linux/drivers/net/ethernet/ibm/ibmveth.c
<<
>>
Prefs
   1/*
   2 * IBM Power Virtual Ethernet Device Driver
   3 *
   4 * This program is free software; you can redistribute it and/or modify
   5 * it under the terms of the GNU General Public License as published by
   6 * the Free Software Foundation; either version 2 of the License, or
   7 * (at your option) any later version.
   8 *
   9 * This program is distributed in the hope that it will be useful,
  10 * but WITHOUT ANY WARRANTY; without even the implied warranty of
  11 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
  12 * GNU General Public License for more details.
  13 *
  14 * You should have received a copy of the GNU General Public License
  15 * along with this program; if not, see <http://www.gnu.org/licenses/>.
  16 *
  17 * Copyright (C) IBM Corporation, 2003, 2010
  18 *
  19 * Authors: Dave Larson <larson1@us.ibm.com>
  20 *          Santiago Leon <santil@linux.vnet.ibm.com>
  21 *          Brian King <brking@linux.vnet.ibm.com>
  22 *          Robert Jennings <rcj@linux.vnet.ibm.com>
  23 *          Anton Blanchard <anton@au.ibm.com>
  24 */
  25
  26#include <linux/module.h>
  27#include <linux/moduleparam.h>
  28#include <linux/types.h>
  29#include <linux/errno.h>
  30#include <linux/dma-mapping.h>
  31#include <linux/kernel.h>
  32#include <linux/netdevice.h>
  33#include <linux/etherdevice.h>
  34#include <linux/skbuff.h>
  35#include <linux/init.h>
  36#include <linux/interrupt.h>
  37#include <linux/mm.h>
  38#include <linux/pm.h>
  39#include <linux/ethtool.h>
  40#include <linux/in.h>
  41#include <linux/ip.h>
  42#include <linux/ipv6.h>
  43#include <linux/slab.h>
  44#include <asm/hvcall.h>
  45#include <linux/atomic.h>
  46#include <asm/vio.h>
  47#include <asm/iommu.h>
  48#include <asm/firmware.h>
  49
  50#include "ibmveth.h"
  51
  52static irqreturn_t ibmveth_interrupt(int irq, void *dev_instance);
  53static void ibmveth_rxq_harvest_buffer(struct ibmveth_adapter *adapter);
  54static unsigned long ibmveth_get_desired_dma(struct vio_dev *vdev);
  55
  56static struct kobj_type ktype_veth_pool;
  57
  58
  59static const char ibmveth_driver_name[] = "ibmveth";
  60static const char ibmveth_driver_string[] = "IBM Power Virtual Ethernet Driver";
  61#define ibmveth_driver_version "1.05"
  62
  63MODULE_AUTHOR("Santiago Leon <santil@linux.vnet.ibm.com>");
  64MODULE_DESCRIPTION("IBM Power Virtual Ethernet Driver");
  65MODULE_LICENSE("GPL");
  66MODULE_VERSION(ibmveth_driver_version);
  67
  68static unsigned int tx_copybreak __read_mostly = 128;
  69module_param(tx_copybreak, uint, 0644);
  70MODULE_PARM_DESC(tx_copybreak,
  71        "Maximum size of packet that is copied to a new buffer on transmit");
  72
  73static unsigned int rx_copybreak __read_mostly = 128;
  74module_param(rx_copybreak, uint, 0644);
  75MODULE_PARM_DESC(rx_copybreak,
  76        "Maximum size of packet that is copied to a new buffer on receive");
  77
  78static unsigned int rx_flush __read_mostly = 0;
  79module_param(rx_flush, uint, 0644);
  80MODULE_PARM_DESC(rx_flush, "Flush receive buffers before use");
  81
  82static bool old_large_send __read_mostly;
  83module_param(old_large_send, bool, S_IRUGO);
  84MODULE_PARM_DESC(old_large_send,
  85        "Use old large send method on firmware that supports the new method");
  86
  87struct ibmveth_stat {
  88        char name[ETH_GSTRING_LEN];
  89        int offset;
  90};
  91
  92#define IBMVETH_STAT_OFF(stat) offsetof(struct ibmveth_adapter, stat)
  93#define IBMVETH_GET_STAT(a, off) *((u64 *)(((unsigned long)(a)) + off))
  94
  95struct ibmveth_stat ibmveth_stats[] = {
  96        { "replenish_task_cycles", IBMVETH_STAT_OFF(replenish_task_cycles) },
  97        { "replenish_no_mem", IBMVETH_STAT_OFF(replenish_no_mem) },
  98        { "replenish_add_buff_failure",
  99                        IBMVETH_STAT_OFF(replenish_add_buff_failure) },
 100        { "replenish_add_buff_success",
 101                        IBMVETH_STAT_OFF(replenish_add_buff_success) },
 102        { "rx_invalid_buffer", IBMVETH_STAT_OFF(rx_invalid_buffer) },
 103        { "rx_no_buffer", IBMVETH_STAT_OFF(rx_no_buffer) },
 104        { "tx_map_failed", IBMVETH_STAT_OFF(tx_map_failed) },
 105        { "tx_send_failed", IBMVETH_STAT_OFF(tx_send_failed) },
 106        { "fw_enabled_ipv4_csum", IBMVETH_STAT_OFF(fw_ipv4_csum_support) },
 107        { "fw_enabled_ipv6_csum", IBMVETH_STAT_OFF(fw_ipv6_csum_support) },
 108        { "tx_large_packets", IBMVETH_STAT_OFF(tx_large_packets) },
 109        { "rx_large_packets", IBMVETH_STAT_OFF(rx_large_packets) },
 110        { "fw_enabled_large_send", IBMVETH_STAT_OFF(fw_large_send_support) }
 111};
 112
 113/* simple methods of getting data from the current rxq entry */
 114static inline u32 ibmveth_rxq_flags(struct ibmveth_adapter *adapter)
 115{
 116        return be32_to_cpu(adapter->rx_queue.queue_addr[adapter->rx_queue.index].flags_off);
 117}
 118
 119static inline int ibmveth_rxq_toggle(struct ibmveth_adapter *adapter)
 120{
 121        return (ibmveth_rxq_flags(adapter) & IBMVETH_RXQ_TOGGLE) >>
 122                        IBMVETH_RXQ_TOGGLE_SHIFT;
 123}
 124
 125static inline int ibmveth_rxq_pending_buffer(struct ibmveth_adapter *adapter)
 126{
 127        return ibmveth_rxq_toggle(adapter) == adapter->rx_queue.toggle;
 128}
 129
 130static inline int ibmveth_rxq_buffer_valid(struct ibmveth_adapter *adapter)
 131{
 132        return ibmveth_rxq_flags(adapter) & IBMVETH_RXQ_VALID;
 133}
 134
 135static inline int ibmveth_rxq_frame_offset(struct ibmveth_adapter *adapter)
 136{
 137        return ibmveth_rxq_flags(adapter) & IBMVETH_RXQ_OFF_MASK;
 138}
 139
 140static inline int ibmveth_rxq_frame_length(struct ibmveth_adapter *adapter)
 141{
 142        return be32_to_cpu(adapter->rx_queue.queue_addr[adapter->rx_queue.index].length);
 143}
 144
 145static inline int ibmveth_rxq_csum_good(struct ibmveth_adapter *adapter)
 146{
 147        return ibmveth_rxq_flags(adapter) & IBMVETH_RXQ_CSUM_GOOD;
 148}
 149
 150/* setup the initial settings for a buffer pool */
 151static void ibmveth_init_buffer_pool(struct ibmveth_buff_pool *pool,
 152                                     u32 pool_index, u32 pool_size,
 153                                     u32 buff_size, u32 pool_active)
 154{
 155        pool->size = pool_size;
 156        pool->index = pool_index;
 157        pool->buff_size = buff_size;
 158        pool->threshold = pool_size * 7 / 8;
 159        pool->active = pool_active;
 160}
 161
 162/* allocate and setup an buffer pool - called during open */
 163static int ibmveth_alloc_buffer_pool(struct ibmveth_buff_pool *pool)
 164{
 165        int i;
 166
 167        pool->free_map = kmalloc(sizeof(u16) * pool->size, GFP_KERNEL);
 168
 169        if (!pool->free_map)
 170                return -1;
 171
 172        pool->dma_addr = kcalloc(pool->size, sizeof(dma_addr_t), GFP_KERNEL);
 173        if (!pool->dma_addr) {
 174                kfree(pool->free_map);
 175                pool->free_map = NULL;
 176                return -1;
 177        }
 178
 179        pool->skbuff = kcalloc(pool->size, sizeof(void *), GFP_KERNEL);
 180
 181        if (!pool->skbuff) {
 182                kfree(pool->dma_addr);
 183                pool->dma_addr = NULL;
 184
 185                kfree(pool->free_map);
 186                pool->free_map = NULL;
 187                return -1;
 188        }
 189
 190        for (i = 0; i < pool->size; ++i)
 191                pool->free_map[i] = i;
 192
 193        atomic_set(&pool->available, 0);
 194        pool->producer_index = 0;
 195        pool->consumer_index = 0;
 196
 197        return 0;
 198}
 199
 200static inline void ibmveth_flush_buffer(void *addr, unsigned long length)
 201{
 202        unsigned long offset;
 203
 204        for (offset = 0; offset < length; offset += SMP_CACHE_BYTES)
 205                asm("dcbfl %0,%1" :: "b" (addr), "r" (offset));
 206}
 207
 208/* replenish the buffers for a pool.  note that we don't need to
 209 * skb_reserve these since they are used for incoming...
