linux/drivers/net/ethernet/ibm/ibmvnic.c
<<
>>
Prefs
   1// SPDX-License-Identifier: GPL-2.0-or-later
   2/**************************************************************************/
   3/*                                                                        */
   4/*  IBM System i and System p Virtual NIC Device Driver                   */
   5/*  Copyright (C) 2014 IBM Corp.                                          */
   6/*  Santiago Leon (santi_leon@yahoo.com)                                  */
   7/*  Thomas Falcon (tlfalcon@linux.vnet.ibm.com)                           */
   8/*  John Allen (jallen@linux.vnet.ibm.com)                                */
   9/*                                                                        */
  10/*                                                                        */
  11/* This module contains the implementation of a virtual ethernet device   */
  12/* for use with IBM i/p Series LPAR Linux. It utilizes the logical LAN    */
  13/* option of the RS/6000 Platform Architecture to interface with virtual  */
  14/* ethernet NICs that are presented to the partition by the hypervisor.   */
  15/*                                                                         */
  16/* Messages are passed between the VNIC driver and the VNIC server using  */
  17/* Command/Response Queues (CRQs) and sub CRQs (sCRQs). CRQs are used to  */
  18/* issue and receive commands that initiate communication with the server */
  19/* on driver initialization. Sub CRQs (sCRQs) are similar to CRQs, but    */
  20/* are used by the driver to notify the server that a packet is           */
  21/* ready for transmission or that a buffer has been added to receive a    */
  22/* packet. Subsequently, sCRQs are used by the server to notify the       */
  23/* driver that a packet transmission has been completed or that a packet  */
  24/* has been received and placed in a waiting buffer.                      */
  25/*                                                                        */
  26/* In lieu of a more conventional "on-the-fly" DMA mapping strategy in    */
  27/* which skbs are DMA mapped and immediately unmapped when the transmit   */
  28/* or receive has been completed, the VNIC driver is required to use      */
  29/* "long term mapping". This entails that large, continuous DMA mapped    */
  30/* buffers are allocated on driver initialization and these buffers are   */
  31/* then continuously reused to pass skbs to and from the VNIC server.     */
  32/*                                                                        */
  33/**************************************************************************/
  34
  35#include <linux/module.h>
  36#include <linux/moduleparam.h>
  37#include <linux/types.h>
  38#include <linux/errno.h>
  39#include <linux/completion.h>
  40#include <linux/ioport.h>
  41#include <linux/dma-mapping.h>
  42#include <linux/kernel.h>
  43#include <linux/netdevice.h>
  44#include <linux/etherdevice.h>
  45#include <linux/skbuff.h>
  46#include <linux/init.h>
  47#include <linux/delay.h>
  48#include <linux/mm.h>
  49#include <linux/ethtool.h>
  50#include <linux/proc_fs.h>
  51#include <linux/if_arp.h>
  52#include <linux/in.h>
  53#include <linux/ip.h>
  54#include <linux/ipv6.h>
  55#include <linux/irq.h>
  56#include <linux/kthread.h>
  57#include <linux/seq_file.h>
  58#include <linux/interrupt.h>
  59#include <net/net_namespace.h>
  60#include <asm/hvcall.h>
  61#include <linux/atomic.h>
  62#include <asm/vio.h>
  63#include <asm/iommu.h>
  64#include <linux/uaccess.h>
  65#include <asm/firmware.h>
  66#include <linux/workqueue.h>
  67#include <linux/if_vlan.h>
  68#include <linux/utsname.h>
  69
  70#include "ibmvnic.h"
  71
  72static const char ibmvnic_driver_name[] = "ibmvnic";
  73static const char ibmvnic_driver_string[] = "IBM System i/p Virtual NIC Driver";
  74
  75MODULE_AUTHOR("Santiago Leon");
  76MODULE_DESCRIPTION("IBM System i/p Virtual NIC Driver");
  77MODULE_LICENSE("GPL");
  78MODULE_VERSION(IBMVNIC_DRIVER_VERSION);
  79
  80static int ibmvnic_version = IBMVNIC_INITIAL_VERSION;
  81static void release_sub_crqs(struct ibmvnic_adapter *, bool);
  82static int ibmvnic_reset_crq(struct ibmvnic_adapter *);
  83static int ibmvnic_send_crq_init(struct ibmvnic_adapter *);
  84static int ibmvnic_reenable_crq_queue(struct ibmvnic_adapter *);
  85static int ibmvnic_send_crq(struct ibmvnic_adapter *, union ibmvnic_crq *);
  86static int send_subcrq_indirect(struct ibmvnic_adapter *, u64, u64, u64);
  87static irqreturn_t ibmvnic_interrupt_rx(int irq, void *instance);
  88static int enable_scrq_irq(struct ibmvnic_adapter *,
  89                           struct ibmvnic_sub_crq_queue *);
  90static int disable_scrq_irq(struct ibmvnic_adapter *,
  91                            struct ibmvnic_sub_crq_queue *);
  92static int pending_scrq(struct ibmvnic_adapter *,
  93                        struct ibmvnic_sub_crq_queue *);
  94static union sub_crq *ibmvnic_next_scrq(struct ibmvnic_adapter *,
  95                                        struct ibmvnic_sub_crq_queue *);
  96static int ibmvnic_poll(struct napi_struct *napi, int data);
  97static void send_query_map(struct ibmvnic_adapter *adapter);
  98static int send_request_map(struct ibmvnic_adapter *, dma_addr_t, u32, u8);
  99static int send_request_unmap(struct ibmvnic_adapter *, u8);
 100static int send_login(struct ibmvnic_adapter *adapter);
 101static void send_query_cap(struct ibmvnic_adapter *adapter);
 102static int init_sub_crqs(struct ibmvnic_adapter *);
 103static int init_sub_crq_irqs(struct ibmvnic_adapter *adapter);
 104static int ibmvnic_reset_init(struct ibmvnic_adapter *, bool reset);
 105static void release_crq_queue(struct ibmvnic_adapter *);
 106static int __ibmvnic_set_mac(struct net_device *, u8 *);
 107static int init_crq_queue(struct ibmvnic_adapter *adapter);
 108static int send_query_phys_parms(struct ibmvnic_adapter *adapter);
 109static void ibmvnic_tx_scrq_clean_buffer(struct ibmvnic_adapter *adapter,
 110                                         struct ibmvnic_sub_crq_queue *tx_scrq);
 111
 112struct ibmvnic_stat {
 113        char name[ETH_GSTRING_LEN];
 114        int offset;
 115};
 116
 117#define IBMVNIC_STAT_OFF(stat) (offsetof(struct ibmvnic_adapter, stats) + \
 118                             offsetof(struct ibmvnic_statistics, stat))
 119#define IBMVNIC_GET_STAT(a, off) (*((u64 *)(((unsigned long)(a)) + (off))))
 120
 121static const struct ibmvnic_stat ibmvnic_stats[] = {
 122        {"rx_packets", IBMVNIC_STAT_OFF(rx_packets)},
 123        {"rx_bytes", IBMVNIC_STAT_OFF(rx_bytes)},
 124        {"tx_packets", IBMVNIC_STAT_OFF(tx_packets)},
 125        {"tx_bytes", IBMVNIC_STAT_OFF(tx_bytes)},
 126        {"ucast_tx_packets", IBMVNIC_STAT_OFF(ucast_tx_packets)},
 127        {"ucast_rx_packets", IBMVNIC_STAT_OFF(ucast_rx_packets)},
 128        {"mcast_tx_packets", IBMVNIC_STAT_OFF(mcast_tx_packets)},
 129        {"mcast_rx_packets", IBMVNIC_STAT_OFF(mcast_rx_packets)},
 130        {"bcast_tx_packets", IBMVNIC_STAT_OFF(bcast_tx_packets)},
 131        {"bcast_rx_packets", IBMVNIC_STAT_OFF(bcast_rx_packets)},
 132        {"align_errors", IBMVNIC_STAT_OFF(align_errors)},
 133        {"fcs_errors", IBMVNIC_STAT_OFF(fcs_errors)},
 134        {"single_collision_frames", IBMVNIC_STAT_OFF(single_collision_frames)},
 135        {"multi_collision_frames", IBMVNIC_STAT_OFF(multi_collision_frames)},
 136        {"sqe_test_errors", IBMVNIC_STAT_OFF(sqe_test_errors)},
 137        {"deferred_tx", IBMVNIC_STAT_OFF(deferred_tx)},
 138        {"late_collisions", IBMVNIC_STAT_OFF(late_collisions)},
 139        {"excess_collisions", IBMVNIC_STAT_OFF(excess_collisions)},
 140        {"internal_mac_tx_errors", IBMVNIC_STAT_OFF(internal_mac_tx_errors)},
 141        {"carrier_sense", IBMVNIC_STAT_OFF(carrier_sense)},
 142        {"too_long_frames", IBMVNIC_STAT_OFF(too_long_frames)},
 143        {"internal_mac_rx_errors", IBMVNIC_STAT_OFF(internal_mac_rx_errors)},
 144};
 145
 146static int send_crq_init_complete(struct ibmvnic_adapter *adapter)
 147{
 148        union ibmvnic_crq crq;
 149
 150        memset(&crq, 0, sizeof(crq));
 151        crq.generic.first = IBMVNIC_CRQ_INIT_CMD;
 152        crq.generic.cmd = IBMVNIC_CRQ_INIT_COMPLETE;
 153
 154        return ibmvnic_send_crq(adapter, &crq);
 155}
 156
 157static int send_version_xchg(struct ibmvnic_adapter *adapter)
 158{
 159        union ibmvnic_crq crq;
 160
 161        memset(&crq, 0, sizeof(crq));
 162        crq.version_exchange.first = IBMVNIC_CRQ_CMD;
 163        crq.version_exchange.cmd = VERSION_EXCHANGE;
 164        crq.version_exchange.version = cpu_to_be16(ibmvnic_version);
 165
 166        return ibmvnic_send_crq(adapter, &crq);
 167}
 168
 169static long h_reg_sub_crq(unsigned long unit_address, unsigned long token,
 170                          unsigned long length, unsigned long *number,
 171                          unsigned long *irq)
 172{
 173        unsigned long retbuf[PLPAR_HCALL_BUFSIZE];
 174        long rc;
 175
 176        rc = plpar_hcall(H_REG_SUB_CRQ, retbuf, unit_address, token, length);
 177        *number = retbuf[0];
 178        *irq = retbuf[1];
 179
 180        return rc;
 181}
 182
 183/**
 184 * ibmvnic_wait_for_completion - Check device state and wait for completion
 185 * @adapter: private device data
 186 * @comp_done: completion structure to wait for
 187 * @timeout: time to wait in milliseconds
 188 *
 189 * Wait for a completion signal or until the timeout limit is reached
 190 * while checking that the device is still active.
 191 */
 192static int ibmvnic_wait_for_completion(struct ibmvnic_adapter *adapter,
 193                                       struct completion *comp_done,
 194                                       unsigned long timeout)
 195{
 196        struct net_device *netdev;
 197        unsigned long div_timeout;
 198        u8 retry;
 199
 200        netdev = adapter->netdev;
 201        retry = 5;
 202        div_timeout = msecs_to_jiffies(timeout / retry);
 203        while (true) {
 204                if (!adapter->crq.active) {
 205                        netdev_err(netdev, "Device down!\n");
 206                        return -ENODEV;
 207                }
 208                if (!retry--)
 209                        break;
 210                if (wait_for_completion_timeout(comp_done, div_timeout))
 211                        return 0;
 212        }
 213        netdev_err(netdev, "Operation timed out.\n");
 214        return -ETIMEDOUT;
 215}
 216
 217static int alloc_long_term_buff(struct ibmvnic_adapter *adapter,
 218                                struct ibmvnic_long_term_buff *ltb, int size)
 219{
 220        struct device *dev = &adapter->vdev->dev;
 221        int rc;
 222
 223        ltb->size = size;
 224        ltb->buff = dma_alloc_coherent(dev, ltb->size, &ltb->addr,
 225                                       GFP_KERNEL);
 226
 227        if (!ltb->buff) {
 228                dev_err(dev, "Couldn't alloc long term buffer\n");
 229                return -ENOMEM;
 230        }
 231        ltb->map_id = adapter->map_id;
 232        adapter->map_id++;
 233
 234        mutex_lock(&adapter->fw_lock);
 235        adapter->fw_done_rc = 0;
 236        reinit_completion(&adapter->fw_done);
 237
 238        rc = send_request_map(adapter, ltb->addr, ltb->size, ltb->map_id);
 239        if (rc) {
 240                dev_err(dev, "send_request_map failed, rc = %d\n", rc);
 241                goto out;
 242        }
 243
 244        rc = ibmvnic_wait_for_completion(adapter, &adapter->fw_done, 10000);
 245        if (rc) {
 246                dev_err(dev,
 247                        "Long term map request aborted or timed out,rc = %d\n",
 248                        rc);
 249                goto out;
 250        }
 251
 252        if (adapter->fw_done_rc) {
 253                dev_err(dev, "Couldn't map long term buffer,rc = %d\n",
 254                        adapter->fw_done_rc);
 255                rc = -1;
 256                goto out;
 257        }
 258        rc = 0;
 259out:
 260        if (rc) {
 261                dma_free_coherent(dev, ltb->size, ltb->buff, ltb->addr);
 262                ltb->buff = NULL;
 263        }
 264        mutex_unlock(&adapter->fw_lock);
 265        return rc;
 266}
 267
 268static void free_long_term_buff(struct ibmvnic_adapter *adapter,
 269                                struct ibmvnic_long_term_buff *ltb)
 270{
 271        struct device *dev = &adapter->vdev->dev;
 272
 273        if (!ltb->buff)
 274                return;
 275
 276        /* VIOS automatically unmaps the long term buffer at remote
 277         * end for the following resets:
 278         * FAILOVER, MOBILITY, TIMEOUT.
 279         */
 280        if (adapter->reset_reason != VNIC_RESET_FAILOVER &&
 281            adapter->reset_reason != VNIC_RESET_MOBILITY &&
 282            adapter->reset_reason != VNIC_RESET_TIMEOUT)
 283                send_request_unmap(adapter, ltb->map_id);
 284        dma_free_coherent(dev, ltb->size, ltb->buff, ltb->addr);
 285        ltb->buff = NULL;
 286        ltb->map_id = 0;
 287}
 288
 289static int reset_long_term_buff(struct ibmvnic_adapter *adapter,
 290                                struct ibmvnic_long_term_buff *ltb)
 291{
 292        struct device *dev = &adapter->vdev->dev;
 293        int rc;
 294
 295        memset(ltb->buff, 0, ltb->size);
 296
 297        mutex_lock(&adapter->fw_lock);
 298        adapter->fw_done_rc = 0;
 299
 300        reinit_completion(&adapter->fw_done);
 301        rc = send_request_map(adapter, ltb->addr, ltb->size, ltb->map_id);
 302        if (rc) {
 303                mutex_unlock(&adapter->fw_lock);
 304                return rc;
 305        }
 306
 307        rc = ibmvnic_wait_for_completion(adapter, &adapter->fw_done, 10000);
 308        if (rc) {
 309                dev_info(dev,
 310                         "Reset failed, long term map request timed out or aborted\n");
 311                mutex_unlock(&adapter->fw_lock);
 312                return rc;
 313        }
 314
 315        if (adapter->fw_done_rc) {
 316                dev_info(dev,
 317                         "Reset failed, attempting to free and reallocate buffer\n");
 318                free_long_term_buff(adapter, ltb);
 319                mutex_unlock(&adapter->fw_lock);
 320                return alloc_long_term_buff(adapter, ltb, ltb->size);
 321        }
 322        mutex_unlock(&adapter->fw_lock);
 323        return 0;
 324}
 325
 326static void deactivate_rx_pools(struct ibmvnic_adapter *adapter)
 327{
 328        int i;
 329
 330        for (i = 0; i < adapter->num_active_rx_pools; i++)
 331                adapter->rx_pool[i].active = 0;
 332}
 333
 334static void replenish_rx_pool(struct ibmvnic_adapter *adapter,
 335                              struct ibmvnic_rx_pool *pool)
 336{
 337        int count = pool->size - atomic_read(&pool->available);
 338        u64 handle = adapter->rx_scrq[pool->index]->handle;
 339        struct device *dev = &adapter->vdev->dev;
 340        struct ibmvnic_ind_xmit_queue *ind_bufp;
 341        struct ibmvnic_sub_crq_queue *rx_scrq;
 342        union sub_crq *sub_crq;
 343        int buffers_added = 0;
 344        unsigned long lpar_rc;
 345        struct sk_buff *skb;
 346        unsigned int offset;
 347        dma_addr_t dma_addr;
 348        unsigned char *dst;
 349        int shift = 0;
 350        int index;
 351        int i;
 352
 353        if (!pool->active)
 354                return;
 355
 356        rx_scrq = adapter->rx_scrq[pool->index];
 357        ind_bufp = &rx_scrq->ind_buf;
 358
 359        /* netdev_skb_alloc() could have failed after we saved a few skbs
 360         * in the indir_buf and we would not have sent them to VIOS yet.
 361         * To account for them, start the loop at ind_bufp->index rather
 362         * than 0. If we pushed all the skbs to VIOS, ind_bufp->index will
 363         * be 0.
 364         */
 365        for (i = ind_bufp->index; i < count; ++i) {
 366                skb = netdev_alloc_skb(adapter->netdev, pool->buff_size);
 367                if (!skb) {
 368                        dev_err(dev, "Couldn't replenish rx buff\n");
 369                        adapter->replenish_no_mem++;
 370                        break;
 371                }
 372
 373                index = pool->free_map[pool->next_free];
 374
 375                if (pool->rx_buff[index].skb)
 376                        dev_err(dev, "Inconsistent free_map!\n");
 377
 378                /* Copy the skb to the long term mapped DMA buffer */
 379                offset = index * pool->buff_size;
 380                dst = pool->long_term_buff.buff + offset;
 381                memset(dst, 0, pool->buff_size);
 382                dma_addr = pool->long_term_buff.addr + offset;
 383                pool->rx_buff[index].data = dst;
 384
 385                pool->free_map[pool->next_free] = IBMVNIC_INVALID_MAP;
 386                pool->rx_buff[index].dma = dma_addr;
 387                pool->rx_buff[index].skb = skb;
 388                pool->rx_buff[index].pool_index = pool->index;
 389                pool->rx_buff[index].size = pool->buff_size;
 390
 391                sub_crq = &ind_bufp->indir_arr[ind_bufp->index++];
 392                memset(sub_crq, 0, sizeof(*sub_crq));
 393                sub_crq->rx_add.first = IBMVNIC_CRQ_CMD;
 394                sub_crq->rx_add.correlator =
 395                    cpu_to_be64((u64)&pool->rx_buff[index]);
 396                sub_crq->rx_add.ioba = cpu_to_be32(dma_addr);
 397                sub_crq->rx_add.map_id = pool->long_term_buff.map_id;
 398
 399                /* The length field of the sCRQ is defined to be 24 bits so the
 400                 * buffer size needs to be left shifted by a byte before it is
 401                 * converted to big endian to prevent the last byte from being
 402                 * truncated.
 403                 */
 404#ifdef __LITTLE_ENDIAN__
 405                shift = 8;
 406#endif
 407                sub_crq->rx_add.len = cpu_to_be32(pool->buff_size << shift);
 408                pool->next_free = (pool->next_free + 1) % pool->size;
 409                if (ind_bufp->index == IBMVNIC_MAX_IND_DESCS ||
 410                    i == count - 1) {
 411                        lpar_rc =
 412                                send_subcrq_indirect(adapter, handle,
 413                                                     (u64)ind_bufp->indir_dma,
 414                                                     (u64)ind_bufp->index);
 415                        if (lpar_rc != H_SUCCESS)
 416                                goto failure;
 417                        buffers_added += ind_bufp->index;
 418                        adapter->replenish_add_buff_success += ind_bufp->index;
 419                        ind_bufp->index = 0;
 420                }
 421        }
 422        atomic_add(buffers_added, &pool->available);
 423        return;
 424
 425failure:
 426        if (lpar_rc != H_PARAMETER && lpar_rc != H_CLOSED)
 427                dev_err_ratelimited(dev, "rx: replenish packet buffer failed\n");
 428        for (i = ind_bufp->index - 1; i >= 0; --i) {
 429                struct ibmvnic_rx_buff *rx_buff;
 430
 431                pool->next_free = pool->next_free == 0 ?
