
2564                 release_crq_queue(adapter);
2565         }
2566
2567         if (adapter->reset_reason != VNIC_RESET_NON_FATAL) {
2568                 /* remove the closed state so when we call open it appears
2569                  * we are coming from the probed state.
2570                  */
2571                 adapter->state = VNIC_PROBED;
2572
2573                 reinit_init_done(adapter);
2574
2575                 if (adapter->reset_reason == VNIC_RESET_CHANGE_PARAM) {
2576                         rc = init_crq_queue(adapter);
2577                 } else if (adapter->reset_reason == VNIC_RESET_MOBILITY) {
2578                         rc = ibmvnic_reenable_crq_queue(adapter);
2579                         release_sub_crqs(adapter, 1);
2580                 } else {
2581                         rc = ibmvnic_reset_crq(adapter);
2582                         if (rc == H_CLOSED || rc == H_SUCCESS) {
2583                                 rc = vio_enable_interrupts(adapter->vdev);
2584                                 if (rc)
2585                                         netdev_err(adapter->netdev,
2586                                                    "Reset failed to enable interrupts. rc=%d\n",
2587                                                    rc);
2588                         }
2589                 }
2590
2591                 if (rc) {
2592                         netdev_err(adapter->netdev,
2593                                    "Reset couldn't initialize crq. rc=%d\n", rc);
2594                         goto out;
2595                 }
2596
2597                 rc = ibmvnic_reset_init(adapter, true);
2598                 if (rc)
2599                         goto out;
2600
2601                 /* If the adapter was in PROBE or DOWN state prior to the reset,
2602                  * exit here.
2603                  */
2604                 if (reset_state == VNIC_PROBED || reset_state == VNIC_DOWN) {
2605                         rc = 0;
2606                         goto out;
2607                 }
2608
2609                 rc = ibmvnic_login(netdev);
2610                 if (rc)
2611                         goto out;
2612
2613                 if (adapter->reset_reason == VNIC_RESET_CHANGE_PARAM) {
2614                         rc = init_resources(adapter);
2615                         if (rc)
2616                                 goto out;
2617                 } else if (adapter->req_rx_queues != old_num_rx_queues ||
2618                     adapter->req_tx_queues != old_num_tx_queues ||
2619                     adapter->req_rx_add_entries_per_subcrq !=
2620                     old_num_rx_slots ||
2621                     adapter->req_tx_entries_per_subcrq !=
2622                     old_num_tx_slots ||
2623                     !adapter->rx_pool ||
2624                     !adapter->tso_pool ||
2625                     !adapter->tx_pool) {
2626                         release_napi(adapter);
2627                         release_vpd_data(adapter);
2628
2629                         rc = init_resources(adapter);
2630                         if (rc)
2631                                 goto out;
2632
2633                 } else {
2634                         rc = init_tx_pools(netdev);
2635                         if (rc) {
2636                                 netdev_dbg(netdev,
2637                                            "init tx pools failed (%d)\n",
2638                                            rc);
2639                                 goto out;
2640                         }
2641
2642                         rc = init_rx_pools(netdev);
2643                         if (rc) {
2644                                 netdev_dbg(netdev,
2645                                            "init rx pools failed (%d)\n",
2646                                            rc);
2647                                 goto out;
2648                         }
2649                 }
2650                 ibmvnic_disable_irqs(adapter);
2651         }
2652         adapter->state = VNIC_CLOSED;
2653
2654         if (reset_state == VNIC_CLOSED) {
2655                 rc = 0;
2656                 goto out;
2657         }
2658
2659         rc = __ibmvnic_open(netdev);
2660         if (rc) {
2661                 rc = IBMVNIC_OPEN_FAILED;
2662                 goto out;
2663         }
2664
2665         /* refresh device's multicast list */
2666         ibmvnic_set_multi(netdev);
2667
2668         if (adapter->reset_reason == VNIC_RESET_FAILOVER ||
2669             adapter->reset_reason == VNIC_RESET_MOBILITY)
2670                 __netdev_notify_peers(netdev);
2671
2672         rc = 0;
2673
2674 out:
2675         /* restore the adapter state if reset failed */
2676         if (rc)
2677                 adapter->state = reset_state;
2678         /* requestor of VNIC_RESET_CHANGE_PARAM should still hold the rtnl lock */
2679         if (!(adapter->reset_reason == VNIC_RESET_CHANGE_PARAM))
2680                 rtnl_unlock();
2681
2682         netdev_dbg(adapter->netdev, "[S:%s FOP:%d] Reset done, rc %d\n",
2683                    adapter_state_to_string(adapter->state),
2684                    adapter->failover_pending, rc);
2685         return rc;
2686 }
2687
2688 static int do_hard_reset(struct ibmvnic_adapter *adapter,
2689                          struct ibmvnic_rwi *rwi, u32 reset_state)
2690 {
2691         struct net_device *netdev = adapter->netdev;
2692         int rc;
2693
2694         netdev_dbg(adapter->netdev, "Hard resetting driver (%s)\n",
2695                    reset_reason_to_string(rwi->reset_reason));
2696
2697         /* read the state and check (again) after getting rtnl */
2698         reset_state = adapter->state;
2699
2700         if (reset_state == VNIC_REMOVING || reset_state == VNIC_REMOVED) {
2701                 rc = -EBUSY;
2702                 goto out;
2703         }
2704
2705         netif_carrier_off(netdev);
2706         adapter->reset_reason = rwi->reset_reason;
2707
2708         ibmvnic_cleanup(netdev);
2709         release_resources(adapter);
2710         release_sub_crqs(adapter, 0);
2711         release_crq_queue(adapter);
2712
2713         /* remove the closed state so when we call open it appears
2714          * we are coming from the probed state.
2715          */
2716         adapter->state = VNIC_PROBED;
2717
2718         reinit_init_done(adapter);
2719
2720         rc = init_crq_queue(adapter);
2721         if (rc) {
2722                 netdev_err(adapter->netdev,
2723                            "Couldn't initialize crq. rc=%d\n", rc);
2724                 goto out;
2725         }
2726
2727         rc = ibmvnic_reset_init(adapter, false);
2728         if (rc)
2729                 goto out;
2730
2731         /* If the adapter was in PROBE or DOWN state prior to the reset,
2732          * exit here.
2733          */
2734         if (reset_state == VNIC_PROBED || reset_state == VNIC_DOWN)
2735                 goto out;
2736
2737         rc = ibmvnic_login(netdev);
2738         if (rc)
2739                 goto out;
2740
2741         rc = init_resources(adapter);
2742         if (rc)
2743                 goto out;
2744
2745         ibmvnic_disable_irqs(adapter);
2746         adapter->state = VNIC_CLOSED;
2747
2748         if (reset_state == VNIC_CLOSED)
2749                 goto out;
2750
2751         rc = __ibmvnic_open(netdev);
2752         if (rc) {
2753                 rc = IBMVNIC_OPEN_FAILED;
2754                 goto out;
2755         }
2756
2757         __netdev_notify_peers(netdev);
2758 out:
2759         /* restore adapter state if reset failed */
2760         if (rc)
2761                 adapter->state = reset_state;
2762         netdev_dbg(adapter->netdev, "[S:%s FOP:%d] Hard reset done, rc %d\n",
2763                    adapter_state_to_string(adapter->state),
2764                    adapter->failover_pending, rc);
2765         return rc;
2766 }
2767
2768 static struct ibmvnic_rwi *get_next_rwi(struct ibmvnic_adapter *adapter)
2769 {
2770         struct ibmvnic_rwi *rwi;
2771         unsigned long flags;
2772
2773         spin_lock_irqsave(&adapter->rwi_lock, flags);
2774
2775         if (!list_empty(&adapter->rwi_list)) {
2776                 rwi = list_first_entry(&adapter->rwi_list, struct ibmvnic_rwi,
2777                                        list);
2778                 list_del(&rwi->list);
2779         } else {
2780                 rwi = NULL;
2781         }
2782
2783         spin_unlock_irqrestore(&adapter->rwi_lock, flags);
2784         return rwi;
2785 }
2786
2787 /**
2788  * do_passive_init - complete probing when partner device is detected.
2789  * @adapter: ibmvnic_adapter struct
2790  *
2791  * If the ibmvnic device does not have a partner device to communicate with at boot
2792  * and that partner device comes online at a later time, this function is called
2793  * to complete the initialization process of ibmvnic device.
2794  * Caller is expected to hold rtnl_lock().
2795  *
2796  * Returns non-zero if sub-CRQs are not initialized properly leaving the device
2797  * in the down state.
2798  * Returns 0 upon success and the device is in PROBED state.
2799  */
2800
2801 static int do_passive_init(struct ibmvnic_adapter *adapter)
2802 {
2803         unsigned long timeout = msecs_to_jiffies(30000);
2804         struct net_device *netdev = adapter->netdev;
2805         struct device *dev = &adapter->vdev->dev;
2806         int rc;
2807
2808         netdev_dbg(netdev, "Partner device found, probing.\n");
2809
2810         adapter->state = VNIC_PROBING;
2811         reinit_completion(&adapter->init_done);
2812         adapter->init_done_rc = 0;
2813         adapter->crq.active = true;
2814
2815         rc = send_crq_init_complete(adapter);
2816         if (rc)
2817                 goto out;
2818
2819         rc = send_version_xchg(adapter);
2820         if (rc)
2821                 netdev_dbg(adapter->netdev, "send_version_xchg failed, rc=%d\n", rc);
2822
2823         if (!wait_for_completion_timeout(&adapter->init_done, timeout)) {
2824                 dev_err(dev, "Initialization sequence timed out\n");
2825                 rc = -ETIMEDOUT;
2826                 goto out;
2827         }
2828
2829         rc = init_sub_crqs(adapter);
2830         if (rc) {
2831                 dev_err(dev, "Initialization of sub crqs failed, rc=%d\n", rc);
2832                 goto out;
2833         }
2834
2835         rc = init_sub_crq_irqs(adapter);
2836         if (rc) {
2837                 dev_err(dev, "Failed to initialize sub crq irqs\n, rc=%d", rc);
2838                 goto init_failed;
2839         }
2840
2841         netdev->mtu = adapter->req_mtu - ETH_HLEN;
2842         netdev->min_mtu = adapter->min_mtu - ETH_HLEN;
2843         netdev->max_mtu = adapter->max_mtu - ETH_HLEN;
2844
2845         adapter->state = VNIC_PROBED;
2846         netdev_dbg(netdev, "Probed successfully. Waiting for signal from partner device.\n");
2847
2848         return 0;
2849
2850 init_failed:
2851         release_sub_crqs(adapter, 1);
2852 out:
2853         adapter->state = VNIC_DOWN;
2854         return rc;
2855 }
2856
2857 static void __ibmvnic_reset(struct work_struct *work)
2858 {
2859         struct ibmvnic_adapter *adapter;
2860         unsigned int timeout = 5000;
2861         struct ibmvnic_rwi *tmprwi;
2862         bool saved_state = false;
2863         struct ibmvnic_rwi *rwi;
2864         unsigned long flags;
2865         struct device *dev;
2866         bool need_reset;
2867         int num_fails = 0;
2868         u32 reset_state;
2869         int rc = 0;
2870
2871         adapter = container_of(work, struct ibmvnic_adapter, ibmvnic_reset);
2872                 dev = &adapter->vdev->dev;
2873
2874         /* Wait for ibmvnic_probe() to complete. If probe is taking too long
2875          * or if another reset is in progress, defer work for now. If probe
2876          * eventually fails it will flush and terminate our work.
2877          *
2878          * Three possibilities here:
2879          * 1. Adpater being removed  - just return
2880          * 2. Timed out on probe or another reset in progress - delay the work
2881          * 3. Completed probe - perform any resets in queue
2882          */
2883         if (adapter->state == VNIC_PROBING &&
2884             !wait_for_completion_timeout(&adapter->probe_done, timeout)) {
2885                 dev_err(dev, "Reset thread timed out on probe");
2886                 queue_delayed_work(system_long_wq,
2887                                    &adapter->ibmvnic_delayed_reset,
2888                                    IBMVNIC_RESET_DELAY);
2889                 return;
2890         }
2891
2892         /* adapter is done with probe (i.e state is never VNIC_PROBING now) */
2893         if (adapter->state == VNIC_REMOVING)
2894                 return;
2895
2896         /* ->rwi_list is stable now (no one else is removing entries) */
2897
2898         /* ibmvnic_probe() may have purged the reset queue after we were
2899          * scheduled to process a reset so there maybe no resets to process.
2900          * Before setting the ->resetting bit though, we have to make sure
2901          * that there is infact a reset to process. Otherwise we may race
2902          * with ibmvnic_open() and end up leaving the vnic down:
2903          *
2904          *      __ibmvnic_reset()           ibmvnic_open()
2905          *      -----------------           --------------
2906          *
2907          *  set ->resetting bit
2908          *                              find ->resetting bit is set
2909          *                              set ->state to IBMVNIC_OPEN (i.e
2910          *                              assume reset will open device)
2911          *                              return
2912          *  find reset queue empty
2913          *  return
2914          *
2915          *      Neither performed vnic login/open and vnic stays down
2916          *
2917          * If we hold the lock and conditionally set the bit, either we
2918          * or ibmvnic_open() will complete the open.
2919          */
2920         need_reset = false;
2921         spin_lock(&adapter->rwi_lock);
2922         if (!list_empty(&adapter->rwi_list)) {
2923                 if (test_and_set_bit_lock(0, &adapter->resetting)) {
2924                         queue_delayed_work(system_long_wq,
2925                                            &adapter->ibmvnic_delayed_reset,
2926                                            IBMVNIC_RESET_DELAY);
2927                 } else {
2928                         need_reset = true;
2929                 }
2930         }
2931         spin_unlock(&adapter->rwi_lock);
2932
2933         if (!need_reset)
2934                 return;
2935
2936         rwi = get_next_rwi(adapter);
2937         while (rwi) {
2938                 spin_lock_irqsave(&adapter->state_lock, flags);
2939
2940                 if (adapter->state == VNIC_REMOVING ||
2941                     adapter->state == VNIC_REMOVED) {
2942                         spin_unlock_irqrestore(&adapter->state_lock, flags);
2943                         kfree(rwi);
2944                         rc = EBUSY;
2945                         break;
2946                 }
2947
2948                 if (!saved_state) {
2949                         reset_state = adapter->state;
2950                         saved_state = true;
2951                 }
2952                 spin_unlock_irqrestore(&adapter->state_lock, flags);
2953
2954                 if (rwi->reset_reason == VNIC_RESET_PASSIVE_INIT) {
2955                         rtnl_lock();
2956                         rc = do_passive_init(adapter);
2957                         rtnl_unlock();
2958                         if (!rc)
2959                                 netif_carrier_on(adapter->netdev);
2960                 } else if (adapter->force_reset_recovery) {
2961                         /* Since we are doing a hard reset now, clear the
2962                          * failover_pending flag so we don't ignore any
2963                          * future MOBILITY or other resets.
2964                          */
2965                         adapter->failover_pending = false;
2966
2967                         /* Transport event occurred during previous reset */
2968                         if (adapter->wait_for_reset) {
2969                                 /* Previous was CHANGE_PARAM; caller locked */
2970                                 adapter->force_reset_recovery = false;
2971                                 rc = do_hard_reset(adapter, rwi, reset_state);
2972                         } else {
2973                                 rtnl_lock();
2974                                 adapter->force_reset_recovery = false;
2975                                 rc = do_hard_reset(adapter, rwi, reset_state);
2976                                 rtnl_unlock();
2977                         }
2978                         if (rc)
2979                                 num_fails++;
2980                         else
2981                                 num_fails = 0;
2982
2983                         /* If auto-priority-failover is enabled we can get
2984                          * back to back failovers during resets, resulting
2985                          * in at least two failed resets (from high-priority
2986                          * backing device to low-priority one and then back)
2987                          * If resets continue to fail beyond that, give the
2988                          * adapter some time to settle down before retrying.
2989                          */
2990                         if (num_fails >= 3) {
2991                                 netdev_dbg(adapter->netdev,
2992                                            "[S:%s] Hard reset failed %d times, waiting 60 secs\n",
2993                                            adapter_state_to_string(adapter->state),
2994                                            num_fails);
2995                                 set_current_state(TASK_UNINTERRUPTIBLE);
2996                                 schedule_timeout(60 * HZ);
2997                         }
2998                 } else {
2999                         rc = do_reset(adapter, rwi, reset_state);
3000                 }
3001                 tmprwi = rwi;
3002                 adapter->last_reset_time = jiffies;
3003
3004                 if (rc)
3005                         netdev_dbg(adapter->netdev, "Reset failed, rc=%d\n", rc);
3006
3007                 rwi = get_next_rwi(adapter);
3008
3009                 /*
3010                  * If there is another reset queued, free the previous rwi
3011                  * and process the new reset even if previous reset failed
3012                  * (the previous reset could have failed because of a fail
3013                  * over for instance, so process the fail over).
3014                  *
3015                  * If there are no resets queued and the previous reset failed,
3016                  * the adapter would be in an undefined state. So retry the
3017                  * previous reset as a hard reset.
3018                  */
3019                 if (rwi)
3020                         kfree(tmprwi);
3021                 else if (rc)
3022                         rwi = tmprwi;
3023
3024                 if (rwi && (rwi->reset_reason == VNIC_RESET_FAILOVER ||
3025                             rwi->reset_reason == VNIC_RESET_MOBILITY || rc))
3026                         adapter->force_reset_recovery = true;
3027         }
3028
3029         if (adapter->wait_for_reset) {
3030                 adapter->reset_done_rc = rc;
3031                 complete(&adapter->reset_done);
3032         }
3033
3034         clear_bit_unlock(0, &adapter->resetting);
3035
3036         netdev_dbg(adapter->netdev,
3037                    "[S:%s FRR:%d WFR:%d] Done processing resets\n",
3038                    adapter_state_to_string(adapter->state),
3039                    adapter->force_reset_recovery,
3040                    adapter->wait_for_reset);
3041 }
3042
3043 static void __ibmvnic_delayed_reset(struct work_struct *work)
3044 {
3045         struct ibmvnic_adapter *adapter;
3046
3047         adapter = container_of(work, struct ibmvnic_adapter,
3048                                ibmvnic_delayed_reset.work);
3049         __ibmvnic_reset(&adapter->ibmvnic_reset);
3050 }
3051
3052 static void flush_reset_queue(struct ibmvnic_adapter *adapter)
3053 {
3054         struct list_head *entry, *tmp_entry;
3055
3056         if (!list_empty(&adapter->rwi_list)) {
3057                 list_for_each_safe(entry, tmp_entry, &adapter->rwi_list) {
3058                         list_del(entry);
3059                         kfree(list_entry(entry, struct ibmvnic_rwi, list));
3060                 }
3061         }
3062 }
3063
3064 static int ibmvnic_reset(struct ibmvnic_adapter *adapter,
3065                          enum ibmvnic_reset_reason reason)
3066 {
3067         struct net_device *netdev = adapter->netdev;
3068         struct ibmvnic_rwi *rwi, *tmp;
3069         unsigned long flags;
3070         int ret;
3071
3072         spin_lock_irqsave(&adapter->rwi_lock, flags);
3073
3074         /* If failover is pending don't schedule any other reset.
3075          * Instead let the failover complete. If there is already a
3076          * a failover reset scheduled, we will detect and drop the
3077          * duplicate reset when walking the ->rwi_list below.