 210 */
 211static void ibmveth_replenish_buffer_pool(struct ibmveth_adapter *adapter,
 212                                          struct ibmveth_buff_pool *pool)
 213{
 214        u32 i;
 215        u32 count = pool->size - atomic_read(&pool->available);
 216        u32 buffers_added = 0;
 217        struct sk_buff *skb;
 218        unsigned int free_index, index;
 219        u64 correlator;
 220        unsigned long lpar_rc;
 221        dma_addr_t dma_addr;
 222
 223        mb();
 224
 225        for (i = 0; i < count; ++i) {
 226                union ibmveth_buf_desc desc;
 227
 228                skb = netdev_alloc_skb(adapter->netdev, pool->buff_size);
 229
 230                if (!skb) {
 231                        netdev_dbg(adapter->netdev,
 232                                   "replenish: unable to allocate skb\n");
 233                        adapter->replenish_no_mem++;
 234                        break;
 235                }
 236
 237                free_index = pool->consumer_index;
 238                pool->consumer_index++;
 239                if (pool->consumer_index >= pool->size)
 240                        pool->consumer_index = 0;
 241                index = pool->free_map[free_index];
 242
 243                BUG_ON(index == IBM_VETH_INVALID_MAP);
 244                BUG_ON(pool->skbuff[index] != NULL);
 245
 246                dma_addr = dma_map_single(&adapter->vdev->dev, skb->data,
 247                                pool->buff_size, DMA_FROM_DEVICE);
 248
 249                if (dma_mapping_error(&adapter->vdev->dev, dma_addr))
 250                        goto failure;
 251
 252                pool->free_map[free_index] = IBM_VETH_INVALID_MAP;
 253                pool->dma_addr[index] = dma_addr;
 254                pool->skbuff[index] = skb;
 255
 256                correlator = ((u64)pool->index << 32) | index;
 257                *(u64 *)skb->data = correlator;
 258
 259                desc.fields.flags_len = IBMVETH_BUF_VALID | pool->buff_size;
 260                desc.fields.address = dma_addr;
 261
 262                if (rx_flush) {
 263                        unsigned int len = min(pool->buff_size,
 264                                                adapter->netdev->mtu +
 265                                                IBMVETH_BUFF_OH);
 266                        ibmveth_flush_buffer(skb->data, len);
 267                }
 268                lpar_rc = h_add_logical_lan_buffer(adapter->vdev->unit_address,
 269                                                   desc.desc);
 270
 271                if (lpar_rc != H_SUCCESS) {
 272                        goto failure;
 273                } else {
 274                        buffers_added++;
 275                        adapter->replenish_add_buff_success++;
 276                }
 277        }
 278
 279        mb();
 280        atomic_add(buffers_added, &(pool->available));
 281        return;
 282
 283failure:
 284        pool->free_map[free_index] = index;
 285        pool->skbuff[index] = NULL;
 286        if (pool->consumer_index == 0)
 287                pool->consumer_index = pool->size - 1;
 288        else
 289                pool->consumer_index--;
 290        if (!dma_mapping_error(&adapter->vdev->dev, dma_addr))
 291                dma_unmap_single(&adapter->vdev->dev,
 292                                 pool->dma_addr[index], pool->buff_size,
 293                                 DMA_FROM_DEVICE);
 294        dev_kfree_skb_any(skb);
 295        adapter->replenish_add_buff_failure++;
 296
 297        mb();
 298        atomic_add(buffers_added, &(pool->available));
 299}
 300
 301/*
 302 * The final 8 bytes of the buffer list is a counter of frames dropped
 303 * because there was not a buffer in the buffer list capable of holding
 304 * the frame.