 432                                  pool->size - 1 : pool->next_free - 1;
 433                sub_crq = &ind_bufp->indir_arr[i];
 434                rx_buff = (struct ibmvnic_rx_buff *)
 435                                be64_to_cpu(sub_crq->rx_add.correlator);
 436                index = (int)(rx_buff - pool->rx_buff);
 437                pool->free_map[pool->next_free] = index;
 438                dev_kfree_skb_any(pool->rx_buff[index].skb);
 439                pool->rx_buff[index].skb = NULL;
 440        }
 441        adapter->replenish_add_buff_failure += ind_bufp->index;
 442        atomic_add(buffers_added, &pool->available);
 443        ind_bufp->index = 0;
 444        if (lpar_rc == H_CLOSED || adapter->failover_pending) {
 445                /* Disable buffer pool replenishment and report carrier off if
 446                 * queue is closed or pending failover.
 447                 * Firmware guarantees that a signal will be sent to the
 448                 * driver, triggering a reset.
 449                 */
 450                deactivate_rx_pools(adapter);
 451                netif_carrier_off(adapter->netdev);
 452        }
 453}
 454
 455static void replenish_pools(struct ibmvnic_adapter *adapter)
 456{
 457        int i;
 458
 459        adapter->replenish_task_cycles++;
 460        for (i = 0; i < adapter->num_active_rx_pools; i++) {
 461                if (adapter->rx_pool[i].active)
 462                        replenish_rx_pool(adapter, &adapter->rx_pool[i]);
 463        }
 464
 465        netdev_dbg(adapter->netdev, "Replenished %d pools\n", i);
 466}
 467
 468static void release_stats_buffers(struct ibmvnic_adapter *adapter)
 469{
 470        kfree(adapter->tx_stats_buffers);
 471        kfree(adapter->rx_stats_buffers);
 472        adapter->tx_stats_buffers = NULL;
 473        adapter->rx_stats_buffers = NULL;
 474}
 475
 476static int init_stats_buffers(struct ibmvnic_adapter *adapter)
 477{
 478        adapter->tx_stats_buffers =
 479                                kcalloc(IBMVNIC_MAX_QUEUES,
 480                                        sizeof(struct ibmvnic_tx_queue_stats),
 481                                        GFP_KERNEL);
 482        if (!adapter->tx_stats_buffers)
 483                return -ENOMEM;
 484
 485        adapter->rx_stats_buffers =
 486                                kcalloc(IBMVNIC_MAX_QUEUES,
 487                                        sizeof(struct ibmvnic_rx_queue_stats),
 488                                        GFP_KERNEL);
 489        if (!adapter->rx_stats_buffers)
 490                return -ENOMEM;
 491
 492        return 0;
 493}
 494
 495static void release_stats_token(struct ibmvnic_adapter *adapter)
 496{
 497        struct device *dev = &adapter->vdev->dev;
 498
 499        if (!adapter->stats_token)
 500                return;
 501
 502        dma_unmap_single(dev, adapter->stats_token,
 503                         sizeof(struct ibmvnic_statistics),
 504                         DMA_FROM_DEVICE);
 505        adapter->stats_token = 0;
 506}
 507
 508static int init_stats_token(struct ibmvnic_adapter *adapter)
 509{
 510        struct device *dev = &adapter->vdev->dev;
 511        dma_addr_t stok;
 512
 513        stok = dma_map_single(dev, &adapter->stats,
 514                              sizeof(struct ibmvnic_statistics),
 515                              DMA_FROM_DEVICE);
 516        if (dma_mapping_error(dev, stok)) {
 517                dev_err(dev, "Couldn't map stats buffer\n");
 518                return -1;
 519        }
 520
 521        adapter->stats_token = stok;
 522        netdev_dbg(adapter->netdev, "Stats token initialized (%llx)\n", stok);
 523        return 0;
 524}
 525
 526static int reset_rx_pools(struct ibmvnic_adapter *adapter)
 527{
 528        struct ibmvnic_rx_pool *rx_pool;
 529        u64 buff_size;
 530        int rx_scrqs;
 531        int i, j, rc;
 532
 533        if (!adapter->rx_pool)
 534                return -1;
 535
 536        buff_size = adapter->cur_rx_buf_sz;
 537        rx_scrqs = adapter->num_active_rx_pools;
 538        for (i = 0; i < rx_scrqs; i++) {
 539                rx_pool = &adapter->rx_pool[i];
 540
 541                netdev_dbg(adapter->netdev, "Re-setting rx_pool[%d]\n", i);
 542
 543                if (rx_pool->buff_size != buff_size) {
 544                        free_long_term_buff(adapter, &rx_pool->long_term_buff);
 545                        rx_pool->buff_size = ALIGN(buff_size, L1_CACHE_BYTES);
 546                        rc = alloc_long_term_buff(adapter,
 547                                                  &rx_pool->long_term_buff,
 548                                                  rx_pool->size *
 549                                                  rx_pool->buff_size);
 550                } else {
 551                        rc = reset_long_term_buff(adapter,
 552                                                  &rx_pool->long_term_buff);
 553                }
 554
 555                if (rc)
 556                        return rc;
 557
 558                for (j = 0; j < rx_pool->size; j++)
 559                        rx_pool->free_map[j] = j;
 560
 561                memset(rx_pool->rx_buff, 0,
 562                       rx_pool->size * sizeof(struct ibmvnic_rx_buff));
 563
 564                atomic_set(&rx_pool->available, 0);
 565                rx_pool->next_alloc = 0;
 566                rx_pool->next_free = 0;
 567                rx_pool->active = 1;
 568        }
 569
 570        return 0;
 571}
 572
 573static void release_rx_pools(struct ibmvnic_adapter *adapter)
 574{
 575        struct ibmvnic_rx_pool *rx_pool;
 576        int i, j;
 577
 578        if (!adapter->rx_pool)
 579                return;
 580
 581        for (i = 0; i < adapter->num_active_rx_pools; i++) {
 582                rx_pool = &adapter->rx_pool[i];
 583
 584                netdev_dbg(adapter->netdev, "Releasing rx_pool[%d]\n", i);
 585
 586                kfree(rx_pool->free_map);
 587                free_long_term_buff(adapter, &rx_pool->long_term_buff);
 588
 589                if (!rx_pool->rx_buff)
 590                        continue;
 591
 592                for (j = 0; j < rx_pool->size; j++) {
 593                        if (rx_pool->rx_buff[j].skb) {
 594                                dev_kfree_skb_any(rx_pool->rx_buff[j].skb);
 595                                rx_pool->rx_buff[j].skb = NULL;
 596                        }
 597                }
 598
 599                kfree(rx_pool->rx_buff);
 600        }
 601
 602        kfree(adapter->rx_pool);
 603        adapter->rx_pool = NULL;
 604        adapter->num_active_rx_pools = 0;
 605}
 606
 607static int init_rx_pools(struct net_device *netdev)
 608{
 609        struct ibmvnic_adapter *adapter = netdev_priv(netdev);
 610        struct device *dev = &adapter->vdev->dev;
 611        struct ibmvnic_rx_pool *rx_pool;
 612        int rxadd_subcrqs;
 613        u64 buff_size;
 614        int i, j;
 615
 616        rxadd_subcrqs = adapter->num_active_rx_scrqs;
 617        buff_size = adapter->cur_rx_buf_sz;
 618
 619        adapter->rx_pool = kcalloc(rxadd_subcrqs,
 620                                   sizeof(struct ibmvnic_rx_pool),
 621                                   GFP_KERNEL);
 622        if (!adapter->rx_pool) {
 623                dev_err(dev, "Failed to allocate rx pools\n");
 624                return -1;
 625        }
 626
 627        adapter->num_active_rx_pools = rxadd_subcrqs;
 628
 629        for (i = 0; i < rxadd_subcrqs; i++) {
 630                rx_pool = &adapter->rx_pool[i];
 631
 632                netdev_dbg(adapter->netdev,
 633                           "Initializing rx_pool[%d], %lld buffs, %lld bytes each\n",
 634                           i, adapter->req_rx_add_entries_per_subcrq,
 635                           buff_size);
 636
 637                rx_pool->size = adapter->req_rx_add_entries_per_subcrq;
 638                rx_pool->index = i;
 639                rx_pool->buff_size = ALIGN(buff_size, L1_CACHE_BYTES);
 640                rx_pool->active = 1;
 641
 642                rx_pool->free_map = kcalloc(rx_pool->size, sizeof(int),
 643                                            GFP_KERNEL);
 644                if (!rx_pool->free_map) {
 645                        release_rx_pools(adapter);
 646                        return -1;
 647                }
 648
 649                rx_pool->rx_buff = kcalloc(rx_pool->size,
 650                                           sizeof(struct ibmvnic_rx_buff),
 651                                           GFP_KERNEL);
 652                if (!rx_pool->rx_buff) {
 653                        dev_err(dev, "Couldn't alloc rx buffers\n");
 654                        release_rx_pools(adapter);
 655                        return -1;
 656                }
 657
 658                if (alloc_long_term_buff(adapter, &rx_pool->long_term_buff,
 659                                         rx_pool->size * rx_pool->buff_size)) {
 660                        release_rx_pools(adapter);
 661                        return -1;
 662                }
 663
 664                for (j = 0; j < rx_pool->size; ++j)
 665                        rx_pool->free_map[j] = j;
 666
 667                atomic_set(&rx_pool->available, 0);
 668                rx_pool->next_alloc = 0;
 669                rx_pool->next_free = 0;
 670        }
 671
 672        return 0;
 673}
 674
 675static int reset_one_tx_pool(struct ibmvnic_adapter *adapter,
 676                             struct ibmvnic_tx_pool *tx_pool)
 677{
 678        int rc, i;
 679
 680        rc = reset_long_term_buff(adapter, &tx_pool->long_term_buff);
 681        if (rc)
 682                return rc;
 683
 684        memset(tx_pool->tx_buff, 0,
 685               tx_pool->num_buffers *
 686               sizeof(struct ibmvnic_tx_buff));
 687
 688        for (i = 0; i < tx_pool->num_buffers; i++)
 689                tx_pool->free_map[i] = i;
 690
 691        tx_pool->consumer_index = 0;
 692        tx_pool->producer_index = 0;
 693
 694        return 0;
 695}
 696
 697static int reset_tx_pools(struct ibmvnic_adapter *adapter)
 698{
 699        int tx_scrqs;
 700        int i, rc;
 701
 702        if (!adapter->tx_pool)
 703                return -1;
 704
 705        tx_scrqs = adapter->num_active_tx_pools;
 706        for (i = 0; i < tx_scrqs; i++) {
 707                ibmvnic_tx_scrq_clean_buffer(adapter, adapter->tx_scrq[i]);
 708                rc = reset_one_tx_pool(adapter, &adapter->tso_pool[i]);
 709                if (rc)
 710                        return rc;
 711                rc = reset_one_tx_pool(adapter, &adapter->tx_pool[i]);
 712                if (rc)
 713                        return rc;
 714        }
 715
 716        return 0;
 717}
 718
 719static void release_vpd_data(struct ibmvnic_adapter *adapter)
 720{
 721        if (!adapter->vpd)
 722                return;
 723
 724        kfree(adapter->vpd->buff);
 725        kfree(adapter->vpd);
 726
 727        adapter->vpd = NULL;
 728}
 729
 730static void release_one_tx_pool(struct ibmvnic_adapter *adapter,
 731                                struct ibmvnic_tx_pool *tx_pool)
 732{
 733        kfree(tx_pool->tx_buff);
 734        kfree(tx_pool->free_map);
 735        free_long_term_buff(adapter, &tx_pool->long_term_buff);
 736}
 737
 738static void release_tx_pools(struct ibmvnic_adapter *adapter)
 739{
 740        int i;
 741
 742        if (!adapter->tx_pool)
 743                return;
 744
 745        for (i = 0; i < adapter->num_active_tx_pools; i++) {
 746                release_one_tx_pool(adapter, &adapter->tx_pool[i]);
 747                release_one_tx_pool(adapter, &adapter->tso_pool[i]);
 748        }
 749
 750        kfree(adapter->tx_pool);
 751        adapter->tx_pool = NULL;
 752        kfree(adapter->tso_pool);
 753        adapter->tso_pool = NULL;
 754        adapter->num_active_tx_pools = 0;
 755}
 756
 757static int init_one_tx_pool(struct net_device *netdev,
 758                            struct ibmvnic_tx_pool *tx_pool,
 759                            int num_entries, int buf_size)
 760{
 761        struct ibmvnic_adapter *adapter = netdev_priv(netdev);
 762        int i;
 763
 764        tx_pool->tx_buff = kcalloc(num_entries,
 765                                   sizeof(struct ibmvnic_tx_buff),
 766                                   GFP_KERNEL);
 767        if (!tx_pool->tx_buff)
 768                return -1;
 769
 770        if (alloc_long_term_buff(adapter, &tx_pool->long_term_buff,
 771                                 num_entries * buf_size))
 772                return -1;
 773
 774        tx_pool->free_map = kcalloc(num_entries, sizeof(int), GFP_KERNEL);
 775        if (!tx_pool->free_map)
 776                return -1;
 777
 778        for (i = 0; i < num_entries; i++)
 779                tx_pool->free_map[i] = i;
 780
 781        tx_pool->consumer_index = 0;
 782        tx_pool->producer_index = 0;
 783        tx_pool->num_buffers = num_entries;
 784        tx_pool->buf_size = buf_size;
 785
 786        return 0;
 787}
 788
 789static int init_tx_pools(struct net_device *netdev)
 790{
 791        struct ibmvnic_adapter *adapter = netdev_priv(netdev);
 792        int tx_subcrqs;
 793        u64 buff_size;
 794        int i, rc;
 795
 796        tx_subcrqs = adapter->num_active_tx_scrqs;
 797        adapter->tx_pool = kcalloc(tx_subcrqs,
 798                                   sizeof(struct ibmvnic_tx_pool), GFP_KERNEL);
 799        if (!adapter->tx_pool)
 800                return -1;
 801
 802        adapter->tso_pool = kcalloc(tx_subcrqs,
 803                                    sizeof(struct ibmvnic_tx_pool), GFP_KERNEL);
 804        if (!adapter->tso_pool) {
 805                kfree(adapter->tx_pool);
 806                adapter->tx_pool = NULL;
 807                return -1;
 808        }
 809
 810        adapter->num_active_tx_pools = tx_subcrqs;
 811
 812        for (i = 0; i < tx_subcrqs; i++) {
 813                buff_size = adapter->req_mtu + VLAN_HLEN;
 814                buff_size = ALIGN(buff_size, L1_CACHE_BYTES);
 815                rc = init_one_tx_pool(netdev, &adapter->tx_pool[i],
 816                                      adapter->req_tx_entries_per_subcrq,
 817                                      buff_size);
 818                if (rc) {
 819                        release_tx_pools(adapter);
 820                        return rc;
 821                }
 822
 823                rc = init_one_tx_pool(netdev, &adapter->tso_pool[i],
 824                                      IBMVNIC_TSO_BUFS,
 825                                      IBMVNIC_TSO_BUF_SZ);
 826                if (rc) {
 827                        release_tx_pools(adapter);
 828                        return rc;
 829                }
 830        }
 831
 832        return 0;
 833}
 834
 835static void ibmvnic_napi_enable(struct ibmvnic_adapter *adapter)
 836{
 837        int i;
 838
 839        if (adapter->napi_enabled)
 840                return;
 841
 842        for (i = 0; i < adapter->req_rx_queues; i++)
 843                napi_enable(&adapter->napi[i]);
 844
 845        adapter->napi_enabled = true;
 846}
 847
 848static void ibmvnic_napi_disable(struct ibmvnic_adapter *adapter)
 849{
 850        int i;
 851
 852        if (!adapter->napi_enabled)
 853                return;
 854
 855        for (i = 0; i < adapter->req_rx_queues; i++) {
 856                netdev_dbg(adapter->netdev, "Disabling napi[%d]\n", i);
 857                napi_disable(&adapter->napi[i]);
 858        }
 859
 860        adapter->napi_enabled = false;
 861}
 862
 863static int init_napi(struct ibmvnic_adapter *adapter)
 864{
 865        int i;
 866
 867        adapter->napi = kcalloc(adapter->req_rx_queues,
 868                                sizeof(struct napi_struct), GFP_KERNEL);
 869        if (!adapter->napi)
 870                return -ENOMEM;
 871
 872        for (i = 0; i < adapter->req_rx_queues; i++) {
 873                netdev_dbg(adapter->netdev, "Adding napi[%d]\n", i);
 874                netif_napi_add(adapter->netdev, &adapter->napi[i],
 875                               ibmvnic_poll, NAPI_POLL_WEIGHT);
 876        }
 877
 878        adapter->num_active_rx_napi = adapter->req_rx_queues;
 879        return 0;
 880}
 881
 882static void release_napi(struct ibmvnic_adapter *adapter)
 883{
 884        int i;
 885
 886        if (!adapter->napi)
 887                return;
 888
 889        for (i = 0; i < adapter->num_active_rx_napi; i++) {
 890                netdev_dbg(adapter->netdev, "Releasing napi[%d]\n", i);
 891                netif_napi_del(&adapter->napi[i]);
 892        }
 893
 894        kfree(adapter->napi);
 895        adapter->napi = NULL;
 896        adapter->num_active_rx_napi = 0;
 897        adapter->napi_enabled = false;
 898}
 899
 900static const char *adapter_state_to_string(enum vnic_state state)
 901{
 902        switch (state) {
 903        case VNIC_PROBING:
 904                return "PROBING";
 905        case VNIC_PROBED:
 906                return "PROBED";
 907        case VNIC_OPENING:
 908                return "OPENING";
 909        case VNIC_OPEN:
 910                return "OPEN";
 911        case VNIC_CLOSING:
 912                return "CLOSING";
 913        case VNIC_CLOSED:
 914                return "CLOSED";
 915        case VNIC_REMOVING:
 916                return "REMOVING";
 917        case VNIC_REMOVED:
 918                return "REMOVED";
 919        case VNIC_DOWN:
 920                return "DOWN";
 921        }
 922        return "UNKNOWN";
 923}
 924
 925static int ibmvnic_login(struct net_device *netdev)
 926{
 927        struct ibmvnic_adapter *adapter = netdev_priv(netdev);
 928        unsigned long timeout = msecs_to_jiffies(20000);
 929        int retry_count = 0;
 930        int retries = 10;
 931        bool retry;
 932        int rc;
 933
 934        do {
 935                retry = false;
 936                if (retry_count > retries) {
 937                        netdev_warn(netdev, "Login attempts exceeded\n");
 938                        return -1;
 939                }
 940
 941                adapter->init_done_rc = 0;
 942                reinit_completion(&adapter->init_done);
 943                rc = send_login(adapter);
 944                if (rc)
 945                        return rc;
 946
 947                if (!wait_for_completion_timeout(&adapter->init_done,
 948                                                 timeout)) {
 949                        netdev_warn(netdev, "Login timed out, retrying...\n");
 950                        retry = true;
 951                        adapter->init_done_rc = 0;
 952                        retry_count++;
 953                        continue;
 954                }
 955
 956                if (adapter->init_done_rc == ABORTED) {
 957                        netdev_warn(netdev, "Login aborted, retrying...\n");
 958                        retry = true;
 959                        adapter->init_done_rc = 0;
 960                        retry_count++;
 961                        /* FW or device may be busy, so
 962                         * wait a bit before retrying login