3078          */
3079         if (adapter->state == VNIC_REMOVING ||
3080             adapter->state == VNIC_REMOVED ||
3081             (adapter->failover_pending && reason != VNIC_RESET_FAILOVER)) {
3082                 ret = EBUSY;
3083                 netdev_dbg(netdev, "Adapter removing or pending failover, skipping reset\n");
3084                 goto err;
3085         }
3086
3087         list_for_each_entry(tmp, &adapter->rwi_list, list) {
3088                 if (tmp->reset_reason == reason) {
3089                         netdev_dbg(netdev, "Skipping matching reset, reason=%s\n",
3090                                    reset_reason_to_string(reason));
3091                         ret = EBUSY;
3092                         goto err;
3093                 }
3094         }
3095
3096         rwi = kzalloc(sizeof(*rwi), GFP_ATOMIC);
3097         if (!rwi) {
3098                 ret = ENOMEM;
3099                 goto err;
3100         }
3101         /* if we just received a transport event,
3102          * flush reset queue and process this reset
3103          */
3104         if (adapter->force_reset_recovery)
3105                 flush_reset_queue(adapter);
3106
3107         rwi->reset_reason = reason;
3108         list_add_tail(&rwi->list, &adapter->rwi_list);
3109         netdev_dbg(adapter->netdev, "Scheduling reset (reason %s)\n",
3110                    reset_reason_to_string(reason));
3111         queue_work(system_long_wq, &adapter->ibmvnic_reset);
3112
3113         ret = 0;
3114 err:
3115         /* ibmvnic_close() below can block, so drop the lock first */
3116         spin_unlock_irqrestore(&adapter->rwi_lock, flags);
3117
3118         if (ret == ENOMEM)
3119                 ibmvnic_close(netdev);
3120
3121         return -ret;
3122 }
3123
3124 static void ibmvnic_tx_timeout(struct net_device *dev, unsigned int txqueue)
3125 {
3126         struct ibmvnic_adapter *adapter = netdev_priv(dev);
3127
3128         if (test_bit(0, &adapter->resetting)) {
3129                 netdev_err(adapter->netdev,
3130                            "Adapter is resetting, skip timeout reset\n");
3131                 return;
3132         }
3133         /* No queuing up reset until at least 5 seconds (default watchdog val)
3134          * after last reset
3135          */
3136         if (time_before(jiffies, (adapter->last_reset_time + dev->watchdog_timeo))) {
3137                 netdev_dbg(dev, "Not yet time to tx timeout.\n");
3138                 return;
3139         }
3140         ibmvnic_reset(adapter, VNIC_RESET_TIMEOUT);
3141 }
3142
3143 static void remove_buff_from_pool(struct ibmvnic_adapter *adapter,
3144                                   struct ibmvnic_rx_buff *rx_buff)
3145 {
3146         struct ibmvnic_rx_pool *pool = &adapter->rx_pool[rx_buff->pool_index];
3147
3148         rx_buff->skb = NULL;
3149
3150         pool->free_map[pool->next_alloc] = (int)(rx_buff - pool->rx_buff);
3151         pool->next_alloc = (pool->next_alloc + 1) % pool->size;
3152
3153         atomic_dec(&pool->available);
3154 }
3155
3156 static int ibmvnic_poll(struct napi_struct *napi, int budget)
3157 {
3158         struct ibmvnic_sub_crq_queue *rx_scrq;
3159         struct ibmvnic_adapter *adapter;
3160         struct net_device *netdev;
3161         int frames_processed;
3162         int scrq_num;
3163
3164         netdev = napi->dev;
3165         adapter = netdev_priv(netdev);
3166         scrq_num = (int)(napi - adapter->napi);
3167         frames_processed = 0;
3168         rx_scrq = adapter->rx_scrq[scrq_num];
3169
3170 restart_poll:
3171         while (frames_processed < budget) {
3172                 struct sk_buff *skb;
3173                 struct ibmvnic_rx_buff *rx_buff;
3174                 union sub_crq *next;
3175                 u32 length;
3176                 u16 offset;
3177                 u8 flags = 0;
3178
3179                 if (unlikely(test_bit(0, &adapter->resetting) &&
3180                              adapter->reset_reason != VNIC_RESET_NON_FATAL)) {
3181                         enable_scrq_irq(adapter, rx_scrq);
3182                         napi_complete_done(napi, frames_processed);
3183                         return frames_processed;
3184                 }
3185
3186                 if (!pending_scrq(adapter, rx_scrq))
3187                         break;
3188                 next = ibmvnic_next_scrq(adapter, rx_scrq);
3189                 rx_buff = (struct ibmvnic_rx_buff *)
3190                           be64_to_cpu(next->rx_comp.correlator);
3191                 /* do error checking */
3192                 if (next->rx_comp.rc) {
3193                         netdev_dbg(netdev, "rx buffer returned with rc %x\n",
3194                                    be16_to_cpu(next->rx_comp.rc));
3195                         /* free the entry */
3196                         next->rx_comp.first = 0;
3197                         dev_kfree_skb_any(rx_buff->skb);
3198                         remove_buff_from_pool(adapter, rx_buff);
3199                         continue;
3200                 } else if (!rx_buff->skb) {
3201                         /* free the entry */
3202                         next->rx_comp.first = 0;
3203                         remove_buff_from_pool(adapter, rx_buff);
3204                         continue;
3205                 }
3206
3207                 length = be32_to_cpu(next->rx_comp.len);
3208                 offset = be16_to_cpu(next->rx_comp.off_frame_data);
3209                 flags = next->rx_comp.flags;
3210                 skb = rx_buff->skb;
3211                 /* load long_term_buff before copying to skb */
3212                 dma_rmb();
3213                 skb_copy_to_linear_data(skb, rx_buff->data + offset,
3214                                         length);
3215
3216                 /* VLAN Header has been stripped by the system firmware and
3217                  * needs to be inserted by the driver
3218                  */
3219                 if (adapter->rx_vlan_header_insertion &&
3220                     (flags & IBMVNIC_VLAN_STRIPPED))
3221                         __vlan_hwaccel_put_tag(skb, htons(ETH_P_8021Q),
3222                                                ntohs(next->rx_comp.vlan_tci));
3223
3224                 /* free the entry */
3225                 next->rx_comp.first = 0;
3226                 remove_buff_from_pool(adapter, rx_buff);
3227
3228                 skb_put(skb, length);
3229                 skb->protocol = eth_type_trans(skb, netdev);
3230                 skb_record_rx_queue(skb, scrq_num);
3231
3232                 if (flags & IBMVNIC_IP_CHKSUM_GOOD &&
3233                     flags & IBMVNIC_TCP_UDP_CHKSUM_GOOD) {
3234                         skb->ip_summed = CHECKSUM_UNNECESSARY;
3235                 }
3236
3237                 length = skb->len;
3238                 napi_gro_receive(napi, skb); /* send it up */
3239                 netdev->stats.rx_packets++;
3240                 netdev->stats.rx_bytes += length;
3241                 adapter->rx_stats_buffers[scrq_num].packets++;
3242                 adapter->rx_stats_buffers[scrq_num].bytes += length;
3243                 frames_processed++;
3244         }
3245
3246         if (adapter->state != VNIC_CLOSING &&
3247             ((atomic_read(&adapter->rx_pool[scrq_num].available) <
3248               adapter->req_rx_add_entries_per_subcrq / 2) ||
3249               frames_processed < budget))
3250                 replenish_rx_pool(adapter, &adapter->rx_pool[scrq_num]);
3251         if (frames_processed < budget) {
3252                 if (napi_complete_done(napi, frames_processed)) {
3253                         enable_scrq_irq(adapter, rx_scrq);
3254                         if (pending_scrq(adapter, rx_scrq)) {
3255                                 if (napi_reschedule(napi)) {
3256                                         disable_scrq_irq(adapter, rx_scrq);
3257                                         goto restart_poll;
3258                                 }
3259                         }
3260                 }
3261         }
3262         return frames_processed;
3263 }
3264
3265 static int wait_for_reset(struct ibmvnic_adapter *adapter)
3266 {
3267         int rc, ret;
3268
3269         adapter->fallback.mtu = adapter->req_mtu;
3270         adapter->fallback.rx_queues = adapter->req_rx_queues;
3271         adapter->fallback.tx_queues = adapter->req_tx_queues;
3272         adapter->fallback.rx_entries = adapter->req_rx_add_entries_per_subcrq;
3273         adapter->fallback.tx_entries = adapter->req_tx_entries_per_subcrq;
3274
3275         reinit_completion(&adapter->reset_done);
3276         adapter->wait_for_reset = true;
3277         rc = ibmvnic_reset(adapter, VNIC_RESET_CHANGE_PARAM);
3278
3279         if (rc) {
3280                 ret = rc;
3281                 goto out;
3282         }
3283         rc = ibmvnic_wait_for_completion(adapter, &adapter->reset_done, 60000);
3284         if (rc) {
3285                 ret = -ENODEV;
3286                 goto out;
3287         }
3288
3289         ret = 0;
3290         if (adapter->reset_done_rc) {
3291                 ret = -EIO;
3292                 adapter->desired.mtu = adapter->fallback.mtu;
3293                 adapter->desired.rx_queues = adapter->fallback.rx_queues;
3294                 adapter->desired.tx_queues = adapter->fallback.tx_queues;
3295                 adapter->desired.rx_entries = adapter->fallback.rx_entries;
3296                 adapter->desired.tx_entries = adapter->fallback.tx_entries;
3297
3298                 reinit_completion(&adapter->reset_done);
3299                 adapter->wait_for_reset = true;
3300                 rc = ibmvnic_reset(adapter, VNIC_RESET_CHANGE_PARAM);
3301                 if (rc) {
3302                         ret = rc;
3303                         goto out;
3304                 }
3305                 rc = ibmvnic_wait_for_completion(adapter, &adapter->reset_done,
3306                                                  60000);
3307                 if (rc) {
3308                         ret = -ENODEV;
3309                         goto out;
3310                 }
3311         }
3312 out:
3313         adapter->wait_for_reset = false;
3314
3315         return ret;
3316 }
3317
3318 static int ibmvnic_change_mtu(struct net_device *netdev, int new_mtu)
3319 {
3320         struct ibmvnic_adapter *adapter = netdev_priv(netdev);
3321
3322         adapter->desired.mtu = new_mtu + ETH_HLEN;
3323
3324         return wait_for_reset(adapter);
3325 }
3326
3327 static netdev_features_t ibmvnic_features_check(struct sk_buff *skb,
3328                                                 struct net_device *dev,
3329                                                 netdev_features_t features)
3330 {
3331         /* Some backing hardware adapters can not
3332          * handle packets with a MSS less than 224
3333          * or with only one segment.
3334          */
3335         if (skb_is_gso(skb)) {
3336                 if (skb_shinfo(skb)->gso_size < 224 ||
3337                     skb_shinfo(skb)->gso_segs == 1)
3338                         features &= ~NETIF_F_GSO_MASK;
3339         }
3340
3341         return features;
3342 }
3343
3344 static const struct net_device_ops ibmvnic_netdev_ops = {
3345         .ndo_open               = ibmvnic_open,
3346         .ndo_stop               = ibmvnic_close,
3347         .ndo_start_xmit         = ibmvnic_xmit,
3348         .ndo_set_rx_mode        = ibmvnic_set_multi,
3349         .ndo_set_mac_address    = ibmvnic_set_mac,
3350         .ndo_validate_addr      = eth_validate_addr,
3351         .ndo_tx_timeout         = ibmvnic_tx_timeout,
3352         .ndo_change_mtu         = ibmvnic_change_mtu,
3353         .ndo_features_check     = ibmvnic_features_check,
3354 };
3355
3356 /* ethtool functions */
3357
3358 static int ibmvnic_get_link_ksettings(struct net_device *netdev,
3359                                       struct ethtool_link_ksettings *cmd)
3360 {
3361         struct ibmvnic_adapter *adapter = netdev_priv(netdev);
3362         int rc;
3363
3364         rc = send_query_phys_parms(adapter);
3365         if (rc) {
3366                 adapter->speed = SPEED_UNKNOWN;
3367                 adapter->duplex = DUPLEX_UNKNOWN;
3368         }
3369         cmd->base.speed = adapter->speed;
3370         cmd->base.duplex = adapter->duplex;
3371         cmd->base.port = PORT_FIBRE;
3372         cmd->base.phy_address = 0;
3373         cmd->base.autoneg = AUTONEG_ENABLE;
3374
3375         return 0;
3376 }
3377
3378 static void ibmvnic_get_drvinfo(struct net_device *netdev,
3379                                 struct ethtool_drvinfo *info)
3380 {
3381         struct ibmvnic_adapter *adapter = netdev_priv(netdev);
3382
3383         strscpy(info->driver, ibmvnic_driver_name, sizeof(info->driver));
3384         strscpy(info->version, IBMVNIC_DRIVER_VERSION, sizeof(info->version));
3385         strscpy(info->fw_version, adapter->fw_version,
3386                 sizeof(info->fw_version));
3387 }
3388
3389 static u32 ibmvnic_get_msglevel(struct net_device *netdev)
3390 {
3391         struct ibmvnic_adapter *adapter = netdev_priv(netdev);
3392
3393         return adapter->msg_enable;
3394 }
3395
3396 static void ibmvnic_set_msglevel(struct net_device *netdev, u32 data)
3397 {
3398         struct ibmvnic_adapter *adapter = netdev_priv(netdev);
3399
3400         adapter->msg_enable = data;
3401 }
3402
3403 static u32 ibmvnic_get_link(struct net_device *netdev)
3404 {
3405         struct ibmvnic_adapter *adapter = netdev_priv(netdev);
3406
3407         /* Don't need to send a query because we request a logical link up at
3408          * init and then we wait for link state indications
3409          */
3410         return adapter->logical_link_state;
3411 }
3412
3413 static void ibmvnic_get_ringparam(struct net_device *netdev,
3414                                   struct ethtool_ringparam *ring,
3415                                   struct kernel_ethtool_ringparam *kernel_ring,
3416                                   struct netlink_ext_ack *extack)
3417 {
3418         struct ibmvnic_adapter *adapter = netdev_priv(netdev);
3419
3420         ring->rx_max_pending = adapter->max_rx_add_entries_per_subcrq;
3421         ring->tx_max_pending = adapter->max_tx_entries_per_subcrq;
3422         ring->rx_mini_max_pending = 0;
3423         ring->rx_jumbo_max_pending = 0;
3424         ring->rx_pending = adapter->req_rx_add_entries_per_subcrq;
3425         ring->tx_pending = adapter->req_tx_entries_per_subcrq;
3426         ring->rx_mini_pending = 0;
3427         ring->rx_jumbo_pending = 0;
3428 }
3429
3430 static int ibmvnic_set_ringparam(struct net_device *netdev,
3431                                  struct ethtool_ringparam *ring,
3432                                  struct kernel_ethtool_ringparam *kernel_ring,
3433                                  struct netlink_ext_ack *extack)
3434 {
3435         struct ibmvnic_adapter *adapter = netdev_priv(netdev);
3436
3437         if (ring->rx_pending > adapter->max_rx_add_entries_per_subcrq  ||
3438             ring->tx_pending > adapter->max_tx_entries_per_subcrq) {
3439                 netdev_err(netdev, "Invalid request.\n");
3440                 netdev_err(netdev, "Max tx buffers = %llu\n",
3441                            adapter->max_rx_add_entries_per_subcrq);
3442                 netdev_err(netdev, "Max rx buffers = %llu\n",
3443                            adapter->max_tx_entries_per_subcrq);
3444                 return -EINVAL;
3445         }
3446
3447         adapter->desired.rx_entries = ring->rx_pending;
3448         adapter->desired.tx_entries = ring->tx_pending;
3449
3450         return wait_for_reset(adapter);
3451 }
3452
3453 static void ibmvnic_get_channels(struct net_device *netdev,
3454                                  struct ethtool_channels *channels)
3455 {
3456         struct ibmvnic_adapter *adapter = netdev_priv(netdev);
3457
3458         channels->max_rx = adapter->max_rx_queues;
3459         channels->max_tx = adapter->max_tx_queues;
3460         channels->max_other = 0;
3461         channels->max_combined = 0;
3462         channels->rx_count = adapter->req_rx_queues;
3463         channels->tx_count = adapter->req_tx_queues;
3464         channels->other_count = 0;
3465         channels->combined_count = 0;
3466 }
3467
3468 static int ibmvnic_set_channels(struct net_device *netdev,
3469                                 struct ethtool_channels *channels)
3470 {
3471         struct ibmvnic_adapter *adapter = netdev_priv(netdev);
3472
3473         adapter->desired.rx_queues = channels->rx_count;
3474         adapter->desired.tx_queues = channels->tx_count;
3475
3476         return wait_for_reset(adapter);
3477 }
3478
3479 static void ibmvnic_get_strings(struct net_device *dev, u32 stringset, u8 *data)
3480 {
3481         struct ibmvnic_adapter *adapter = netdev_priv(dev);
3482         int i;
3483
3484         if (stringset != ETH_SS_STATS)
3485                 return;
3486
3487         for (i = 0; i < ARRAY_SIZE(ibmvnic_stats); i++, data += ETH_GSTRING_LEN)
3488                 memcpy(data, ibmvnic_stats[i].name, ETH_GSTRING_LEN);
3489
3490         for (i = 0; i < adapter->req_tx_queues; i++) {
3491                 snprintf(data, ETH_GSTRING_LEN, "tx%d_packets", i);
3492                 data += ETH_GSTRING_LEN;
3493
3494                 snprintf(data, ETH_GSTRING_LEN, "tx%d_bytes", i);
3495                 data += ETH_GSTRING_LEN;
3496
3497                 snprintf(data, ETH_GSTRING_LEN, "tx%d_dropped_packets", i);
3498                 data += ETH_GSTRING_LEN;
3499         }
3500
3501         for (i = 0; i < adapter->req_rx_queues; i++) {
3502                 snprintf(data, ETH_GSTRING_LEN, "rx%d_packets", i);
3503                 data += ETH_GSTRING_LEN;
3504
3505                 snprintf(data, ETH_GSTRING_LEN, "rx%d_bytes", i);
3506                 data += ETH_GSTRING_LEN;
3507
3508                 snprintf(data, ETH_GSTRING_LEN, "rx%d_interrupts", i);
3509                 data += ETH_GSTRING_LEN;
3510         }
3511 }
3512
3513 static int ibmvnic_get_sset_count(struct net_device *dev, int sset)
3514 {
3515         struct ibmvnic_adapter *adapter = netdev_priv(dev);
3516
3517         switch (sset) {
3518         case ETH_SS_STATS:
3519                 return ARRAY_SIZE(ibmvnic_stats) +
3520                        adapter->req_tx_queues * NUM_TX_STATS +
3521                        adapter->req_rx_queues * NUM_RX_STATS;
3522         default:
3523                 return -EOPNOTSUPP;
3524         }