 305 */
 306static void ibmveth_update_rx_no_buffer(struct ibmveth_adapter *adapter)
 307{
 308        __be64 *p = adapter->buffer_list_addr + 4096 - 8;
 309
 310        adapter->rx_no_buffer = be64_to_cpup(p);
 311}
 312
 313/* replenish routine */
 314static void ibmveth_replenish_task(struct ibmveth_adapter *adapter)
 315{
 316        int i;
 317
 318        adapter->replenish_task_cycles++;
 319
 320        for (i = (IBMVETH_NUM_BUFF_POOLS - 1); i >= 0; i--) {
 321                struct ibmveth_buff_pool *pool = &adapter->rx_buff_pool[i];
 322
 323                if (pool->active &&
 324                    (atomic_read(&pool->available) < pool->threshold))
 325                        ibmveth_replenish_buffer_pool(adapter, pool);
 326        }
 327
 328        ibmveth_update_rx_no_buffer(adapter);
 329}
 330
 331/* empty and free ana buffer pool - also used to do cleanup in error paths */
 332static void ibmveth_free_buffer_pool(struct ibmveth_adapter *adapter,
 333                                     struct ibmveth_buff_pool *pool)
 334{
 335        int i;
 336
 337        kfree(pool->free_map);
 338        pool->free_map = NULL;
 339
 340        if (pool->skbuff && pool->dma_addr) {
 341                for (i = 0; i < pool->size; ++i) {
 342                        struct sk_buff *skb = pool->skbuff[i];
 343                        if (skb) {
 344                                dma_unmap_single(&adapter->vdev->dev,
 345                                                 pool->dma_addr[i],
 346                                                 pool->buff_size,
 347                                                 DMA_FROM_DEVICE);
 348                                dev_kfree_skb_any(skb);
 349                                pool->skbuff[i] = NULL;
 350                        }
 351                }
 352        }
 353
 354        if (pool->dma_addr) {
 355                kfree(pool->dma_addr);
 356                pool->dma_addr = NULL;
 357        }
 358
 359        if (pool->skbuff) {
 360                kfree(pool->skbuff);
 361                pool->skbuff = NULL;
 362        }
 363}
 364
 365/* remove a buffer from a pool */
 366static void ibmveth_remove_buffer_from_pool(struct ibmveth_adapter *adapter,
 367                                            u64 correlator)
 368{
 369        unsigned int pool  = correlator >> 32;
 370        unsigned int index = correlator & 0xffffffffUL;
 371        unsigned int free_index;
 372        struct sk_buff *skb;
 373
 374        BUG_ON(pool >= IBMVETH_NUM_BUFF_POOLS);
 375        BUG_ON(index >= adapter->rx_buff_pool[pool].size);
 376
 377        skb = adapter->rx_buff_pool[pool].skbuff[index];
 378
 379        BUG_ON(skb == NULL);
 380
 381        adapter->rx_buff_pool[pool].skbuff[index] = NULL;
 382
 383        dma_unmap_single(&adapter->vdev->dev,
 384                         adapter->rx_buff_pool[pool].dma_addr[index],
 385                         adapter->rx_buff_pool[pool].buff_size,
 386                         DMA_FROM_DEVICE);
 387
 388        free_index = adapter->rx_buff_pool[pool].producer_index;
 389        adapter->rx_buff_pool[pool].producer_index++;
 390        if (adapter->rx_buff_pool[pool].producer_index >=
 391            adapter->rx_buff_pool[pool].size)
 392                adapter->rx_buff_pool[pool].producer_index = 0;
 393        adapter->rx_buff_pool[pool].free_map[free_index] = index;
 394
 395        mb();
 396
 397        atomic_dec(&(adapter->rx_buff_pool[pool].available));
 398}
 399
 400/* get the current buffer on the rx queue */
 401static inline struct sk_buff *ibmveth_rxq_get_buffer(struct ibmveth_adapter *adapter)
 402{
 403        u64 correlator = adapter->rx_queue.queue_addr[adapter->rx_queue.index].correlator;
 404        unsigned int pool = correlator >> 32;
 405        unsigned int index = correlator & 0xffffffffUL;
 406
 407        BUG_ON(pool >= IBMVETH_NUM_BUFF_POOLS);
 408        BUG_ON(index >= adapter->rx_buff_pool[pool].size);
 409
 410        return adapter->rx_buff_pool[pool].skbuff[index];
 411}
 412
 413/* recycle the current buffer on the rx queue */
 414static int ibmveth_rxq_recycle_buffer(struct ibmveth_adapter *adapter)
 415{
 416        u32 q_index = adapter->rx_queue.index;
 417        u64 correlator = adapter->rx_queue.queue_addr[q_index].correlator;
 418        unsigned int pool = correlator >> 32;
 419        unsigned int index = correlator & 0xffffffffUL;
 420        union ibmveth_buf_desc desc;
 421        unsigned long lpar_rc;
 422        int ret = 1;
 423
 424        BUG_ON(pool >= IBMVETH_NUM_BUFF_POOLS);
 425        BUG_ON(index >= adapter->rx_buff_pool[pool].size);
 426
 427        if (!adapter->rx_buff_pool[pool].active) {
 428                ibmveth_rxq_harvest_buffer(adapter);
 429                ibmveth_free_buffer_pool(adapter, &adapter->rx_buff_pool[pool]);
 430                goto out;
 431        }
 432
 433        desc.fields.flags_len = IBMVETH_BUF_VALID |
 434                adapter->rx_buff_pool[pool].buff_size;
 435        desc.fields.address = adapter->rx_buff_pool[pool].dma_addr[index];
 436
 437        lpar_rc = h_add_logical_lan_buffer(adapter->vdev->unit_address, desc.desc);
 438
 439        if (lpar_rc != H_SUCCESS) {
 440                netdev_dbg(adapter->netdev, "h_add_logical_lan_buffer failed "
 441                           "during recycle rc=%ld", lpar_rc);
 442                ibmveth_remove_buffer_from_pool(adapter, adapter->rx_queue.queue_addr[adapter->rx_queue.index].correlator);
 443                ret = 0;
 444        }
 445
 446        if (++adapter->rx_queue.index == adapter->rx_queue.num_slots) {
 447                adapter->rx_queue.index = 0;
 448                adapter->rx_queue.toggle = !adapter->rx_queue.toggle;
 449        }
 450
 451out:
 452        return ret;
 453}
 454
 455static void ibmveth_rxq_harvest_buffer(struct ibmveth_adapter *adapter)
 456{
 457        ibmveth_remove_buffer_from_pool(adapter, adapter->rx_queue.queue_addr[adapter->rx_queue.index].correlator);
 458
 459        if (++adapter->rx_queue.index == adapter->rx_queue.num_slots) {
 460                adapter->rx_queue.index = 0;
 461                adapter->rx_queue.toggle = !adapter->rx_queue.toggle;
 462        }
 463}
 464
 465static void ibmveth_cleanup(struct ibmveth_adapter *adapter)
 466{
 467        int i;
 468        struct device *dev = &adapter->vdev->dev;
 469
 470        if (adapter->buffer_list_addr != NULL) {
 471                if (!dma_mapping_error(dev, adapter->buffer_list_dma)) {
 472                        dma_unmap_single(dev, adapter->buffer_list_dma, 4096,
 473                                        DMA_BIDIRECTIONAL);
 474                        adapter->buffer_list_dma = DMA_ERROR_CODE;
 475                }
 476                free_page((unsigned long)adapter->buffer_list_addr);
 477                adapter->buffer_list_addr = NULL;
 478        }
 479
 480        if (adapter->filter_list_addr != NULL) {
 481                if (!dma_mapping_error(dev, adapter->filter_list_dma)) {
 482                        dma_unmap_single(dev, adapter->filter_list_dma, 4096,
 483                                        DMA_BIDIRECTIONAL);
 484                        adapter->filter_list_dma = DMA_ERROR_CODE;
 485                }
 486                free_page((unsigned long)adapter->filter_list_addr);
 487                adapter->filter_list_addr = NULL;
 488        }
 489
 490        if (adapter->rx_queue.queue_addr != NULL) {
 491                dma_free_coherent(dev, adapter->rx_queue.queue_len,
 492                                  adapter->rx_queue.queue_addr,
 493                                  adapter->rx_queue.queue_dma);
 494                adapter->rx_queue.queue_addr = NULL;
 495        }
 496
 497        for (i = 0; i < IBMVETH_NUM_BUFF_POOLS; i++)
 498                if (adapter->rx_buff_pool[i].active)
 499                        ibmveth_free_buffer_pool(adapter,
 500                                                 &adapter->rx_buff_pool[i]);
 501
 502        if (adapter->bounce_buffer != NULL) {
 503                if (!dma_mapping_error(dev, adapter->bounce_buffer_dma)) {
 504                        dma_unmap_single(&adapter->vdev->dev,
 505                                        adapter->bounce_buffer_dma,
 506                                        adapter->netdev->mtu + IBMVETH_BUFF_OH,
 507                                        DMA_BIDIRECTIONAL);
 508                        adapter->bounce_buffer_dma = DMA_ERROR_CODE;
 509                }
 510                kfree(adapter->bounce_buffer);
 511                adapter->bounce_buffer = NULL;
 512        }
 513}
 514
 515static int ibmveth_register_logical_lan(struct ibmveth_adapter *adapter,
 516        union ibmveth_buf_desc rxq_desc, u64 mac_address)
 517{
 518        int rc, try_again = 1;
 519
 520        /*
 521         * After a kexec the adapter will still be open, so our attempt to
 522         * open it will fail. So if we get a failure we free the adapter and
 523         * try again, but only once.