 963                         */
 964                        msleep(500);
 965                } else if (adapter->init_done_rc == PARTIALSUCCESS) {
 966                        retry_count++;
 967                        release_sub_crqs(adapter, 1);
 968
 969                        retry = true;
 970                        netdev_dbg(netdev,
 971                                   "Received partial success, retrying...\n");
 972                        adapter->init_done_rc = 0;
 973                        reinit_completion(&adapter->init_done);
 974                        send_query_cap(adapter);
 975                        if (!wait_for_completion_timeout(&adapter->init_done,
 976                                                         timeout)) {
 977                                netdev_warn(netdev,
 978                                            "Capabilities query timed out\n");
 979                                return -1;
 980                        }
 981
 982                        rc = init_sub_crqs(adapter);
 983                        if (rc) {
 984                                netdev_warn(netdev,
 985                                            "SCRQ initialization failed\n");
 986                                return -1;
 987                        }
 988
 989                        rc = init_sub_crq_irqs(adapter);
 990                        if (rc) {
 991                                netdev_warn(netdev,
 992                                            "SCRQ irq initialization failed\n");
 993                                return -1;
 994                        }
 995                } else if (adapter->init_done_rc) {
 996                        netdev_warn(netdev, "Adapter login failed\n");
 997                        return -1;
 998                }
 999        } while (retry);
1000
1001        __ibmvnic_set_mac(netdev, adapter->mac_addr);
1002
1003        netdev_dbg(netdev, "[S:%s] Login succeeded\n", adapter_state_to_string(adapter->state));
1004        return 0;
1005}
1006
1007static void release_login_buffer(struct ibmvnic_adapter *adapter)
1008{
1009        kfree(adapter->login_buf);
1010        adapter->login_buf = NULL;
1011}
1012
1013static void release_login_rsp_buffer(struct ibmvnic_adapter *adapter)
1014{
1015        kfree(adapter->login_rsp_buf);
1016        adapter->login_rsp_buf = NULL;
1017}
1018
1019static void release_resources(struct ibmvnic_adapter *adapter)
1020{
1021        release_vpd_data(adapter);
1022
1023        release_tx_pools(adapter);
1024        release_rx_pools(adapter);
1025
1026        release_napi(adapter);
1027        release_login_buffer(adapter);
1028        release_login_rsp_buffer(adapter);
1029}
1030
1031static int set_link_state(struct ibmvnic_adapter *adapter, u8 link_state)
1032{
1033        struct net_device *netdev = adapter->netdev;
1034        unsigned long timeout = msecs_to_jiffies(20000);
1035        union ibmvnic_crq crq;
1036        bool resend;
1037        int rc;
1038
1039        netdev_dbg(netdev, "setting link state %d\n", link_state);
1040
1041        memset(&crq, 0, sizeof(crq));
1042        crq.logical_link_state.first = IBMVNIC_CRQ_CMD;
1043        crq.logical_link_state.cmd = LOGICAL_LINK_STATE;
1044        crq.logical_link_state.link_state = link_state;
1045
1046        do {
1047                resend = false;
1048
1049                reinit_completion(&adapter->init_done);
1050                rc = ibmvnic_send_crq(adapter, &crq);
1051                if (rc) {
1052                        netdev_err(netdev, "Failed to set link state\n");
1053                        return rc;
1054                }
1055
1056                if (!wait_for_completion_timeout(&adapter->init_done,
1057                                                 timeout)) {
1058                        netdev_err(netdev, "timeout setting link state\n");
1059                        return -1;
1060                }
1061
1062                if (adapter->init_done_rc == PARTIALSUCCESS) {
1063                        /* Partuial success, delay and re-send */
1064                        mdelay(1000);
1065                        resend = true;
1066                } else if (adapter->init_done_rc) {
1067                        netdev_warn(netdev, "Unable to set link state, rc=%d\n",
1068                                    adapter->init_done_rc);
1069                        return adapter->init_done_rc;
1070                }
1071        } while (resend);
1072
1073        return 0;
1074}
1075
1076static int set_real_num_queues(struct net_device *netdev)
1077{
1078        struct ibmvnic_adapter *adapter = netdev_priv(netdev);
1079        int rc;
1080
1081        netdev_dbg(netdev, "Setting real tx/rx queues (%llx/%llx)\n",
1082                   adapter->req_tx_queues, adapter->req_rx_queues);
1083
1084        rc = netif_set_real_num_tx_queues(netdev, adapter->req_tx_queues);
1085        if (rc) {
1086                netdev_err(netdev, "failed to set the number of tx queues\n");
1087                return rc;
1088        }
1089
1090        rc = netif_set_real_num_rx_queues(netdev, adapter->req_rx_queues);
1091        if (rc)
1092                netdev_err(netdev, "failed to set the number of rx queues\n");
1093
1094        return rc;
1095}
1096
1097static int ibmvnic_get_vpd(struct ibmvnic_adapter *adapter)
1098{
1099        struct device *dev = &adapter->vdev->dev;
1100        union ibmvnic_crq crq;
1101        int len = 0;
1102        int rc;
1103
1104        if (adapter->vpd->buff)
1105                len = adapter->vpd->len;
1106
1107        mutex_lock(&adapter->fw_lock);
1108        adapter->fw_done_rc = 0;
1109        reinit_completion(&adapter->fw_done);
1110
1111        crq.get_vpd_size.first = IBMVNIC_CRQ_CMD;
1112        crq.get_vpd_size.cmd = GET_VPD_SIZE;
1113        rc = ibmvnic_send_crq(adapter, &crq);
1114        if (rc) {
1115                mutex_unlock(&adapter->fw_lock);
1116                return rc;
1117        }
1118
1119        rc = ibmvnic_wait_for_completion(adapter, &adapter->fw_done, 10000);
1120        if (rc) {
1121                dev_err(dev, "Could not retrieve VPD size, rc = %d\n", rc);
1122                mutex_unlock(&adapter->fw_lock);
1123                return rc;
1124        }
1125        mutex_unlock(&adapter->fw_lock);
1126
1127        if (!adapter->vpd->len)
1128                return -ENODATA;
1129
1130        if (!adapter->vpd->buff)
1131                adapter->vpd->buff = kzalloc(adapter->vpd->len, GFP_KERNEL);
1132        else if (adapter->vpd->len != len)
1133                adapter->vpd->buff =
1134                        krealloc(adapter->vpd->buff,
1135                                 adapter->vpd->len, GFP_KERNEL);
1136
1137        if (!adapter->vpd->buff) {
1138                dev_err(dev, "Could allocate VPD buffer\n");
1139                return -ENOMEM;
1140        }
1141
1142        adapter->vpd->dma_addr =
1143                dma_map_single(dev, adapter->vpd->buff, adapter->vpd->len,
1144                               DMA_FROM_DEVICE);
1145        if (dma_mapping_error(dev, adapter->vpd->dma_addr)) {
1146                dev_err(dev, "Could not map VPD buffer\n");
1147                kfree(adapter->vpd->buff);
1148                adapter->vpd->buff = NULL;
1149                return -ENOMEM;
1150        }
1151
1152        mutex_lock(&adapter->fw_lock);
1153        adapter->fw_done_rc = 0;
1154        reinit_completion(&adapter->fw_done);
1155
1156        crq.get_vpd.first = IBMVNIC_CRQ_CMD;
1157        crq.get_vpd.cmd = GET_VPD;
1158        crq.get_vpd.ioba = cpu_to_be32(adapter->vpd->dma_addr);
1159        crq.get_vpd.len = cpu_to_be32((u32)adapter->vpd->len);
1160        rc = ibmvnic_send_crq(adapter, &crq);
1161        if (rc) {
1162                kfree(adapter->vpd->buff);
1163                adapter->vpd->buff = NULL;
1164                mutex_unlock(&adapter->fw_lock);
1165                return rc;
1166        }
1167
1168        rc = ibmvnic_wait_for_completion(adapter, &adapter->fw_done, 10000);
1169        if (rc) {
1170                dev_err(dev, "Unable to retrieve VPD, rc = %d\n", rc);
1171                kfree(adapter->vpd->buff);
1172                adapter->vpd->buff = NULL;
1173                mutex_unlock(&adapter->fw_lock);
1174                return rc;
1175        }
1176
1177        mutex_unlock(&adapter->fw_lock);
1178        return 0;
1179}
1180
1181static int init_resources(struct ibmvnic_adapter *adapter)
1182{
1183        struct net_device *netdev = adapter->netdev;
1184        int rc;
1185
1186        rc = set_real_num_queues(netdev);
1187        if (rc)
1188                return rc;
1189
1190        adapter->vpd = kzalloc(sizeof(*adapter->vpd), GFP_KERNEL);
1191        if (!adapter->vpd)
1192                return -ENOMEM;
1193
1194        /* Vital Product Data (VPD) */
1195        rc = ibmvnic_get_vpd(adapter);
1196        if (rc) {
1197                netdev_err(netdev, "failed to initialize Vital Product Data (VPD)\n");
1198                return rc;
1199        }
1200
1201        adapter->map_id = 1;
1202
1203        rc = init_napi(adapter);
1204        if (rc)
1205                return rc;
1206
1207        send_query_map(adapter);
1208
1209        rc = init_rx_pools(netdev);
1210        if (rc)
1211                return rc;
1212
1213        rc = init_tx_pools(netdev);
1214        return rc;
1215}
1216
1217static int __ibmvnic_open(struct net_device *netdev)
1218{
1219        struct ibmvnic_adapter *adapter = netdev_priv(netdev);
1220        enum vnic_state prev_state = adapter->state;
1221        int i, rc;
1222
1223        adapter->state = VNIC_OPENING;
1224        replenish_pools(adapter);
1225        ibmvnic_napi_enable(adapter);
1226
1227        /* We're ready to receive frames, enable the sub-crq interrupts and
1228         * set the logical link state to up
1229         */
1230        for (i = 0; i < adapter->req_rx_queues; i++) {
1231                netdev_dbg(netdev, "Enabling rx_scrq[%d] irq\n", i);
1232                if (prev_state == VNIC_CLOSED)
1233                        enable_irq(adapter->rx_scrq[i]->irq);
1234                enable_scrq_irq(adapter, adapter->rx_scrq[i]);
1235        }
1236
1237        for (i = 0; i < adapter->req_tx_queues; i++) {
1238                netdev_dbg(netdev, "Enabling tx_scrq[%d] irq\n", i);
1239                if (prev_state == VNIC_CLOSED)
1240                        enable_irq(adapter->tx_scrq[i]->irq);
1241                enable_scrq_irq(adapter, adapter->tx_scrq[i]);
1242                netdev_tx_reset_queue(netdev_get_tx_queue(netdev, i));
1243        }
1244
1245        rc = set_link_state(adapter, IBMVNIC_LOGICAL_LNK_UP);
1246        if (rc) {
1247                ibmvnic_napi_disable(adapter);
1248                release_resources(adapter);
1249                return rc;
1250        }
1251
1252        netif_tx_start_all_queues(netdev);
1253
1254        if (prev_state == VNIC_CLOSED) {
1255                for (i = 0; i < adapter->req_rx_queues; i++)
1256                        napi_schedule(&adapter->napi[i]);
1257        }
1258
1259        adapter->state = VNIC_OPEN;
1260        return rc;
1261}
1262
1263static int ibmvnic_open(struct net_device *netdev)
1264{
1265        struct ibmvnic_adapter *adapter = netdev_priv(netdev);
1266        int rc;
1267
1268        ASSERT_RTNL();
1269
1270        /* If device failover is pending or we are about to reset, just set
1271         * device state and return. Device operation will be handled by reset
1272         * routine.
1273         *
1274         * It should be safe to overwrite the adapter->state here. Since
1275         * we hold the rtnl, either the reset has not actually started or
1276         * the rtnl got dropped during the set_link_state() in do_reset().
1277         * In the former case, no one else is changing the state (again we
1278         * have the rtnl) and in the latter case, do_reset() will detect and
1279         * honor our setting below.
1280         */
1281        if (adapter->failover_pending || (test_bit(0, &adapter->resetting))) {
1282                netdev_dbg(netdev, "[S:%s FOP:%d] Resetting, deferring open\n",
1283                           adapter_state_to_string(adapter->state),
1284                           adapter->failover_pending);
1285                adapter->state = VNIC_OPEN;
1286                rc = 0;
1287                goto out;
1288        }
1289
1290        if (adapter->state != VNIC_CLOSED) {
1291                rc = ibmvnic_login(netdev);
1292                if (rc)
1293                        goto out;
1294
1295                rc = init_resources(adapter);
1296                if (rc) {
1297                        netdev_err(netdev, "failed to initialize resources\n");
1298                        release_resources(adapter);
1299                        goto out;
1300                }
1301        }
1302
1303        rc = __ibmvnic_open(netdev);
1304
1305out:
1306        /* If open failed and there is a pending failover or in-progress reset,
1307         * set device state and return. Device operation will be handled by
1308         * reset routine. See also comments above regarding rtnl.
1309         */
1310        if (rc &&
1311            (adapter->failover_pending || (test_bit(0, &adapter->resetting)))) {
1312                adapter->state = VNIC_OPEN;
1313                rc = 0;
1314        }
1315        return rc;
1316}
1317
1318static void clean_rx_pools(struct ibmvnic_adapter *adapter)
1319{
1320        struct ibmvnic_rx_pool *rx_pool;
1321        struct ibmvnic_rx_buff *rx_buff;
1322        u64 rx_entries;
1323        int rx_scrqs;
1324        int i, j;
1325
1326        if (!adapter->rx_pool)
1327                return;
1328
1329        rx_scrqs = adapter->num_active_rx_pools;
1330        rx_entries = adapter->req_rx_add_entries_per_subcrq;
1331
1332        /* Free any remaining skbs in the rx buffer pools */
1333        for (i = 0; i < rx_scrqs; i++) {
1334                rx_pool = &adapter->rx_pool[i];
1335                if (!rx_pool || !rx_pool->rx_buff)
1336                        continue;
1337
1338                netdev_dbg(adapter->netdev, "Cleaning rx_pool[%d]\n", i);
1339                for (j = 0; j < rx_entries; j++) {
1340                        rx_buff = &rx_pool->rx_buff[j];
1341                        if (rx_buff && rx_buff->skb) {
1342                                dev_kfree_skb_any(rx_buff->skb);
1343                                rx_buff->skb = NULL;
1344                        }
1345                }
1346        }
1347}
1348
1349static void clean_one_tx_pool(struct ibmvnic_adapter *adapter,
1350                              struct ibmvnic_tx_pool *tx_pool)
1351{
1352        struct ibmvnic_tx_buff *tx_buff;
1353        u64 tx_entries;
1354        int i;
1355
1356        if (!tx_pool || !tx_pool->tx_buff)
1357                return;
1358
1359        tx_entries = tx_pool->num_buffers;
1360
1361        for (i = 0; i < tx_entries; i++) {
1362                tx_buff = &tx_pool->tx_buff[i];
1363                if (tx_buff && tx_buff->skb) {
1364                        dev_kfree_skb_any(tx_buff->skb);
1365                        tx_buff->skb = NULL;
1366                }
1367        }
1368}
1369
1370static void clean_tx_pools(struct ibmvnic_adapter *adapter)
1371{
1372        int tx_scrqs;
1373        int i;
1374
1375        if (!adapter->tx_pool || !adapter->tso_pool)
1376                return;
1377
1378        tx_scrqs = adapter->num_active_tx_pools;
1379
1380        /* Free any remaining skbs in the tx buffer pools */
1381        for (i = 0; i < tx_scrqs; i++) {
1382                netdev_dbg(adapter->netdev, "Cleaning tx_pool[%d]\n", i);
1383                clean_one_tx_pool(adapter, &adapter->tx_pool[i]);
1384                clean_one_tx_pool(adapter, &adapter->tso_pool[i]);
1385        }
1386}
1387
1388static void ibmvnic_disable_irqs(struct ibmvnic_adapter *adapter)
1389{
1390        struct net_device *netdev = adapter->netdev;
1391        int i;
1392
1393        if (adapter->tx_scrq) {
1394                for (i = 0; i < adapter->req_tx_queues; i++)
1395                        if (adapter->tx_scrq[i]->irq) {
1396                                netdev_dbg(netdev,
1397                                           "Disabling tx_scrq[%d] irq\n", i);
1398                                disable_scrq_irq(adapter, adapter->tx_scrq[i]);
1399                                disable_irq(adapter->tx_scrq[i]->irq);
1400                        }
1401        }
1402
1403        if (adapter->rx_scrq) {
1404                for (i = 0; i < adapter->req_rx_queues; i++) {
1405                        if (adapter->rx_scrq[i]->irq) {
1406                                netdev_dbg(netdev,
1407                                           "Disabling rx_scrq[%d] irq\n", i);
1408                                disable_scrq_irq(adapter, adapter->rx_scrq[i]);
1409                                disable_irq(adapter->rx_scrq[i]->irq);
1410                        }
1411                }
1412        }
1413}
1414
1415static void ibmvnic_cleanup(struct net_device *netdev)
1416{
1417        struct ibmvnic_adapter *adapter = netdev_priv(netdev);
1418
1419        /* ensure that transmissions are stopped if called by do_reset */
1420        if (test_bit(0, &adapter->resetting))
1421                netif_tx_disable(netdev);
1422        else
1423                netif_tx_stop_all_queues(netdev);
1424
1425        ibmvnic_napi_disable(adapter);
1426        ibmvnic_disable_irqs(adapter);
1427
1428        clean_rx_pools(adapter);
1429        clean_tx_pools(adapter);
1430}
1431
1432static int __ibmvnic_close(struct net_device *netdev)
1433{
1434        struct ibmvnic_adapter *adapter = netdev_priv(netdev);
1435        int rc = 0;
1436
1437        adapter->state = VNIC_CLOSING;
1438        rc = set_link_state(adapter, IBMVNIC_LOGICAL_LNK_DN);
1439        adapter->state = VNIC_CLOSED;
1440        return rc;
1441}
1442
1443static int ibmvnic_close(struct net_device *netdev)
1444{
1445        struct ibmvnic_adapter *adapter = netdev_priv(netdev);
1446        int rc;
1447
1448        netdev_dbg(netdev, "[S:%s FOP:%d FRR:%d] Closing\n",
1449                   adapter_state_to_string(adapter->state),
1450                   adapter->failover_pending,
1451                   adapter->force_reset_recovery);
1452
1453        /* If device failover is pending, just set device state and return.
1454         * Device operation will be handled by reset routine.
1455         */
1456        if (adapter->failover_pending) {
1457                adapter->state = VNIC_CLOSED;
1458                return 0;
1459        }
1460
1461        rc = __ibmvnic_close(netdev);
1462        ibmvnic_cleanup(netdev);
1463
1464        return rc;
1465}
1466
1467/**
1468 * build_hdr_data - creates L2/L3/L4 header data buffer
1469 * @hdr_field: bitfield determining needed headers
1470 * @skb: socket buffer
1471 * @hdr_len: array of header lengths
1472 * @hdr_data: buffer to write the header to
1473 *
1474 * Reads hdr_field to determine which headers are needed by firmware.
1475 * Builds a buffer containing these headers.  Saves individual header
1476 * lengths and total buffer length to be used to build descriptors.