3525 }
3526
3527 static void ibmvnic_get_ethtool_stats(struct net_device *dev,
3528                                       struct ethtool_stats *stats, u64 *data)
3529 {
3530         struct ibmvnic_adapter *adapter = netdev_priv(dev);
3531         union ibmvnic_crq crq;
3532         int i, j;
3533         int rc;
3534
3535         memset(&crq, 0, sizeof(crq));
3536         crq.request_statistics.first = IBMVNIC_CRQ_CMD;
3537         crq.request_statistics.cmd = REQUEST_STATISTICS;
3538         crq.request_statistics.ioba = cpu_to_be32(adapter->stats_token);
3539         crq.request_statistics.len =
3540             cpu_to_be32(sizeof(struct ibmvnic_statistics));
3541
3542         /* Wait for data to be written */
3543         reinit_completion(&adapter->stats_done);
3544         rc = ibmvnic_send_crq(adapter, &crq);
3545         if (rc)
3546                 return;
3547         rc = ibmvnic_wait_for_completion(adapter, &adapter->stats_done, 10000);
3548         if (rc)
3549                 return;
3550
3551         for (i = 0; i < ARRAY_SIZE(ibmvnic_stats); i++)
3552                 data[i] = be64_to_cpu(IBMVNIC_GET_STAT
3553                                       (adapter, ibmvnic_stats[i].offset));
3554
3555         for (j = 0; j < adapter->req_tx_queues; j++) {
3556                 data[i] = adapter->tx_stats_buffers[j].packets;
3557                 i++;
3558                 data[i] = adapter->tx_stats_buffers[j].bytes;
3559                 i++;
3560                 data[i] = adapter->tx_stats_buffers[j].dropped_packets;
3561                 i++;
3562         }
3563
3564         for (j = 0; j < adapter->req_rx_queues; j++) {
3565                 data[i] = adapter->rx_stats_buffers[j].packets;
3566                 i++;
3567                 data[i] = adapter->rx_stats_buffers[j].bytes;
3568                 i++;
3569                 data[i] = adapter->rx_stats_buffers[j].interrupts;
3570                 i++;
3571         }
3572 }
3573
3574 static const struct ethtool_ops ibmvnic_ethtool_ops = {
3575         .get_drvinfo            = ibmvnic_get_drvinfo,
3576         .get_msglevel           = ibmvnic_get_msglevel,
3577         .set_msglevel           = ibmvnic_set_msglevel,
3578         .get_link               = ibmvnic_get_link,
3579         .get_ringparam          = ibmvnic_get_ringparam,
3580         .set_ringparam          = ibmvnic_set_ringparam,
3581         .get_channels           = ibmvnic_get_channels,
3582         .set_channels           = ibmvnic_set_channels,
3583         .get_strings            = ibmvnic_get_strings,
3584         .get_sset_count         = ibmvnic_get_sset_count,
3585         .get_ethtool_stats      = ibmvnic_get_ethtool_stats,
3586         .get_link_ksettings     = ibmvnic_get_link_ksettings,
3587 };
3588
3589 /* Routines for managing CRQs/sCRQs  */
3590
3591 static int reset_one_sub_crq_queue(struct ibmvnic_adapter *adapter,
3592                                    struct ibmvnic_sub_crq_queue *scrq)
3593 {
3594         int rc;
3595
3596         if (!scrq) {
3597                 netdev_dbg(adapter->netdev, "Invalid scrq reset.\n");
3598                 return -EINVAL;
3599         }
3600
3601         if (scrq->irq) {
3602                 free_irq(scrq->irq, scrq);
3603                 irq_dispose_mapping(scrq->irq);
3604                 scrq->irq = 0;
3605         }
3606
3607         if (scrq->msgs) {
3608                 memset(scrq->msgs, 0, 4 * PAGE_SIZE);
3609                 atomic_set(&scrq->used, 0);
3610                 scrq->cur = 0;
3611                 scrq->ind_buf.index = 0;
3612         } else {
3613                 netdev_dbg(adapter->netdev, "Invalid scrq reset\n");
3614                 return -EINVAL;
3615         }
3616
3617         rc = h_reg_sub_crq(adapter->vdev->unit_address, scrq->msg_token,
3618                            4 * PAGE_SIZE, &scrq->crq_num, &scrq->hw_irq);
3619         return rc;
3620 }
3621
3622 static int reset_sub_crq_queues(struct ibmvnic_adapter *adapter)
3623 {
3624         int i, rc;
3625
3626         if (!adapter->tx_scrq || !adapter->rx_scrq)
3627                 return -EINVAL;
3628
3629         for (i = 0; i < adapter->req_tx_queues; i++) {
3630                 netdev_dbg(adapter->netdev, "Re-setting tx_scrq[%d]\n", i);
3631                 rc = reset_one_sub_crq_queue(adapter, adapter->tx_scrq[i]);
3632                 if (rc)
3633                         return rc;
3634         }
3635
3636         for (i = 0; i < adapter->req_rx_queues; i++) {
3637                 netdev_dbg(adapter->netdev, "Re-setting rx_scrq[%d]\n", i);
3638                 rc = reset_one_sub_crq_queue(adapter, adapter->rx_scrq[i]);
3639                 if (rc)
3640                         return rc;
3641         }
3642
3643         return rc;
3644 }
3645
3646 static void release_sub_crq_queue(struct ibmvnic_adapter *adapter,
3647                                   struct ibmvnic_sub_crq_queue *scrq,
3648                                   bool do_h_free)
3649 {
3650         struct device *dev = &adapter->vdev->dev;
3651         long rc;
3652
3653         netdev_dbg(adapter->netdev, "Releasing sub-CRQ\n");
3654
3655         if (do_h_free) {
3656                 /* Close the sub-crqs */
3657                 do {
3658                         rc = plpar_hcall_norets(H_FREE_SUB_CRQ,
3659                                                 adapter->vdev->unit_address,
3660                                                 scrq->crq_num);
3661                 } while (rc == H_BUSY || H_IS_LONG_BUSY(rc));
3662
3663                 if (rc) {
3664                         netdev_err(adapter->netdev,
3665                                    "Failed to release sub-CRQ %16lx, rc = %ld\n",
3666                                    scrq->crq_num, rc);
3667                 }
3668         }
3669
3670         dma_free_coherent(dev,
3671                           IBMVNIC_IND_ARR_SZ,
3672                           scrq->ind_buf.indir_arr,
3673                           scrq->ind_buf.indir_dma);
3674
3675         dma_unmap_single(dev, scrq->msg_token, 4 * PAGE_SIZE,
3676                          DMA_BIDIRECTIONAL);
3677         free_pages((unsigned long)scrq->msgs, 2);
3678         kfree(scrq);
3679 }
3680
3681 static struct ibmvnic_sub_crq_queue *init_sub_crq_queue(struct ibmvnic_adapter
3682                                                         *adapter)
3683 {
3684         struct device *dev = &adapter->vdev->dev;
3685         struct ibmvnic_sub_crq_queue *scrq;
3686         int rc;
3687
3688         scrq = kzalloc(sizeof(*scrq), GFP_KERNEL);
3689         if (!scrq)
3690                 return NULL;
3691
3692         scrq->msgs =
3693                 (union sub_crq *)__get_free_pages(GFP_KERNEL | __GFP_ZERO, 2);
3694         if (!scrq->msgs) {
3695                 dev_warn(dev, "Couldn't allocate crq queue messages page\n");
3696                 goto zero_page_failed;
3697         }
3698
3699         scrq->msg_token = dma_map_single(dev, scrq->msgs, 4 * PAGE_SIZE,
3700                                          DMA_BIDIRECTIONAL);
3701         if (dma_mapping_error(dev, scrq->msg_token)) {
3702                 dev_warn(dev, "Couldn't map crq queue messages page\n");
3703                 goto map_failed;
3704         }
3705
3706         rc = h_reg_sub_crq(adapter->vdev->unit_address, scrq->msg_token,
3707                            4 * PAGE_SIZE, &scrq->crq_num, &scrq->hw_irq);
3708
3709         if (rc == H_RESOURCE)
3710                 rc = ibmvnic_reset_crq(adapter);
3711
3712         if (rc == H_CLOSED) {
3713                 dev_warn(dev, "Partner adapter not ready, waiting.\n");
3714         } else if (rc) {
3715                 dev_warn(dev, "Error %d registering sub-crq\n", rc);
3716                 goto reg_failed;
3717         }
3718
3719         scrq->adapter = adapter;
3720         scrq->size = 4 * PAGE_SIZE / sizeof(*scrq->msgs);
3721         scrq->ind_buf.index = 0;
3722
3723         scrq->ind_buf.indir_arr =
3724                 dma_alloc_coherent(dev,
3725                                    IBMVNIC_IND_ARR_SZ,
3726                                    &scrq->ind_buf.indir_dma,
3727                                    GFP_KERNEL);
3728
3729         if (!scrq->ind_buf.indir_arr)
3730                 goto indir_failed;
3731
3732         spin_lock_init(&scrq->lock);
3733
3734         netdev_dbg(adapter->netdev,
3735                    "sub-crq initialized, num %lx, hw_irq=%lx, irq=%x\n",
3736                    scrq->crq_num, scrq->hw_irq, scrq->irq);
3737
3738         return scrq;
3739
3740 indir_failed:
3741         do {
3742                 rc = plpar_hcall_norets(H_FREE_SUB_CRQ,
3743                                         adapter->vdev->unit_address,
3744                                         scrq->crq_num);
3745         } while (rc == H_BUSY || rc == H_IS_LONG_BUSY(rc));
3746 reg_failed:
3747         dma_unmap_single(dev, scrq->msg_token, 4 * PAGE_SIZE,
3748                          DMA_BIDIRECTIONAL);
3749 map_failed:
3750         free_pages((unsigned long)scrq->msgs, 2);
3751 zero_page_failed:
3752         kfree(scrq);
3753
3754         return NULL;
3755 }
3756
3757 static void release_sub_crqs(struct ibmvnic_adapter *adapter, bool do_h_free)
3758 {
3759         int i;
3760
3761         if (adapter->tx_scrq) {
3762                 for (i = 0; i < adapter->num_active_tx_scrqs; i++) {
3763                         if (!adapter->tx_scrq[i])
3764                                 continue;
3765
3766                         netdev_dbg(adapter->netdev, "Releasing tx_scrq[%d]\n",
3767                                    i);
3768                         ibmvnic_tx_scrq_clean_buffer(adapter, adapter->tx_scrq[i]);
3769                         if (adapter->tx_scrq[i]->irq) {
3770                                 free_irq(adapter->tx_scrq[i]->irq,
3771                                          adapter->tx_scrq[i]);
3772                                 irq_dispose_mapping(adapter->tx_scrq[i]->irq);
3773                                 adapter->tx_scrq[i]->irq = 0;
3774                         }
3775
3776                         release_sub_crq_queue(adapter, adapter->tx_scrq[i],
3777                                               do_h_free);
3778                 }
3779
3780                 kfree(adapter->tx_scrq);
3781                 adapter->tx_scrq = NULL;
3782                 adapter->num_active_tx_scrqs = 0;
3783         }
3784
3785         if (adapter->rx_scrq) {
3786                 for (i = 0; i < adapter->num_active_rx_scrqs; i++) {
3787                         if (!adapter->rx_scrq[i])
3788                                 continue;
3789
3790                         netdev_dbg(adapter->netdev, "Releasing rx_scrq[%d]\n",
3791                                    i);
3792                         if (adapter->rx_scrq[i]->irq) {
3793                                 free_irq(adapter->rx_scrq[i]->irq,
3794                                          adapter->rx_scrq[i]);
3795                                 irq_dispose_mapping(adapter->rx_scrq[i]->irq);
3796                                 adapter->rx_scrq[i]->irq = 0;
3797                         }
3798
3799                         release_sub_crq_queue(adapter, adapter->rx_scrq[i],
3800                                               do_h_free);
3801                 }
3802
3803                 kfree(adapter->rx_scrq);
3804                 adapter->rx_scrq = NULL;
3805                 adapter->num_active_rx_scrqs = 0;
3806         }
3807 }
3808
3809 static int disable_scrq_irq(struct ibmvnic_adapter *adapter,
3810                             struct ibmvnic_sub_crq_queue *scrq)
3811 {
3812         struct device *dev = &adapter->vdev->dev;
3813         unsigned long rc;
3814
3815         rc = plpar_hcall_norets(H_VIOCTL, adapter->vdev->unit_address,
3816                                 H_DISABLE_VIO_INTERRUPT, scrq->hw_irq, 0, 0);
3817         if (rc)
3818                 dev_err(dev, "Couldn't disable scrq irq 0x%lx. rc=%ld\n",
3819                         scrq->hw_irq, rc);
3820         return rc;
3821 }
3822
3823 /* We can not use the IRQ chip EOI handler because that has the
3824  * unintended effect of changing the interrupt priority.
3825  */
3826 static void ibmvnic_xics_eoi(struct device *dev, struct ibmvnic_sub_crq_queue *scrq)
3827 {
3828         u64 val = 0xff000000 | scrq->hw_irq;
3829         unsigned long rc;
3830
3831         rc = plpar_hcall_norets(H_EOI, val);
3832         if (rc)
3833                 dev_err(dev, "H_EOI FAILED irq 0x%llx. rc=%ld\n", val, rc);
3834 }
3835
3836 /* Due to a firmware bug, the hypervisor can send an interrupt to a
3837  * transmit or receive queue just prior to a partition migration.
3838  * Force an EOI after migration.
3839  */
3840 static void ibmvnic_clear_pending_interrupt(struct device *dev,
3841                                             struct ibmvnic_sub_crq_queue *scrq)
3842 {
3843         if (!xive_enabled())
3844                 ibmvnic_xics_eoi(dev, scrq);
3845 }
3846
3847 static int enable_scrq_irq(struct ibmvnic_adapter *adapter,
3848                            struct ibmvnic_sub_crq_queue *scrq)
3849 {
3850         struct device *dev = &adapter->vdev->dev;
3851         unsigned long rc;
3852
3853         if (scrq->hw_irq > 0x100000000ULL) {
3854                 dev_err(dev, "bad hw_irq = %lx\n", scrq->hw_irq);
3855                 return 1;
3856         }
3857
3858         if (test_bit(0, &adapter->resetting) &&
3859             adapter->reset_reason == VNIC_RESET_MOBILITY) {
3860                 ibmvnic_clear_pending_interrupt(dev, scrq);
3861         }
3862
3863         rc = plpar_hcall_norets(H_VIOCTL, adapter->vdev->unit_address,
3864                                 H_ENABLE_VIO_INTERRUPT, scrq->hw_irq, 0, 0);
3865         if (rc)
3866                 dev_err(dev, "Couldn't enable scrq irq 0x%lx. rc=%ld\n",
3867                         scrq->hw_irq, rc);
3868         return rc;
3869 }
3870
3871 static int ibmvnic_complete_tx(struct ibmvnic_adapter *adapter,
3872                                struct ibmvnic_sub_crq_queue *scrq)
3873 {
3874         struct device *dev = &adapter->vdev->dev;
3875         struct ibmvnic_tx_pool *tx_pool;
3876         struct ibmvnic_tx_buff *txbuff;
3877         struct netdev_queue *txq;
3878         union sub_crq *next;
3879         int index;
3880         int i;
3881
3882 restart_loop:
3883         while (pending_scrq(adapter, scrq)) {
3884                 unsigned int pool = scrq->pool_index;
3885                 int num_entries = 0;
3886                 int total_bytes = 0;
3887                 int num_packets = 0;
3888
3889                 next = ibmvnic_next_scrq(adapter, scrq);
3890                 for (i = 0; i < next->tx_comp.num_comps; i++) {
3891                         index = be32_to_cpu(next->tx_comp.correlators[i]);
3892                         if (index & IBMVNIC_TSO_POOL_MASK) {
3893                                 tx_pool = &adapter->tso_pool[pool];
3894                                 index &= ~IBMVNIC_TSO_POOL_MASK;
3895                         } else {
3896                                 tx_pool = &adapter->tx_pool[pool];
3897                         }
3898
3899                         txbuff = &tx_pool->tx_buff[index];
3900                         num_packets++;
3901                         num_entries += txbuff->num_entries;
3902                         if (txbuff->skb) {
3903                                 total_bytes += txbuff->skb->len;
3904                                 if (next->tx_comp.rcs[i]) {
3905                                         dev_err(dev, "tx error %x\n",
3906                                                 next->tx_comp.rcs[i]);
3907                                         dev_kfree_skb_irq(txbuff->skb);
3908                                 } else {
3909                                         dev_consume_skb_irq(txbuff->skb);
3910                                 }
3911                                 txbuff->skb = NULL;
3912                         } else {
3913                                 netdev_warn(adapter->netdev,
3914                                             "TX completion received with NULL socket buffer\n");
3915                         }
3916                         tx_pool->free_map[tx_pool->producer_index] = index;
3917                         tx_pool->producer_index =
3918                                 (tx_pool->producer_index + 1) %
3919                                         tx_pool->num_buffers;
3920                 }
3921                 /* remove tx_comp scrq*/
3922                 next->tx_comp.first = 0;
3923
3924                 txq = netdev_get_tx_queue(adapter->netdev, scrq->pool_index);
3925                 netdev_tx_completed_queue(txq, num_packets, total_bytes);
3926
3927                 if (atomic_sub_return(num_entries, &scrq->used) <=
3928                     (adapter->req_tx_entries_per_subcrq / 2) &&
3929                     __netif_subqueue_stopped(adapter->netdev,
3930                                              scrq->pool_index)) {
3931                         rcu_read_lock();
3932                         if (adapter->tx_queues_active) {
3933                                 netif_wake_subqueue(adapter->netdev,
3934                                                     scrq->pool_index);
3935                                 netdev_dbg(adapter->netdev,
3936                                            "Started queue %d\n",
3937                                            scrq->pool_index);
3938                         }
3939                         rcu_read_unlock();
3940                 }
3941         }
3942
3943         enable_scrq_irq(adapter, scrq);
3944
3945         if (pending_scrq(adapter, scrq)) {
3946                 disable_scrq_irq(adapter, scrq);
3947                 goto restart_loop;
3948         }
3949
3950         return 0;
3951 }
3952
3953 static irqreturn_t ibmvnic_interrupt_tx(int irq, void *instance)
3954 {
3955         struct ibmvnic_sub_crq_queue *scrq = instance;
3956         struct ibmvnic_adapter *adapter = scrq->adapter;
3957
3958         disable_scrq_irq(adapter, scrq);
3959         ibmvnic_complete_tx(adapter, scrq);
3960
3961         return IRQ_HANDLED;
3962 }
3963
3964 static irqreturn_t ibmvnic_interrupt_rx(int irq, void *instance)
3965 {
3966         struct ibmvnic_sub_crq_queue *scrq = instance;
3967         struct ibmvnic_adapter *adapter = scrq->adapter;
3968
3969         /* When booting a kdump kernel we can hit pending interrupts
3970          * prior to completing driver initialization.