 524         */
 525retry:
 526        rc = h_register_logical_lan(adapter->vdev->unit_address,
 527                                    adapter->buffer_list_dma, rxq_desc.desc,
 528                                    adapter->filter_list_dma, mac_address);
 529
 530        if (rc != H_SUCCESS && try_again) {
 531                do {
 532                        rc = h_free_logical_lan(adapter->vdev->unit_address);
 533                } while (H_IS_LONG_BUSY(rc) || (rc == H_BUSY));
 534
 535                try_again = 0;
 536                goto retry;
 537        }
 538
 539        return rc;
 540}
 541
 542static u64 ibmveth_encode_mac_addr(u8 *mac)
 543{
 544        int i;
 545        u64 encoded = 0;
 546
 547        for (i = 0; i < ETH_ALEN; i++)
 548                encoded = (encoded << 8) | mac[i];
 549
 550        return encoded;
 551}
 552
 553static int ibmveth_open(struct net_device *netdev)
 554{
 555        struct ibmveth_adapter *adapter = netdev_priv(netdev);
 556        u64 mac_address;
 557        int rxq_entries = 1;
 558        unsigned long lpar_rc;
 559        int rc;
 560        union ibmveth_buf_desc rxq_desc;
 561        int i;
 562        struct device *dev;
 563
 564        netdev_dbg(netdev, "open starting\n");
 565
 566        napi_enable(&adapter->napi);
 567
 568        for(i = 0; i < IBMVETH_NUM_BUFF_POOLS; i++)
 569                rxq_entries += adapter->rx_buff_pool[i].size;
 570
 571        adapter->buffer_list_addr = (void*) get_zeroed_page(GFP_KERNEL);
 572        adapter->filter_list_addr = (void*) get_zeroed_page(GFP_KERNEL);
 573
 574        if (!adapter->buffer_list_addr || !adapter->filter_list_addr) {
 575                netdev_err(netdev, "unable to allocate filter or buffer list "
 576                           "pages\n");
 577                rc = -ENOMEM;
 578                goto err_out;
 579        }
 580
 581        dev = &adapter->vdev->dev;
 582
 583        adapter->rx_queue.queue_len = sizeof(struct ibmveth_rx_q_entry) *
 584                                                rxq_entries;
 585        adapter->rx_queue.queue_addr =
 586                dma_alloc_coherent(dev, adapter->rx_queue.queue_len,
 587                                   &adapter->rx_queue.queue_dma, GFP_KERNEL);
 588        if (!adapter->rx_queue.queue_addr) {
 589                rc = -ENOMEM;
 590                goto err_out;
 591        }
 592
 593        adapter->buffer_list_dma = dma_map_single(dev,
 594                        adapter->buffer_list_addr, 4096, DMA_BIDIRECTIONAL);
 595        adapter->filter_list_dma = dma_map_single(dev,
 596                        adapter->filter_list_addr, 4096, DMA_BIDIRECTIONAL);
 597
 598        if ((dma_mapping_error(dev, adapter->buffer_list_dma)) ||
 599            (dma_mapping_error(dev, adapter->filter_list_dma))) {
 600                netdev_err(netdev, "unable to map filter or buffer list "
 601                           "pages\n");
 602                rc = -ENOMEM;
 603                goto err_out;
 604        }
 605
 606        adapter->rx_queue.index = 0;
 607        adapter->rx_queue.num_slots = rxq_entries;
 608        adapter->rx_queue.toggle = 1;
 609
 610        mac_address = ibmveth_encode_mac_addr(netdev->dev_addr);
 611
 612        rxq_desc.fields.flags_len = IBMVETH_BUF_VALID |
 613                                        adapter->rx_queue.queue_len;
 614        rxq_desc.fields.address = adapter->rx_queue.queue_dma;
 615
 616        netdev_dbg(netdev, "buffer list @ 0x%p\n", adapter->buffer_list_addr);
 617        netdev_dbg(netdev, "filter list @ 0x%p\n", adapter->filter_list_addr);
 618        netdev_dbg(netdev, "receive q   @ 0x%p\n", adapter->rx_queue.queue_addr);
 619
 620        h_vio_signal(adapter->vdev->unit_address, VIO_IRQ_DISABLE);
 621
 622        lpar_rc = ibmveth_register_logical_lan(adapter, rxq_desc, mac_address);
 623
 624        if (lpar_rc != H_SUCCESS) {
 625                netdev_err(netdev, "h_register_logical_lan failed with %ld\n",
 626                           lpar_rc);
 627                netdev_err(netdev, "buffer TCE:0x%llx filter TCE:0x%llx rxq "
 628                           "desc:0x%llx MAC:0x%llx\n",
 629                                     adapter->buffer_list_dma,
 630                                     adapter->filter_list_dma,
 631                                     rxq_desc.desc,
 632                                     mac_address);
 633                rc = -ENONET;
 634                goto err_out;
 635        }
 636
 637        for (i = 0; i < IBMVETH_NUM_BUFF_POOLS; i++) {
 638                if (!adapter->rx_buff_pool[i].active)
 639                        continue;
 640                if (ibmveth_alloc_buffer_pool(&adapter->rx_buff_pool[i])) {
 641                        netdev_err(netdev, "unable to alloc pool\n");
 642                        adapter->rx_buff_pool[i].active = 0;
 643                        rc = -ENOMEM;
 644                        goto err_out;
 645                }
 646        }
 647
 648        netdev_dbg(netdev, "registering irq 0x%x\n", netdev->irq);
 649        rc = request_irq(netdev->irq, ibmveth_interrupt, 0, netdev->name,
 650                         netdev);
 651        if (rc != 0) {
 652                netdev_err(netdev, "unable to request irq 0x%x, rc %d\n",
 653                           netdev->irq, rc);
 654                do {
 655                        lpar_rc = h_free_logical_lan(adapter->vdev->unit_address);
 656                } while (H_IS_LONG_BUSY(lpar_rc) || (lpar_rc == H_BUSY));
 657
 658                goto err_out;
 659        }
 660
 661        adapter->bounce_buffer =
 662            kmalloc(netdev->mtu + IBMVETH_BUFF_OH, GFP_KERNEL);
 663        if (!adapter->bounce_buffer) {
 664                rc = -ENOMEM;
 665                goto err_out_free_irq;
 666        }
 667        adapter->bounce_buffer_dma =
 668            dma_map_single(&adapter->vdev->dev, adapter->bounce_buffer,
 669                           netdev->mtu + IBMVETH_BUFF_OH, DMA_BIDIRECTIONAL);
 670        if (dma_mapping_error(dev, adapter->bounce_buffer_dma)) {
 671                netdev_err(netdev, "unable to map bounce buffer\n");
 672                rc = -ENOMEM;
 673                goto err_out_free_irq;
 674        }
 675
 676        netdev_dbg(netdev, "initial replenish cycle\n");