1477 */
1478static int build_hdr_data(u8 hdr_field, struct sk_buff *skb,
1479                          int *hdr_len, u8 *hdr_data)
1480{
1481        int len = 0;
1482        u8 *hdr;
1483
1484        if (skb_vlan_tagged(skb) && !skb_vlan_tag_present(skb))
1485                hdr_len[0] = sizeof(struct vlan_ethhdr);
1486        else
1487                hdr_len[0] = sizeof(struct ethhdr);
1488
1489        if (skb->protocol == htons(ETH_P_IP)) {
1490                hdr_len[1] = ip_hdr(skb)->ihl * 4;
1491                if (ip_hdr(skb)->protocol == IPPROTO_TCP)
1492                        hdr_len[2] = tcp_hdrlen(skb);
1493                else if (ip_hdr(skb)->protocol == IPPROTO_UDP)
1494                        hdr_len[2] = sizeof(struct udphdr);
1495        } else if (skb->protocol == htons(ETH_P_IPV6)) {
1496                hdr_len[1] = sizeof(struct ipv6hdr);
1497                if (ipv6_hdr(skb)->nexthdr == IPPROTO_TCP)
1498                        hdr_len[2] = tcp_hdrlen(skb);
1499                else if (ipv6_hdr(skb)->nexthdr == IPPROTO_UDP)
1500                        hdr_len[2] = sizeof(struct udphdr);
1501        } else if (skb->protocol == htons(ETH_P_ARP)) {
1502                hdr_len[1] = arp_hdr_len(skb->dev);
1503                hdr_len[2] = 0;
1504        }
1505
1506        memset(hdr_data, 0, 120);
1507        if ((hdr_field >> 6) & 1) {
1508                hdr = skb_mac_header(skb);
1509                memcpy(hdr_data, hdr, hdr_len[0]);
1510                len += hdr_len[0];
1511        }
1512
1513        if ((hdr_field >> 5) & 1) {
1514                hdr = skb_network_header(skb);
1515                memcpy(hdr_data + len, hdr, hdr_len[1]);
1516                len += hdr_len[1];
1517        }
1518
1519        if ((hdr_field >> 4) & 1) {
1520                hdr = skb_transport_header(skb);
1521                memcpy(hdr_data + len, hdr, hdr_len[2]);
1522                len += hdr_len[2];
1523        }
1524        return len;
1525}
1526
1527/**
1528 * create_hdr_descs - create header and header extension descriptors
1529 * @hdr_field: bitfield determining needed headers
1530 * @hdr_data: buffer containing header data
1531 * @len: length of data buffer
1532 * @hdr_len: array of individual header lengths
1533 * @scrq_arr: descriptor array
1534 *
1535 * Creates header and, if needed, header extension descriptors and
1536 * places them in a descriptor array, scrq_arr
1537 */
1538
1539static int create_hdr_descs(u8 hdr_field, u8 *hdr_data, int len, int *hdr_len,
1540                            union sub_crq *scrq_arr)
1541{
1542        union sub_crq hdr_desc;
1543        int tmp_len = len;
1544        int num_descs = 0;
1545        u8 *data, *cur;
1546        int tmp;
1547
1548        while (tmp_len > 0) {
1549                cur = hdr_data + len - tmp_len;
1550
1551                memset(&hdr_desc, 0, sizeof(hdr_desc));
1552                if (cur != hdr_data) {
1553                        data = hdr_desc.hdr_ext.data;
1554                        tmp = tmp_len > 29 ? 29 : tmp_len;
1555                        hdr_desc.hdr_ext.first = IBMVNIC_CRQ_CMD;
1556                        hdr_desc.hdr_ext.type = IBMVNIC_HDR_EXT_DESC;
1557                        hdr_desc.hdr_ext.len = tmp;
1558                } else {
1559                        data = hdr_desc.hdr.data;
1560                        tmp = tmp_len > 24 ? 24 : tmp_len;
1561                        hdr_desc.hdr.first = IBMVNIC_CRQ_CMD;
1562                        hdr_desc.hdr.type = IBMVNIC_HDR_DESC;
1563                        hdr_desc.hdr.len = tmp;
1564                        hdr_desc.hdr.l2_len = (u8)hdr_len[0];
1565                        hdr_desc.hdr.l3_len = cpu_to_be16((u16)hdr_len[1]);
1566                        hdr_desc.hdr.l4_len = (u8)hdr_len[2];
1567                        hdr_desc.hdr.flag = hdr_field << 1;
1568                }
1569                memcpy(data, cur, tmp);
1570                tmp_len -= tmp;
1571                *scrq_arr = hdr_desc;
1572                scrq_arr++;
1573                num_descs++;
1574        }
1575
1576        return num_descs;
1577}
1578
1579/**
1580 * build_hdr_descs_arr - build a header descriptor array
1581 * @skb: tx socket buffer
1582 * @indir_arr: indirect array
1583 * @num_entries: number of descriptors to be sent
1584 * @hdr_field: bit field determining which headers will be sent
1585 *
1586 * This function will build a TX descriptor array with applicable
1587 * L2/L3/L4 packet header descriptors to be sent by send_subcrq_indirect.
1588 */
1589
1590static void build_hdr_descs_arr(struct sk_buff *skb,
1591                                union sub_crq *indir_arr,
1592                                int *num_entries, u8 hdr_field)
1593{
1594        int hdr_len[3] = {0, 0, 0};
1595        u8 hdr_data[140] = {0};
1596        int tot_len;
1597
1598        tot_len = build_hdr_data(hdr_field, skb, hdr_len,
1599                                 hdr_data);
1600        *num_entries += create_hdr_descs(hdr_field, hdr_data, tot_len, hdr_len,
1601                                         indir_arr + 1);
1602}
1603
1604static int ibmvnic_xmit_workarounds(struct sk_buff *skb,
1605                                    struct net_device *netdev)
1606{
1607        /* For some backing devices, mishandling of small packets
1608         * can result in a loss of connection or TX stall. Device
1609         * architects recommend that no packet should be smaller
1610         * than the minimum MTU value provided to the driver, so
1611         * pad any packets to that length
1612         */
1613        if (skb->len < netdev->min_mtu)
1614                return skb_put_padto(skb, netdev->min_mtu);
1615
1616        return 0;
1617}
1618
1619static void ibmvnic_tx_scrq_clean_buffer(struct ibmvnic_adapter *adapter,
1620                                         struct ibmvnic_sub_crq_queue *tx_scrq)
1621{
1622        struct ibmvnic_ind_xmit_queue *ind_bufp;
1623        struct ibmvnic_tx_buff *tx_buff;
1624        struct ibmvnic_tx_pool *tx_pool;
1625        union sub_crq tx_scrq_entry;
1626        int queue_num;
1627        int entries;
1628        int index;
1629        int i;
1630
1631        ind_bufp = &tx_scrq->ind_buf;
1632        entries = (u64)ind_bufp->index;
1633        queue_num = tx_scrq->pool_index;
1634
1635        for (i = entries - 1; i >= 0; --i) {
1636                tx_scrq_entry = ind_bufp->indir_arr[i];
1637                if (tx_scrq_entry.v1.type != IBMVNIC_TX_DESC)
1638                        continue;
1639                index = be32_to_cpu(tx_scrq_entry.v1.correlator);
1640                if (index & IBMVNIC_TSO_POOL_MASK) {
1641                        tx_pool = &adapter->tso_pool[queue_num];
1642                        index &= ~IBMVNIC_TSO_POOL_MASK;
1643                } else {
1644                        tx_pool = &adapter->tx_pool[queue_num];
1645                }
1646                tx_pool->free_map[tx_pool->consumer_index] = index;
1647                tx_pool->consumer_index = tx_pool->consumer_index == 0 ?
1648                                          tx_pool->num_buffers - 1 :
1649                                          tx_pool->consumer_index - 1;
1650                tx_buff = &tx_pool->tx_buff[index];
1651                adapter->netdev->stats.tx_packets--;
1652                adapter->netdev->stats.tx_bytes -= tx_buff->skb->len;
1653                adapter->tx_stats_buffers[queue_num].packets--;
1654                adapter->tx_stats_buffers[queue_num].bytes -=
1655                                                tx_buff->skb->len;
1656                dev_kfree_skb_any(tx_buff->skb);
1657                tx_buff->skb = NULL;
1658                adapter->netdev->stats.tx_dropped++;
1659        }
1660        ind_bufp->index = 0;
1661        if (atomic_sub_return(entries, &tx_scrq->used) <=
1662            (adapter->req_tx_entries_per_subcrq / 2) &&
1663            __netif_subqueue_stopped(adapter->netdev, queue_num) &&
1664            !test_bit(0, &adapter->resetting)) {
1665                netif_wake_subqueue(adapter->netdev, queue_num);
1666                netdev_dbg(adapter->netdev, "Started queue %d\n",
1667                           queue_num);
1668        }
1669}
1670
1671static int ibmvnic_tx_scrq_flush(struct ibmvnic_adapter *adapter,
1672                                 struct ibmvnic_sub_crq_queue *tx_scrq)
1673{
1674        struct ibmvnic_ind_xmit_queue *ind_bufp;
1675        u64 dma_addr;
1676        u64 entries;
1677        u64 handle;
1678        int rc;
1679
1680        ind_bufp = &tx_scrq->ind_buf;
1681        dma_addr = (u64)ind_bufp->indir_dma;
1682        entries = (u64)ind_bufp->index;
1683        handle = tx_scrq->handle;
1684
1685        if (!entries)
1686                return 0;
1687        rc = send_subcrq_indirect(adapter, handle, dma_addr, entries);
1688        if (rc)
1689                ibmvnic_tx_scrq_clean_buffer(adapter, tx_scrq);
1690        else
1691                ind_bufp->index = 0;
1692        return 0;
1693}
1694
1695static netdev_tx_t ibmvnic_xmit(struct sk_buff *skb, struct net_device *netdev)
1696{
1697        struct ibmvnic_adapter *adapter = netdev_priv(netdev);
1698        int queue_num = skb_get_queue_mapping(skb);
1699        u8 *hdrs = (u8 *)&adapter->tx_rx_desc_req;
1700        struct device *dev = &adapter->vdev->dev;
1701        struct ibmvnic_ind_xmit_queue *ind_bufp;
1702        struct ibmvnic_tx_buff *tx_buff = NULL;
1703        struct ibmvnic_sub_crq_queue *tx_scrq;
1704        struct ibmvnic_tx_pool *tx_pool;
1705        unsigned int tx_send_failed = 0;
1706        netdev_tx_t ret = NETDEV_TX_OK;
1707        unsigned int tx_map_failed = 0;
1708        union sub_crq indir_arr[16];
1709        unsigned int tx_dropped = 0;
1710        unsigned int tx_packets = 0;
1711        unsigned int tx_bytes = 0;
1712        dma_addr_t data_dma_addr;
1713        struct netdev_queue *txq;
1714        unsigned long lpar_rc;
1715        union sub_crq tx_crq;
1716        unsigned int offset;
1717        int num_entries = 1;
1718        unsigned char *dst;
1719        int index = 0;
1720        u8 proto = 0;
1721
1722        tx_scrq = adapter->tx_scrq[queue_num];
1723        txq = netdev_get_tx_queue(netdev, queue_num);
1724        ind_bufp = &tx_scrq->ind_buf;
1725
1726        if (test_bit(0, &adapter->resetting)) {
1727                if (!netif_subqueue_stopped(netdev, skb))
1728                        netif_stop_subqueue(netdev, queue_num);
1729                dev_kfree_skb_any(skb);
1730
1731                tx_send_failed++;
1732                tx_dropped++;
1733                ret = NETDEV_TX_OK;
1734                goto out;
1735        }
1736
1737        if (ibmvnic_xmit_workarounds(skb, netdev)) {
1738                tx_dropped++;
1739                tx_send_failed++;
1740                ret = NETDEV_TX_OK;
1741                ibmvnic_tx_scrq_flush(adapter, tx_scrq);
1742                goto out;
1743        }
1744        if (skb_is_gso(skb))
1745                tx_pool = &adapter->tso_pool[queue_num];
1746        else
1747                tx_pool = &adapter->tx_pool[queue_num];
1748
1749        index = tx_pool->free_map[tx_pool->consumer_index];
1750
1751        if (index == IBMVNIC_INVALID_MAP) {
1752                dev_kfree_skb_any(skb);
1753                tx_send_failed++;
1754                tx_dropped++;
1755                ibmvnic_tx_scrq_flush(adapter, tx_scrq);
1756                ret = NETDEV_TX_OK;
1757                goto out;
1758        }
1759
1760        tx_pool->free_map[tx_pool->consumer_index] = IBMVNIC_INVALID_MAP;
1761
1762        offset = index * tx_pool->buf_size;
1763        dst = tx_pool->long_term_buff.buff + offset;
1764        memset(dst, 0, tx_pool->buf_size);
1765        data_dma_addr = tx_pool->long_term_buff.addr + offset;
1766
1767        if (skb_shinfo(skb)->nr_frags) {
1768                int cur, i;
1769
1770                /* Copy the head */
1771                skb_copy_from_linear_data(skb, dst, skb_headlen(skb));
1772                cur = skb_headlen(skb);
1773
1774                /* Copy the frags */
1775                for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
1776                        const skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
1777
1778                        memcpy(dst + cur, skb_frag_address(frag),
1779                               skb_frag_size(frag));
1780                        cur += skb_frag_size(frag);
1781                }
1782        } else {
1783                skb_copy_from_linear_data(skb, dst, skb->len);
1784        }
1785
1786        /* post changes to long_term_buff *dst before VIOS accessing it */
1787        dma_wmb();
1788
1789        tx_pool->consumer_index =
1790            (tx_pool->consumer_index + 1) % tx_pool->num_buffers;
1791
1792        tx_buff = &tx_pool->tx_buff[index];
1793        tx_buff->skb = skb;
1794        tx_buff->index = index;
1795        tx_buff->pool_index = queue_num;
1796
1797        memset(&tx_crq, 0, sizeof(tx_crq));
1798        tx_crq.v1.first = IBMVNIC_CRQ_CMD;
1799        tx_crq.v1.type = IBMVNIC_TX_DESC;
1800        tx_crq.v1.n_crq_elem = 1;
1801        tx_crq.v1.n_sge = 1;
1802        tx_crq.v1.flags1 = IBMVNIC_TX_COMP_NEEDED;
1803
1804        if (skb_is_gso(skb))
1805                tx_crq.v1.correlator =
1806                        cpu_to_be32(index | IBMVNIC_TSO_POOL_MASK);
1807        else
1808                tx_crq.v1.correlator = cpu_to_be32(index);
1809        tx_crq.v1.dma_reg = cpu_to_be16(tx_pool->long_term_buff.map_id);
1810        tx_crq.v1.sge_len = cpu_to_be32(skb->len);
1811        tx_crq.v1.ioba = cpu_to_be64(data_dma_addr);
1812
1813        if (adapter->vlan_header_insertion && skb_vlan_tag_present(skb)) {
1814                tx_crq.v1.flags2 |= IBMVNIC_TX_VLAN_INSERT;
1815                tx_crq.v1.vlan_id = cpu_to_be16(skb->vlan_tci);
1816        }
1817
1818        if (skb->protocol == htons(ETH_P_IP)) {
1819                tx_crq.v1.flags1 |= IBMVNIC_TX_PROT_IPV4;
1820                proto = ip_hdr(skb)->protocol;
1821        } else if (skb->protocol == htons(ETH_P_IPV6)) {
1822                tx_crq.v1.flags1 |= IBMVNIC_TX_PROT_IPV6;
1823                proto = ipv6_hdr(skb)->nexthdr;
1824        }
1825
1826        if (proto == IPPROTO_TCP)
1827                tx_crq.v1.flags1 |= IBMVNIC_TX_PROT_TCP;
1828        else if (proto == IPPROTO_UDP)
1829                tx_crq.v1.flags1 |= IBMVNIC_TX_PROT_UDP;
1830
1831        if (skb->ip_summed == CHECKSUM_PARTIAL) {
1832                tx_crq.v1.flags1 |= IBMVNIC_TX_CHKSUM_OFFLOAD;
1833                hdrs += 2;
1834        }
1835        if (skb_is_gso(skb)) {
1836                tx_crq.v1.flags1 |= IBMVNIC_TX_LSO;
1837                tx_crq.v1.mss = cpu_to_be16(skb_shinfo(skb)->gso_size);
1838                hdrs += 2;
1839        }
1840
1841        if ((*hdrs >> 7) & 1)
1842                build_hdr_descs_arr(skb, indir_arr, &num_entries, *hdrs);
1843
1844        tx_crq.v1.n_crq_elem = num_entries;
1845        tx_buff->num_entries = num_entries;
1846        /* flush buffer if current entry can not fit */
1847        if (num_entries + ind_bufp->index > IBMVNIC_MAX_IND_DESCS) {
1848                lpar_rc = ibmvnic_tx_scrq_flush(adapter, tx_scrq);
1849                if (lpar_rc != H_SUCCESS)
1850                        goto tx_flush_err;
1851        }
1852
1853        indir_arr[0] = tx_crq;
1854        memcpy(&ind_bufp->indir_arr[ind_bufp->index], &indir_arr[0],
1855               num_entries * sizeof(struct ibmvnic_generic_scrq));
1856        ind_bufp->index += num_entries;
1857        if (__netdev_tx_sent_queue(txq, skb->len,
1858                                   netdev_xmit_more() &&
1859                                   ind_bufp->index < IBMVNIC_MAX_IND_DESCS)) {
1860                lpar_rc = ibmvnic_tx_scrq_flush(adapter, tx_scrq);
1861                if (lpar_rc != H_SUCCESS)
1862                        goto tx_err;
1863        }
1864
1865        if (atomic_add_return(num_entries, &tx_scrq->used)
1866                                        >= adapter->req_tx_entries_per_subcrq) {
1867                netdev_dbg(netdev, "Stopping queue %d\n", queue_num);
1868                netif_stop_subqueue(netdev, queue_num);
1869        }
1870
1871        tx_packets++;
1872        tx_bytes += skb->len;
1873        txq->trans_start = jiffies;
1874        ret = NETDEV_TX_OK;
1875        goto out;
1876
1877tx_flush_err:
1878        dev_kfree_skb_any(skb);
1879        tx_buff->skb = NULL;
1880        tx_pool->consumer_index = tx_pool->consumer_index == 0 ?
1881                                  tx_pool->num_buffers - 1 :
1882                                  tx_pool->consumer_index - 1;
1883        tx_dropped++;
1884tx_err:
1885        if (lpar_rc != H_CLOSED && lpar_rc != H_PARAMETER)
1886                dev_err_ratelimited(dev, "tx: send failed\n");
1887
1888        if (lpar_rc == H_CLOSED || adapter->failover_pending) {
1889                /* Disable TX and report carrier off if queue is closed
1890                 * or pending failover.
1891                 * Firmware guarantees that a signal will be sent to the
1892                 * driver, triggering a reset or some other action.