3971          */
3972         if (unlikely(adapter->state != VNIC_OPEN))
3973                 return IRQ_NONE;
3974
3975         adapter->rx_stats_buffers[scrq->scrq_num].interrupts++;
3976
3977         if (napi_schedule_prep(&adapter->napi[scrq->scrq_num])) {
3978                 disable_scrq_irq(adapter, scrq);
3979                 __napi_schedule(&adapter->napi[scrq->scrq_num]);
3980         }
3981
3982         return IRQ_HANDLED;
3983 }
3984
3985 static int init_sub_crq_irqs(struct ibmvnic_adapter *adapter)
3986 {
3987         struct device *dev = &adapter->vdev->dev;
3988         struct ibmvnic_sub_crq_queue *scrq;
3989         int i = 0, j = 0;
3990         int rc = 0;
3991
3992         for (i = 0; i < adapter->req_tx_queues; i++) {
3993                 netdev_dbg(adapter->netdev, "Initializing tx_scrq[%d] irq\n",
3994                            i);
3995                 scrq = adapter->tx_scrq[i];
3996                 scrq->irq = irq_create_mapping(NULL, scrq->hw_irq);
3997
3998                 if (!scrq->irq) {
3999                         rc = -EINVAL;
4000                         dev_err(dev, "Error mapping irq\n");
4001                         goto req_tx_irq_failed;
4002                 }
4003
4004                 snprintf(scrq->name, sizeof(scrq->name), "ibmvnic-%x-tx%d",
4005                          adapter->vdev->unit_address, i);
4006                 rc = request_irq(scrq->irq, ibmvnic_interrupt_tx,
4007                                  0, scrq->name, scrq);
4008
4009                 if (rc) {
4010                         dev_err(dev, "Couldn't register tx irq 0x%x. rc=%d\n",
4011                                 scrq->irq, rc);
4012                         irq_dispose_mapping(scrq->irq);
4013                         goto req_tx_irq_failed;
4014                 }
4015         }
4016
4017         for (i = 0; i < adapter->req_rx_queues; i++) {
4018                 netdev_dbg(adapter->netdev, "Initializing rx_scrq[%d] irq\n",
4019                            i);
4020                 scrq = adapter->rx_scrq[i];
4021                 scrq->irq = irq_create_mapping(NULL, scrq->hw_irq);
4022                 if (!scrq->irq) {
4023                         rc = -EINVAL;
4024                         dev_err(dev, "Error mapping irq\n");
4025                         goto req_rx_irq_failed;
4026                 }
4027                 snprintf(scrq->name, sizeof(scrq->name), "ibmvnic-%x-rx%d",
4028                          adapter->vdev->unit_address, i);
4029                 rc = request_irq(scrq->irq, ibmvnic_interrupt_rx,
4030                                  0, scrq->name, scrq);
4031                 if (rc) {
4032                         dev_err(dev, "Couldn't register rx irq 0x%x. rc=%d\n",
4033                                 scrq->irq, rc);
4034                         irq_dispose_mapping(scrq->irq);
4035                         goto req_rx_irq_failed;
4036                 }
4037         }
4038         return rc;
4039
4040 req_rx_irq_failed:
4041         for (j = 0; j < i; j++) {
4042                 free_irq(adapter->rx_scrq[j]->irq, adapter->rx_scrq[j]);
4043                 irq_dispose_mapping(adapter->rx_scrq[j]->irq);
4044         }
4045         i = adapter->req_tx_queues;
4046 req_tx_irq_failed:
4047         for (j = 0; j < i; j++) {
4048                 free_irq(adapter->tx_scrq[j]->irq, adapter->tx_scrq[j]);
4049                 irq_dispose_mapping(adapter->tx_scrq[j]->irq);
4050         }
4051         release_sub_crqs(adapter, 1);
4052         return rc;
4053 }
4054
4055 static int init_sub_crqs(struct ibmvnic_adapter *adapter)
4056 {
4057         struct device *dev = &adapter->vdev->dev;
4058         struct ibmvnic_sub_crq_queue **allqueues;
4059         int registered_queues = 0;
4060         int total_queues;
4061         int more = 0;
4062         int i;
4063
4064         total_queues = adapter->req_tx_queues + adapter->req_rx_queues;
4065
4066         allqueues = kcalloc(total_queues, sizeof(*allqueues), GFP_KERNEL);
4067         if (!allqueues)
4068                 return -ENOMEM;
4069
4070         for (i = 0; i < total_queues; i++) {
4071                 allqueues[i] = init_sub_crq_queue(adapter);
4072                 if (!allqueues[i]) {
4073                         dev_warn(dev, "Couldn't allocate all sub-crqs\n");
4074                         break;
4075                 }
4076                 registered_queues++;
4077         }
4078
4079         /* Make sure we were able to register the minimum number of queues */
4080         if (registered_queues <
4081             adapter->min_tx_queues + adapter->min_rx_queues) {
4082                 dev_err(dev, "Fatal: Couldn't init  min number of sub-crqs\n");
4083                 goto tx_failed;
4084         }
4085
4086         /* Distribute the failed allocated queues*/
4087         for (i = 0; i < total_queues - registered_queues + more ; i++) {
4088                 netdev_dbg(adapter->netdev, "Reducing number of queues\n");
4089                 switch (i % 3) {
4090                 case 0:
4091                         if (adapter->req_rx_queues > adapter->min_rx_queues)
4092                                 adapter->req_rx_queues--;
4093                         else
4094                                 more++;
4095                         break;
4096                 case 1:
4097                         if (adapter->req_tx_queues > adapter->min_tx_queues)
4098                                 adapter->req_tx_queues--;
4099                         else
4100                                 more++;
4101                         break;
4102                 }
4103         }
4104
4105         adapter->tx_scrq = kcalloc(adapter->req_tx_queues,
4106                                    sizeof(*adapter->tx_scrq), GFP_KERNEL);
4107         if (!adapter->tx_scrq)
4108                 goto tx_failed;
4109
4110         for (i = 0; i < adapter->req_tx_queues; i++) {
4111                 adapter->tx_scrq[i] = allqueues[i];
4112                 adapter->tx_scrq[i]->pool_index = i;
4113                 adapter->num_active_tx_scrqs++;
4114         }
4115
4116         adapter->rx_scrq = kcalloc(adapter->req_rx_queues,
4117                                    sizeof(*adapter->rx_scrq), GFP_KERNEL);
4118         if (!adapter->rx_scrq)
4119                 goto rx_failed;
4120
4121         for (i = 0; i < adapter->req_rx_queues; i++) {
4122                 adapter->rx_scrq[i] = allqueues[i + adapter->req_tx_queues];
4123                 adapter->rx_scrq[i]->scrq_num = i;
4124                 adapter->num_active_rx_scrqs++;
4125         }
4126
4127         kfree(allqueues);
4128         return 0;
4129
4130 rx_failed:
4131         kfree(adapter->tx_scrq);
4132         adapter->tx_scrq = NULL;
4133 tx_failed:
4134         for (i = 0; i < registered_queues; i++)
4135                 release_sub_crq_queue(adapter, allqueues[i], 1);
4136         kfree(allqueues);
4137         return -ENOMEM;
4138 }
4139
4140 static void send_request_cap(struct ibmvnic_adapter *adapter, int retry)
4141 {
4142         struct device *dev = &adapter->vdev->dev;
4143         union ibmvnic_crq crq;
4144         int max_entries;
4145         int cap_reqs;
4146
4147         /* We send out 6 or 7 REQUEST_CAPABILITY CRQs below (depending on
4148          * the PROMISC flag). Initialize this count upfront. When the tasklet
4149          * receives a response to all of these, it will send the next protocol
4150          * message (QUERY_IP_OFFLOAD).
4151          */
4152         if (!(adapter->netdev->flags & IFF_PROMISC) ||
4153             adapter->promisc_supported)
4154                 cap_reqs = 7;
4155         else
4156                 cap_reqs = 6;
4157
4158         if (!retry) {
4159                 /* Sub-CRQ entries are 32 byte long */
4160                 int entries_page = 4 * PAGE_SIZE / (sizeof(u64) * 4);
4161
4162                 atomic_set(&adapter->running_cap_crqs, cap_reqs);
4163
4164                 if (adapter->min_tx_entries_per_subcrq > entries_page ||
4165                     adapter->min_rx_add_entries_per_subcrq > entries_page) {
4166                         dev_err(dev, "Fatal, invalid entries per sub-crq\n");
4167                         return;
4168                 }
4169
4170                 if (adapter->desired.mtu)
4171                         adapter->req_mtu = adapter->desired.mtu;
4172                 else
4173                         adapter->req_mtu = adapter->netdev->mtu + ETH_HLEN;
4174
4175                 if (!adapter->desired.tx_entries)
4176                         adapter->desired.tx_entries =
4177                                         adapter->max_tx_entries_per_subcrq;
4178                 if (!adapter->desired.rx_entries)
4179                         adapter->desired.rx_entries =
4180                                         adapter->max_rx_add_entries_per_subcrq;
4181
4182                 max_entries = IBMVNIC_LTB_SET_SIZE /
4183                               (adapter->req_mtu + IBMVNIC_BUFFER_HLEN);
4184
4185                 if ((adapter->req_mtu + IBMVNIC_BUFFER_HLEN) *
4186                         adapter->desired.tx_entries > IBMVNIC_LTB_SET_SIZE) {
4187                         adapter->desired.tx_entries = max_entries;
4188                 }
4189
4190                 if ((adapter->req_mtu + IBMVNIC_BUFFER_HLEN) *
4191                         adapter->desired.rx_entries > IBMVNIC_LTB_SET_SIZE) {
4192                         adapter->desired.rx_entries = max_entries;
4193                 }
4194
4195                 if (adapter->desired.tx_entries)
4196                         adapter->req_tx_entries_per_subcrq =
4197                                         adapter->desired.tx_entries;
4198                 else
4199                         adapter->req_tx_entries_per_subcrq =
4200                                         adapter->max_tx_entries_per_subcrq;
4201
4202                 if (adapter->desired.rx_entries)
4203                         adapter->req_rx_add_entries_per_subcrq =
4204                                         adapter->desired.rx_entries;
4205                 else
4206                         adapter->req_rx_add_entries_per_subcrq =
4207                                         adapter->max_rx_add_entries_per_subcrq;
4208
4209                 if (adapter->desired.tx_queues)
4210                         adapter->req_tx_queues =
4211                                         adapter->desired.tx_queues;
4212                 else
4213                         adapter->req_tx_queues =
4214                                         adapter->opt_tx_comp_sub_queues;
4215
4216                 if (adapter->desired.rx_queues)
4217                         adapter->req_rx_queues =
4218                                         adapter->desired.rx_queues;
4219                 else
4220                         adapter->req_rx_queues =
4221                                         adapter->opt_rx_comp_queues;
4222
4223                 adapter->req_rx_add_queues = adapter->max_rx_add_queues;
4224         } else {
4225                 atomic_add(cap_reqs, &adapter->running_cap_crqs);
4226         }
4227         memset(&crq, 0, sizeof(crq));
4228         crq.request_capability.first = IBMVNIC_CRQ_CMD;
4229         crq.request_capability.cmd = REQUEST_CAPABILITY;
4230
4231         crq.request_capability.capability = cpu_to_be16(REQ_TX_QUEUES);
4232         crq.request_capability.number = cpu_to_be64(adapter->req_tx_queues);
4233         cap_reqs--;
4234         ibmvnic_send_crq(adapter, &crq);
4235
4236         crq.request_capability.capability = cpu_to_be16(REQ_RX_QUEUES);
4237         crq.request_capability.number = cpu_to_be64(adapter->req_rx_queues);
4238         cap_reqs--;
4239         ibmvnic_send_crq(adapter, &crq);
4240
4241         crq.request_capability.capability = cpu_to_be16(REQ_RX_ADD_QUEUES);
4242         crq.request_capability.number = cpu_to_be64(adapter->req_rx_add_queues);
4243         cap_reqs--;
4244         ibmvnic_send_crq(adapter, &crq);
4245
4246         crq.request_capability.capability =
4247             cpu_to_be16(REQ_TX_ENTRIES_PER_SUBCRQ);
4248         crq.request_capability.number =
4249             cpu_to_be64(adapter->req_tx_entries_per_subcrq);
4250         cap_reqs--;
4251         ibmvnic_send_crq(adapter, &crq);
4252
4253         crq.request_capability.capability =
4254             cpu_to_be16(REQ_RX_ADD_ENTRIES_PER_SUBCRQ);
4255         crq.request_capability.number =
4256             cpu_to_be64(adapter->req_rx_add_entries_per_subcrq);
4257         cap_reqs--;
4258         ibmvnic_send_crq(adapter, &crq);
4259
4260         crq.request_capability.capability = cpu_to_be16(REQ_MTU);
4261         crq.request_capability.number = cpu_to_be64(adapter->req_mtu);
4262         cap_reqs--;
4263         ibmvnic_send_crq(adapter, &crq);
4264
4265         if (adapter->netdev->flags & IFF_PROMISC) {
4266                 if (adapter->promisc_supported) {
4267                         crq.request_capability.capability =
4268                             cpu_to_be16(PROMISC_REQUESTED);
4269                         crq.request_capability.number = cpu_to_be64(1);
4270                         cap_reqs--;
4271                         ibmvnic_send_crq(adapter, &crq);
4272                 }
4273         } else {
4274                 crq.request_capability.capability =
4275                     cpu_to_be16(PROMISC_REQUESTED);
4276                 crq.request_capability.number = cpu_to_be64(0);
4277                 cap_reqs--;
4278                 ibmvnic_send_crq(adapter, &crq);
4279         }
4280
4281         /* Keep at end to catch any discrepancy between expected and actual
4282          * CRQs sent.
4283          */
4284         WARN_ON(cap_reqs != 0);
4285 }
4286
4287 static int pending_scrq(struct ibmvnic_adapter *adapter,
4288                         struct ibmvnic_sub_crq_queue *scrq)
4289 {
4290         union sub_crq *entry = &scrq->msgs[scrq->cur];
4291         int rc;
4292
4293         rc = !!(entry->generic.first & IBMVNIC_CRQ_CMD_RSP);
4294
4295         /* Ensure that the SCRQ valid flag is loaded prior to loading the
4296          * contents of the SCRQ descriptor
4297          */
4298         dma_rmb();
4299
4300         return rc;
4301 }
4302
4303 static union sub_crq *ibmvnic_next_scrq(struct ibmvnic_adapter *adapter,
4304                                         struct ibmvnic_sub_crq_queue *scrq)
4305 {
4306         union sub_crq *entry;
4307         unsigned long flags;
4308
4309         spin_lock_irqsave(&scrq->lock, flags);
4310         entry = &scrq->msgs[scrq->cur];
4311         if (entry->generic.first & IBMVNIC_CRQ_CMD_RSP) {
4312                 if (++scrq->cur == scrq->size)
4313                         scrq->cur = 0;
4314         } else {
4315                 entry = NULL;
4316         }
4317         spin_unlock_irqrestore(&scrq->lock, flags);
4318
4319         /* Ensure that the SCRQ valid flag is loaded prior to loading the
4320          * contents of the SCRQ descriptor
4321          */
4322         dma_rmb();
4323
4324         return entry;
4325 }
4326
4327 static union ibmvnic_crq *ibmvnic_next_crq(struct ibmvnic_adapter *adapter)
4328 {
4329         struct ibmvnic_crq_queue *queue = &adapter->crq;
4330         union ibmvnic_crq *crq;
4331
4332         crq = &queue->msgs[queue->cur];
4333         if (crq->generic.first & IBMVNIC_CRQ_CMD_RSP) {
4334                 if (++queue->cur == queue->size)
4335                         queue->cur = 0;
4336         } else {
4337                 crq = NULL;
4338         }
4339
4340         return crq;
4341 }
4342
4343 static void print_subcrq_error(struct device *dev, int rc, const char *func)
4344 {
4345         switch (rc) {
4346         case H_PARAMETER:
4347                 dev_warn_ratelimited(dev,
4348                                      "%s failed: Send request is malformed or adapter failover pending. (rc=%d)\n",
4349                                      func, rc);
4350                 break;
4351         case H_CLOSED:
4352                 dev_warn_ratelimited(dev,
4353                                      "%s failed: Backing queue closed. Adapter is down or failover pending. (rc=%d)\n",
4354                                      func, rc);
4355                 break;
4356         default:
4357                 dev_err_ratelimited(dev, "%s failed: (rc=%d)\n", func, rc);
4358                 break;
4359         }
4360 }
4361
4362 static int send_subcrq_indirect(struct ibmvnic_adapter *adapter,
4363                                 u64 remote_handle, u64 ioba, u64 num_entries)
4364 {
4365         unsigned int ua = adapter->vdev->unit_address;
4366         struct device *dev = &adapter->vdev->dev;
4367         int rc;
4368
4369         /* Make sure the hypervisor sees the complete request */
4370         dma_wmb();
4371         rc = plpar_hcall_norets(H_SEND_SUB_CRQ_INDIRECT, ua,
4372                                 cpu_to_be64(remote_handle),
4373                                 ioba, num_entries);
4374
4375         if (rc)
4376                 print_subcrq_error(dev, rc, __func__);
4377
4378         return rc;
4379 }
4380
4381 static int ibmvnic_send_crq(struct ibmvnic_adapter *adapter,
4382                             union ibmvnic_crq *crq)
4383 {
4384         unsigned int ua = adapter->vdev->unit_address;
4385         struct device *dev = &adapter->vdev->dev;
4386         u64 *u64_crq = (u64 *)crq;
4387         int rc;
4388
4389         netdev_dbg(adapter->netdev, "Sending CRQ: %016lx %016lx\n",
4390                    (unsigned long)cpu_to_be64(u64_crq[0]),
4391                    (unsigned long)cpu_to_be64(u64_crq[1]));
4392
4393         if (!adapter->crq.active &&
4394             crq->generic.first != IBMVNIC_CRQ_INIT_CMD) {
4395                 dev_warn(dev, "Invalid request detected while CRQ is inactive, possible device state change during reset\n");
4396                 return -EINVAL;
4397         }
4398
4399         /* Make sure the hypervisor sees the complete request */
4400         dma_wmb();
4401
4402         rc = plpar_hcall_norets(H_SEND_CRQ, ua,
4403                                 cpu_to_be64(u64_crq[0]),
4404                                 cpu_to_be64(u64_crq[1]));
4405
4406         if (rc) {
4407                 if (rc == H_CLOSED) {
4408                         dev_warn(dev, "CRQ Queue closed\n");
4409                         /* do not reset, report the fail, wait for passive init from server */
4410                 }
4411
4412                 dev_warn(dev, "Send error (rc=%d)\n", rc);
4413         }
4414
4415         return rc;
4416 }
4417
4418 static int ibmvnic_send_crq_init(struct ibmvnic_adapter *adapter)
4419 {
4420         struct device *dev = &adapter->vdev->dev;
4421         union ibmvnic_crq crq;
4422         int retries = 100;
4423         int rc;
4424
4425         memset(&crq, 0, sizeof(crq));
4426         crq.generic.first = IBMVNIC_CRQ_INIT_CMD;
4427         crq.generic.cmd = IBMVNIC_CRQ_INIT;
4428         netdev_dbg(adapter->netdev, "Sending CRQ init\n");
4429
4430         do {
4431                 rc = ibmvnic_send_crq(adapter, &crq);
4432                 if (rc != H_CLOSED)
4433                         break;
4434                 retries--;
4435                 msleep(50);
4436
4437         } while (retries > 0);
4438
4439         if (rc) {
4440                 dev_err(dev, "Failed to send init request, rc = %d\n", rc);
4441                 return rc;
4442         }
4443
4444         return 0;
4445 }
4446
4447 struct vnic_login_client_data {
4448         u8      type;
4449         __be16  len;
4450         char    name[];
4451 } __packed;
4452
4453 static int vnic_client_data_len(struct ibmvnic_adapter *adapter)
4454 {
4455         int len;
4456
4457         /* Calculate the amount of buffer space needed for the
4458          * vnic client data in the login buffer. There are four entries,
4459          * OS name, LPAR name, device name, and a null last entry.