 677        ibmveth_interrupt(netdev->irq, netdev);
 678
 679        netif_start_queue(netdev);
 680
 681        netdev_dbg(netdev, "open complete\n");
 682
 683        return 0;
 684
 685err_out_free_irq:
 686        free_irq(netdev->irq, netdev);
 687err_out:
 688        ibmveth_cleanup(adapter);
 689        napi_disable(&adapter->napi);
 690        return rc;
 691}
 692
 693static int ibmveth_close(struct net_device *netdev)
 694{
 695        struct ibmveth_adapter *adapter = netdev_priv(netdev);
 696        long lpar_rc;
 697
 698        netdev_dbg(netdev, "close starting\n");
 699
 700        napi_disable(&adapter->napi);
 701
 702        if (!adapter->pool_config)
 703                netif_stop_queue(netdev);
 704
 705        h_vio_signal(adapter->vdev->unit_address, VIO_IRQ_DISABLE);
 706
 707        do {
 708                lpar_rc = h_free_logical_lan(adapter->vdev->unit_address);
 709        } while (H_IS_LONG_BUSY(lpar_rc) || (lpar_rc == H_BUSY));
 710
 711        if (lpar_rc != H_SUCCESS) {
 712                netdev_err(netdev, "h_free_logical_lan failed with %lx, "
 713                           "continuing with close\n", lpar_rc);
 714        }
 715
 716        free_irq(netdev->irq, netdev);
 717
 718        ibmveth_update_rx_no_buffer(adapter);
 719
 720        ibmveth_cleanup(adapter);
 721
 722        netdev_dbg(netdev, "close complete\n");
 723
 724        return 0;
 725}
 726
 727static int netdev_get_settings(struct net_device *dev, struct ethtool_cmd *cmd)
 728{
 729        cmd->supported = (SUPPORTED_1000baseT_Full | SUPPORTED_Autoneg |
 730                                SUPPORTED_FIBRE);
 731        cmd->advertising = (ADVERTISED_1000baseT_Full | ADVERTISED_Autoneg |
 732                                ADVERTISED_FIBRE);
 733        ethtool_cmd_speed_set(cmd, SPEED_1000);
 734        cmd->duplex = DUPLEX_FULL;
 735        cmd->port = PORT_FIBRE;
 736        cmd->phy_address = 0;
 737        cmd->transceiver = XCVR_INTERNAL;
 738        cmd->autoneg = AUTONEG_ENABLE;
 739        cmd->maxtxpkt = 0;
 740        cmd->maxrxpkt = 1;
 741        return 0;
 742}
 743
 744static void netdev_get_drvinfo(struct net_device *dev,
 745                               struct ethtool_drvinfo *info)
 746{
 747        strlcpy(info->driver, ibmveth_driver_name, sizeof(info->driver));
 748        strlcpy(info->version, ibmveth_driver_version, sizeof(info->version));
 749}
 750
 751static netdev_features_t ibmveth_fix_features(struct net_device *dev,
 752        netdev_features_t features)
 753{
 754        /*
 755         * Since the ibmveth firmware interface does not have the
 756         * concept of separate tx/rx checksum offload enable, if rx
 757         * checksum is disabled we also have to disable tx checksum
 758         * offload. Once we disable rx checksum offload, we are no
 759         * longer allowed to send tx buffers that are not properly
 760         * checksummed.
 761         */
 762
 763        if (!(features & NETIF_F_RXCSUM))
 764                features &= ~NETIF_F_CSUM_MASK;
 765
 766        return features;
 767}
 768
 769static int ibmveth_set_csum_offload(struct net_device *dev, u32 data)
 770{
 771        struct ibmveth_adapter *adapter = netdev_priv(dev);
 772        unsigned long set_attr, clr_attr, ret_attr;
 773        unsigned long set_attr6, clr_attr6;
 774        long ret, ret4, ret6;
 775        int rc1 = 0, rc2 = 0;
 776        int restart = 0;
 777
 778        if (netif_running(dev)) {
 779                restart = 1;
 780                adapter->pool_config = 1;
 781                ibmveth_close(dev);
 782                adapter->pool_config = 0;
 783        }
 784
 785        set_attr = 0;
 786        clr_attr = 0;
 787        set_attr6 = 0;
 788        clr_attr6 = 0;
 789
 790        if (data) {
 791                set_attr = IBMVETH_ILLAN_IPV4_TCP_CSUM;
 792                set_attr6 = IBMVETH_ILLAN_IPV6_TCP_CSUM;
 793        } else {
 794                clr_attr = IBMVETH_ILLAN_IPV4_TCP_CSUM;
 795                clr_attr6 = IBMVETH_ILLAN_IPV6_TCP_CSUM;
 796        }
 797
 798        ret = h_illan_attributes(adapter->vdev->unit_address, 0, 0, &ret_attr);
 799
 800        if (ret == H_SUCCESS && !(ret_attr & IBMVETH_ILLAN_ACTIVE_TRUNK) &&
 801            !(ret_attr & IBMVETH_ILLAN_TRUNK_PRI_MASK) &&
 802            (ret_attr & IBMVETH_ILLAN_PADDED_PKT_CSUM)) {
 803                ret4 = h_illan_attributes(adapter->vdev->unit_address, clr_attr,
 804                                         set_attr, &ret_attr);
 805
 806                if (ret4 != H_SUCCESS) {
 807                        netdev_err(dev, "unable to change IPv4 checksum "
 808                                        "offload settings. %d rc=%ld\n",
 809                                        data, ret4);
 810
 811                        h_illan_attributes(adapter->vdev->unit_address,
 812                                           set_attr, clr_attr, &ret_attr);
 813
 814                        if (data == 1)
 815                                dev->features &= ~NETIF_F_IP_CSUM;
 816
 817                } else {
 818                        adapter->fw_ipv4_csum_support = data;
 819                }
 820
 821                ret6 = h_illan_attributes(adapter->vdev->unit_address,
 822                                         clr_attr6, set_attr6, &ret_attr);
 823
 824                if (ret6 != H_SUCCESS) {
 825                        netdev_err(dev, "unable to change IPv6 checksum "
 826                                        "offload settings. %d rc=%ld\n",
 827                                        data, ret6);
 828
 829                        h_illan_attributes(adapter->vdev->unit_address,
 830                                           set_attr6, clr_attr6, &ret_attr);
 831
 832                        if (data == 1)
 833                                dev->features &= ~NETIF_F_IPV6_CSUM;
 834
 835                } else
 836                        adapter->fw_ipv6_csum_support = data;
 837
 838                if (ret4 == H_SUCCESS || ret6 == H_SUCCESS)
 839                        adapter->rx_csum = data;
 840                else
 841                        rc1 = -EIO;
 842        } else {
 843                rc1 = -EIO;
 844                netdev_err(dev, "unable to change checksum offload settings."