1893                 */
1894                netif_tx_stop_all_queues(netdev);
1895                netif_carrier_off(netdev);
1896        }
1897out:
1898        netdev->stats.tx_dropped += tx_dropped;
1899        netdev->stats.tx_bytes += tx_bytes;
1900        netdev->stats.tx_packets += tx_packets;
1901        adapter->tx_send_failed += tx_send_failed;
1902        adapter->tx_map_failed += tx_map_failed;
1903        adapter->tx_stats_buffers[queue_num].packets += tx_packets;
1904        adapter->tx_stats_buffers[queue_num].bytes += tx_bytes;
1905        adapter->tx_stats_buffers[queue_num].dropped_packets += tx_dropped;
1906
1907        return ret;
1908}
1909
1910static void ibmvnic_set_multi(struct net_device *netdev)
1911{
1912        struct ibmvnic_adapter *adapter = netdev_priv(netdev);
1913        struct netdev_hw_addr *ha;
1914        union ibmvnic_crq crq;
1915
1916        memset(&crq, 0, sizeof(crq));
1917        crq.request_capability.first = IBMVNIC_CRQ_CMD;
1918        crq.request_capability.cmd = REQUEST_CAPABILITY;
1919
1920        if (netdev->flags & IFF_PROMISC) {
1921                if (!adapter->promisc_supported)
1922                        return;
1923        } else {
1924                if (netdev->flags & IFF_ALLMULTI) {
1925                        /* Accept all multicast */
1926                        memset(&crq, 0, sizeof(crq));
1927                        crq.multicast_ctrl.first = IBMVNIC_CRQ_CMD;
1928                        crq.multicast_ctrl.cmd = MULTICAST_CTRL;
1929                        crq.multicast_ctrl.flags = IBMVNIC_ENABLE_ALL;
1930                        ibmvnic_send_crq(adapter, &crq);
1931                } else if (netdev_mc_empty(netdev)) {
1932                        /* Reject all multicast */
1933                        memset(&crq, 0, sizeof(crq));
1934                        crq.multicast_ctrl.first = IBMVNIC_CRQ_CMD;
1935                        crq.multicast_ctrl.cmd = MULTICAST_CTRL;
1936                        crq.multicast_ctrl.flags = IBMVNIC_DISABLE_ALL;
1937                        ibmvnic_send_crq(adapter, &crq);
1938                } else {
1939                        /* Accept one or more multicast(s) */
1940                        netdev_for_each_mc_addr(ha, netdev) {
1941                                memset(&crq, 0, sizeof(crq));
1942                                crq.multicast_ctrl.first = IBMVNIC_CRQ_CMD;
1943                                crq.multicast_ctrl.cmd = MULTICAST_CTRL;
1944                                crq.multicast_ctrl.flags = IBMVNIC_ENABLE_MC;
1945                                ether_addr_copy(&crq.multicast_ctrl.mac_addr[0],
1946                                                ha->addr);
1947                                ibmvnic_send_crq(adapter, &crq);
1948                        }
1949                }
1950        }
1951}
1952
1953static int __ibmvnic_set_mac(struct net_device *netdev, u8 *dev_addr)
1954{
1955        struct ibmvnic_adapter *adapter = netdev_priv(netdev);
1956        union ibmvnic_crq crq;
1957        int rc;
1958
1959        if (!is_valid_ether_addr(dev_addr)) {
1960                rc = -EADDRNOTAVAIL;
1961                goto err;
1962        }
1963
1964        memset(&crq, 0, sizeof(crq));
1965        crq.change_mac_addr.first = IBMVNIC_CRQ_CMD;
1966        crq.change_mac_addr.cmd = CHANGE_MAC_ADDR;
1967        ether_addr_copy(&crq.change_mac_addr.mac_addr[0], dev_addr);
1968
1969        mutex_lock(&adapter->fw_lock);
1970        adapter->fw_done_rc = 0;
1971        reinit_completion(&adapter->fw_done);
1972
1973        rc = ibmvnic_send_crq(adapter, &crq);
1974        if (rc) {
1975                rc = -EIO;
1976                mutex_unlock(&adapter->fw_lock);
1977                goto err;
1978        }
1979
1980        rc = ibmvnic_wait_for_completion(adapter, &adapter->fw_done, 10000);
1981        /* netdev->dev_addr is changed in handle_change_mac_rsp function */
1982        if (rc || adapter->fw_done_rc) {
1983                rc = -EIO;
1984                mutex_unlock(&adapter->fw_lock);
1985                goto err;
1986        }
1987        mutex_unlock(&adapter->fw_lock);
1988        return 0;
1989err:
1990        ether_addr_copy(adapter->mac_addr, netdev->dev_addr);
1991        return rc;
1992}
1993
1994static int ibmvnic_set_mac(struct net_device *netdev, void *p)
1995{
1996        struct ibmvnic_adapter *adapter = netdev_priv(netdev);
1997        struct sockaddr *addr = p;
1998        int rc;
1999
2000        rc = 0;
2001        if (!is_valid_ether_addr(addr->sa_data))
2002                return -EADDRNOTAVAIL;
2003
2004        ether_addr_copy(adapter->mac_addr, addr->sa_data);
2005        if (adapter->state != VNIC_PROBED)
2006                rc = __ibmvnic_set_mac(netdev, addr->sa_data);
2007
2008        return rc;
2009}
2010
2011static const char *reset_reason_to_string(enum ibmvnic_reset_reason reason)
2012{
2013        switch (reason) {
2014        case VNIC_RESET_FAILOVER:
2015                return "FAILOVER";
2016        case VNIC_RESET_MOBILITY:
2017                return "MOBILITY";
2018        case VNIC_RESET_FATAL:
2019                return "FATAL";
2020        case VNIC_RESET_NON_FATAL:
2021                return "NON_FATAL";
2022        case VNIC_RESET_TIMEOUT:
2023                return "TIMEOUT";
2024        case VNIC_RESET_CHANGE_PARAM:
2025                return "CHANGE_PARAM";
2026        case VNIC_RESET_PASSIVE_INIT:
2027                return "PASSIVE_INIT";
2028        }
2029        return "UNKNOWN";
2030}
2031
2032/*
2033 * do_reset returns zero if we are able to keep processing reset events, or
2034 * non-zero if we hit a fatal error and must halt.
2035 */
2036static int do_reset(struct ibmvnic_adapter *adapter,
2037                    struct ibmvnic_rwi *rwi, u32 reset_state)
2038{
2039        u64 old_num_rx_queues, old_num_tx_queues;
2040        u64 old_num_rx_slots, old_num_tx_slots;
2041        struct net_device *netdev = adapter->netdev;
2042        int rc;
2043
2044        netdev_dbg(adapter->netdev,
2045                   "[S:%s FOP:%d] Reset reason: %s, reset_state: %s\n",
2046                   adapter_state_to_string(adapter->state),
2047                   adapter->failover_pending,
2048                   reset_reason_to_string(rwi->reset_reason),
2049                   adapter_state_to_string(reset_state));
2050
2051        adapter->reset_reason = rwi->reset_reason;
2052        /* requestor of VNIC_RESET_CHANGE_PARAM already has the rtnl lock */
2053        if (!(adapter->reset_reason == VNIC_RESET_CHANGE_PARAM))
2054                rtnl_lock();
2055
2056        /* Now that we have the rtnl lock, clear any pending failover.
2057         * This will ensure ibmvnic_open() has either completed or will
2058         * block until failover is complete.
2059         */
2060        if (rwi->reset_reason == VNIC_RESET_FAILOVER)
2061                adapter->failover_pending = false;
2062
2063        /* read the state and check (again) after getting rtnl */
2064        reset_state = adapter->state;
2065
2066        if (reset_state == VNIC_REMOVING || reset_state == VNIC_REMOVED) {
2067                rc = -EBUSY;
2068                goto out;
2069        }
2070
2071        netif_carrier_off(netdev);
2072
2073        old_num_rx_queues = adapter->req_rx_queues;
2074        old_num_tx_queues = adapter->req_tx_queues;
2075        old_num_rx_slots = adapter->req_rx_add_entries_per_subcrq;
2076        old_num_tx_slots = adapter->req_tx_entries_per_subcrq;
2077
2078        ibmvnic_cleanup(netdev);
2079
2080        if (reset_state == VNIC_OPEN &&
2081            adapter->reset_reason != VNIC_RESET_MOBILITY &&
2082            adapter->reset_reason != VNIC_RESET_FAILOVER) {
2083                if (adapter->reset_reason == VNIC_RESET_CHANGE_PARAM) {
2084                        rc = __ibmvnic_close(netdev);
2085                        if (rc)
2086                                goto out;
2087                } else {
2088                        adapter->state = VNIC_CLOSING;
2089
2090                        /* Release the RTNL lock before link state change and
2091                         * re-acquire after the link state change to allow
2092                         * linkwatch_event to grab the RTNL lock and run during
2093                         * a reset.
2094                         */
2095                        rtnl_unlock();
2096                        rc = set_link_state(adapter, IBMVNIC_LOGICAL_LNK_DN);
2097                        rtnl_lock();
2098                        if (rc)
2099                                goto out;
2100
2101                        if (adapter->state == VNIC_OPEN) {
2102                                /* When we dropped rtnl, ibmvnic_open() got
2103                                 * it and noticed that we are resetting and
2104                                 * set the adapter state to OPEN. Update our
2105                                 * new "target" state, and resume the reset
2106                                 * from VNIC_CLOSING state.
2107                                 */
2108                                netdev_dbg(netdev,
2109                                           "Open changed state from %s, updating.\n",
2110                                           adapter_state_to_string(reset_state));
2111                                reset_state = VNIC_OPEN;
2112                                adapter->state = VNIC_CLOSING;
2113                        }
2114
2115                        if (adapter->state != VNIC_CLOSING) {
2116                                /* If someone else changed the adapter state
2117                                 * when we dropped the rtnl, fail the reset
2118                                 */
2119                                rc = -1;
2120                                goto out;
2121                        }
2122                        adapter->state = VNIC_CLOSED;
2123                }
2124        }
2125
2126        if (adapter->reset_reason == VNIC_RESET_CHANGE_PARAM) {
2127                release_resources(adapter);
2128                release_sub_crqs(adapter, 1);
2129                release_crq_queue(adapter);
2130        }
2131
2132        if (adapter->reset_reason != VNIC_RESET_NON_FATAL) {
2133                /* remove the closed state so when we call open it appears
2134                 * we are coming from the probed state.
2135                 */
2136                adapter->state = VNIC_PROBED;
2137
2138                if (adapter->reset_reason == VNIC_RESET_CHANGE_PARAM) {
2139                        rc = init_crq_queue(adapter);
2140                } else if (adapter->reset_reason == VNIC_RESET_MOBILITY) {
2141                        rc = ibmvnic_reenable_crq_queue(adapter);
2142                        release_sub_crqs(adapter, 1);
2143                } else {
2144                        rc = ibmvnic_reset_crq(adapter);
2145                        if (rc == H_CLOSED || rc == H_SUCCESS) {
2146                                rc = vio_enable_interrupts(adapter->vdev);
2147                                if (rc)
2148                                        netdev_err(adapter->netdev,
2149                                                   "Reset failed to enable interrupts. rc=%d\n",
2150                                                   rc);
2151                        }
2152                }
2153
2154                if (rc) {
2155                        netdev_err(adapter->netdev,
2156                                   "Reset couldn't initialize crq. rc=%d\n", rc);
2157                        goto out;
2158                }
2159
2160                rc = ibmvnic_reset_init(adapter, true);
2161                if (rc) {
2162                        rc = IBMVNIC_INIT_FAILED;
2163                        goto out;
2164                }
2165
2166                /* If the adapter was in PROBE or DOWN state prior to the reset,
2167                 * exit here.
2168                 */
2169                if (reset_state == VNIC_PROBED || reset_state == VNIC_DOWN) {
2170                        rc = 0;
2171                        goto out;
2172                }
2173
2174                rc = ibmvnic_login(netdev);
2175                if (rc)
2176                        goto out;
2177
2178                if (adapter->reset_reason == VNIC_RESET_CHANGE_PARAM) {
2179                        rc = init_resources(adapter);
2180                        if (rc)
2181                                goto out;
2182                } else if (adapter->req_rx_queues != old_num_rx_queues ||
2183                    adapter->req_tx_queues != old_num_tx_queues ||
2184                    adapter->req_rx_add_entries_per_subcrq !=
2185                    old_num_rx_slots ||
2186                    adapter->req_tx_entries_per_subcrq !=
2187                    old_num_tx_slots ||
2188                    !adapter->rx_pool ||
2189                    !adapter->tso_pool ||
2190                    !adapter->tx_pool) {
2191                        release_rx_pools(adapter);
2192                        release_tx_pools(adapter);
2193                        release_napi(adapter);
2194                        release_vpd_data(adapter);
2195
2196                        rc = init_resources(adapter);
2197                        if (rc)
2198                                goto out;
2199
2200                } else {
2201                        rc = reset_tx_pools(adapter);
2202                        if (rc) {
2203                                netdev_dbg(adapter->netdev, "reset tx pools failed (%d)\n",
2204                                           rc);
2205                                goto out;
2206                        }
2207
2208                        rc = reset_rx_pools(adapter);
2209                        if (rc) {
2210                                netdev_dbg(adapter->netdev, "reset rx pools failed (%d)\n",
2211                                           rc);
2212                                goto out;
2213                        }
2214                }
2215                ibmvnic_disable_irqs(adapter);
2216        }
2217        adapter->state = VNIC_CLOSED;
2218
2219        if (reset_state == VNIC_CLOSED) {
2220                rc = 0;
2221                goto out;
2222        }
2223
2224        rc = __ibmvnic_open(netdev);
2225        if (rc) {
2226                rc = IBMVNIC_OPEN_FAILED;
2227                goto out;
2228        }
2229
2230        /* refresh device's multicast list */
2231        ibmvnic_set_multi(netdev);
2232
2233        if (adapter->reset_reason == VNIC_RESET_FAILOVER ||
2234            adapter->reset_reason == VNIC_RESET_MOBILITY)
2235                __netdev_notify_peers(netdev);
2236
2237        rc = 0;
2238
2239out:
2240        /* restore the adapter state if reset failed */
2241        if (rc)
2242                adapter->state = reset_state;
2243        /* requestor of VNIC_RESET_CHANGE_PARAM should still hold the rtnl lock */
2244        if (!(adapter->reset_reason == VNIC_RESET_CHANGE_PARAM))
2245                rtnl_unlock();
2246
2247        netdev_dbg(adapter->netdev, "[S:%s FOP:%d] Reset done, rc %d\n",
2248                   adapter_state_to_string(adapter->state),
2249                   adapter->failover_pending, rc);
2250        return rc;
2251}
2252
2253static int do_hard_reset(struct ibmvnic_adapter *adapter,
2254                         struct ibmvnic_rwi *rwi, u32 reset_state)
2255{
2256        struct net_device *netdev = adapter->netdev;
2257        int rc;
2258
2259        netdev_dbg(adapter->netdev, "Hard resetting driver (%s)\n",
2260                   reset_reason_to_string(rwi->reset_reason));
2261
2262        /* read the state and check (again) after getting rtnl */
2263        reset_state = adapter->state;
2264
2265        if (reset_state == VNIC_REMOVING || reset_state == VNIC_REMOVED) {
2266                rc = -EBUSY;
2267                goto out;
2268        }
2269
2270        netif_carrier_off(netdev);
2271        adapter->reset_reason = rwi->reset_reason;
2272
2273        ibmvnic_cleanup(netdev);
2274        release_resources(adapter);
2275        release_sub_crqs(adapter, 0);
2276        release_crq_queue(adapter);
2277
2278        /* remove the closed state so when we call open it appears
2279         * we are coming from the probed state.
2280         */
2281        adapter->state = VNIC_PROBED;
2282
2283        reinit_completion(&adapter->init_done);
2284        rc = init_crq_queue(adapter);
2285        if (rc) {
2286                netdev_err(adapter->netdev,
2287                           "Couldn't initialize crq. rc=%d\n", rc);
2288                goto out;
2289        }
2290
2291        rc = ibmvnic_reset_init(adapter, false);
2292        if (rc)
2293                goto out;
2294
2295        /* If the adapter was in PROBE or DOWN state prior to the reset,
2296         * exit here.
2297         */
2298        if (reset_state == VNIC_PROBED || reset_state == VNIC_DOWN)
2299                goto out;
2300
2301        rc = ibmvnic_login(netdev);
2302        if (rc)
2303                goto out;
2304
2305        rc = init_resources(adapter);
2306        if (rc)
2307                goto out;
2308
2309        ibmvnic_disable_irqs(adapter);
2310        adapter->state = VNIC_CLOSED;
2311
2312        if (reset_state == VNIC_CLOSED)
2313                goto out;
2314
2315        rc = __ibmvnic_open(netdev);
2316        if (rc) {
2317                rc = IBMVNIC_OPEN_FAILED;
2318                goto out;
2319        }
2320
2321        __netdev_notify_peers(netdev);
2322out:
2323        /* restore adapter state if reset failed */
2324        if (rc)
2325                adapter->state = reset_state;
2326        netdev_dbg(adapter->netdev, "[S:%s FOP:%d] Hard reset done, rc %d\n",
2327                   adapter_state_to_string(adapter->state),
2328                   adapter->failover_pending, rc);
2329        return rc;
2330}
2331
2332static struct ibmvnic_rwi *get_next_rwi(struct ibmvnic_adapter *adapter)
2333{
2334        struct ibmvnic_rwi *rwi;
2335        unsigned long flags;
2336
2337        spin_lock_irqsave(&adapter->rwi_lock, flags);
2338
2339        if (!list_empty(&adapter->rwi_list)) {
2340                rwi = list_first_entry(&adapter->rwi_list, struct ibmvnic_rwi,
2341                                       list);
2342                list_del(&rwi->list);
2343        } else {
2344                rwi = NULL;
2345        }
2346
2347        spin_unlock_irqrestore(&adapter->rwi_lock, flags);
2348        return rwi;
2349}
2350
2351/**
2352 * do_passive_init - complete probing when partner device is detected.
2353 * @adapter: ibmvnic_adapter struct
2354 *
2355 * If the ibmvnic device does not have a partner device to communicate with at boot
2356 * and that partner device comes online at a later time, this function is called
2357 * to complete the initialization process of ibmvnic device.
2358 * Caller is expected to hold rtnl_lock().
2359 *
2360 * Returns non-zero if sub-CRQs are not initialized properly leaving the device
2361 * in the down state.
2362 * Returns 0 upon success and the device is in PROBED state.
2363 */
2364
2365static int do_passive_init(struct ibmvnic_adapter *adapter)
2366{
2367        unsigned long timeout = msecs_to_jiffies(30000);
2368        struct net_device *netdev = adapter->netdev;
2369        struct device *dev = &adapter->vdev->dev;
2370        int rc;
2371
2372        netdev_dbg(netdev, "Partner device found, probing.\n");
2373
2374        adapter->state = VNIC_PROBING;
2375        reinit_completion(&adapter->init_done);
2376        adapter->init_done_rc = 0;
2377        adapter->crq.active = true;
2378
2379        rc = send_crq_init_complete(adapter);
2380        if (rc)
2381                goto out;
2382
2383        rc = send_version_xchg(adapter);
2384        if (rc)
2385                netdev_dbg(adapter->netdev, "send_version_xchg failed, rc=%d\n", rc);
2386
2387        if (!wait_for_completion_timeout(&adapter->init_done, timeout)) {
2388                dev_err(dev, "Initialization sequence timed out\n");
2389                rc = -ETIMEDOUT;
2390                goto out;
2391        }
2392
2393        rc = init_sub_crqs(adapter);
2394        if (rc) {
2395                dev_err(dev, "Initialization of sub crqs failed, rc=%d\n", rc);
2396                goto out;
2397        }
2398
2399        rc = init_sub_crq_irqs(adapter);
2400        if (rc) {
2401                dev_err(dev, "Failed to initialize sub crq irqs\n, rc=%d", rc);
2402                goto init_failed;
2403        }
2404
2405        netdev->mtu = adapter->req_mtu - ETH_HLEN;
2406        netdev->min_mtu = adapter->min_mtu - ETH_HLEN;
2407        netdev->max_mtu = adapter->max_mtu - ETH_HLEN;
2408
2409        adapter->state = VNIC_PROBED;
2410        netdev_dbg(netdev, "Probed successfully. Waiting for signal from partner device.\n");
2411
2412        return 0;
2413
2414init_failed:
2415        release_sub_crqs(adapter, 1);
2416out:
2417        adapter->state = VNIC_DOWN;
2418        return rc;
2419}
2420
2421static void __ibmvnic_reset(struct work_struct *work)
2422{
2423        struct ibmvnic_adapter *adapter;
2424        bool saved_state = false;
2425        struct ibmvnic_rwi *tmprwi;
2426        struct ibmvnic_rwi *rwi;
2427        unsigned long flags;
2428        u32 reset_state;
2429        int rc = 0;
2430
2431        adapter = container_of(work, struct ibmvnic_adapter, ibmvnic_reset);
2432
2433        if (test_and_set_bit_lock(0, &adapter->resetting)) {
2434                queue_delayed_work(system_long_wq,
2435                                   &adapter->ibmvnic_delayed_reset,
2436                                   IBMVNIC_RESET_DELAY);
2437                return;
2438        }
2439
2440        rwi = get_next_rwi(adapter);
2441        while (rwi) {
2442                spin_lock_irqsave(&adapter->state_lock, flags);
2443
2444                if (adapter->state == VNIC_REMOVING ||
2445                    adapter->state == VNIC_REMOVED) {
2446                        spin_unlock_irqrestore(&adapter->state_lock, flags);
2447                        kfree(rwi);
2448                        rc = EBUSY;
2449                        break;
2450                }
2451
2452                if (!saved_state) {
2453                        reset_state = adapter->state;
2454                        saved_state = true;
2455                }
2456                spin_unlock_irqrestore(&adapter->state_lock, flags);
2457
2458                if (rwi->reset_reason == VNIC_RESET_PASSIVE_INIT) {
2459                        rtnl_lock();
2460                        rc = do_passive_init(adapter);
2461                        rtnl_unlock();
2462                        if (!rc)
2463                                netif_carrier_on(adapter->netdev);
2464                } else if (adapter->force_reset_recovery) {
2465                        /* Since we are doing a hard reset now, clear the
2466                         * failover_pending flag so we don't ignore any
2467                         * future MOBILITY or other resets.