4460          */
4461         len = 4 * sizeof(struct vnic_login_client_data);
4462         len += 6; /* "Linux" plus NULL */
4463         len += strlen(utsname()->nodename) + 1;
4464         len += strlen(adapter->netdev->name) + 1;
4465
4466         return len;
4467 }
4468
4469 static void vnic_add_client_data(struct ibmvnic_adapter *adapter,
4470                                  struct vnic_login_client_data *vlcd)
4471 {
4472         const char *os_name = "Linux";
4473         int len;
4474
4475         /* Type 1 - LPAR OS */
4476         vlcd->type = 1;
4477         len = strlen(os_name) + 1;
4478         vlcd->len = cpu_to_be16(len);
4479         strscpy(vlcd->name, os_name, len);
4480         vlcd = (struct vnic_login_client_data *)(vlcd->name + len);
4481
4482         /* Type 2 - LPAR name */
4483         vlcd->type = 2;
4484         len = strlen(utsname()->nodename) + 1;
4485         vlcd->len = cpu_to_be16(len);
4486         strscpy(vlcd->name, utsname()->nodename, len);
4487         vlcd = (struct vnic_login_client_data *)(vlcd->name + len);
4488
4489         /* Type 3 - device name */
4490         vlcd->type = 3;
4491         len = strlen(adapter->netdev->name) + 1;
4492         vlcd->len = cpu_to_be16(len);
4493         strscpy(vlcd->name, adapter->netdev->name, len);
4494 }
4495
4496 static int send_login(struct ibmvnic_adapter *adapter)
4497 {
4498         struct ibmvnic_login_rsp_buffer *login_rsp_buffer;
4499         struct ibmvnic_login_buffer *login_buffer;
4500         struct device *dev = &adapter->vdev->dev;
4501         struct vnic_login_client_data *vlcd;
4502         dma_addr_t rsp_buffer_token;
4503         dma_addr_t buffer_token;
4504         size_t rsp_buffer_size;
4505         union ibmvnic_crq crq;
4506         int client_data_len;
4507         size_t buffer_size;
4508         __be64 *tx_list_p;
4509         __be64 *rx_list_p;
4510         int rc;
4511         int i;
4512
4513         if (!adapter->tx_scrq || !adapter->rx_scrq) {
4514                 netdev_err(adapter->netdev,
4515                            "RX or TX queues are not allocated, device login failed\n");
4516                 return -ENOMEM;
4517         }
4518
4519         release_login_buffer(adapter);
4520         release_login_rsp_buffer(adapter);
4521
4522         client_data_len = vnic_client_data_len(adapter);
4523
4524         buffer_size =
4525             sizeof(struct ibmvnic_login_buffer) +
4526             sizeof(u64) * (adapter->req_tx_queues + adapter->req_rx_queues) +
4527             client_data_len;
4528
4529         login_buffer = kzalloc(buffer_size, GFP_ATOMIC);
4530         if (!login_buffer)
4531                 goto buf_alloc_failed;
4532
4533         buffer_token = dma_map_single(dev, login_buffer, buffer_size,
4534                                       DMA_TO_DEVICE);
4535         if (dma_mapping_error(dev, buffer_token)) {
4536                 dev_err(dev, "Couldn't map login buffer\n");
4537                 goto buf_map_failed;
4538         }
4539
4540         rsp_buffer_size = sizeof(struct ibmvnic_login_rsp_buffer) +
4541                           sizeof(u64) * adapter->req_tx_queues +
4542                           sizeof(u64) * adapter->req_rx_queues +
4543                           sizeof(u64) * adapter->req_rx_queues +
4544                           sizeof(u8) * IBMVNIC_TX_DESC_VERSIONS;
4545
4546         login_rsp_buffer = kmalloc(rsp_buffer_size, GFP_ATOMIC);
4547         if (!login_rsp_buffer)
4548                 goto buf_rsp_alloc_failed;
4549
4550         rsp_buffer_token = dma_map_single(dev, login_rsp_buffer,
4551                                           rsp_buffer_size, DMA_FROM_DEVICE);
4552         if (dma_mapping_error(dev, rsp_buffer_token)) {
4553                 dev_err(dev, "Couldn't map login rsp buffer\n");
4554                 goto buf_rsp_map_failed;
4555         }
4556
4557         adapter->login_buf = login_buffer;
4558         adapter->login_buf_token = buffer_token;
4559         adapter->login_buf_sz = buffer_size;
4560         adapter->login_rsp_buf = login_rsp_buffer;
4561         adapter->login_rsp_buf_token = rsp_buffer_token;
4562         adapter->login_rsp_buf_sz = rsp_buffer_size;
4563
4564         login_buffer->len = cpu_to_be32(buffer_size);
4565         login_buffer->version = cpu_to_be32(INITIAL_VERSION_LB);
4566         login_buffer->num_txcomp_subcrqs = cpu_to_be32(adapter->req_tx_queues);
4567         login_buffer->off_txcomp_subcrqs =
4568             cpu_to_be32(sizeof(struct ibmvnic_login_buffer));
4569         login_buffer->num_rxcomp_subcrqs = cpu_to_be32(adapter->req_rx_queues);
4570         login_buffer->off_rxcomp_subcrqs =
4571             cpu_to_be32(sizeof(struct ibmvnic_login_buffer) +
4572                         sizeof(u64) * adapter->req_tx_queues);
4573         login_buffer->login_rsp_ioba = cpu_to_be32(rsp_buffer_token);
4574         login_buffer->login_rsp_len = cpu_to_be32(rsp_buffer_size);
4575
4576         tx_list_p = (__be64 *)((char *)login_buffer +
4577                                       sizeof(struct ibmvnic_login_buffer));
4578         rx_list_p = (__be64 *)((char *)login_buffer +
4579                                       sizeof(struct ibmvnic_login_buffer) +
4580                                       sizeof(u64) * adapter->req_tx_queues);
4581
4582         for (i = 0; i < adapter->req_tx_queues; i++) {
4583                 if (adapter->tx_scrq[i]) {
4584                         tx_list_p[i] =
4585                                 cpu_to_be64(adapter->tx_scrq[i]->crq_num);
4586                 }
4587         }
4588
4589         for (i = 0; i < adapter->req_rx_queues; i++) {
4590                 if (adapter->rx_scrq[i]) {
4591                         rx_list_p[i] =
4592                                 cpu_to_be64(adapter->rx_scrq[i]->crq_num);
4593                 }
4594         }
4595
4596         /* Insert vNIC login client data */
4597         vlcd = (struct vnic_login_client_data *)
4598                 ((char *)rx_list_p + (sizeof(u64) * adapter->req_rx_queues));
4599         login_buffer->client_data_offset =
4600                         cpu_to_be32((char *)vlcd - (char *)login_buffer);
4601         login_buffer->client_data_len = cpu_to_be32(client_data_len);
4602
4603         vnic_add_client_data(adapter, vlcd);
4604
4605         netdev_dbg(adapter->netdev, "Login Buffer:\n");
4606         for (i = 0; i < (adapter->login_buf_sz - 1) / 8 + 1; i++) {
4607                 netdev_dbg(adapter->netdev, "%016lx\n",
4608                            ((unsigned long *)(adapter->login_buf))[i]);
4609         }
4610
4611         memset(&crq, 0, sizeof(crq));
4612         crq.login.first = IBMVNIC_CRQ_CMD;
4613         crq.login.cmd = LOGIN;
4614         crq.login.ioba = cpu_to_be32(buffer_token);
4615         crq.login.len = cpu_to_be32(buffer_size);
4616
4617         adapter->login_pending = true;
4618         rc = ibmvnic_send_crq(adapter, &crq);
4619         if (rc) {
4620                 adapter->login_pending = false;
4621                 netdev_err(adapter->netdev, "Failed to send login, rc=%d\n", rc);
4622                 goto buf_rsp_map_failed;
4623         }
4624
4625         return 0;
4626
4627 buf_rsp_map_failed:
4628         kfree(login_rsp_buffer);
4629         adapter->login_rsp_buf = NULL;
4630 buf_rsp_alloc_failed:
4631         dma_unmap_single(dev, buffer_token, buffer_size, DMA_TO_DEVICE);
4632 buf_map_failed:
4633         kfree(login_buffer);
4634         adapter->login_buf = NULL;
4635 buf_alloc_failed:
4636         return -ENOMEM;
4637 }
4638
4639 static int send_request_map(struct ibmvnic_adapter *adapter, dma_addr_t addr,
4640                             u32 len, u8 map_id)
4641 {
4642         union ibmvnic_crq crq;
4643
4644         memset(&crq, 0, sizeof(crq));
4645         crq.request_map.first = IBMVNIC_CRQ_CMD;
4646         crq.request_map.cmd = REQUEST_MAP;
4647         crq.request_map.map_id = map_id;
4648         crq.request_map.ioba = cpu_to_be32(addr);
4649         crq.request_map.len = cpu_to_be32(len);
4650         return ibmvnic_send_crq(adapter, &crq);
4651 }
4652
4653 static int send_request_unmap(struct ibmvnic_adapter *adapter, u8 map_id)
4654 {
4655         union ibmvnic_crq crq;
4656
4657         memset(&crq, 0, sizeof(crq));
4658         crq.request_unmap.first = IBMVNIC_CRQ_CMD;
4659         crq.request_unmap.cmd = REQUEST_UNMAP;
4660         crq.request_unmap.map_id = map_id;
4661         return ibmvnic_send_crq(adapter, &crq);
4662 }
4663
4664 static void send_query_map(struct ibmvnic_adapter *adapter)
4665 {
4666         union ibmvnic_crq crq;
4667
4668         memset(&crq, 0, sizeof(crq));
4669         crq.query_map.first = IBMVNIC_CRQ_CMD;
4670         crq.query_map.cmd = QUERY_MAP;
4671         ibmvnic_send_crq(adapter, &crq);
4672 }
4673
4674 /* Send a series of CRQs requesting various capabilities of the VNIC server */
4675 static void send_query_cap(struct ibmvnic_adapter *adapter)
4676 {
4677         union ibmvnic_crq crq;
4678         int cap_reqs;
4679
4680         /* We send out 25 QUERY_CAPABILITY CRQs below.  Initialize this count
4681          * upfront. When the tasklet receives a response to all of these, it
4682          * can send out the next protocol messaage (REQUEST_CAPABILITY).
4683          */
4684         cap_reqs = 25;
4685
4686         atomic_set(&adapter->running_cap_crqs, cap_reqs);
4687
4688         memset(&crq, 0, sizeof(crq));
4689         crq.query_capability.first = IBMVNIC_CRQ_CMD;
4690         crq.query_capability.cmd = QUERY_CAPABILITY;
4691
4692         crq.query_capability.capability = cpu_to_be16(MIN_TX_QUEUES);
4693         ibmvnic_send_crq(adapter, &crq);
4694         cap_reqs--;
4695
4696         crq.query_capability.capability = cpu_to_be16(MIN_RX_QUEUES);
4697         ibmvnic_send_crq(adapter, &crq);
4698         cap_reqs--;
4699
4700         crq.query_capability.capability = cpu_to_be16(MIN_RX_ADD_QUEUES);
4701         ibmvnic_send_crq(adapter, &crq);
4702         cap_reqs--;
4703
4704         crq.query_capability.capability = cpu_to_be16(MAX_TX_QUEUES);
4705         ibmvnic_send_crq(adapter, &crq);
4706         cap_reqs--;
4707
4708         crq.query_capability.capability = cpu_to_be16(MAX_RX_QUEUES);
4709         ibmvnic_send_crq(adapter, &crq);
4710         cap_reqs--;
4711
4712         crq.query_capability.capability = cpu_to_be16(MAX_RX_ADD_QUEUES);
4713         ibmvnic_send_crq(adapter, &crq);
4714         cap_reqs--;
4715
4716         crq.query_capability.capability =
4717             cpu_to_be16(MIN_TX_ENTRIES_PER_SUBCRQ);
4718         ibmvnic_send_crq(adapter, &crq);
4719         cap_reqs--;
4720
4721         crq.query_capability.capability =
4722             cpu_to_be16(MIN_RX_ADD_ENTRIES_PER_SUBCRQ);
4723         ibmvnic_send_crq(adapter, &crq);
4724         cap_reqs--;
4725
4726         crq.query_capability.capability =
4727             cpu_to_be16(MAX_TX_ENTRIES_PER_SUBCRQ);
4728         ibmvnic_send_crq(adapter, &crq);
4729         cap_reqs--;
4730
4731         crq.query_capability.capability =
4732             cpu_to_be16(MAX_RX_ADD_ENTRIES_PER_SUBCRQ);
4733         ibmvnic_send_crq(adapter, &crq);
4734         cap_reqs--;
4735
4736         crq.query_capability.capability = cpu_to_be16(TCP_IP_OFFLOAD);
4737         ibmvnic_send_crq(adapter, &crq);
4738         cap_reqs--;
4739
4740         crq.query_capability.capability = cpu_to_be16(PROMISC_SUPPORTED);
4741         ibmvnic_send_crq(adapter, &crq);
4742         cap_reqs--;
4743
4744         crq.query_capability.capability = cpu_to_be16(MIN_MTU);
4745         ibmvnic_send_crq(adapter, &crq);
4746         cap_reqs--;
4747
4748         crq.query_capability.capability = cpu_to_be16(MAX_MTU);
4749         ibmvnic_send_crq(adapter, &crq);
4750         cap_reqs--;
4751
4752         crq.query_capability.capability = cpu_to_be16(MAX_MULTICAST_FILTERS);
4753         ibmvnic_send_crq(adapter, &crq);
4754         cap_reqs--;
4755
4756         crq.query_capability.capability = cpu_to_be16(VLAN_HEADER_INSERTION);
4757         ibmvnic_send_crq(adapter, &crq);
4758         cap_reqs--;
4759
4760         crq.query_capability.capability = cpu_to_be16(RX_VLAN_HEADER_INSERTION);
4761         ibmvnic_send_crq(adapter, &crq);
4762         cap_reqs--;
4763
4764         crq.query_capability.capability = cpu_to_be16(MAX_TX_SG_ENTRIES);
4765         ibmvnic_send_crq(adapter, &crq);
4766         cap_reqs--;
4767
4768         crq.query_capability.capability = cpu_to_be16(RX_SG_SUPPORTED);
4769         ibmvnic_send_crq(adapter, &crq);
4770         cap_reqs--;
4771
4772         crq.query_capability.capability = cpu_to_be16(OPT_TX_COMP_SUB_QUEUES);
4773         ibmvnic_send_crq(adapter, &crq);
4774         cap_reqs--;
4775
4776         crq.query_capability.capability = cpu_to_be16(OPT_RX_COMP_QUEUES);
4777         ibmvnic_send_crq(adapter, &crq);
4778         cap_reqs--;
4779
4780         crq.query_capability.capability =
4781                         cpu_to_be16(OPT_RX_BUFADD_Q_PER_RX_COMP_Q);
4782         ibmvnic_send_crq(adapter, &crq);
4783         cap_reqs--;
4784
4785         crq.query_capability.capability =
4786                         cpu_to_be16(OPT_TX_ENTRIES_PER_SUBCRQ);
4787         ibmvnic_send_crq(adapter, &crq);
4788         cap_reqs--;
4789
4790         crq.query_capability.capability =
4791                         cpu_to_be16(OPT_RXBA_ENTRIES_PER_SUBCRQ);
4792         ibmvnic_send_crq(adapter, &crq);
4793         cap_reqs--;
4794
4795         crq.query_capability.capability = cpu_to_be16(TX_RX_DESC_REQ);
4796
4797         ibmvnic_send_crq(adapter, &crq);
4798         cap_reqs--;
4799
4800         /* Keep at end to catch any discrepancy between expected and actual
4801          * CRQs sent.
4802          */
4803         WARN_ON(cap_reqs != 0);
4804 }
4805
4806 static void send_query_ip_offload(struct ibmvnic_adapter *adapter)
4807 {
4808         int buf_sz = sizeof(struct ibmvnic_query_ip_offload_buffer);
4809         struct device *dev = &adapter->vdev->dev;
4810         union ibmvnic_crq crq;
4811
4812         adapter->ip_offload_tok =
4813                 dma_map_single(dev,
4814                                &adapter->ip_offload_buf,
4815                                buf_sz,
4816                                DMA_FROM_DEVICE);
4817
4818         if (dma_mapping_error(dev, adapter->ip_offload_tok)) {
4819                 if (!firmware_has_feature(FW_FEATURE_CMO))
4820                         dev_err(dev, "Couldn't map offload buffer\n");
4821                 return;
4822         }
4823
4824         memset(&crq, 0, sizeof(crq));
4825         crq.query_ip_offload.first = IBMVNIC_CRQ_CMD;
4826         crq.query_ip_offload.cmd = QUERY_IP_OFFLOAD;
4827         crq.query_ip_offload.len = cpu_to_be32(buf_sz);
4828         crq.query_ip_offload.ioba =
4829             cpu_to_be32(adapter->ip_offload_tok);
4830
4831         ibmvnic_send_crq(adapter, &crq);
4832 }
4833
4834 static void send_control_ip_offload(struct ibmvnic_adapter *adapter)
4835 {
4836         struct ibmvnic_control_ip_offload_buffer *ctrl_buf = &adapter->ip_offload_ctrl;
4837         struct ibmvnic_query_ip_offload_buffer *buf = &adapter->ip_offload_buf;
4838         struct device *dev = &adapter->vdev->dev;
4839         netdev_features_t old_hw_features = 0;
4840         union ibmvnic_crq crq;
4841
4842         adapter->ip_offload_ctrl_tok =
4843                 dma_map_single(dev,
4844                                ctrl_buf,
4845                                sizeof(adapter->ip_offload_ctrl),
4846                                DMA_TO_DEVICE);
4847
4848         if (dma_mapping_error(dev, adapter->ip_offload_ctrl_tok)) {
4849                 dev_err(dev, "Couldn't map ip offload control buffer\n");
4850                 return;
4851         }
4852
4853         ctrl_buf->len = cpu_to_be32(sizeof(adapter->ip_offload_ctrl));
4854         ctrl_buf->version = cpu_to_be32(INITIAL_VERSION_IOB);
4855         ctrl_buf->ipv4_chksum = buf->ipv4_chksum;
4856         ctrl_buf->ipv6_chksum = buf->ipv6_chksum;
4857         ctrl_buf->tcp_ipv4_chksum = buf->tcp_ipv4_chksum;
4858         ctrl_buf->udp_ipv4_chksum = buf->udp_ipv4_chksum;
4859         ctrl_buf->tcp_ipv6_chksum = buf->tcp_ipv6_chksum;
4860         ctrl_buf->udp_ipv6_chksum = buf->udp_ipv6_chksum;
4861         ctrl_buf->large_tx_ipv4 = buf->large_tx_ipv4;
4862         ctrl_buf->large_tx_ipv6 = buf->large_tx_ipv6;
4863
4864         /* large_rx disabled for now, additional features needed */
4865         ctrl_buf->large_rx_ipv4 = 0;
4866         ctrl_buf->large_rx_ipv6 = 0;
4867
4868         if (adapter->state != VNIC_PROBING) {
4869                 old_hw_features = adapter->netdev->hw_features;
4870                 adapter->netdev->hw_features = 0;
4871         }
4872
4873         adapter->netdev->hw_features = NETIF_F_SG | NETIF_F_GSO | NETIF_F_GRO;
4874
4875         if (buf->tcp_ipv4_chksum || buf->udp_ipv4_chksum)
4876                 adapter->netdev->hw_features |= NETIF_F_IP_CSUM;
4877
4878         if (buf->tcp_ipv6_chksum || buf->udp_ipv6_chksum)
4879                 adapter->netdev->hw_features |= NETIF_F_IPV6_CSUM;
4880
4881         if ((adapter->netdev->features &
4882             (NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM)))
4883                 adapter->netdev->hw_features |= NETIF_F_RXCSUM;
4884
4885         if (buf->large_tx_ipv4)
4886                 adapter->netdev->hw_features |= NETIF_F_TSO;
4887         if (buf->large_tx_ipv6)
4888                 adapter->netdev->hw_features |= NETIF_F_TSO6;
4889
4890         if (adapter->state == VNIC_PROBING) {
4891                 adapter->netdev->features |= adapter->netdev->hw_features;
4892         } else if (old_hw_features != adapter->netdev->hw_features) {
4893                 netdev_features_t tmp = 0;
4894
4895                 /* disable features no longer supported */
4896                 adapter->netdev->features &= adapter->netdev->hw_features;
4897                 /* turn on features now supported if previously enabled */
4898                 tmp = (old_hw_features ^ adapter->netdev->hw_features) &
4899                         adapter->netdev->hw_features;
4900                 adapter->netdev->features |=
4901                                 tmp & adapter->netdev->wanted_features;
4902         }
4903
4904         memset(&crq, 0, sizeof(crq));
4905         crq.control_ip_offload.first = IBMVNIC_CRQ_CMD;
4906         crq.control_ip_offload.cmd = CONTROL_IP_OFFLOAD;
4907         crq.control_ip_offload.len =
4908             cpu_to_be32(sizeof(adapter->ip_offload_ctrl));
4909         crq.control_ip_offload.ioba = cpu_to_be32(adapter->ip_offload_ctrl_tok);
4910         ibmvnic_send_crq(adapter, &crq);
4911 }
4912
4913 static void handle_vpd_size_rsp(union ibmvnic_crq *crq,
4914                                 struct ibmvnic_adapter *adapter)
4915 {
4916         struct device *dev = &adapter->vdev->dev;
4917
4918         if (crq->get_vpd_size_rsp.rc.code) {
4919                 dev_err(dev, "Error retrieving VPD size, rc=%x\n",
4920                         crq->get_vpd_size_rsp.rc.code);
4921                 complete(&adapter->fw_done);
4922                 return;
4923         }
4924
4925         adapter->vpd->len = be64_to_cpu(crq->get_vpd_size_rsp.len);
4926         complete(&adapter->fw_done);
4927 }
4928
4929 static void handle_vpd_rsp(union ibmvnic_crq *crq,
4930                            struct ibmvnic_adapter *adapter)
4931 {
4932         struct device *dev = &adapter->vdev->dev;
4933         unsigned char *substr = NULL;
4934         u8 fw_level_len = 0;
4935
4936         memset(adapter->fw_version, 0, 32);
4937