 845                                     " %d rc=%ld ret_attr=%lx\n", data, ret,
 846                                     ret_attr);
 847        }
 848
 849        if (restart)
 850                rc2 = ibmveth_open(dev);
 851
 852        return rc1 ? rc1 : rc2;
 853}
 854
 855static int ibmveth_set_tso(struct net_device *dev, u32 data)
 856{
 857        struct ibmveth_adapter *adapter = netdev_priv(dev);
 858        unsigned long set_attr, clr_attr, ret_attr;
 859        long ret1, ret2;
 860        int rc1 = 0, rc2 = 0;
 861        int restart = 0;
 862
 863        if (netif_running(dev)) {
 864                restart = 1;
 865                adapter->pool_config = 1;
 866                ibmveth_close(dev);
 867                adapter->pool_config = 0;
 868        }
 869
 870        set_attr = 0;
 871        clr_attr = 0;
 872
 873        if (data)
 874                set_attr = IBMVETH_ILLAN_LRG_SR_ENABLED;
 875        else
 876                clr_attr = IBMVETH_ILLAN_LRG_SR_ENABLED;
 877
 878        ret1 = h_illan_attributes(adapter->vdev->unit_address, 0, 0, &ret_attr);
 879
 880        if (ret1 == H_SUCCESS && (ret_attr & IBMVETH_ILLAN_LRG_SND_SUPPORT) &&
 881            !old_large_send) {
 882                ret2 = h_illan_attributes(adapter->vdev->unit_address, clr_attr,
 883                                          set_attr, &ret_attr);
 884
 885                if (ret2 != H_SUCCESS) {
 886                        netdev_err(dev, "unable to change tso settings. %d rc=%ld\n",
 887                                   data, ret2);
 888
 889                        h_illan_attributes(adapter->vdev->unit_address,
 890                                           set_attr, clr_attr, &ret_attr);
 891
 892                        if (data == 1)
 893                                dev->features &= ~(NETIF_F_TSO | NETIF_F_TSO6);
 894                        rc1 = -EIO;
 895
 896                } else {
 897                        adapter->fw_large_send_support = data;
 898                        adapter->large_send = data;
 899                }
 900        } else {
 901                /* Older firmware version of large send offload does not
 902                 * support tcp6/ipv6
 903                 */
 904                if (data == 1) {
 905                        dev->features &= ~NETIF_F_TSO6;
 906                        netdev_info(dev, "TSO feature requires all partitions to have updated driver");
 907                }
 908                adapter->large_send = data;
 909        }
 910
 911        if (restart)
 912                rc2 = ibmveth_open(dev);
 913
 914        return rc1 ? rc1 : rc2;
 915}
 916
 917static int ibmveth_set_features(struct net_device *dev,
 918        netdev_features_t features)
 919{
 920        struct ibmveth_adapter *adapter = netdev_priv(dev);
 921        int rx_csum = !!(features & NETIF_F_RXCSUM);
 922        int large_send = !!(features & (NETIF_F_TSO | NETIF_F_TSO6));
 923        int rc1 = 0, rc2 = 0;
 924
 925        if (rx_csum != adapter->rx_csum) {
 926                rc1 = ibmveth_set_csum_offload(dev, rx_csum);
 927                if (rc1 && !adapter->rx_csum)
 928                        dev->features =
 929                                features & ~(NETIF_F_CSUM_MASK |
 930                                             NETIF_F_RXCSUM);
 931        }
 932
 933        if (large_send != adapter->large_send) {
 934                rc2 = ibmveth_set_tso(dev, large_send);
 935                if (rc2 && !adapter->large_send)
 936                        dev->features =
 937                                features & ~(NETIF_F_TSO | NETIF_F_TSO6);
 938        }
 939
 940        return rc1 ? rc1 : rc2;
 941}
 942
 943static void ibmveth_get_strings(struct net_device *dev, u32 stringset, u8 *data)
 944{
 945        int i;
 946
 947        if (stringset != ETH_SS_STATS)
 948                return;
 949
 950        for (i = 0; i < ARRAY_SIZE(ibmveth_stats); i++, data += ETH_GSTRING_LEN)
 951                memcpy(data, ibmveth_stats[i].name, ETH_GSTRING_LEN);
 952}
 953
 954static int ibmveth_get_sset_count(struct net_device *dev, int sset)
 955{
 956        switch (sset) {
 957        case ETH_SS_STATS:
 958                return ARRAY_SIZE(ibmveth_stats);
 959        default:
 960                return -EOPNOTSUPP;
 961        }
 962}
 963
 964static void ibmveth_get_ethtool_stats(struct net_device *dev,
 965                                      struct ethtool_stats *stats, u64 *data)
 966{
 967        int i;
 968        struct ibmveth_adapter *adapter = netdev_priv(dev);
 969
 970        for (i = 0; i < ARRAY_SIZE(ibmveth_stats); i++)
 971                data[i] = IBMVETH_GET_STAT(adapter, ibmveth_stats[i].offset);
 972}
 973
 974static const struct ethtool_ops netdev_ethtool_ops = {
 975        .get_drvinfo            = netdev_get_drvinfo,
 976        .get_settings           = netdev_get_settings,
 977        .get_link               = ethtool_op_get_link,
 978        .get_strings            = ibmveth_get_strings,
 979        .get_sset_count         = ibmveth_get_sset_count,
 980        .get_ethtool_stats      = ibmveth_get_ethtool_stats,
 981};
 982
 983static int ibmveth_ioctl(struct net_device *dev, struct ifreq *ifr, int cmd)
 984{
 985        return -EOPNOTSUPP;
 986}
 987
 988#define page_offset(v) ((unsigned long)(v) & ((1 << 12) - 1))
 989
 990static int ibmveth_send(struct ibmveth_adapter *adapter,
 991                        union ibmveth_buf_desc *descs, unsigned long mss)
 992{
 993        unsigned long correlator;
 994        unsigned int retry_count;
 995        unsigned long ret;
 996
 997        /*
 998         * The retry count sets a maximum for the number of broadcast and
 999         * multicast destinations within the system.
1000         */
1001        retry_count = 1024;
1002        correlator = 0;
1003        do {
1004                ret = h_send_logical_lan(adapter->vdev->unit_address,
1005                                             descs[0].desc, descs[1].desc,
1006                                             descs[2].desc, descs[3].desc,
1007                                             descs[4].desc, descs[5].desc,
1008                                             correlator, &correlator, mss,
1009                                             adapter->fw_large_send_support);
1010        } while ((ret == H_BUSY) && (retry_count--));
1011
1012        if (ret != H_SUCCESS && ret != H_DROPPED) {
1013                netdev_err(adapter->netdev, "tx: h_send_logical_lan failed "
1014                           "with rc=%ld\n", ret);
1015                return 1;
1016        }
1017
1018        return 0;
1019}
1020
1021static netdev_tx_t ibmveth_start_xmit(struct sk_buff *skb,
1022                                      struct net_device *netdev)
1023{
1024        struct ibmveth_adapter *adapter = netdev_priv(netdev);
1025        unsigned int desc_flags;
1026        union ibmveth_buf_desc descs[6];
1027        int last, i;
1028        int force_bounce = 0;
1029        dma_addr_t dma_addr;
1030        unsigned long mss = 0;
1031
1032        /*
1033         * veth handles a maximum of 6 segments including the header, so
1034         * we have to linearize the skb if there are more than this.