2468                         */
2469                        adapter->failover_pending = false;
2470
2471                        /* Transport event occurred during previous reset */
2472                        if (adapter->wait_for_reset) {
2473                                /* Previous was CHANGE_PARAM; caller locked */
2474                                adapter->force_reset_recovery = false;
2475                                rc = do_hard_reset(adapter, rwi, reset_state);
2476                        } else {
2477                                rtnl_lock();
2478                                adapter->force_reset_recovery = false;
2479                                rc = do_hard_reset(adapter, rwi, reset_state);
2480                                rtnl_unlock();
2481                        }
2482                        if (rc) {
2483                                /* give backing device time to settle down */
2484                                netdev_dbg(adapter->netdev,
2485                                           "[S:%s] Hard reset failed, waiting 60 secs\n",
2486                                           adapter_state_to_string(adapter->state));
2487                                set_current_state(TASK_UNINTERRUPTIBLE);
2488                                schedule_timeout(60 * HZ);
2489                        }
2490                } else {
2491                        rc = do_reset(adapter, rwi, reset_state);
2492                }
2493                tmprwi = rwi;
2494                adapter->last_reset_time = jiffies;
2495
2496                if (rc)
2497                        netdev_dbg(adapter->netdev, "Reset failed, rc=%d\n", rc);
2498
2499                rwi = get_next_rwi(adapter);
2500
2501                /*
2502                 * If there is another reset queued, free the previous rwi
2503                 * and process the new reset even if previous reset failed
2504                 * (the previous reset could have failed because of a fail
2505                 * over for instance, so process the fail over).
2506                 *
2507                 * If there are no resets queued and the previous reset failed,
2508                 * the adapter would be in an undefined state. So retry the
2509                 * previous reset as a hard reset.
2510                 */
2511                if (rwi)
2512                        kfree(tmprwi);
2513                else if (rc)
2514                        rwi = tmprwi;
2515
2516                if (rwi && (rwi->reset_reason == VNIC_RESET_FAILOVER ||
2517                            rwi->reset_reason == VNIC_RESET_MOBILITY || rc))
2518                        adapter->force_reset_recovery = true;
2519        }
2520
2521        if (adapter->wait_for_reset) {
2522                adapter->reset_done_rc = rc;
2523                complete(&adapter->reset_done);
2524        }
2525
2526        clear_bit_unlock(0, &adapter->resetting);
2527
2528        netdev_dbg(adapter->netdev,
2529                   "[S:%s FRR:%d WFR:%d] Done processing resets\n",
2530                   adapter_state_to_string(adapter->state),
2531                   adapter->force_reset_recovery,
2532                   adapter->wait_for_reset);
2533}
2534
2535static void __ibmvnic_delayed_reset(struct work_struct *work)
2536{
2537        struct ibmvnic_adapter *adapter;
2538
2539        adapter = container_of(work, struct ibmvnic_adapter,
2540                               ibmvnic_delayed_reset.work);
2541        __ibmvnic_reset(&adapter->ibmvnic_reset);
2542}
2543
2544static int ibmvnic_reset(struct ibmvnic_adapter *adapter,
2545                         enum ibmvnic_reset_reason reason)
2546{
2547        struct list_head *entry, *tmp_entry;
2548        struct ibmvnic_rwi *rwi, *tmp;
2549        struct net_device *netdev = adapter->netdev;
2550        unsigned long flags;
2551        int ret;
2552
2553        spin_lock_irqsave(&adapter->rwi_lock, flags);
2554
2555        /* If failover is pending don't schedule any other reset.
2556         * Instead let the failover complete. If there is already a
2557         * a failover reset scheduled, we will detect and drop the
2558         * duplicate reset when walking the ->rwi_list below.
2559         */
2560        if (adapter->state == VNIC_REMOVING ||
2561            adapter->state == VNIC_REMOVED ||
2562            (adapter->failover_pending && reason != VNIC_RESET_FAILOVER)) {
2563                ret = EBUSY;
2564                netdev_dbg(netdev, "Adapter removing or pending failover, skipping reset\n");
2565                goto err;
2566        }
2567
2568        if (adapter->state == VNIC_PROBING) {
2569                netdev_warn(netdev, "Adapter reset during probe\n");
2570                adapter->init_done_rc = EAGAIN;
2571                ret = EAGAIN;
2572                goto err;
2573        }
2574
2575        list_for_each_entry(tmp, &adapter->rwi_list, list) {
2576                if (tmp->reset_reason == reason) {
2577                        netdev_dbg(netdev, "Skipping matching reset, reason=%s\n",
2578                                   reset_reason_to_string(reason));
2579                        ret = EBUSY;
2580                        goto err;
2581                }
2582        }
2583
2584        rwi = kzalloc(sizeof(*rwi), GFP_ATOMIC);
2585        if (!rwi) {
2586                ret = ENOMEM;
2587                goto err;
2588        }
2589        /* if we just received a transport event,
2590         * flush reset queue and process this reset
2591         */
2592        if (adapter->force_reset_recovery && !list_empty(&adapter->rwi_list)) {
2593                list_for_each_safe(entry, tmp_entry, &adapter->rwi_list)
2594                        list_del(entry);
2595        }
2596        rwi->reset_reason = reason;
2597        list_add_tail(&rwi->list, &adapter->rwi_list);
2598        netdev_dbg(adapter->netdev, "Scheduling reset (reason %s)\n",
2599                   reset_reason_to_string(reason));
2600        queue_work(system_long_wq, &adapter->ibmvnic_reset);
2601
2602        ret = 0;
2603err:
2604        /* ibmvnic_close() below can block, so drop the lock first */
2605        spin_unlock_irqrestore(&adapter->rwi_lock, flags);
2606
2607        if (ret == ENOMEM)
2608                ibmvnic_close(netdev);
2609
2610        return -ret;
2611}
2612
2613static void ibmvnic_tx_timeout(struct net_device *dev, unsigned int txqueue)
2614{
2615        struct ibmvnic_adapter *adapter = netdev_priv(dev);
2616
2617        if (test_bit(0, &adapter->resetting)) {
2618                netdev_err(adapter->netdev,
2619                           "Adapter is resetting, skip timeout reset\n");
2620                return;
2621        }
2622        /* No queuing up reset until at least 5 seconds (default watchdog val)
2623         * after last reset
2624         */
2625        if (time_before(jiffies, (adapter->last_reset_time + dev->watchdog_timeo))) {
2626                netdev_dbg(dev, "Not yet time to tx timeout.\n");
2627                return;
2628        }
2629        ibmvnic_reset(adapter, VNIC_RESET_TIMEOUT);
2630}
2631
2632static void remove_buff_from_pool(struct ibmvnic_adapter *adapter,
2633                                  struct ibmvnic_rx_buff *rx_buff)
2634{
2635        struct ibmvnic_rx_pool *pool = &adapter->rx_pool[rx_buff->pool_index];
2636
2637        rx_buff->skb = NULL;
2638
2639        pool->free_map[pool->next_alloc] = (int)(rx_buff - pool->rx_buff);
2640        pool->next_alloc = (pool->next_alloc + 1) % pool->size;
2641
2642        atomic_dec(&pool->available);
2643}
2644
2645static int ibmvnic_poll(struct napi_struct *napi, int budget)
2646{
2647        struct ibmvnic_sub_crq_queue *rx_scrq;
2648        struct ibmvnic_adapter *adapter;
2649        struct net_device *netdev;
2650        int frames_processed;
2651        int scrq_num;
2652
2653        netdev = napi->dev;
2654        adapter = netdev_priv(netdev);
2655        scrq_num = (int)(napi - adapter->napi);
2656        frames_processed = 0;
2657        rx_scrq = adapter->rx_scrq[scrq_num];
2658
2659restart_poll:
2660        while (frames_processed < budget) {
2661                struct sk_buff *skb;
2662                struct ibmvnic_rx_buff *rx_buff;
2663                union sub_crq *next;
2664                u32 length;
2665                u16 offset;
2666                u8 flags = 0;
2667
2668                if (unlikely(test_bit(0, &adapter->resetting) &&
2669                             adapter->reset_reason != VNIC_RESET_NON_FATAL)) {
2670                        enable_scrq_irq(adapter, rx_scrq);
2671                        napi_complete_done(napi, frames_processed);
2672                        return frames_processed;
2673                }
2674
2675                if (!pending_scrq(adapter, rx_scrq))
2676                        break;
2677                next = ibmvnic_next_scrq(adapter, rx_scrq);
2678                rx_buff = (struct ibmvnic_rx_buff *)
2679                          be64_to_cpu(next->rx_comp.correlator);
2680                /* do error checking */
2681                if (next->rx_comp.rc) {
2682                        netdev_dbg(netdev, "rx buffer returned with rc %x\n",
2683                                   be16_to_cpu(next->rx_comp.rc));
2684                        /* free the entry */
2685                        next->rx_comp.first = 0;
2686                        dev_kfree_skb_any(rx_buff->skb);
2687                        remove_buff_from_pool(adapter, rx_buff);
2688                        continue;
2689                } else if (!rx_buff->skb) {
2690                        /* free the entry */
2691                        next->rx_comp.first = 0;
2692                        remove_buff_from_pool(adapter, rx_buff);
2693                        continue;
2694                }
2695
2696                length = be32_to_cpu(next->rx_comp.len);
2697                offset = be16_to_cpu(next->rx_comp.off_frame_data);
2698                flags = next->rx_comp.flags;
2699                skb = rx_buff->skb;
2700                /* load long_term_buff before copying to skb */
2701                dma_rmb();
2702                skb_copy_to_linear_data(skb, rx_buff->data + offset,
2703                                        length);
2704
2705                /* VLAN Header has been stripped by the system firmware and
2706                 * needs to be inserted by the driver
2707                 */
2708                if (adapter->rx_vlan_header_insertion &&
2709                    (flags & IBMVNIC_VLAN_STRIPPED))
2710                        __vlan_hwaccel_put_tag(skb, htons(ETH_P_8021Q),
2711                                               ntohs(next->rx_comp.vlan_tci));
2712
2713                /* free the entry */
2714                next->rx_comp.first = 0;
2715                remove_buff_from_pool(adapter, rx_buff);
2716
2717                skb_put(skb, length);
2718                skb->protocol = eth_type_trans(skb, netdev);
2719                skb_record_rx_queue(skb, scrq_num);
2720
2721                if (flags & IBMVNIC_IP_CHKSUM_GOOD &&
2722                    flags & IBMVNIC_TCP_UDP_CHKSUM_GOOD) {
2723                        skb->ip_summed = CHECKSUM_UNNECESSARY;
2724                }
2725
2726                length = skb->len;
2727                napi_gro_receive(napi, skb); /* send it up */
2728                netdev->stats.rx_packets++;
2729                netdev->stats.rx_bytes += length;
2730                adapter->rx_stats_buffers[scrq_num].packets++;
2731                adapter->rx_stats_buffers[scrq_num].bytes += length;
2732                frames_processed++;
2733        }
2734
2735        if (adapter->state != VNIC_CLOSING &&
2736            ((atomic_read(&adapter->rx_pool[scrq_num].available) <
2737              adapter->req_rx_add_entries_per_subcrq / 2) ||
2738              frames_processed < budget))
2739                replenish_rx_pool(adapter, &adapter->rx_pool[scrq_num]);
2740        if (frames_processed < budget) {
2741                if (napi_complete_done(napi, frames_processed)) {
2742                        enable_scrq_irq(adapter, rx_scrq);
2743                        if (pending_scrq(adapter, rx_scrq)) {
2744                                if (napi_reschedule(napi)) {
2745                                        disable_scrq_irq(adapter, rx_scrq);
2746                                        goto restart_poll;
2747                                }
2748                        }
2749                }
2750        }
2751        return frames_processed;
2752}
2753
2754static int wait_for_reset(struct ibmvnic_adapter *adapter)
2755{
2756        int rc, ret;
2757
2758        adapter->fallback.mtu = adapter->req_mtu;
2759        adapter->fallback.rx_queues = adapter->req_rx_queues;
2760        adapter->fallback.tx_queues = adapter->req_tx_queues;
2761        adapter->fallback.rx_entries = adapter->req_rx_add_entries_per_subcrq;
2762        adapter->fallback.tx_entries = adapter->req_tx_entries_per_subcrq;
2763
2764        reinit_completion(&adapter->reset_done);
2765        adapter->wait_for_reset = true;
2766        rc = ibmvnic_reset(adapter, VNIC_RESET_CHANGE_PARAM);
2767
2768        if (rc) {
2769                ret = rc;
2770                goto out;
2771        }
2772        rc = ibmvnic_wait_for_completion(adapter, &adapter->reset_done, 60000);
2773        if (rc) {
2774                ret = -ENODEV;
2775                goto out;
2776        }
2777
2778        ret = 0;
2779        if (adapter->reset_done_rc) {
2780                ret = -EIO;
2781                adapter->desired.mtu = adapter->fallback.mtu;
2782                adapter->desired.rx_queues = adapter->fallback.rx_queues;
2783                adapter->desired.tx_queues = adapter->fallback.tx_queues;
2784                adapter->desired.rx_entries = adapter->fallback.rx_entries;
2785                adapter->desired.tx_entries = adapter->fallback.tx_entries;
2786
2787                reinit_completion(&adapter->reset_done);
2788                adapter->wait_for_reset = true;
2789                rc = ibmvnic_reset(adapter, VNIC_RESET_CHANGE_PARAM);
2790                if (rc) {
2791                        ret = rc;
2792                        goto out;
2793                }
2794                rc = ibmvnic_wait_for_completion(adapter, &adapter->reset_done,
2795                                                 60000);
2796                if (rc) {
2797                        ret = -ENODEV;
2798                        goto out;
2799                }
2800        }
2801out:
2802        adapter->wait_for_reset = false;
2803
2804        return ret;
2805}
2806
2807static int ibmvnic_change_mtu(struct net_device *netdev, int new_mtu)
2808{
2809        struct ibmvnic_adapter *adapter = netdev_priv(netdev);
2810
2811        adapter->desired.mtu = new_mtu + ETH_HLEN;
2812
2813        return wait_for_reset(adapter);
2814}
2815
2816static netdev_features_t ibmvnic_features_check(struct sk_buff *skb,
2817                                                struct net_device *dev,
2818                                                netdev_features_t features)
2819{
2820        /* Some backing hardware adapters can not
2821         * handle packets with a MSS less than 224
2822         * or with only one segment.