4938         dma_unmap_single(dev, adapter->vpd->dma_addr, adapter->vpd->len,
4939                          DMA_FROM_DEVICE);
4940
4941         if (crq->get_vpd_rsp.rc.code) {
4942                 dev_err(dev, "Error retrieving VPD from device, rc=%x\n",
4943                         crq->get_vpd_rsp.rc.code);
4944                 goto complete;
4945         }
4946
4947         /* get the position of the firmware version info
4948          * located after the ASCII 'RM' substring in the buffer
4949          */
4950         substr = strnstr(adapter->vpd->buff, "RM", adapter->vpd->len);
4951         if (!substr) {
4952                 dev_info(dev, "Warning - No FW level has been provided in the VPD buffer by the VIOS Server\n");
4953                 goto complete;
4954         }
4955
4956         /* get length of firmware level ASCII substring */
4957         if ((substr + 2) < (adapter->vpd->buff + adapter->vpd->len)) {
4958                 fw_level_len = *(substr + 2);
4959         } else {
4960                 dev_info(dev, "Length of FW substr extrapolated VDP buff\n");
4961                 goto complete;
4962         }
4963
4964         /* copy firmware version string from vpd into adapter */
4965         if ((substr + 3 + fw_level_len) <
4966             (adapter->vpd->buff + adapter->vpd->len)) {
4967                 strncpy((char *)adapter->fw_version, substr + 3, fw_level_len);
4968         } else {
4969                 dev_info(dev, "FW substr extrapolated VPD buff\n");
4970         }
4971
4972 complete:
4973         if (adapter->fw_version[0] == '\0')
4974                 strscpy((char *)adapter->fw_version, "N/A", sizeof(adapter->fw_version));
4975         complete(&adapter->fw_done);
4976 }
4977
4978 static void handle_query_ip_offload_rsp(struct ibmvnic_adapter *adapter)
4979 {
4980         struct device *dev = &adapter->vdev->dev;
4981         struct ibmvnic_query_ip_offload_buffer *buf = &adapter->ip_offload_buf;
4982         int i;
4983
4984         dma_unmap_single(dev, adapter->ip_offload_tok,
4985                          sizeof(adapter->ip_offload_buf), DMA_FROM_DEVICE);
4986
4987         netdev_dbg(adapter->netdev, "Query IP Offload Buffer:\n");
4988         for (i = 0; i < (sizeof(adapter->ip_offload_buf) - 1) / 8 + 1; i++)
4989                 netdev_dbg(adapter->netdev, "%016lx\n",
4990                            ((unsigned long *)(buf))[i]);
4991
4992         netdev_dbg(adapter->netdev, "ipv4_chksum = %d\n", buf->ipv4_chksum);
4993         netdev_dbg(adapter->netdev, "ipv6_chksum = %d\n", buf->ipv6_chksum);
4994         netdev_dbg(adapter->netdev, "tcp_ipv4_chksum = %d\n",
4995                    buf->tcp_ipv4_chksum);
4996         netdev_dbg(adapter->netdev, "tcp_ipv6_chksum = %d\n",
4997                    buf->tcp_ipv6_chksum);
4998         netdev_dbg(adapter->netdev, "udp_ipv4_chksum = %d\n",
4999                    buf->udp_ipv4_chksum);
5000         netdev_dbg(adapter->netdev, "udp_ipv6_chksum = %d\n",
5001                    buf->udp_ipv6_chksum);
5002         netdev_dbg(adapter->netdev, "large_tx_ipv4 = %d\n",
5003                    buf->large_tx_ipv4);
5004         netdev_dbg(adapter->netdev, "large_tx_ipv6 = %d\n",
5005                    buf->large_tx_ipv6);
5006         netdev_dbg(adapter->netdev, "large_rx_ipv4 = %d\n",
5007                    buf->large_rx_ipv4);
5008         netdev_dbg(adapter->netdev, "large_rx_ipv6 = %d\n",
5009                    buf->large_rx_ipv6);
5010         netdev_dbg(adapter->netdev, "max_ipv4_hdr_sz = %d\n",
5011                    buf->max_ipv4_header_size);
5012         netdev_dbg(adapter->netdev, "max_ipv6_hdr_sz = %d\n",
5013                    buf->max_ipv6_header_size);
5014         netdev_dbg(adapter->netdev, "max_tcp_hdr_size = %d\n",
5015                    buf->max_tcp_header_size);
5016         netdev_dbg(adapter->netdev, "max_udp_hdr_size = %d\n",
5017                    buf->max_udp_header_size);
5018         netdev_dbg(adapter->netdev, "max_large_tx_size = %d\n",
5019                    buf->max_large_tx_size);
5020         netdev_dbg(adapter->netdev, "max_large_rx_size = %d\n",
5021                    buf->max_large_rx_size);
5022         netdev_dbg(adapter->netdev, "ipv6_ext_hdr = %d\n",
5023                    buf->ipv6_extension_header);
5024         netdev_dbg(adapter->netdev, "tcp_pseudosum_req = %d\n",
5025                    buf->tcp_pseudosum_req);
5026         netdev_dbg(adapter->netdev, "num_ipv6_ext_hd = %d\n",
5027                    buf->num_ipv6_ext_headers);
5028         netdev_dbg(adapter->netdev, "off_ipv6_ext_hd = %d\n",
5029                    buf->off_ipv6_ext_headers);
5030
5031         send_control_ip_offload(adapter);
5032 }
5033
5034 static const char *ibmvnic_fw_err_cause(u16 cause)
5035 {
5036         switch (cause) {
5037         case ADAPTER_PROBLEM:
5038                 return "adapter problem";
5039         case BUS_PROBLEM:
5040                 return "bus problem";
5041         case FW_PROBLEM:
5042                 return "firmware problem";
5043         case DD_PROBLEM:
5044                 return "device driver problem";
5045         case EEH_RECOVERY:
5046                 return "EEH recovery";
5047         case FW_UPDATED:
5048                 return "firmware updated";
5049         case LOW_MEMORY:
5050                 return "low Memory";
5051         default:
5052                 return "unknown";
5053         }
5054 }
5055
5056 static void handle_error_indication(union ibmvnic_crq *crq,
5057                                     struct ibmvnic_adapter *adapter)
5058 {
5059         struct device *dev = &adapter->vdev->dev;
5060         u16 cause;
5061
5062         cause = be16_to_cpu(crq->error_indication.error_cause);
5063
5064         dev_warn_ratelimited(dev,
5065                              "Firmware reports %serror, cause: %s. Starting recovery...\n",
5066                              crq->error_indication.flags
5067                                 & IBMVNIC_FATAL_ERROR ? "FATAL " : "",
5068                              ibmvnic_fw_err_cause(cause));
5069
5070         if (crq->error_indication.flags & IBMVNIC_FATAL_ERROR)
5071                 ibmvnic_reset(adapter, VNIC_RESET_FATAL);
5072         else
5073                 ibmvnic_reset(adapter, VNIC_RESET_NON_FATAL);
5074 }
5075
5076 static int handle_change_mac_rsp(union ibmvnic_crq *crq,
5077                                  struct ibmvnic_adapter *adapter)
5078 {
5079         struct net_device *netdev = adapter->netdev;
5080         struct device *dev = &adapter->vdev->dev;
5081         long rc;
5082
5083         rc = crq->change_mac_addr_rsp.rc.code;
5084         if (rc) {
5085                 dev_err(dev, "Error %ld in CHANGE_MAC_ADDR_RSP\n", rc);
5086                 goto out;
5087         }
5088         /* crq->change_mac_addr.mac_addr is the requested one
5089          * crq->change_mac_addr_rsp.mac_addr is the returned valid one.
5090          */
5091         eth_hw_addr_set(netdev, &crq->change_mac_addr_rsp.mac_addr[0]);
5092         ether_addr_copy(adapter->mac_addr,
5093                         &crq->change_mac_addr_rsp.mac_addr[0]);
5094 out:
5095         complete(&adapter->fw_done);
5096         return rc;
5097 }
5098
5099 static void handle_request_cap_rsp(union ibmvnic_crq *crq,
5100                                    struct ibmvnic_adapter *adapter)
5101 {
5102         struct device *dev = &adapter->vdev->dev;
5103         u64 *req_value;
5104         char *name;
5105
5106         atomic_dec(&adapter->running_cap_crqs);
5107         netdev_dbg(adapter->netdev, "Outstanding request-caps: %d\n",
5108                    atomic_read(&adapter->running_cap_crqs));
5109         switch (be16_to_cpu(crq->request_capability_rsp.capability)) {
5110         case REQ_TX_QUEUES:
5111                 req_value = &adapter->req_tx_queues;
5112                 name = "tx";
5113                 break;
5114         case REQ_RX_QUEUES:
5115                 req_value = &adapter->req_rx_queues;
5116                 name = "rx";
5117                 break;
5118         case REQ_RX_ADD_QUEUES:
5119                 req_value = &adapter->req_rx_add_queues;
5120                 name = "rx_add";
5121                 break;
5122         case REQ_TX_ENTRIES_PER_SUBCRQ:
5123                 req_value = &adapter->req_tx_entries_per_subcrq;
5124                 name = "tx_entries_per_subcrq";
5125                 break;
5126         case REQ_RX_ADD_ENTRIES_PER_SUBCRQ:
5127                 req_value = &adapter->req_rx_add_entries_per_subcrq;
5128                 name = "rx_add_entries_per_subcrq";
5129                 break;
5130         case REQ_MTU:
5131                 req_value = &adapter->req_mtu;
5132                 name = "mtu";
5133                 break;
5134         case PROMISC_REQUESTED:
5135                 req_value = &adapter->promisc;
5136                 name = "promisc";
5137                 break;
5138         default:
5139                 dev_err(dev, "Got invalid cap request rsp %d\n",
5140                         crq->request_capability.capability);
5141                 return;
5142         }
5143
5144         switch (crq->request_capability_rsp.rc.code) {
5145         case SUCCESS:
5146                 break;
5147         case PARTIALSUCCESS:
5148                 dev_info(dev, "req=%lld, rsp=%ld in %s queue, retrying.\n",
5149                          *req_value,
5150                          (long)be64_to_cpu(crq->request_capability_rsp.number),
5151                          name);
5152
5153                 if (be16_to_cpu(crq->request_capability_rsp.capability) ==
5154                     REQ_MTU) {
5155                         pr_err("mtu of %llu is not supported. Reverting.\n",
5156                                *req_value);
5157                         *req_value = adapter->fallback.mtu;
5158                 } else {
5159                         *req_value =
5160                                 be64_to_cpu(crq->request_capability_rsp.number);
5161                 }
5162
5163                 send_request_cap(adapter, 1);
5164                 return;
5165         default:
5166                 dev_err(dev, "Error %d in request cap rsp\n",
5167                         crq->request_capability_rsp.rc.code);
5168                 return;
5169         }
5170
5171         /* Done receiving requested capabilities, query IP offload support */
5172         if (atomic_read(&adapter->running_cap_crqs) == 0)
5173                 send_query_ip_offload(adapter);
5174 }
5175
5176 static int handle_login_rsp(union ibmvnic_crq *login_rsp_crq,
5177                             struct ibmvnic_adapter *adapter)
5178 {
5179         struct device *dev = &adapter->vdev->dev;
5180         struct net_device *netdev = adapter->netdev;
5181         struct ibmvnic_login_rsp_buffer *login_rsp = adapter->login_rsp_buf;
5182         struct ibmvnic_login_buffer *login = adapter->login_buf;
5183         u64 *tx_handle_array;
5184         u64 *rx_handle_array;
5185         int num_tx_pools;
5186         int num_rx_pools;
5187         u64 *size_array;
5188         int i;
5189
5190         /* CHECK: Test/set of login_pending does not need to be atomic
5191          * because only ibmvnic_tasklet tests/clears this.
5192          */
5193         if (!adapter->login_pending) {
5194                 netdev_warn(netdev, "Ignoring unexpected login response\n");
5195                 return 0;
5196         }
5197         adapter->login_pending = false;
5198
5199         dma_unmap_single(dev, adapter->login_buf_token, adapter->login_buf_sz,
5200                          DMA_TO_DEVICE);
5201         dma_unmap_single(dev, adapter->login_rsp_buf_token,
5202                          adapter->login_rsp_buf_sz, DMA_FROM_DEVICE);
5203
5204         /* If the number of queues requested can't be allocated by the
5205          * server, the login response will return with code 1. We will need
5206          * to resend the login buffer with fewer queues requested.
5207          */
5208         if (login_rsp_crq->generic.rc.code) {
5209                 adapter->init_done_rc = login_rsp_crq->generic.rc.code;
5210                 complete(&adapter->init_done);
5211                 return 0;
5212         }
5213
5214         if (adapter->failover_pending) {
5215                 adapter->init_done_rc = -EAGAIN;
5216                 netdev_dbg(netdev, "Failover pending, ignoring login response\n");
5217                 complete(&adapter->init_done);
5218                 /* login response buffer will be released on reset */
5219                 return 0;
5220         }
5221
5222         netdev->mtu = adapter->req_mtu - ETH_HLEN;
5223
5224         netdev_dbg(adapter->netdev, "Login Response Buffer:\n");
5225         for (i = 0; i < (adapter->login_rsp_buf_sz - 1) / 8 + 1; i++) {
5226                 netdev_dbg(adapter->netdev, "%016lx\n",
5227                            ((unsigned long *)(adapter->login_rsp_buf))[i]);
5228         }
5229
5230         /* Sanity checks */
5231         if (login->num_txcomp_subcrqs != login_rsp->num_txsubm_subcrqs ||
5232             (be32_to_cpu(login->num_rxcomp_subcrqs) *
5233              adapter->req_rx_add_queues !=
5234              be32_to_cpu(login_rsp->num_rxadd_subcrqs))) {
5235                 dev_err(dev, "FATAL: Inconsistent login and login rsp\n");
5236                 ibmvnic_reset(adapter, VNIC_RESET_FATAL);
5237                 return -EIO;
5238         }
5239         size_array = (u64 *)((u8 *)(adapter->login_rsp_buf) +
5240                 be32_to_cpu(adapter->login_rsp_buf->off_rxadd_buff_size));
5241         /* variable buffer sizes are not supported, so just read the
5242          * first entry.
5243          */
5244         adapter->cur_rx_buf_sz = be64_to_cpu(size_array[0]);
5245
5246         num_tx_pools = be32_to_cpu(adapter->login_rsp_buf->num_txsubm_subcrqs);
5247         num_rx_pools = be32_to_cpu(adapter->login_rsp_buf->num_rxadd_subcrqs);
5248
5249         tx_handle_array = (u64 *)((u8 *)(adapter->login_rsp_buf) +
5250                                   be32_to_cpu(adapter->login_rsp_buf->off_txsubm_subcrqs));
5251         rx_handle_array = (u64 *)((u8 *)(adapter->login_rsp_buf) +
5252                                   be32_to_cpu(adapter->login_rsp_buf->off_rxadd_subcrqs));
5253
5254         for (i = 0; i < num_tx_pools; i++)
5255                 adapter->tx_scrq[i]->handle = tx_handle_array[i];
5256
5257         for (i = 0; i < num_rx_pools; i++)
5258                 adapter->rx_scrq[i]->handle = rx_handle_array[i];
5259
5260         adapter->num_active_tx_scrqs = num_tx_pools;
5261         adapter->num_active_rx_scrqs = num_rx_pools;
5262         release_login_rsp_buffer(adapter);
5263         release_login_buffer(adapter);
5264         complete(&adapter->init_done);
5265
5266         return 0;
5267 }
5268
5269 static void handle_request_unmap_rsp(union ibmvnic_crq *crq,
5270                                      struct ibmvnic_adapter *adapter)
5271 {
5272         struct device *dev = &adapter->vdev->dev;
5273         long rc;
5274
5275         rc = crq->request_unmap_rsp.rc.code;
5276         if (rc)
5277                 dev_err(dev, "Error %ld in REQUEST_UNMAP_RSP\n", rc);
5278 }
5279
5280 static void handle_query_map_rsp(union ibmvnic_crq *crq,
5281                                  struct ibmvnic_adapter *adapter)
5282 {
5283         struct net_device *netdev = adapter->netdev;
5284         struct device *dev = &adapter->vdev->dev;
5285         long rc;
5286
5287         rc = crq->query_map_rsp.rc.code;
5288         if (rc) {
5289                 dev_err(dev, "Error %ld in QUERY_MAP_RSP\n", rc);
5290                 return;
5291         }
5292         netdev_dbg(netdev, "page_size = %d\ntot_pages = %u\nfree_pages = %u\n",
5293                    crq->query_map_rsp.page_size,
5294                    __be32_to_cpu(crq->query_map_rsp.tot_pages),
5295                    __be32_to_cpu(crq->query_map_rsp.free_pages));
5296 }
5297
5298 static void handle_query_cap_rsp(union ibmvnic_crq *crq,
5299                                  struct ibmvnic_adapter *adapter)
5300 {
5301         struct net_device *netdev = adapter->netdev;
5302         struct device *dev = &adapter->vdev->dev;
5303         long rc;
5304
5305         atomic_dec(&adapter->running_cap_crqs);
5306         netdev_dbg(netdev, "Outstanding queries: %d\n",
5307                    atomic_read(&adapter->running_cap_crqs));
5308         rc = crq->query_capability.rc.code;
5309         if (rc) {
5310                 dev_err(dev, "Error %ld in QUERY_CAP_RSP\n", rc);
5311                 goto out;
5312         }
5313
5314         switch (be16_to_cpu(crq->query_capability.capability)) {
5315         case MIN_TX_QUEUES:
5316                 adapter->min_tx_queues =
5317                     be64_to_cpu(crq->query_capability.number);
5318                 netdev_dbg(netdev, "min_tx_queues = %lld\n",
5319                            adapter->min_tx_queues);
5320                 break;
5321         case MIN_RX_QUEUES:
5322                 adapter->min_rx_queues =
5323                     be64_to_cpu(crq->query_capability.number);
5324                 netdev_dbg(netdev, "min_rx_queues = %lld\n",
5325                            adapter->min_rx_queues);
5326                 break;
5327         case MIN_RX_ADD_QUEUES:
5328                 adapter->min_rx_add_queues =
5329                     be64_to_cpu(crq->query_capability.number);
5330                 netdev_dbg(netdev, "min_rx_add_queues = %lld\n",
5331                            adapter->min_rx_add_queues);
5332                 break;
5333         case MAX_TX_QUEUES:
5334                 adapter->max_tx_queues =
5335                     be64_to_cpu(crq->query_capability.number);
5336                 netdev_dbg(netdev, "max_tx_queues = %lld\n",
5337                            adapter->max_tx_queues);
5338                 break;
5339         case MAX_RX_QUEUES:
5340                 adapter->max_rx_queues =
5341                     be64_to_cpu(crq->query_capability.number);
5342                 netdev_dbg(netdev, "max_rx_queues = %lld\n",
5343                            adapter->max_rx_queues);
5344                 break;
5345         case MAX_RX_ADD_QUEUES:
5346                 adapter->max_rx_add_queues =
5347                     be64_to_cpu(crq->query_capability.number);
5348                 netdev_dbg(netdev, "max_rx_add_queues = %lld\n",
5349                            adapter->max_rx_add_queues);
5350                 break;
5351         case MIN_TX_ENTRIES_PER_SUBCRQ:
5352                 adapter->min_tx_entries_per_subcrq =
5353                     be64_to_cpu(crq->query_capability.number);
5354                 netdev_dbg(netdev, "min_tx_entries_per_subcrq = %lld\n",
5355                            adapter->min_tx_entries_per_subcrq);
5356                 break;
5357         case MIN_RX_ADD_ENTRIES_PER_SUBCRQ:
5358                 adapter->min_rx_add_entries_per_subcrq =
5359                     be64_to_cpu(crq->query_capability.number);
5360                 netdev_dbg(netdev, "min_rx_add_entrs_per_subcrq = %lld\n",
5361                            adapter->min_rx_add_entries_per_subcrq);
5362                 break;
5363         case MAX_TX_ENTRIES_PER_SUBCRQ:
5364                 adapter->max_tx_entries_per_subcrq =
5365                     be64_to_cpu(crq->query_capability.number);
5366                 netdev_dbg(netdev, "max_tx_entries_per_subcrq = %lld\n",
5367                            adapter->max_tx_entries_per_subcrq);
5368                 break;
5369         case MAX_RX_ADD_ENTRIES_PER_SUBCRQ:
5370                 adapter->max_rx_add_entries_per_subcrq =