1035         */
1036        if (skb_shinfo(skb)->nr_frags > 5 && __skb_linearize(skb)) {
1037                netdev->stats.tx_dropped++;
1038                goto out;
1039        }
1040
1041        /* veth can't checksum offload UDP */
1042        if (skb->ip_summed == CHECKSUM_PARTIAL &&
1043            ((skb->protocol == htons(ETH_P_IP) &&
1044              ip_hdr(skb)->protocol != IPPROTO_TCP) ||
1045             (skb->protocol == htons(ETH_P_IPV6) &&
1046              ipv6_hdr(skb)->nexthdr != IPPROTO_TCP)) &&
1047            skb_checksum_help(skb)) {
1048
1049                netdev_err(netdev, "tx: failed to checksum packet\n");
1050                netdev->stats.tx_dropped++;
1051                goto out;
1052        }
1053
1054        desc_flags = IBMVETH_BUF_VALID;
1055
1056        if (skb_is_gso(skb) && adapter->fw_large_send_support)
1057                desc_flags |= IBMVETH_BUF_LRG_SND;
1058
1059        if (skb->ip_summed == CHECKSUM_PARTIAL) {
1060                unsigned char *buf = skb_transport_header(skb) +
1061                                                skb->csum_offset;
1062
1063                desc_flags |= (IBMVETH_BUF_NO_CSUM | IBMVETH_BUF_CSUM_GOOD);
1064
1065                /* Need to zero out the checksum */
1066                buf[0] = 0;
1067                buf[1] = 0;
1068        }
1069
1070retry_bounce:
1071        memset(descs, 0, sizeof(descs));
1072
1073        /*
1074         * If a linear packet is below the rx threshold then
1075         * copy it into the static bounce buffer. This avoids the
1076         * cost of a TCE insert and remove.
1077         */
1078        if (force_bounce || (!skb_is_nonlinear(skb) &&
1079                                (skb->len < tx_copybreak))) {
1080                skb_copy_from_linear_data(skb, adapter->bounce_buffer,
1081                                          skb->len);
1082
1083                descs[0].fields.flags_len = desc_flags | skb->len;
1084                descs[0].fields.address = adapter->bounce_buffer_dma;
1085
1086                if (ibmveth_send(adapter, descs, 0)) {
1087                        adapter->tx_send_failed++;
1088                        netdev->stats.tx_dropped++;
1089                } else {
1090                        netdev->stats.tx_packets++;
1091                        netdev->stats.tx_bytes += skb->len;
1092                }
1093
1094                goto out;
1095        }
1096
1097        /* Map the header */
1098        dma_addr = dma_map_single(&adapter->vdev->dev, skb->data,
1099                                  skb_headlen(skb), DMA_TO_DEVICE);
1100        if (dma_mapping_error(&adapter->vdev->dev, dma_addr))
1101                goto map_failed;
1102
1103        descs[0].fields.flags_len = desc_flags | skb_headlen(skb);
1104        descs[0].fields.address = dma_addr;
1105
1106        /* Map the frags */
1107        for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
1108                const skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
1109
1110                dma_addr = skb_frag_dma_map(&adapter->vdev->dev, frag, 0,
1111                                            skb_frag_size(frag), DMA_TO_DEVICE);
1112
1113                if (dma_mapping_error(&adapter->vdev->dev, dma_addr))
1114                        goto map_failed_frags;
1115
1116                descs[i+1].fields.flags_len = desc_flags | skb_frag_size(frag);
1117                descs[i+1].fields.address = dma_addr;
1118        }
1119
1120        if (skb_is_gso(skb)) {
1121                if (adapter->fw_large_send_support) {
1122                        mss = (unsigned long)skb_shinfo(skb)->gso_size;
1123                        adapter->tx_large_packets++;
1124                } else if (!skb_is_gso_v6(skb)) {
1125                        /* Put -1 in the IP checksum to tell phyp it
1126                         * is a largesend packet. Put the mss in
1127                         * the TCP checksum.
1128                         */
1129                        ip_hdr(skb)->check = 0xffff;
1130                        tcp_hdr(skb)->check =
1131                                cpu_to_be16(skb_shinfo(skb)->gso_size);
1132                        adapter->tx_large_packets++;
1133                }
1134        }
1135
1136        if (ibmveth_send(adapter, descs, mss)) {
1137                adapter->tx_send_failed++;
1138                netdev->stats.tx_dropped++;
1139        } else {
1140                netdev->stats.tx_packets++;
1141                netdev->stats.tx_bytes += skb->len;
1142        }
1143
1144        dma_unmap_single(&adapter->vdev->dev,
1145                         descs[0].fields.address,
1146                         descs[0].fields.flags_len & IBMVETH_BUF_LEN_MASK,
1147                         DMA_TO_DEVICE);
1148
1149        for (i = 1; i < skb_shinfo(skb)->nr_frags + 1; i++)
1150                dma_unmap_page(&adapter->vdev->dev, descs[i].fields.address,
1151                               descs[i].fields.flags_len & IBMVETH_BUF_LEN_MASK,
1152                               DMA_TO_DEVICE);
1153
1154out:
1155        dev_consume_skb_any(skb);
1156        return NETDEV_TX_OK;
1157
1158map_failed_frags:
1159        last = i+1;
1160        for (i = 0; i < last; i++)
1161                dma_unmap_page(&adapter->vdev->dev, descs[i].fields.address,
1162                               descs[i].fields.flags_len & IBMVETH_BUF_LEN_MASK,
1163                               DMA_TO_DEVICE);
1164
1165map_failed:
1166        if (!firmware_has_feature(FW_FEATURE_CMO))
1167                netdev_err(netdev, "tx: unable to map xmit buffer\n");
1168        adapter->tx_map_failed++;
1169        if (skb_linearize(skb)) {
1170                netdev->stats.tx_dropped++;
1171                goto out;
1172        }
1173        force_bounce = 1;
1174        goto retry_bounce;
1175}
1176
1177static int ibmveth_poll(struct napi_struct *napi, int budget)
1178{
1179        struct ibmveth_adapter *adapter =
1180                        container_of(napi, struct ibmveth_adapter, napi);
1181        struct net_device *netdev = adapter->netdev;
1182        int frames_processed = 0;
1183        unsigned long lpar_rc;
1184        struct iphdr *iph;
1185
1186restart_poll:
1187        while (frames_processed < budget) {
1188                if (!ibmveth_rxq_pending_buffer(adapter))
1189                        break;
1190
1191                smp_rmb();
1192                if (!ibmveth_rxq_buffer_valid(adapter)) {
1193                        wmb(); /* suggested by larson1 */
1194                        adapter->rx_invalid_buffer++;
1195                        netdev_dbg(netdev, "recycling invalid buffer\n");
1196                        ibmveth_rxq_recycle_buffer(adapter);
1197                } else {
1198                        struct sk_buff *skb, *new_skb;
1199                        int length = ibmveth_rxq_frame_length(adapter);
1200                        int offset = ibmveth_rxq_frame_offset(adapter);
1201                        int csum_good = ibmveth_rxq_csum_good(adapter);
1202
1203                        skb = ibmveth_rxq_get_buffer(adapter);
1204
1205                        new_skb = NULL;
1206                        if (length < rx_copybreak)
1207                                new_skb = netdev_alloc_skb(netdev, length);
1208
1209                        if (new_skb) {
1210                                skb_copy_to_linear_data(new_skb,
1211                                                        skb->data + offset,
1212                                                        length);
1213                                if (rx_flush)
1214                                        ibmveth_flush_buffer(skb->data,
1215                                                length + offset);
1216                                if (!ibmveth_rxq_recycle_buffer(adapter))
1217                                        kfree_skb(skb);
1218                                skb = new_skb;
1219                        } else {
1220                                ibmveth_rxq_harvest_buffer(adapter);
1221                                skb_reserve(skb, offset);
1222                        }
1223
1224                        skb_put(skb, length);
1225                        skb->protocol = eth_type_trans(skb, netdev);
1226
1227                        if (csum_good) {
1228                                skb->ip_summed = CHECKSUM_UNNECESSARY;
1229                                if (be16_to_cpu(skb->protocol) == ETH_P_IP) {
1230                                        iph = (struct iphdr *)skb->data;
1231
1232                                        /* If the IP checksum is not offloaded and if the packet
1233                                         *  is large send, the checksum must be rebuilt.