2823         */
2824        if (skb_is_gso(skb)) {
2825                if (skb_shinfo(skb)->gso_size < 224 ||
2826                    skb_shinfo(skb)->gso_segs == 1)
2827                        features &= ~NETIF_F_GSO_MASK;
2828        }
2829
2830        return features;
2831}
2832
2833static const struct net_device_ops ibmvnic_netdev_ops = {
2834        .ndo_open               = ibmvnic_open,
2835        .ndo_stop               = ibmvnic_close,
2836        .ndo_start_xmit         = ibmvnic_xmit,
2837        .ndo_set_rx_mode        = ibmvnic_set_multi,
2838        .ndo_set_mac_address    = ibmvnic_set_mac,
2839        .ndo_validate_addr      = eth_validate_addr,
2840        .ndo_tx_timeout         = ibmvnic_tx_timeout,
2841        .ndo_change_mtu         = ibmvnic_change_mtu,
2842        .ndo_features_check     = ibmvnic_features_check,
2843};
2844
2845/* ethtool functions */
2846
2847static int ibmvnic_get_link_ksettings(struct net_device *netdev,
2848                                      struct ethtool_link_ksettings *cmd)
2849{
2850        struct ibmvnic_adapter *adapter = netdev_priv(netdev);
2851        int rc;
2852
2853        rc = send_query_phys_parms(adapter);
2854        if (rc) {
2855                adapter->speed = SPEED_UNKNOWN;
2856                adapter->duplex = DUPLEX_UNKNOWN;
2857        }
2858        cmd->base.speed = adapter->speed;
2859        cmd->base.duplex = adapter->duplex;
2860        cmd->base.port = PORT_FIBRE;
2861        cmd->base.phy_address = 0;
2862        cmd->base.autoneg = AUTONEG_ENABLE;
2863
2864        return 0;
2865}
2866
2867static void ibmvnic_get_drvinfo(struct net_device *netdev,
2868                                struct ethtool_drvinfo *info)
2869{
2870        struct ibmvnic_adapter *adapter = netdev_priv(netdev);
2871
2872        strscpy(info->driver, ibmvnic_driver_name, sizeof(info->driver));
2873        strscpy(info->version, IBMVNIC_DRIVER_VERSION, sizeof(info->version));
2874        strscpy(info->fw_version, adapter->fw_version,
2875                sizeof(info->fw_version));
2876}
2877
2878static u32 ibmvnic_get_msglevel(struct net_device *netdev)
2879{
2880        struct ibmvnic_adapter *adapter = netdev_priv(netdev);
2881
2882        return adapter->msg_enable;
2883}
2884
2885static void ibmvnic_set_msglevel(struct net_device *netdev, u32 data)
2886{
2887        struct ibmvnic_adapter *adapter = netdev_priv(netdev);
2888
2889        adapter->msg_enable = data;
2890}
2891
2892static u32 ibmvnic_get_link(struct net_device *netdev)
2893{
2894        struct ibmvnic_adapter *adapter = netdev_priv(netdev);
2895
2896        /* Don't need to send a query because we request a logical link up at
2897         * init and then we wait for link state indications
2898         */
2899        return adapter->logical_link_state;
2900}
2901
2902static void ibmvnic_get_ringparam(struct net_device *netdev,
2903                                  struct ethtool_ringparam *ring)
2904{
2905        struct ibmvnic_adapter *adapter = netdev_priv(netdev);
2906
2907        if (adapter->priv_flags & IBMVNIC_USE_SERVER_MAXES) {
2908                ring->rx_max_pending = adapter->max_rx_add_entries_per_subcrq;
2909                ring->tx_max_pending = adapter->max_tx_entries_per_subcrq;
2910        } else {
2911                ring->rx_max_pending = IBMVNIC_MAX_QUEUE_SZ;
2912                ring->tx_max_pending = IBMVNIC_MAX_QUEUE_SZ;
2913        }
2914        ring->rx_mini_max_pending = 0;
2915        ring->rx_jumbo_max_pending = 0;
2916        ring->rx_pending = adapter->req_rx_add_entries_per_subcrq;
2917        ring->tx_pending = adapter->req_tx_entries_per_subcrq;
2918        ring->rx_mini_pending = 0;
2919        ring->rx_jumbo_pending = 0;
2920}
2921
2922static int ibmvnic_set_ringparam(struct net_device *netdev,
2923                                 struct ethtool_ringparam *ring)
2924{
2925        struct ibmvnic_adapter *adapter = netdev_priv(netdev);
2926        int ret;
2927
2928        ret = 0;
2929        adapter->desired.rx_entries = ring->rx_pending;
2930        adapter->desired.tx_entries = ring->tx_pending;
2931
2932        ret = wait_for_reset(adapter);
2933
2934        if (!ret &&
2935            (adapter->req_rx_add_entries_per_subcrq != ring->rx_pending ||
2936             adapter->req_tx_entries_per_subcrq != ring->tx_pending))
2937                netdev_info(netdev,
2938                            "Could not match full ringsize request. Requested: RX %d, TX %d; Allowed: RX %llu, TX %llu\n",
2939                            ring->rx_pending, ring->tx_pending,
2940                            adapter->req_rx_add_entries_per_subcrq,
2941                            adapter->req_tx_entries_per_subcrq);
2942        return ret;
2943}
2944
2945static void ibmvnic_get_channels(struct net_device *netdev,
2946                                 struct ethtool_channels *channels)
2947{
2948        struct ibmvnic_adapter *adapter = netdev_priv(netdev);
2949
2950        if (adapter->priv_flags & IBMVNIC_USE_SERVER_MAXES) {
2951                channels->max_rx = adapter->max_rx_queues;
2952                channels->max_tx = adapter->max_tx_queues;
2953        } else {
2954                channels->max_rx = IBMVNIC_MAX_QUEUES;
2955                channels->max_tx = IBMVNIC_MAX_QUEUES;
2956        }
2957
2958        channels->max_other = 0;
2959        channels->max_combined = 0;
2960        channels->rx_count = adapter->req_rx_queues;
2961        channels->tx_count = adapter->req_tx_queues;
2962        channels->other_count = 0;
2963        channels->combined_count = 0;
2964}
2965
2966static int ibmvnic_set_channels(struct net_device *netdev,
2967                                struct ethtool_channels *channels)
2968{
2969        struct ibmvnic_adapter *adapter = netdev_priv(netdev);
2970        int ret;
2971
2972        ret = 0;
2973        adapter->desired.rx_queues = channels->rx_count;
2974        adapter->desired.tx_queues = channels->tx_count;
2975
2976        ret = wait_for_reset(adapter);
2977
2978        if (!ret &&
2979            (adapter->req_rx_queues != channels->rx_count ||
2980             adapter->req_tx_queues != channels->tx_count))
2981                netdev_info(netdev,
2982                            "Could not match full channels request. Requested: RX %d, TX %d; Allowed: RX %llu, TX %llu\n",
2983                            channels->rx_count, channels->tx_count,
2984                            adapter->req_rx_queues, adapter->req_tx_queues);
2985        return ret;
2986}
2987
2988static void ibmvnic_get_strings(struct net_device *dev, u32 stringset, u8 *data)
2989{
2990        struct ibmvnic_adapter *adapter = netdev_priv(dev);
2991        int i;
2992
2993        switch (stringset) {
2994        case ETH_SS_STATS:
2995                for (i = 0; i < ARRAY_SIZE(ibmvnic_stats);
2996                                i++, data += ETH_GSTRING_LEN)
2997                        memcpy(data, ibmvnic_stats[i].name, ETH_GSTRING_LEN);
2998
2999                for (i = 0; i < adapter->req_tx_queues; i++) {
3000                        snprintf(data, ETH_GSTRING_LEN, "tx%d_packets", i);
3001                        data += ETH_GSTRING_LEN;
3002
3003                        snprintf(data, ETH_GSTRING_LEN, "tx%d_bytes", i);
3004                        data += ETH_GSTRING_LEN;
3005
3006                        snprintf(data, ETH_GSTRING_LEN,
3007                                 "tx%d_dropped_packets", i);
3008                        data += ETH_GSTRING_LEN;
3009                }
3010
3011                for (i = 0; i < adapter->req_rx_queues; i++) {
3012                        snprintf(data, ETH_GSTRING_LEN, "rx%d_packets", i);
3013                        data += ETH_GSTRING_LEN;
3014
3015                        snprintf(data, ETH_GSTRING_LEN, "rx%d_bytes", i);
3016                        data += ETH_GSTRING_LEN;
3017
3018                        snprintf(data, ETH_GSTRING_LEN, "rx%d_interrupts", i);
3019                        data += ETH_GSTRING_LEN;
3020                }
3021                break;
3022
3023        case ETH_SS_PRIV_FLAGS:
3024                for (i = 0; i < ARRAY_SIZE(ibmvnic_priv_flags); i++)
3025                        strcpy(data + i * ETH_GSTRING_LEN,
3026                               ibmvnic_priv_flags[i]);
3027                break;
3028        default:
3029                return;
3030        }
3031}
3032
3033static int ibmvnic_get_sset_count(struct net_device *dev, int sset)
3034{
3035        struct ibmvnic_adapter *adapter = netdev_priv(dev);
3036
3037        switch (sset) {
3038        case ETH_SS_STATS:
3039                return ARRAY_SIZE(ibmvnic_stats) +
3040                       adapter->req_tx_queues * NUM_TX_STATS +
3041                       adapter->req_rx_queues * NUM_RX_STATS;
3042        case ETH_SS_PRIV_FLAGS:
3043                return ARRAY_SIZE(ibmvnic_priv_flags);
3044        default:
3045                return -EOPNOTSUPP;
3046        }
3047}
3048
3049static void ibmvnic_get_ethtool_stats(struct net_device *dev,
3050                                      struct ethtool_stats *stats, u64 *data)
3051{
3052        struct ibmvnic_adapter *adapter = netdev_priv(dev);
3053        union ibmvnic_crq crq;
3054        int i, j;
3055        int rc;
3056
3057        memset(&crq, 0, sizeof(crq));
3058        crq.request_statistics.first = IBMVNIC_CRQ_CMD;
3059        crq.request_statistics.cmd = REQUEST_STATISTICS;
3060        crq.request_statistics.ioba = cpu_to_be32(adapter->stats_token);
3061        crq.request_statistics.len =
3062            cpu_to_be32(sizeof(struct ibmvnic_statistics));
3063
3064        /* Wait for data to be written */
3065        reinit_completion(&adapter->stats_done);
3066        rc = ibmvnic_send_crq(adapter, &crq);
3067        if (rc)
3068                return;
3069        rc = ibmvnic_wait_for_completion(adapter, &adapter->stats_done, 10000);
3070        if (rc)
3071                return;
3072
3073        for (i = 0; i < ARRAY_SIZE(ibmvnic_stats); i++)
3074                data[i] = be64_to_cpu(IBMVNIC_GET_STAT
3075                                      (adapter, ibmvnic_stats[i].offset));
3076
3077        for (j = 0; j < adapter->req_tx_queues; j++) {
3078                data[i] = adapter->tx_stats_buffers[j].packets;
3079                i++;
3080                data[i] = adapter->tx_stats_buffers[j].bytes;
3081                i++;
3082                data[i] = adapter->tx_stats_buffers[j].dropped_packets;
3083                i++;
3084        }
3085
3086        for (j = 0; j < adapter->req_rx_queues; j++) {
3087                data[i] = adapter->rx_stats_buffers[j].packets;
3088                i++;
3089                data[i] = adapter->rx_stats_buffers[j].bytes;
3090                i++;
3091                data[i] = adapter->rx_stats_buffers[j].interrupts;
3092                i++;
3093        }
3094}
3095
3096static u32 ibmvnic_get_priv_flags(struct net_device *netdev)
3097{
3098        struct ibmvnic_adapter *adapter = netdev_priv(netdev);
3099
3100        return adapter->priv_flags;
3101}
3102
3103static int ibmvnic_set_priv_flags(struct net_device *netdev, u32 flags)
3104{
3105        struct ibmvnic_adapter *adapter = netdev_priv(netdev);
3106        bool which_maxes = !!(flags & IBMVNIC_USE_SERVER_MAXES);
3107
3108        if (which_maxes)
3109                adapter->priv_flags |= IBMVNIC_USE_SERVER_MAXES;
3110        else
3111                adapter->priv_flags &= ~IBMVNIC_USE_SERVER_MAXES;
3112
3113        return 0;
3114}
3115
3116static const struct ethtool_ops ibmvnic_ethtool_ops = {
3117        .get_drvinfo            = ibmvnic_get_drvinfo,
3118        .get_msglevel           = ibmvnic_get_msglevel,
3119        .set_msglevel           = ibmvnic_set_msglevel,
3120        .get_link               = ibmvnic_get_link,
3121        .get_ringparam          = ibmvnic_get_ringparam,
3122        .set_ringparam          = ibmvnic_set_ringparam,
3123        .get_channels           = ibmvnic_get_channels,
3124        .set_channels           = ibmvnic_set_channels,
3125        .get_strings            = ibmvnic_get_strings,
3126        .get_sset_count         = ibmvnic_get_sset_count,
3127        .get_ethtool_stats      = ibmvnic_get_ethtool_stats,
3128        .get_link_ksettings     = ibmvnic_get_link_ksettings,
3129        .get_priv_flags         = ibmvnic_get_priv_flags,
3130        .set_priv_flags         = ibmvnic_set_priv_flags,
3131};
3132
3133/* Routines for managing CRQs/sCRQs  */
3134
3135static int reset_one_sub_crq_queue(struct ibmvnic_adapter *adapter,
3136                                   struct ibmvnic_sub_crq_queue *scrq)
3137{
3138        int rc;
3139
3140        if (!scrq) {
3141                netdev_dbg(adapter->netdev, "Invalid scrq reset.\n");
3142                return -EINVAL;
3143        }
3144
3145        if (scrq->irq) {
3146                free_irq(scrq->irq, scrq);
3147                irq_dispose_mapping(scrq->irq);
3148                scrq->irq = 0;
3149        }
3150
3151        if (scrq->msgs) {
3152                memset(scrq->msgs, 0, 4 * PAGE_SIZE);
3153                atomic_set(&scrq->used, 0);
3154                scrq->cur = 0;
3155                scrq->ind_buf.index = 0;
3156        } else {
3157                netdev_dbg(adapter->netdev, "Invalid scrq reset\n");
3158                return -EINVAL;
3159        }
3160
3161        rc = h_reg_sub_crq(adapter->vdev->unit_address, scrq->msg_token,
3162                           4 * PAGE_SIZE, &scrq->crq_num, &scrq->hw_irq);
3163        return rc;
3164}
3165
3166static int reset_sub_crq_queues(struct ibmvnic_adapter *adapter)
3167{
3168        int i, rc;
3169
3170        if (!adapter->tx_scrq || !adapter->rx_scrq)
3171                return -EINVAL;
3172
3173        for (i = 0; i < adapter->req_tx_queues; i++) {
3174                netdev_dbg(adapter->netdev, "Re-setting tx_scrq[%d]\n", i);
3175                rc = reset_one_sub_crq_queue(adapter, adapter->tx_scrq[i]);
3176                if (rc)
3177                        return rc;
3178        }
3179
3180        for (i = 0; i < adapter->req_rx_queues; i++) {
3181                netdev_dbg(adapter->netdev, "Re-setting rx_scrq[%d]\n", i);
3182                rc = reset_one_sub_crq_queue(adapter, adapter->rx_scrq[i]);
3183                if (rc)
3184                        return rc;
3185        }
3186
3187        return rc;
3188}
3189
3190static void release_sub_crq_queue(struct ibmvnic_adapter *adapter,
3191                                  struct ibmvnic_sub_crq_queue *scrq,
3192                                  bool do_h_free)
3193{
3194        struct device *dev = &adapter->vdev->dev;
3195        long rc;
3196
3197        netdev_dbg(adapter->netdev, "Releasing sub-CRQ\n");
3198
3199        if (do_h_free) {
3200                /* Close the sub-crqs */
3201                do {
3202                        rc = plpar_hcall_norets(H_FREE_SUB_CRQ,
3203                                                adapter->vdev->unit_address,
3204                                                scrq->crq_num);
3205                } while (rc == H_BUSY || H_IS_LONG_BUSY(rc));
3206
3207                if (rc) {
3208                        netdev_err(adapter->netdev,
3209                                   "Failed to release sub-CRQ %16lx, rc = %ld\n",
3210                                   scrq->crq_num, rc);
3211                }
3212        }
3213
3214        dma_free_coherent(dev,
3215                          IBMVNIC_IND_ARR_SZ,
3216                          scrq->ind_buf.indir_arr,
3217                          scrq->ind_buf.indir_dma);
3218
3219        dma_unmap_single(dev, scrq->msg_token, 4 * PAGE_SIZE,
3220                         DMA_BIDIRECTIONAL);
3221        free_pages((unsigned long)scrq->msgs, 2);
3222        kfree(scrq);
3223}
3224
3225static struct ibmvnic_sub_crq_queue *init_sub_crq_queue(struct ibmvnic_adapter
3226                                                        *adapter)
3227{
3228        struct device *dev = &adapter->vdev->dev;
3229        struct ibmvnic_sub_crq_queue *scrq;
3230        int rc;
3231
3232        scrq = kzalloc(sizeof(*scrq), GFP_KERNEL);
3233        if (!scrq)
3234                return NULL;
3235
3236        scrq->msgs =
3237                (union sub_crq *)__get_free_pages(GFP_KERNEL | __GFP_ZERO, 2);
3238        if (!scrq->msgs) {
3239                dev_warn(dev, "Couldn't allocate crq queue messages page\n");
3240                goto zero_page_failed;
3241        }
3242
3243        scrq->msg_token = dma_map_single(dev, scrq->msgs, 4 * PAGE_SIZE,
3244                                         DMA_BIDIRECTIONAL);
3245        if (dma_mapping_error(dev, scrq->msg_token)) {
3246                dev_warn(dev, "Couldn't map crq queue messages page\n");
3247                goto map_failed;
3248        }
3249
3250        rc = h_reg_sub_crq(adapter->vdev->unit_address, scrq->msg_token,
3251                           4 * PAGE_SIZE, &scrq->crq_num, &scrq->hw_irq);
3252
3253        if (rc == H_RESOURCE)
3254                rc = ibmvnic_reset_crq(adapter);
3255
3256        if (rc == H_CLOSED) {
3257                dev_warn(dev, "Partner adapter not ready, waiting.\n");
3258        } else if (rc) {
3259                dev_warn(dev, "Error %d registering sub-crq\n", rc);
3260                goto reg_failed;
3261        }
3262
3263        scrq->adapter = adapter;
3264        scrq->size = 4 * PAGE_SIZE / sizeof(*scrq->msgs);
3265        scrq->ind_buf.index = 0;
3266
3267        scrq->ind_buf.indir_arr =
3268                dma_alloc_coherent(dev,
3269                                   IBMVNIC_IND_ARR_SZ,
3270                                   &scrq->ind_buf.indir_dma,
3271                                   GFP_KERNEL);
3272
3273        if (!scrq->ind_buf.indir_arr)
3274                goto indir_failed;
3275
3276        spin_lock_init(&scrq->lock);
3277
3278        netdev_dbg(adapter->netdev,
3279                   "sub-crq initialized, num %lx, hw_irq=%lx, irq=%x\n",
3280                   scrq->crq_num, scrq->hw_irq, scrq->irq);
3281
3282        return scrq;
3283
3284indir_failed:
3285        do {
3286                rc = plpar_hcall_norets(H_FREE_SUB_CRQ,
3287                                        adapter->vdev->unit_address,
3288                                        scrq->crq_num);
3289        } while (rc == H_BUSY || rc == H_IS_LONG_BUSY(rc));
3290reg_failed:
3291        dma_unmap_single(dev, scrq->msg_token, 4 * PAGE_SIZE,
3292                         DMA_BIDIRECTIONAL);
3293map_failed:
3294        free_pages((unsigned long)scrq->msgs, 2);
3295zero_page_failed:
3296        kfree(scrq);
3297
3298        return NULL;
3299}
3300
3301static void release_sub_crqs(struct ibmvnic_adapter *adapter, bool do_h_free)
3302{
3303        int i;
3304
3305        if (adapter->tx_scrq) {
3306                for (i = 0; i < adapter->num_active_tx_scrqs; i++) {
3307                        if (!adapter->tx_scrq[i])
3308                                continue;
3309
3310                        netdev_dbg(adapter->netdev, "Releasing tx_scrq[%d]\n",
3311                                   i);
3312                        ibmvnic_tx_scrq_clean_buffer(adapter, adapter->tx_scrq[i]);
3313                        if (adapter->tx_scrq[i]->irq) {
3314                                free_irq(adapter->tx_scrq[i]->irq,
3315                                         adapter->tx_scrq[i]);
3316                                irq_dispose_mapping(adapter->tx_scrq[i]->irq);
3317                                adapter->tx_scrq[i]->irq = 0;
3318                        }
3319
3320                        release_sub_crq_queue(adapter, adapter->tx_scrq[i],
3321                                              do_h_free);
3322                }
3323
3324                kfree(adapter->tx_scrq);
3325                adapter->tx_scrq = NULL;
3326                adapter->num_active_tx_scrqs = 0;
3327        }
3328
3329        if (adapter->rx_scrq) {
3330                for (i = 0; i < adapter->num_active_rx_scrqs; i++) {
3331                        if (!adapter->rx_scrq[i])
3332                                continue;
3333
3334                        netdev_dbg(adapter->netdev, "Releasing rx_scrq[%d]\n",
3335                                   i);
3336                        if (adapter->rx_scrq[i]->irq) {
3337                                free_irq(adapter->rx_scrq[i]->irq,
3338                                         adapter->rx_scrq[i]);
3339                                irq_dispose_mapping(adapter->rx_scrq[i]->irq);
3340                                adapter->rx_scrq[i]->irq = 0;
3341                        }
3342
3343                        release_sub_crq_queue(adapter, adapter->rx_scrq[i],
3344                                              do_h_free);
3345                }
3346
3347                kfree(adapter->rx_scrq);
3348                adapter->rx_scrq = NULL;
3349                adapter->num_active_rx_scrqs = 0;
3350        }
3351}
3352
3353static int disable_scrq_irq(struct ibmvnic_adapter *adapter,
3354                            struct ibmvnic_sub_crq_queue *scrq)
3355{
3356        struct device *dev = &adapter->vdev->dev;
3357        unsigned long rc;
3358
3359        rc = plpar_hcall_norets(H_VIOCTL, adapter->vdev->unit_address,
3360                                H_DISABLE_VIO_INTERRUPT, scrq->hw_irq, 0, 0);
3361        if (rc)
3362                dev_err(dev, "Couldn't disable scrq irq 0x%lx. rc=%ld\n",
3363                        scrq->hw_irq, rc);
3364        return rc;
3365}
3366
3367static int enable_scrq_irq(struct ibmvnic_adapter *adapter,
3368                           struct ibmvnic_sub_crq_queue *scrq)
3369{
3370        struct device *dev = &adapter->vdev->dev;
3371        unsigned long rc;
3372
3373        if (scrq->hw_irq > 0x100000000ULL) {
3374                dev_err(dev, "bad hw_irq = %lx\n", scrq->hw_irq);
3375                return 1;
3376        }
3377
3378        if (test_bit(0, &adapter->resetting) &&
3379            adapter->reset_reason == VNIC_RESET_MOBILITY) {
3380                u64 val = (0xff000000) | scrq->hw_irq;
3381
3382                rc = plpar_hcall_norets(H_EOI, val);
3383                /* H_EOI would fail with rc = H_FUNCTION when running
3384                 * in XIVE mode which is expected, but not an error.