5371                     be64_to_cpu(crq->query_capability.number);
5372                 netdev_dbg(netdev, "max_rx_add_entrs_per_subcrq = %lld\n",
5373                            adapter->max_rx_add_entries_per_subcrq);
5374                 break;
5375         case TCP_IP_OFFLOAD:
5376                 adapter->tcp_ip_offload =
5377                     be64_to_cpu(crq->query_capability.number);
5378                 netdev_dbg(netdev, "tcp_ip_offload = %lld\n",
5379                            adapter->tcp_ip_offload);
5380                 break;
5381         case PROMISC_SUPPORTED:
5382                 adapter->promisc_supported =
5383                     be64_to_cpu(crq->query_capability.number);
5384                 netdev_dbg(netdev, "promisc_supported = %lld\n",
5385                            adapter->promisc_supported);
5386                 break;
5387         case MIN_MTU:
5388                 adapter->min_mtu = be64_to_cpu(crq->query_capability.number);
5389                 netdev->min_mtu = adapter->min_mtu - ETH_HLEN;
5390                 netdev_dbg(netdev, "min_mtu = %lld\n", adapter->min_mtu);
5391                 break;
5392         case MAX_MTU:
5393                 adapter->max_mtu = be64_to_cpu(crq->query_capability.number);
5394                 netdev->max_mtu = adapter->max_mtu - ETH_HLEN;
5395                 netdev_dbg(netdev, "max_mtu = %lld\n", adapter->max_mtu);
5396                 break;
5397         case MAX_MULTICAST_FILTERS:
5398                 adapter->max_multicast_filters =
5399                     be64_to_cpu(crq->query_capability.number);
5400                 netdev_dbg(netdev, "max_multicast_filters = %lld\n",
5401                            adapter->max_multicast_filters);
5402                 break;
5403         case VLAN_HEADER_INSERTION:
5404                 adapter->vlan_header_insertion =
5405                     be64_to_cpu(crq->query_capability.number);
5406                 if (adapter->vlan_header_insertion)
5407                         netdev->features |= NETIF_F_HW_VLAN_STAG_TX;
5408                 netdev_dbg(netdev, "vlan_header_insertion = %lld\n",
5409                            adapter->vlan_header_insertion);
5410                 break;
5411         case RX_VLAN_HEADER_INSERTION:
5412                 adapter->rx_vlan_header_insertion =
5413                     be64_to_cpu(crq->query_capability.number);
5414                 netdev_dbg(netdev, "rx_vlan_header_insertion = %lld\n",
5415                            adapter->rx_vlan_header_insertion);
5416                 break;
5417         case MAX_TX_SG_ENTRIES:
5418                 adapter->max_tx_sg_entries =
5419                     be64_to_cpu(crq->query_capability.number);
5420                 netdev_dbg(netdev, "max_tx_sg_entries = %lld\n",
5421                            adapter->max_tx_sg_entries);
5422                 break;
5423         case RX_SG_SUPPORTED:
5424                 adapter->rx_sg_supported =
5425                     be64_to_cpu(crq->query_capability.number);
5426                 netdev_dbg(netdev, "rx_sg_supported = %lld\n",
5427                            adapter->rx_sg_supported);
5428                 break;
5429         case OPT_TX_COMP_SUB_QUEUES:
5430                 adapter->opt_tx_comp_sub_queues =
5431                     be64_to_cpu(crq->query_capability.number);
5432                 netdev_dbg(netdev, "opt_tx_comp_sub_queues = %lld\n",
5433                            adapter->opt_tx_comp_sub_queues);
5434                 break;
5435         case OPT_RX_COMP_QUEUES:
5436                 adapter->opt_rx_comp_queues =
5437                     be64_to_cpu(crq->query_capability.number);
5438                 netdev_dbg(netdev, "opt_rx_comp_queues = %lld\n",
5439                            adapter->opt_rx_comp_queues);
5440                 break;
5441         case OPT_RX_BUFADD_Q_PER_RX_COMP_Q:
5442                 adapter->opt_rx_bufadd_q_per_rx_comp_q =
5443                     be64_to_cpu(crq->query_capability.number);
5444                 netdev_dbg(netdev, "opt_rx_bufadd_q_per_rx_comp_q = %lld\n",
5445                            adapter->opt_rx_bufadd_q_per_rx_comp_q);
5446                 break;
5447         case OPT_TX_ENTRIES_PER_SUBCRQ:
5448                 adapter->opt_tx_entries_per_subcrq =
5449                     be64_to_cpu(crq->query_capability.number);
5450                 netdev_dbg(netdev, "opt_tx_entries_per_subcrq = %lld\n",
5451                            adapter->opt_tx_entries_per_subcrq);
5452                 break;
5453         case OPT_RXBA_ENTRIES_PER_SUBCRQ:
5454                 adapter->opt_rxba_entries_per_subcrq =
5455                     be64_to_cpu(crq->query_capability.number);
5456                 netdev_dbg(netdev, "opt_rxba_entries_per_subcrq = %lld\n",
5457                            adapter->opt_rxba_entries_per_subcrq);
5458                 break;
5459         case TX_RX_DESC_REQ:
5460                 adapter->tx_rx_desc_req = crq->query_capability.number;
5461                 netdev_dbg(netdev, "tx_rx_desc_req = %llx\n",
5462                            adapter->tx_rx_desc_req);
5463                 break;
5464
5465         default:
5466                 netdev_err(netdev, "Got invalid cap rsp %d\n",
5467                            crq->query_capability.capability);
5468         }
5469
5470 out:
5471         if (atomic_read(&adapter->running_cap_crqs) == 0)
5472                 send_request_cap(adapter, 0);
5473 }
5474
5475 static int send_query_phys_parms(struct ibmvnic_adapter *adapter)
5476 {
5477         union ibmvnic_crq crq;
5478         int rc;
5479
5480         memset(&crq, 0, sizeof(crq));
5481         crq.query_phys_parms.first = IBMVNIC_CRQ_CMD;
5482         crq.query_phys_parms.cmd = QUERY_PHYS_PARMS;
5483
5484         mutex_lock(&adapter->fw_lock);
5485         adapter->fw_done_rc = 0;
5486         reinit_completion(&adapter->fw_done);
5487
5488         rc = ibmvnic_send_crq(adapter, &crq);
5489         if (rc) {
5490                 mutex_unlock(&adapter->fw_lock);
5491                 return rc;
5492         }
5493
5494         rc = ibmvnic_wait_for_completion(adapter, &adapter->fw_done, 10000);
5495         if (rc) {
5496                 mutex_unlock(&adapter->fw_lock);
5497                 return rc;
5498         }
5499
5500         mutex_unlock(&adapter->fw_lock);
5501         return adapter->fw_done_rc ? -EIO : 0;
5502 }
5503
5504 static int handle_query_phys_parms_rsp(union ibmvnic_crq *crq,
5505                                        struct ibmvnic_adapter *adapter)
5506 {
5507         struct net_device *netdev = adapter->netdev;
5508         int rc;
5509         __be32 rspeed = cpu_to_be32(crq->query_phys_parms_rsp.speed);
5510
5511         rc = crq->query_phys_parms_rsp.rc.code;
5512         if (rc) {
5513                 netdev_err(netdev, "Error %d in QUERY_PHYS_PARMS\n", rc);
5514                 return rc;
5515         }
5516         switch (rspeed) {
5517         case IBMVNIC_10MBPS:
5518                 adapter->speed = SPEED_10;
5519                 break;
5520         case IBMVNIC_100MBPS:
5521                 adapter->speed = SPEED_100;
5522                 break;
5523         case IBMVNIC_1GBPS:
5524                 adapter->speed = SPEED_1000;
5525                 break;
5526         case IBMVNIC_10GBPS:
5527                 adapter->speed = SPEED_10000;
5528                 break;
5529         case IBMVNIC_25GBPS:
5530                 adapter->speed = SPEED_25000;
5531                 break;
5532         case IBMVNIC_40GBPS:
5533                 adapter->speed = SPEED_40000;
5534                 break;
5535         case IBMVNIC_50GBPS:
5536                 adapter->speed = SPEED_50000;
5537                 break;
5538         case IBMVNIC_100GBPS:
5539                 adapter->speed = SPEED_100000;
5540                 break;
5541         case IBMVNIC_200GBPS:
5542                 adapter->speed = SPEED_200000;
5543                 break;
5544         default:
5545                 if (netif_carrier_ok(netdev))
5546                         netdev_warn(netdev, "Unknown speed 0x%08x\n", rspeed);
5547                 adapter->speed = SPEED_UNKNOWN;
5548         }
5549         if (crq->query_phys_parms_rsp.flags1 & IBMVNIC_FULL_DUPLEX)
5550                 adapter->duplex = DUPLEX_FULL;
5551         else if (crq->query_phys_parms_rsp.flags1 & IBMVNIC_HALF_DUPLEX)
5552                 adapter->duplex = DUPLEX_HALF;
5553         else
5554                 adapter->duplex = DUPLEX_UNKNOWN;
5555
5556         return rc;
5557 }
5558
5559 static void ibmvnic_handle_crq(union ibmvnic_crq *crq,
5560                                struct ibmvnic_adapter *adapter)
5561 {
5562         struct ibmvnic_generic_crq *gen_crq = &crq->generic;
5563         struct net_device *netdev = adapter->netdev;
5564         struct device *dev = &adapter->vdev->dev;
5565         u64 *u64_crq = (u64 *)crq;
5566         long rc;
5567
5568         netdev_dbg(netdev, "Handling CRQ: %016lx %016lx\n",
5569                    (unsigned long)cpu_to_be64(u64_crq[0]),
5570                    (unsigned long)cpu_to_be64(u64_crq[1]));
5571         switch (gen_crq->first) {
5572         case IBMVNIC_CRQ_INIT_RSP:
5573                 switch (gen_crq->cmd) {
5574                 case IBMVNIC_CRQ_INIT:
5575                         dev_info(dev, "Partner initialized\n");
5576                         adapter->from_passive_init = true;
5577                         /* Discard any stale login responses from prev reset.
5578                          * CHECK: should we clear even on INIT_COMPLETE?
5579                          */
5580                         adapter->login_pending = false;
5581
5582                         if (adapter->state == VNIC_DOWN)
5583                                 rc = ibmvnic_reset(adapter, VNIC_RESET_PASSIVE_INIT);
5584                         else
5585                                 rc = ibmvnic_reset(adapter, VNIC_RESET_FAILOVER);
5586
5587                         if (rc && rc != -EBUSY) {
5588                                 /* We were unable to schedule the failover
5589                                  * reset either because the adapter was still
5590                                  * probing (eg: during kexec) or we could not
5591                                  * allocate memory. Clear the failover_pending
5592                                  * flag since no one else will. We ignore
5593                                  * EBUSY because it means either FAILOVER reset
5594                                  * is already scheduled or the adapter is
5595                                  * being removed.
5596                                  */
5597                                 netdev_err(netdev,
5598                                            "Error %ld scheduling failover reset\n",
5599                                            rc);
5600                                 adapter->failover_pending = false;
5601                         }
5602
5603                         if (!completion_done(&adapter->init_done)) {
5604                                 if (!adapter->init_done_rc)
5605                                         adapter->init_done_rc = -EAGAIN;
5606                                 complete(&adapter->init_done);
5607                         }
5608
5609                         break;
5610                 case IBMVNIC_CRQ_INIT_COMPLETE:
5611                         dev_info(dev, "Partner initialization complete\n");
5612                         adapter->crq.active = true;
5613                         send_version_xchg(adapter);
5614                         break;
5615                 default:
5616                         dev_err(dev, "Unknown crq cmd: %d\n", gen_crq->cmd);
5617                 }
5618                 return;
5619         case IBMVNIC_CRQ_XPORT_EVENT:
5620                 netif_carrier_off(netdev);
5621                 adapter->crq.active = false;
5622                 /* terminate any thread waiting for a response
5623                  * from the device
5624                  */
5625                 if (!completion_done(&adapter->fw_done)) {
5626                         adapter->fw_done_rc = -EIO;
5627                         complete(&adapter->fw_done);
5628                 }
5629
5630                 /* if we got here during crq-init, retry crq-init */
5631                 if (!completion_done(&adapter->init_done)) {
5632                         adapter->init_done_rc = -EAGAIN;
5633                         complete(&adapter->init_done);
5634                 }
5635
5636                 if (!completion_done(&adapter->stats_done))
5637                         complete(&adapter->stats_done);
5638                 if (test_bit(0, &adapter->resetting))
5639                         adapter->force_reset_recovery = true;
5640                 if (gen_crq->cmd == IBMVNIC_PARTITION_MIGRATED) {
5641                         dev_info(dev, "Migrated, re-enabling adapter\n");
5642                         ibmvnic_reset(adapter, VNIC_RESET_MOBILITY);
5643                 } else if (gen_crq->cmd == IBMVNIC_DEVICE_FAILOVER) {
5644                         dev_info(dev, "Backing device failover detected\n");
5645                         adapter->failover_pending = true;
5646                 } else {
5647                         /* The adapter lost the connection */
5648                         dev_err(dev, "Virtual Adapter failed (rc=%d)\n",
5649                                 gen_crq->cmd);
5650                         ibmvnic_reset(adapter, VNIC_RESET_FATAL);
5651                 }
5652                 return;
5653         case IBMVNIC_CRQ_CMD_RSP:
5654                 break;
5655         default:
5656                 dev_err(dev, "Got an invalid msg type 0x%02x\n",
5657                         gen_crq->first);
5658                 return;
5659         }
5660
5661         switch (gen_crq->cmd) {
5662         case VERSION_EXCHANGE_RSP:
5663                 rc = crq->version_exchange_rsp.rc.code;
5664                 if (rc) {
5665                         dev_err(dev, "Error %ld in VERSION_EXCHG_RSP\n", rc);
5666                         break;
5667                 }
5668                 ibmvnic_version =
5669                             be16_to_cpu(crq->version_exchange_rsp.version);
5670                 dev_info(dev, "Partner protocol version is %d\n",
5671                          ibmvnic_version);
5672                 send_query_cap(adapter);
5673                 break;
5674         case QUERY_CAPABILITY_RSP:
5675                 handle_query_cap_rsp(crq, adapter);
5676                 break;
5677         case QUERY_MAP_RSP:
5678                 handle_query_map_rsp(crq, adapter);
5679                 break;
5680         case REQUEST_MAP_RSP:
5681                 adapter->fw_done_rc = crq->request_map_rsp.rc.code;
5682                 complete(&adapter->fw_done);
5683                 break;
5684         case REQUEST_UNMAP_RSP:
5685                 handle_request_unmap_rsp(crq, adapter);
5686                 break;
5687         case REQUEST_CAPABILITY_RSP:
5688                 handle_request_cap_rsp(crq, adapter);
5689                 break;
5690         case LOGIN_RSP:
5691                 netdev_dbg(netdev, "Got Login Response\n");
5692                 handle_login_rsp(crq, adapter);
5693                 break;
5694         case LOGICAL_LINK_STATE_RSP:
5695                 netdev_dbg(netdev,
5696                            "Got Logical Link State Response, state: %d rc: %d\n",
5697                            crq->logical_link_state_rsp.link_state,
5698                            crq->logical_link_state_rsp.rc.code);
5699                 adapter->logical_link_state =
5700                     crq->logical_link_state_rsp.link_state;
5701                 adapter->init_done_rc = crq->logical_link_state_rsp.rc.code;
5702                 complete(&adapter->init_done);
5703                 break;
5704         case LINK_STATE_INDICATION:
5705                 netdev_dbg(netdev, "Got Logical Link State Indication\n");
5706                 adapter->phys_link_state =
5707                     crq->link_state_indication.phys_link_state;
5708                 adapter->logical_link_state =
5709                     crq->link_state_indication.logical_link_state;
5710                 if (adapter->phys_link_state && adapter->logical_link_state)
5711                         netif_carrier_on(netdev);
5712                 else
5713                         netif_carrier_off(netdev);
5714                 break;
5715         case CHANGE_MAC_ADDR_RSP:
5716                 netdev_dbg(netdev, "Got MAC address change Response\n");
5717                 adapter->fw_done_rc = handle_change_mac_rsp(crq, adapter);
5718                 break;
5719         case ERROR_INDICATION:
5720                 netdev_dbg(netdev, "Got Error Indication\n");
5721                 handle_error_indication(crq, adapter);
5722                 break;
5723         case REQUEST_STATISTICS_RSP:
5724                 netdev_dbg(netdev, "Got Statistics Response\n");
5725                 complete(&adapter->stats_done);
5726                 break;
5727         case QUERY_IP_OFFLOAD_RSP:
5728                 netdev_dbg(netdev, "Got Query IP offload Response\n");
5729                 handle_query_ip_offload_rsp(adapter);
5730                 break;
5731         case MULTICAST_CTRL_RSP:
5732                 netdev_dbg(netdev, "Got multicast control Response\n");
5733                 break;
5734         case CONTROL_IP_OFFLOAD_RSP:
5735                 netdev_dbg(netdev, "Got Control IP offload Response\n");
5736                 dma_unmap_single(dev, adapter->ip_offload_ctrl_tok,
5737                                  sizeof(adapter->ip_offload_ctrl),
5738                                  DMA_TO_DEVICE);
5739                 complete(&adapter->init_done);
5740                 break;
5741         case COLLECT_FW_TRACE_RSP:
5742                 netdev_dbg(netdev, "Got Collect firmware trace Response\n");
5743                 complete(&adapter->fw_done);
5744                 break;
5745         case GET_VPD_SIZE_RSP:
5746                 handle_vpd_size_rsp(crq, adapter);
5747                 break;
5748         case GET_VPD_RSP:
5749                 handle_vpd_rsp(crq, adapter);
5750                 break;
5751         case QUERY_PHYS_PARMS_RSP:
5752                 adapter->fw_done_rc = handle_query_phys_parms_rsp(crq, adapter);
5753                 complete(&adapter->fw_done);
5754                 break;
5755         default:
5756                 netdev_err(netdev, "Got an invalid cmd type 0x%02x\n",
5757                            gen_crq->cmd);
5758         }
5759 }
5760
5761 static irqreturn_t ibmvnic_interrupt(int irq, void *instance)
5762 {
5763         struct ibmvnic_adapter *adapter = instance;
5764
5765         tasklet_schedule(&adapter->tasklet);
5766         return IRQ_HANDLED;
5767 }
5768
5769 static void ibmvnic_tasklet(struct tasklet_struct *t)
5770 {
5771         struct ibmvnic_adapter *adapter = from_tasklet(adapter, t, tasklet);
5772         struct ibmvnic_crq_queue *queue = &adapter->crq;
5773         union ibmvnic_crq *crq;
5774         unsigned long flags;
5775
5776         spin_lock_irqsave(&queue->lock, flags);
5777
5778         /* Pull all the valid messages off the CRQ */
5779         while ((crq = ibmvnic_next_crq(adapter)) != NULL) {
5780                 /* This barrier makes sure ibmvnic_next_crq()'s
5781                  * crq->generic.first & IBMVNIC_CRQ_CMD_RSP is loaded
5782                  * before ibmvnic_handle_crq()'s
5783                  * switch(gen_crq->first) and switch(gen_crq->cmd).