3385                 */
3386                if (rc && (rc != H_FUNCTION))
3387                        dev_err(dev, "H_EOI FAILED irq 0x%llx. rc=%ld\n",
3388                                val, rc);
3389        }
3390
3391        rc = plpar_hcall_norets(H_VIOCTL, adapter->vdev->unit_address,
3392                                H_ENABLE_VIO_INTERRUPT, scrq->hw_irq, 0, 0);
3393        if (rc)
3394                dev_err(dev, "Couldn't enable scrq irq 0x%lx. rc=%ld\n",
3395                        scrq->hw_irq, rc);
3396        return rc;
3397}
3398
3399static int ibmvnic_complete_tx(struct ibmvnic_adapter *adapter,
3400                               struct ibmvnic_sub_crq_queue *scrq)
3401{
3402        struct device *dev = &adapter->vdev->dev;
3403        struct ibmvnic_tx_pool *tx_pool;
3404        struct ibmvnic_tx_buff *txbuff;
3405        struct netdev_queue *txq;
3406        union sub_crq *next;
3407        int index;
3408        int i;
3409
3410restart_loop:
3411        while (pending_scrq(adapter, scrq)) {
3412                unsigned int pool = scrq->pool_index;
3413                int num_entries = 0;
3414                int total_bytes = 0;
3415                int num_packets = 0;
3416
3417                next = ibmvnic_next_scrq(adapter, scrq);
3418                for (i = 0; i < next->tx_comp.num_comps; i++) {
3419                        index = be32_to_cpu(next->tx_comp.correlators[i]);
3420                        if (index & IBMVNIC_TSO_POOL_MASK) {
3421                                tx_pool = &adapter->tso_pool[pool];
3422                                index &= ~IBMVNIC_TSO_POOL_MASK;
3423                        } else {
3424                                tx_pool = &adapter->tx_pool[pool];
3425                        }
3426
3427                        txbuff = &tx_pool->tx_buff[index];
3428                        num_packets++;
3429                        num_entries += txbuff->num_entries;
3430                        if (txbuff->skb) {
3431                                total_bytes += txbuff->skb->len;
3432                                if (next->tx_comp.rcs[i]) {
3433                                        dev_err(dev, "tx error %x\n",
3434                                                next->tx_comp.rcs[i]);
3435                                        dev_kfree_skb_irq(txbuff->skb);
3436                                } else {
3437                                        dev_consume_skb_irq(txbuff->skb);
3438                                }
3439                                txbuff->skb = NULL;
3440                        } else {
3441                                netdev_warn(adapter->netdev,
3442                                            "TX completion received with NULL socket buffer\n");
3443                        }
3444                        tx_pool->free_map[tx_pool->producer_index] = index;
3445                        tx_pool->producer_index =
3446                                (tx_pool->producer_index + 1) %
3447                                        tx_pool->num_buffers;
3448                }
3449                /* remove tx_comp scrq*/
3450                next->tx_comp.first = 0;
3451
3452                txq = netdev_get_tx_queue(adapter->netdev, scrq->pool_index);
3453                netdev_tx_completed_queue(txq, num_packets, total_bytes);
3454
3455                if (atomic_sub_return(num_entries, &scrq->used) <=
3456                    (adapter->req_tx_entries_per_subcrq / 2) &&
3457                    __netif_subqueue_stopped(adapter->netdev,
3458                                             scrq->pool_index)) {
3459                        netif_wake_subqueue(adapter->netdev, scrq->pool_index);
3460                        netdev_dbg(adapter->netdev, "Started queue %d\n",
3461                                   scrq->pool_index);
3462                }
3463        }
3464
3465        enable_scrq_irq(adapter, scrq);
3466
3467        if (pending_scrq(adapter, scrq)) {
3468                disable_scrq_irq(adapter, scrq);
3469                goto restart_loop;
3470        }
3471
3472        return 0;
3473}
3474
3475static irqreturn_t ibmvnic_interrupt_tx(int irq, void *instance)
3476{
3477        struct ibmvnic_sub_crq_queue *scrq = instance;
3478        struct ibmvnic_adapter *adapter = scrq->adapter;
3479
3480        disable_scrq_irq(adapter, scrq);
3481        ibmvnic_complete_tx(adapter, scrq);
3482
3483        return IRQ_HANDLED;
3484}
3485
3486static irqreturn_t ibmvnic_interrupt_rx(int irq, void *instance)
3487{
3488        struct ibmvnic_sub_crq_queue *scrq = instance;
3489        struct ibmvnic_adapter *adapter = scrq->adapter;
3490
3491        /* When booting a kdump kernel we can hit pending interrupts
3492         * prior to completing driver initialization.
3493         */
3494        if (unlikely(adapter->state != VNIC_OPEN))
3495                return IRQ_NONE;
3496
3497        adapter->rx_stats_buffers[scrq->scrq_num].interrupts++;
3498
3499        if (napi_schedule_prep(&adapter->napi[scrq->scrq_num])) {
3500                disable_scrq_irq(adapter, scrq);
3501                __napi_schedule(&adapter->napi[scrq->scrq_num]);
3502        }
3503
3504        return IRQ_HANDLED;
3505}
3506
3507static int init_sub_crq_irqs(struct ibmvnic_adapter *adapter)
3508{
3509        struct device *dev = &adapter->vdev->dev;
3510        struct ibmvnic_sub_crq_queue *scrq;
3511        int i = 0, j = 0;
3512        int rc = 0;
3513
3514        for (i = 0; i < adapter->req_tx_queues; i++) {
3515                netdev_dbg(adapter->netdev, "Initializing tx_scrq[%d] irq\n",
3516                           i);
3517                scrq = adapter->tx_scrq[i];
3518                scrq->irq = irq_create_mapping(NULL, scrq->hw_irq);
3519
3520                if (!scrq->irq) {
3521                        rc = -EINVAL;
3522                        dev_err(dev, "Error mapping irq\n");
3523                        goto req_tx_irq_failed;
3524                }
3525
3526                snprintf(scrq->name, sizeof(scrq->name), "ibmvnic-%x-tx%d",
3527                         adapter->vdev->unit_address, i);
3528                rc = request_irq(scrq->irq, ibmvnic_interrupt_tx,
3529                                 0, scrq->name, scrq);
3530
3531                if (rc) {
3532                        dev_err(dev, "Couldn't register tx irq 0x%x. rc=%d\n",
3533                                scrq->irq, rc);
3534                        irq_dispose_mapping(scrq->irq);
3535                        goto req_tx_irq_failed;
3536                }
3537        }
3538
3539        for (i = 0; i < adapter->req_rx_queues; i++) {
3540                netdev_dbg(adapter->netdev, "Initializing rx_scrq[%d] irq\n",
3541                           i);
3542                scrq = adapter->rx_scrq[i];
3543                scrq->irq = irq_create_mapping(NULL, scrq->hw_irq);
3544                if (!scrq->irq) {
3545                        rc = -EINVAL;
3546                        dev_err(dev, "Error mapping irq\n");
3547                        goto req_rx_irq_failed;
3548                }
3549                snprintf(scrq->name, sizeof(scrq->name), "ibmvnic-%x-rx%d",
3550                         adapter->vdev->unit_address, i);
3551                rc = request_irq(scrq->irq, ibmvnic_interrupt_rx,
3552                                 0, scrq->name, scrq);
3553                if (rc) {
3554                        dev_err(dev, "Couldn't register rx irq 0x%x. rc=%d\n",
3555                                scrq->irq, rc);
3556                        irq_dispose_mapping(scrq->irq);
3557                        goto req_rx_irq_failed;
3558                }
3559        }
3560        return rc;
3561
3562req_rx_irq_failed:
3563        for (j = 0; j < i; j++) {
3564                free_irq(adapter->rx_scrq[j]->irq, adapter->rx_scrq[j]);
3565                irq_dispose_mapping(adapter->rx_scrq[j]->irq);
3566        }
3567        i = adapter->req_tx_queues;
3568req_tx_irq_failed:
3569        for (j = 0; j < i; j++) {
3570                free_irq(adapter->tx_scrq[j]->irq, adapter->tx_scrq[j]);
3571                irq_dispose_mapping(adapter->tx_scrq[j]->irq);
3572        }
3573        release_sub_crqs(adapter, 1);
3574        return rc;
3575}
3576
3577static int init_sub_crqs(struct ibmvnic_adapter *adapter)
3578{
3579        struct device *dev = &adapter->vdev->dev;
3580        struct ibmvnic_sub_crq_queue **allqueues;
3581        int registered_queues = 0;
3582        int total_queues;
3583        int more = 0;
3584        int i;
3585
3586        total_queues = adapter->req_tx_queues + adapter->req_rx_queues;
3587
3588        allqueues = kcalloc(total_queues, sizeof(*allqueues), GFP_KERNEL);
3589        if (!allqueues)
3590                return -1;
3591
3592        for (i = 0; i < total_queues; i++) {
3593                allqueues[i] = init_sub_crq_queue(adapter);
3594                if (!allqueues[i]) {
3595                        dev_warn(dev, "Couldn't allocate all sub-crqs\n");
3596                        break;
3597                }
3598                registered_queues++;
3599        }
3600
3601        /* Make sure we were able to register the minimum number of queues */
3602        if (registered_queues <
3603            adapter->min_tx_queues + adapter->min_rx_queues) {
3604                dev_err(dev, "Fatal: Couldn't init  min number of sub-crqs\n");
3605                goto tx_failed;
3606        }
3607
3608        /* Distribute the failed allocated queues*/
3609        for (i = 0; i < total_queues - registered_queues + more ; i++) {
3610                netdev_dbg(adapter->netdev, "Reducing number of queues\n");
3611                switch (i % 3) {
3612                case 0:
3613                        if (adapter->req_rx_queues > adapter->min_rx_queues)
3614                                adapter->req_rx_queues--;
3615                        else
3616                                more++;
3617                        break;
3618                case 1:
3619                        if (adapter->req_tx_queues > adapter->min_tx_queues)
3620                                adapter->req_tx_queues--;
3621                        else
3622                                more++;
3623                        break;
3624                }
3625        }
3626
3627        adapter->tx_scrq = kcalloc(adapter->req_tx_queues,
3628                                   sizeof(*adapter->tx_scrq), GFP_KERNEL);
3629        if (!adapter->tx_scrq)
3630                goto tx_failed;
3631
3632        for (i = 0; i < adapter->req_tx_queues; i++) {
3633                adapter->tx_scrq[i] = allqueues[i];
3634                adapter->tx_scrq[i]->pool_index = i;
3635                adapter->num_active_tx_scrqs++;
3636        }
3637
3638        adapter->rx_scrq = kcalloc(adapter->req_rx_queues,
3639                                   sizeof(*adapter->rx_scrq), GFP_KERNEL);
3640        if (!adapter->rx_scrq)
3641                goto rx_failed;
3642
3643        for (i = 0; i < adapter->req_rx_queues; i++) {
3644                adapter->rx_scrq[i] = allqueues[i + adapter->req_tx_queues];
3645                adapter->rx_scrq[i]->scrq_num = i;
3646                adapter->num_active_rx_scrqs++;
3647        }
3648
3649        kfree(allqueues);
3650        return 0;
3651
3652rx_failed:
3653        kfree(adapter->tx_scrq);
3654        adapter->tx_scrq = NULL;
3655tx_failed:
3656        for (i = 0; i < registered_queues; i++)
3657                release_sub_crq_queue(adapter, allqueues[i], 1);
3658        kfree(allqueues);
3659        return -1;
3660}
3661
3662static void send_request_cap(struct ibmvnic_adapter *adapter, int retry)
3663{
3664        struct device *dev = &adapter->vdev->dev;
3665        union ibmvnic_crq crq;
3666        int max_entries;
3667
3668        if (!retry) {
3669                /* Sub-CRQ entries are 32 byte long */
3670                int entries_page = 4 * PAGE_SIZE / (sizeof(u64) * 4);
3671
3672                if (adapter->min_tx_entries_per_subcrq > entries_page ||
3673                    adapter->min_rx_add_entries_per_subcrq > entries_page) {
3674                        dev_err(dev, "Fatal, invalid entries per sub-crq\n");
3675                        return;
3676                }
3677
3678                if (adapter->desired.mtu)
3679                        adapter->req_mtu = adapter->desired.mtu;
3680                else
3681                        adapter->req_mtu = adapter->netdev->mtu + ETH_HLEN;
3682
3683                if (!adapter->desired.tx_entries)
3684                        adapter->desired.tx_entries =
3685                                        adapter->max_tx_entries_per_subcrq;
3686                if (!adapter->desired.rx_entries)
3687                        adapter->desired.rx_entries =
3688                                        adapter->max_rx_add_entries_per_subcrq;
3689
3690                max_entries = IBMVNIC_MAX_LTB_SIZE /
3691                              (adapter->req_mtu + IBMVNIC_BUFFER_HLEN);
3692
3693                if ((adapter->req_mtu + IBMVNIC_BUFFER_HLEN) *
3694                        adapter->desired.tx_entries > IBMVNIC_MAX_LTB_SIZE) {
3695                        adapter->desired.tx_entries = max_entries;
3696                }
3697
3698                if ((adapter->req_mtu + IBMVNIC_BUFFER_HLEN) *
3699                        adapter->desired.rx_entries > IBMVNIC_MAX_LTB_SIZE) {
3700                        adapter->desired.rx_entries = max_entries;
3701                }
3702
3703                if (adapter->desired.tx_entries)
3704                        adapter->req_tx_entries_per_subcrq =
3705                                        adapter->desired.tx_entries;
3706                else
3707                        adapter->req_tx_entries_per_subcrq =
3708                                        adapter->max_tx_entries_per_subcrq;
3709
3710                if (adapter->desired.rx_entries)
3711                        adapter->req_rx_add_entries_per_subcrq =
3712                                        adapter->desired.rx_entries;
3713                else
3714                        adapter->req_rx_add_entries_per_subcrq =
3715                                        adapter->max_rx_add_entries_per_subcrq;
3716
3717                if (adapter->desired.tx_queues)
3718                        adapter->req_tx_queues =
3719                                        adapter->desired.tx_queues;
3720                else
3721                        adapter->req_tx_queues =
3722                                        adapter->opt_tx_comp_sub_queues;
3723
3724                if (adapter->desired.rx_queues)
3725                        adapter->req_rx_queues =
3726                                        adapter->desired.rx_queues;
3727                else
3728                        adapter->req_rx_queues =
3729                                        adapter->opt_rx_comp_queues;
3730
3731                adapter->req_rx_add_queues = adapter->max_rx_add_queues;
3732        }
3733
3734        memset(&crq, 0, sizeof(crq));
3735        crq.request_capability.first = IBMVNIC_CRQ_CMD;
3736        crq.request_capability.cmd = REQUEST_CAPABILITY;
3737
3738        crq.request_capability.capability = cpu_to_be16(REQ_TX_QUEUES);
3739        crq.request_capability.number = cpu_to_be64(adapter->req_tx_queues);
3740        atomic_inc(&adapter->running_cap_crqs);
3741        ibmvnic_send_crq(adapter, &crq);
3742
3743        crq.request_capability.capability = cpu_to_be16(REQ_RX_QUEUES);
3744        crq.request_capability.number = cpu_to_be64(adapter->req_rx_queues);
3745        atomic_inc(&adapter->running_cap_crqs);
3746        ibmvnic_send_crq(adapter, &crq);
3747
3748        crq.request_capability.capability = cpu_to_be16(REQ_RX_ADD_QUEUES);
3749        crq.request_capability.number = cpu_to_be64(adapter->req_rx_add_queues);
3750        atomic_inc(&adapter->running_cap_crqs);
3751        ibmvnic_send_crq(adapter, &crq);
3752
3753        crq.request_capability.capability =
3754            cpu_to_be16(REQ_TX_ENTRIES_PER_SUBCRQ);
3755        crq.request_capability.number =
3756            cpu_to_be64(adapter->req_tx_entries_per_subcrq);
3757        atomic_inc(&adapter->running_cap_crqs);
3758        ibmvnic_send_crq(adapter, &crq);
3759
3760        crq.request_capability.capability =
3761            cpu_to_be16(REQ_RX_ADD_ENTRIES_PER_SUBCRQ);
3762        crq.request_capability.number =
3763            cpu_to_be64(adapter->req_rx_add_entries_per_subcrq);
3764        atomic_inc(&adapter->running_cap_crqs);
3765        ibmvnic_send_crq(adapter, &crq);
3766
3767        crq.request_capability.capability = cpu_to_be16(REQ_MTU);
3768        crq.request_capability.number = cpu_to_be64(adapter->req_mtu);
3769        atomic_inc(&adapter->running_cap_crqs);
3770        ibmvnic_send_crq(adapter, &crq);
3771
3772        if (adapter->netdev->flags & IFF_PROMISC) {
3773                if (adapter->promisc_supported) {
3774                        crq.request_capability.capability =
3775                            cpu_to_be16(PROMISC_REQUESTED);
3776                        crq.request_capability.number = cpu_to_be64(1);
3777                        atomic_inc(&adapter->running_cap_crqs);
3778                        ibmvnic_send_crq(adapter, &crq);
3779                }
3780        } else {
3781                crq.request_capability.capability =
3782                    cpu_to_be16(PROMISC_REQUESTED);
3783                crq.request_capability.number = cpu_to_be64(0);
3784                atomic_inc(&adapter->running_cap_crqs);
3785                ibmvnic_send_crq(adapter, &crq);
3786        }
3787}
3788
3789static int pending_scrq(struct ibmvnic_adapter *adapter,
3790                        struct ibmvnic_sub_crq_queue *scrq)
3791{
3792        union sub_crq *entry = &scrq->msgs[scrq->cur];
3793        int rc;
3794
3795        rc = !!(entry->generic.first & IBMVNIC_CRQ_CMD_RSP);
3796
3797        /* Ensure that the SCRQ valid flag is loaded prior to loading the
3798         * contents of the SCRQ descriptor
3799         */
3800        dma_rmb();
3801
3802        return rc;
3803}
3804
3805static union sub_crq *ibmvnic_next_scrq(struct ibmvnic_adapter *adapter,
3806                                        struct ibmvnic_sub_crq_queue *scrq)
3807{
3808        union sub_crq *entry;
3809        unsigned long flags;
3810
3811        spin_lock_irqsave(&scrq->lock, flags);
3812        entry = &scrq->msgs[scrq->cur];
3813        if (entry->generic.first & IBMVNIC_CRQ_CMD_RSP) {
3814                if (++scrq->cur == scrq->size)
3815                        scrq->cur = 0;
3816        } else {
3817                entry = NULL;
3818        }
3819        spin_unlock_irqrestore(&scrq->lock, flags);
3820
3821        /* Ensure that the SCRQ valid flag is loaded prior to loading the
3822         * contents of the SCRQ descriptor
3823         */
3824        dma_rmb();
3825
3826        return entry;
3827}
3828
3829static union ibmvnic_crq *ibmvnic_next_crq(struct ibmvnic_adapter *adapter)
3830{
3831        struct ibmvnic_crq_queue *queue = &adapter->crq;
3832        union ibmvnic_crq *crq;
3833
3834        crq = &queue->msgs[queue->cur];
3835        if (crq->generic.first & IBMVNIC_CRQ_CMD_RSP) {
3836                if (++queue->cur == queue->size)
3837                        queue->cur = 0;
3838        } else {
3839                crq = NULL;
3840        }
3841
3842        return crq;
3843}
3844
3845static void print_subcrq_error(struct device *dev, int rc, const char *func)
3846{
3847        switch (rc) {
3848        case H_PARAMETER:
3849                dev_warn_ratelimited(dev,
3850                                     "%s failed: Send request is malformed or adapter failover pending. (rc=%d)\n",
3851                                     func, rc);
3852                break;
3853        case H_CLOSED:
3854                dev_warn_ratelimited(dev,
3855                                     "%s failed: Backing queue closed. Adapter is down or failover pending. (rc=%d)\n",
3856                                     func, rc);
3857                break;
3858        default:
3859                dev_err_ratelimited(dev, "%s failed: (rc=%d)\n", func, rc);
3860                break;
3861        }
3862}
3863
3864static int send_subcrq_indirect(struct ibmvnic_adapter *adapter,
3865                                u64 remote_handle, u64 ioba, u64 num_entries)
3866{
3867        unsigned int ua = adapter->vdev->unit_address;
3868        struct device *dev = &adapter->vdev->dev;
3869        int rc;
3870
3871        /* Make sure the hypervisor sees the complete request */
3872        dma_wmb();