5784                  */
5785                 dma_rmb();
5786                 ibmvnic_handle_crq(crq, adapter);
5787                 crq->generic.first = 0;
5788         }
5789
5790         spin_unlock_irqrestore(&queue->lock, flags);
5791 }
5792
5793 static int ibmvnic_reenable_crq_queue(struct ibmvnic_adapter *adapter)
5794 {
5795         struct vio_dev *vdev = adapter->vdev;
5796         int rc;
5797
5798         do {
5799                 rc = plpar_hcall_norets(H_ENABLE_CRQ, vdev->unit_address);
5800         } while (rc == H_IN_PROGRESS || rc == H_BUSY || H_IS_LONG_BUSY(rc));
5801
5802         if (rc)
5803                 dev_err(&vdev->dev, "Error enabling adapter (rc=%d)\n", rc);
5804
5805         return rc;
5806 }
5807
5808 static int ibmvnic_reset_crq(struct ibmvnic_adapter *adapter)
5809 {
5810         struct ibmvnic_crq_queue *crq = &adapter->crq;
5811         struct device *dev = &adapter->vdev->dev;
5812         struct vio_dev *vdev = adapter->vdev;
5813         int rc;
5814
5815         /* Close the CRQ */
5816         do {
5817                 rc = plpar_hcall_norets(H_FREE_CRQ, vdev->unit_address);
5818         } while (rc == H_BUSY || H_IS_LONG_BUSY(rc));
5819
5820         /* Clean out the queue */
5821         if (!crq->msgs)
5822                 return -EINVAL;
5823
5824         memset(crq->msgs, 0, PAGE_SIZE);
5825         crq->cur = 0;
5826         crq->active = false;
5827
5828         /* And re-open it again */
5829         rc = plpar_hcall_norets(H_REG_CRQ, vdev->unit_address,
5830                                 crq->msg_token, PAGE_SIZE);
5831
5832         if (rc == H_CLOSED)
5833                 /* Adapter is good, but other end is not ready */
5834                 dev_warn(dev, "Partner adapter not ready\n");
5835         else if (rc != 0)
5836                 dev_warn(dev, "Couldn't register crq (rc=%d)\n", rc);
5837
5838         return rc;
5839 }
5840
5841 static void release_crq_queue(struct ibmvnic_adapter *adapter)
5842 {
5843         struct ibmvnic_crq_queue *crq = &adapter->crq;
5844         struct vio_dev *vdev = adapter->vdev;
5845         long rc;
5846
5847         if (!crq->msgs)
5848                 return;
5849
5850         netdev_dbg(adapter->netdev, "Releasing CRQ\n");
5851         free_irq(vdev->irq, adapter);
5852         tasklet_kill(&adapter->tasklet);
5853         do {
5854                 rc = plpar_hcall_norets(H_FREE_CRQ, vdev->unit_address);
5855         } while (rc == H_BUSY || H_IS_LONG_BUSY(rc));
5856
5857         dma_unmap_single(&vdev->dev, crq->msg_token, PAGE_SIZE,
5858                          DMA_BIDIRECTIONAL);
5859         free_page((unsigned long)crq->msgs);
5860         crq->msgs = NULL;
5861         crq->active = false;
5862 }
5863
5864 static int init_crq_queue(struct ibmvnic_adapter *adapter)
5865 {
5866         struct ibmvnic_crq_queue *crq = &adapter->crq;
5867         struct device *dev = &adapter->vdev->dev;
5868         struct vio_dev *vdev = adapter->vdev;
5869         int rc, retrc = -ENOMEM;
5870
5871         if (crq->msgs)
5872                 return 0;
5873
5874         crq->msgs = (union ibmvnic_crq *)get_zeroed_page(GFP_KERNEL);
5875         /* Should we allocate more than one page? */
5876
5877         if (!crq->msgs)
5878                 return -ENOMEM;
5879
5880         crq->size = PAGE_SIZE / sizeof(*crq->msgs);
5881         crq->msg_token = dma_map_single(dev, crq->msgs, PAGE_SIZE,
5882                                         DMA_BIDIRECTIONAL);
5883         if (dma_mapping_error(dev, crq->msg_token))
5884                 goto map_failed;
5885
5886         rc = plpar_hcall_norets(H_REG_CRQ, vdev->unit_address,
5887                                 crq->msg_token, PAGE_SIZE);
5888
5889         if (rc == H_RESOURCE)
5890                 /* maybe kexecing and resource is busy. try a reset */
5891                 rc = ibmvnic_reset_crq(adapter);
5892         retrc = rc;
5893
5894         if (rc == H_CLOSED) {
5895                 dev_warn(dev, "Partner adapter not ready\n");
5896         } else if (rc) {
5897                 dev_warn(dev, "Error %d opening adapter\n", rc);
5898                 goto reg_crq_failed;
5899         }
5900
5901         retrc = 0;
5902
5903         tasklet_setup(&adapter->tasklet, (void *)ibmvnic_tasklet);
5904
5905         netdev_dbg(adapter->netdev, "registering irq 0x%x\n", vdev->irq);
5906         snprintf(crq->name, sizeof(crq->name), "ibmvnic-%x",
5907                  adapter->vdev->unit_address);
5908         rc = request_irq(vdev->irq, ibmvnic_interrupt, 0, crq->name, adapter);
5909         if (rc) {
5910                 dev_err(dev, "Couldn't register irq 0x%x. rc=%d\n",
5911                         vdev->irq, rc);
5912                 goto req_irq_failed;
5913         }
5914
5915         rc = vio_enable_interrupts(vdev);
5916         if (rc) {
5917                 dev_err(dev, "Error %d enabling interrupts\n", rc);
5918                 goto req_irq_failed;
5919         }
5920
5921         crq->cur = 0;
5922         spin_lock_init(&crq->lock);
5923
5924         /* process any CRQs that were queued before we enabled interrupts */
5925         tasklet_schedule(&adapter->tasklet);
5926
5927         return retrc;
5928
5929 req_irq_failed:
5930         tasklet_kill(&adapter->tasklet);
5931         do {
5932                 rc = plpar_hcall_norets(H_FREE_CRQ, vdev->unit_address);
5933         } while (rc == H_BUSY || H_IS_LONG_BUSY(rc));
5934 reg_crq_failed:
5935         dma_unmap_single(dev, crq->msg_token, PAGE_SIZE, DMA_BIDIRECTIONAL);
5936 map_failed:
5937         free_page((unsigned long)crq->msgs);
5938         crq->msgs = NULL;
5939         return retrc;
5940 }
5941
5942 static int ibmvnic_reset_init(struct ibmvnic_adapter *adapter, bool reset)
5943 {
5944         struct device *dev = &adapter->vdev->dev;
5945         unsigned long timeout = msecs_to_jiffies(20000);
5946         u64 old_num_rx_queues = adapter->req_rx_queues;
5947         u64 old_num_tx_queues = adapter->req_tx_queues;
5948         int rc;
5949
5950         adapter->from_passive_init = false;
5951
5952         rc = ibmvnic_send_crq_init(adapter);
5953         if (rc) {
5954                 dev_err(dev, "Send crq init failed with error %d\n", rc);
5955                 return rc;
5956         }
5957
5958         if (!wait_for_completion_timeout(&adapter->init_done, timeout)) {
5959                 dev_err(dev, "Initialization sequence timed out\n");
5960                 return -ETIMEDOUT;
5961         }
5962
5963         if (adapter->init_done_rc) {
5964                 release_crq_queue(adapter);
5965                 dev_err(dev, "CRQ-init failed, %d\n", adapter->init_done_rc);
5966                 return adapter->init_done_rc;
5967         }
5968
5969         if (adapter->from_passive_init) {
5970                 adapter->state = VNIC_OPEN;
5971                 adapter->from_passive_init = false;
5972                 dev_err(dev, "CRQ-init failed, passive-init\n");
5973                 return -EINVAL;
5974         }
5975
5976         if (reset &&
5977             test_bit(0, &adapter->resetting) && !adapter->wait_for_reset &&
5978             adapter->reset_reason != VNIC_RESET_MOBILITY) {
5979                 if (adapter->req_rx_queues != old_num_rx_queues ||
5980                     adapter->req_tx_queues != old_num_tx_queues) {
5981                         release_sub_crqs(adapter, 0);
5982                         rc = init_sub_crqs(adapter);
5983                 } else {
5984                         rc = reset_sub_crq_queues(adapter);
5985                 }
5986         } else {
5987                 rc = init_sub_crqs(adapter);
5988         }
5989
5990         if (rc) {
5991                 dev_err(dev, "Initialization of sub crqs failed\n");
5992                 release_crq_queue(adapter);
5993                 return rc;
5994         }
5995
5996         rc = init_sub_crq_irqs(adapter);
5997         if (rc) {
5998                 dev_err(dev, "Failed to initialize sub crq irqs\n");
5999                 release_crq_queue(adapter);
6000         }
6001
6002         return rc;
6003 }
6004
6005 static struct device_attribute dev_attr_failover;
6006
6007 static int ibmvnic_probe(struct vio_dev *dev, const struct vio_device_id *id)
6008 {
6009         struct ibmvnic_adapter *adapter;
6010         struct net_device *netdev;
6011         unsigned char *mac_addr_p;
6012         unsigned long flags;
6013         bool init_success;
6014         int rc;
6015
6016         dev_dbg(&dev->dev, "entering ibmvnic_probe for UA 0x%x\n",
6017                 dev->unit_address);
6018
6019         mac_addr_p = (unsigned char *)vio_get_attribute(dev,
6020                                                         VETH_MAC_ADDR, NULL);
6021         if (!mac_addr_p) {
6022                 dev_err(&dev->dev,
6023                         "(%s:%3.3d) ERROR: Can't find MAC_ADDR attribute\n",
6024                         __FILE__, __LINE__);
6025                 return 0;
6026         }
6027
6028         netdev = alloc_etherdev_mq(sizeof(struct ibmvnic_adapter),
6029                                    IBMVNIC_MAX_QUEUES);
6030         if (!netdev)
6031                 return -ENOMEM;
6032
6033         adapter = netdev_priv(netdev);
6034         adapter->state = VNIC_PROBING;
6035         dev_set_drvdata(&dev->dev, netdev);
6036         adapter->vdev = dev;
6037         adapter->netdev = netdev;
6038         adapter->login_pending = false;
6039         memset(&adapter->map_ids, 0, sizeof(adapter->map_ids));
6040         /* map_ids start at 1, so ensure map_id 0 is always "in-use" */
6041         bitmap_set(adapter->map_ids, 0, 1);
6042
6043         ether_addr_copy(adapter->mac_addr, mac_addr_p);
6044         eth_hw_addr_set(netdev, adapter->mac_addr);
6045         netdev->irq = dev->irq;
6046         netdev->netdev_ops = &ibmvnic_netdev_ops;
6047         netdev->ethtool_ops = &ibmvnic_ethtool_ops;
6048         SET_NETDEV_DEV(netdev, &dev->dev);
6049
6050         INIT_WORK(&adapter->ibmvnic_reset, __ibmvnic_reset);
6051         INIT_DELAYED_WORK(&adapter->ibmvnic_delayed_reset,
6052                           __ibmvnic_delayed_reset);
6053         INIT_LIST_HEAD(&adapter->rwi_list);
6054         spin_lock_init(&adapter->rwi_lock);
6055         spin_lock_init(&adapter->state_lock);
6056         mutex_init(&adapter->fw_lock);
6057         init_completion(&adapter->probe_done);
6058         init_completion(&adapter->init_done);
6059         init_completion(&adapter->fw_done);
6060         init_completion(&adapter->reset_done);
6061         init_completion(&adapter->stats_done);
6062         clear_bit(0, &adapter->resetting);
6063         adapter->prev_rx_buf_sz = 0;
6064         adapter->prev_mtu = 0;
6065
6066         init_success = false;
6067         do {
6068                 reinit_init_done(adapter);
6069
6070                 /* clear any failovers we got in the previous pass
6071                  * since we are reinitializing the CRQ
6072                  */
6073                 adapter->failover_pending = false;
6074
6075                 /* If we had already initialized CRQ, we may have one or
6076                  * more resets queued already. Discard those and release
6077                  * the CRQ before initializing the CRQ again.
6078                  */
6079                 release_crq_queue(adapter);
6080
6081                 /* Since we are still in PROBING state, __ibmvnic_reset()
6082                  * will not access the ->rwi_list and since we released CRQ,
6083                  * we won't get _new_ transport events. But there maybe an
6084                  * ongoing ibmvnic_reset() call. So serialize access to
6085                  * rwi_list. If we win the race, ibvmnic_reset() could add
6086                  * a reset after we purged but thats ok - we just may end
6087                  * up with an extra reset (i.e similar to having two or more
6088                  * resets in the queue at once).
6089                  * CHECK.
6090                  */
6091                 spin_lock_irqsave(&adapter->rwi_lock, flags);
6092                 flush_reset_queue(adapter);
6093                 spin_unlock_irqrestore(&adapter->rwi_lock, flags);
6094
6095                 rc = init_crq_queue(adapter);
6096                 if (rc) {
6097                         dev_err(&dev->dev, "Couldn't initialize crq. rc=%d\n",
6098                                 rc);
6099                         goto ibmvnic_init_fail;
6100                 }
6101
6102                 rc = ibmvnic_reset_init(adapter, false);
6103         } while (rc == -EAGAIN);
6104
6105         /* We are ignoring the error from ibmvnic_reset_init() assuming that the
6106          * partner is not ready. CRQ is not active. When the partner becomes
6107          * ready, we will do the passive init reset.
6108          */
6109
6110         if (!rc)
6111                 init_success = true;
6112
6113         rc = init_stats_buffers(adapter);
6114         if (rc)
6115                 goto ibmvnic_init_fail;
6116
6117         rc = init_stats_token(adapter);
6118         if (rc)
6119                 goto ibmvnic_stats_fail;
6120
6121         rc = device_create_file(&dev->dev, &dev_attr_failover);
6122         if (rc)
6123                 goto ibmvnic_dev_file_err;
6124
6125         netif_carrier_off(netdev);
6126
6127         if (init_success) {
6128                 adapter->state = VNIC_PROBED;
6129                 netdev->mtu = adapter->req_mtu - ETH_HLEN;
6130                 netdev->min_mtu = adapter->min_mtu - ETH_HLEN;
6131                 netdev->max_mtu = adapter->max_mtu - ETH_HLEN;
6132         } else {
6133                 adapter->state = VNIC_DOWN;
6134         }
6135
6136         adapter->wait_for_reset = false;
6137         adapter->last_reset_time = jiffies;
6138
6139         rc = register_netdev(netdev);
6140         if (rc) {
6141                 dev_err(&dev->dev, "failed to register netdev rc=%d\n", rc);
6142                 goto ibmvnic_register_fail;
6143         }
6144         dev_info(&dev->dev, "ibmvnic registered\n");
6145
6146         complete(&adapter->probe_done);
6147
6148         return 0;
6149
6150 ibmvnic_register_fail:
6151         device_remove_file(&dev->dev, &dev_attr_failover);
6152
6153 ibmvnic_dev_file_err:
6154         release_stats_token(adapter);
6155
6156 ibmvnic_stats_fail:
6157         release_stats_buffers(adapter);
6158
6159 ibmvnic_init_fail:
6160         release_sub_crqs(adapter, 1);
6161         release_crq_queue(adapter);
6162
6163         /* cleanup worker thread after releasing CRQ so we don't get
6164          * transport events (i.e new work items for the worker thread).
6165          */
6166         adapter->state = VNIC_REMOVING;
6167         complete(&adapter->probe_done);
6168         flush_work(&adapter->ibmvnic_reset);
6169         flush_delayed_work(&adapter->ibmvnic_delayed_reset);
6170
6171         flush_reset_queue(adapter);
6172
6173         mutex_destroy(&adapter->fw_lock);
6174         free_netdev(netdev);
6175
6176         return rc;
6177 }
6178
6179 static void ibmvnic_remove(struct vio_dev *dev)
6180 {
6181         struct net_device *netdev = dev_get_drvdata(&dev->dev);
6182         struct ibmvnic_adapter *adapter = netdev_priv(netdev);
6183         unsigned long flags;
6184
6185         spin_lock_irqsave(&adapter->state_lock, flags);
6186
6187         /* If ibmvnic_reset() is scheduling a reset, wait for it to
6188          * finish. Then, set the state to REMOVING to prevent it from
6189          * scheduling any more work and to have reset functions ignore
6190          * any resets that have already been scheduled. Drop the lock
6191          * after setting state, so __ibmvnic_reset() which is called
6192          * from the flush_work() below, can make progress.
6193          */
6194         spin_lock(&adapter->rwi_lock);
6195         adapter->state = VNIC_REMOVING;
6196         spin_unlock(&adapter->rwi_lock);
6197
6198         spin_unlock_irqrestore(&adapter->state_lock, flags);
6199
6200         flush_work(&adapter->ibmvnic_reset);
6201         flush_delayed_work(&adapter->ibmvnic_delayed_reset);
6202
6203         rtnl_lock();
6204         unregister_netdevice(netdev);
6205
6206         release_resources(adapter);
6207         release_rx_pools(adapter);
6208         release_tx_pools(adapter);
6209         release_sub_crqs(adapter, 1);
6210         release_crq_queue(adapter);
6211
6212         release_stats_token(adapter);
6213         release_stats_buffers(adapter);
6214
6215         adapter->state = VNIC_REMOVED;
6216
6217         rtnl_unlock();
6218         mutex_destroy(&adapter->fw_lock);
6219         device_remove_file(&dev->dev, &dev_attr_failover);
6220         free_netdev(netdev);
6221         dev_set_drvdata(&dev->dev, NULL);
6222 }
6223
6224 static ssize_t failover_store(struct device *dev, struct device_attribute *attr,
6225                               const char *buf, size_t count)
6226 {
6227         struct net_device *netdev = dev_get_drvdata(dev);
6228         struct ibmvnic_adapter *adapter = netdev_priv(netdev);
6229         unsigned long retbuf[PLPAR_HCALL_BUFSIZE];
6230         __be64 session_token;
6231         long rc;
6232
6233         if (!sysfs_streq(buf, "1"))
6234                 return -EINVAL;
6235
6236         rc = plpar_hcall(H_VIOCTL, retbuf, adapter->vdev->unit_address,
6237                          H_GET_SESSION_TOKEN, 0, 0, 0);
6238         if (rc) {
6239                 netdev_err(netdev, "Couldn't retrieve session token, rc %ld\n",
6240                            rc);
6241                 goto last_resort;
6242         }
6243
6244         session_token = (__be64)retbuf[0];
6245         netdev_dbg(netdev, "Initiating client failover, session id %llx\n",
6246                    be64_to_cpu(session_token));
6247         rc = plpar_hcall_norets(H_VIOCTL, adapter->vdev->unit_address,
6248                                 H_SESSION_ERR_DETECTED, session_token, 0, 0);
6249         if (rc) {
6250                 netdev_err(netdev,
6251                            "H_VIOCTL initiated failover failed, rc %ld\n",
6252                            rc);
6253                 goto last_resort;
6254         }
6255
6256         return count;
6257
6258 last_resort:
6259         netdev_dbg(netdev, "Trying to send CRQ_CMD, the last resort\n");
6260         ibmvnic_reset(adapter, VNIC_RESET_FAILOVER);
6261
6262         return count;
6263 }
6264 static DEVICE_ATTR_WO(failover);
6265
6266 static unsigned long ibmvnic_get_desired_dma(struct vio_dev *vdev)
6267 {
6268         struct net_device *netdev = dev_get_drvdata(&vdev->dev);
6269         struct ibmvnic_adapter *adapter;
6270         struct iommu_table *tbl;
6271         unsigned long ret = 0;
6272         int i;
6273
6274         tbl = get_iommu_table_base(&vdev->dev);
6275
6276         /* netdev inits at probe time along with the structures we need below*/
6277         if (!netdev)
6278                 return IOMMU_PAGE_ALIGN(IBMVNIC_IO_ENTITLEMENT_DEFAULT, tbl);
6279
6280         adapter = netdev_priv(netdev);
6281
6282         ret += PAGE_SIZE; /* the crq message queue */
6283         ret += IOMMU_PAGE_ALIGN(sizeof(struct ibmvnic_statistics), tbl);
6284
6285         for (i = 0; i < adapter->req_tx_queues + adapter->req_rx_queues; i++)
6286                 ret += 4 * PAGE_SIZE; /* the scrq message queue */
6287
6288         for (i = 0; i < adapter->num_active_rx_pools; i++)
6289                 ret += adapter->rx_pool[i].size *
6290                     IOMMU_PAGE_ALIGN(adapter->rx_pool[i].buff_size, tbl);
6291
6292         return ret;
6293 }
6294
6295 static int ibmvnic_resume(struct device *dev)
6296 {
6297         struct net_device *netdev = dev_get_drvdata(dev);
6298         struct ibmvnic_adapter *adapter = netdev_priv(netdev);
6299
6300         if (adapter->state != VNIC_OPEN)
6301                 return 0;
6302
6303         tasklet_schedule(&adapter->tasklet);
6304
6305         return 0;
6306 }
6307
6308 static const struct vio_device_id ibmvnic_device_table[] = {
6309         {"network", "IBM,vnic"},
6310         {"", "" }
6311 };
6312 MODULE_DEVICE_TABLE(vio, ibmvnic_device_table);
6313
6314 static const struct dev_pm_ops ibmvnic_pm_ops = {
6315         .resume = ibmvnic_resume
6316 };
6317
6318 static struct vio_driver ibmvnic_driver = {
6319         .id_table       = ibmvnic_device_table,
6320         .probe          = ibmvnic_probe,
6321         .remove         = ibmvnic_remove,
6322         .get_desired_dma = ibmvnic_get_desired_dma,
6323         .name           = ibmvnic_driver_name,
6324         .pm             = &ibmvnic_pm_ops,
6325 };
6326
6327 /* module functions */
6328 static int __init ibmvnic_module_init(void)
6329 {
6330         pr_info("%s: %s %s\n", ibmvnic_driver_name, ibmvnic_driver_string,
6331                 IBMVNIC_DRIVER_VERSION);
6332
6333         return vio_register_driver(&ibmvnic_driver);
6334 }
6335
6336 static void __exit ibmvnic_module_exit(void)
6337 {
6338         vio_unregister_driver(&ibmvnic_driver);
6339 }
6340
6341 module_init(ibmvnic_module_init);
6342 module_exit(ibmvnic_module_exit);