1234                                         */
1235                                        if (iph->check == 0xffff) {
1236                                                iph->check = 0;
1237                                                iph->check = ip_fast_csum((unsigned char *)iph, iph->ihl);
1238                                                adapter->rx_large_packets++;
1239                                        }
1240                                }
1241                        }
1242
1243                        napi_gro_receive(napi, skb);    /* send it up */
1244
1245                        netdev->stats.rx_packets++;
1246                        netdev->stats.rx_bytes += length;
1247                        frames_processed++;
1248                }
1249        }
1250
1251        ibmveth_replenish_task(adapter);
1252
1253        if (frames_processed < budget) {
1254                napi_complete(napi);
1255
1256                /* We think we are done - reenable interrupts,
1257                 * then check once more to make sure we are done.
1258                 */
1259                lpar_rc = h_vio_signal(adapter->vdev->unit_address,
1260                                       VIO_IRQ_ENABLE);
1261
1262                BUG_ON(lpar_rc != H_SUCCESS);
1263
1264                if (ibmveth_rxq_pending_buffer(adapter) &&
1265                    napi_reschedule(napi)) {
1266                        lpar_rc = h_vio_signal(adapter->vdev->unit_address,
1267                                               VIO_IRQ_DISABLE);
1268                        goto restart_poll;
1269                }
1270        }
1271
1272        return frames_processed;
1273}
1274
1275static irqreturn_t ibmveth_interrupt(int irq, void *dev_instance)
1276{
1277        struct net_device *netdev = dev_instance;
1278        struct ibmveth_adapter *adapter = netdev_priv(netdev);
1279        unsigned long lpar_rc;
1280
1281        if (napi_schedule_prep(&adapter->napi)) {
1282                lpar_rc = h_vio_signal(adapter->vdev->unit_address,
1283                                       VIO_IRQ_DISABLE);
1284                BUG_ON(lpar_rc != H_SUCCESS);
1285                __napi_schedule(&adapter->napi);
1286        }
1287        return IRQ_HANDLED;
1288}
1289
1290static void ibmveth_set_multicast_list(struct net_device *netdev)
1291{
1292        struct ibmveth_adapter *adapter = netdev_priv(netdev);
1293        unsigned long lpar_rc;
1294
1295        if ((netdev->flags & IFF_PROMISC) ||
1296            (netdev_mc_count(netdev) > adapter->mcastFilterSize)) {
1297                lpar_rc = h_multicast_ctrl(adapter->vdev->unit_address,
1298                                           IbmVethMcastEnableRecv |
1299                                           IbmVethMcastDisableFiltering,
1300                                           0);
1301                if (lpar_rc != H_SUCCESS) {
1302                        netdev_err(netdev, "h_multicast_ctrl rc=%ld when "
1303                                   "entering promisc mode\n", lpar_rc);
1304                }
1305        } else {
1306                struct netdev_hw_addr *ha;
1307                /* clear the filter table & disable filtering */
1308                lpar_rc = h_multicast_ctrl(adapter->vdev->unit_address,
1309                                           IbmVethMcastEnableRecv |
1310                                           IbmVethMcastDisableFiltering |
1311                                           IbmVethMcastClearFilterTable,
1312                                           0);
1313                if (lpar_rc != H_SUCCESS) {
1314                        netdev_err(netdev, "h_multicast_ctrl rc=%ld when "
1315                                   "attempting to clear filter table\n",
1316                                   lpar_rc);
1317                }
1318                /* add the addresses to the filter table */
1319                netdev_for_each_mc_addr(ha, netdev) {
1320                        /* add the multicast address to the filter table */
1321                        u64 mcast_addr;
1322                        mcast_addr = ibmveth_encode_mac_addr(ha->addr);
1323                        lpar_rc = h_multicast_ctrl(adapter->vdev->unit_address,
1324                                                   IbmVethMcastAddFilter,
1325                                                   mcast_addr);
1326                        if (lpar_rc != H_SUCCESS) {
1327                                netdev_err(netdev, "h_multicast_ctrl rc=%ld "
1328                                           "when adding an entry to the filter "
1329                                           "table\n", lpar_rc);
1330                        }
1331                }
1332
1333                /* re-enable filtering */
1334                lpar_rc = h_multicast_ctrl(adapter->vdev->unit_address,
1335                                           IbmVethMcastEnableFiltering,
1336                                           0);
1337                if (lpar_rc != H_SUCCESS) {
1338                        netdev_err(netdev, "h_multicast_ctrl rc=%ld when "
1339                                   "enabling filtering\n", lpar_rc);
1340                }
1341        }
1342}
1343
1344static int ibmveth_change_mtu(struct net_device *dev, int new_mtu)
1345{
1346        struct ibmveth_adapter *adapter = netdev_priv(dev);
1347        struct vio_dev *viodev = adapter->vdev;
1348        int new_mtu_oh = new_mtu + IBMVETH_BUFF_OH;
1349        int i, rc;
1350        int need_restart = 0;
1351
1352        if (new_mtu < IBMVETH_MIN_MTU)
1353                return -EINVAL;
1354
1355        for (i = 0; i < IBMVETH_NUM_BUFF_POOLS; i++)
1356                if (new_mtu_oh <= adapter->rx_buff_pool[i].buff_size)
1357                        break;
1358
1359        if (i == IBMVETH_NUM_BUFF_POOLS)
1360                return -EINVAL;
1361
1362        /* Deactivate all the buffer pools so that the next loop can activate
1363           only the buffer pools necessary to hold the new MTU */
1364        if (netif_running(adapter->netdev)) {
1365                need_restart = 1;
1366                adapter->pool_config = 1;
1367                ibmveth_close(adapter->netdev);
1368                adapter->pool_config = 0;
1369        }
1370
1371        /* Look for an active buffer pool that can hold the new MTU */
1372        for (i = 0; i < IBMVETH_NUM_BUFF_POOLS; i++) {
1373                adapter->rx_buff_pool[i].active = 1;
1374
1375                if (new_mtu_oh <= adapter->rx_buff_pool[i].buff_size) {
1376                        dev->mtu = new_mtu;
1377                        vio_cmo_set_dev_desired(viodev,
1378                                                ibmveth_get_desired_dma
1379                                                (viodev));
1380                        if (need_restart) {
1381                                return ibmveth_open(adapter->netdev);
1382                        }
1383                        return 0;
1384                }
1385        }
1386
1387        if (need_restart && (rc = ibmveth_open(adapter->netdev)))
1388                return rc;
1389
1390        return -EINVAL;
1391}
1392
1393#ifdef CONFIG_NET_POLL_CONTROLLER
1394static void ibmveth_poll_controller(struct net_device *dev)
1395{
1396        ibmveth_replenish_task(netdev_priv(dev));
1397        ibmveth_interrupt(dev->irq, dev);
1398}
1399#endif
1400
1401/**
1402 * ibmveth_get_desired_dma - Calculate IO memory desired by the driver
1403 *
1404 * @vdev: struct vio_dev for the device whose desired IO mem is to be returned
1405 *
1406 * Return value:
1407 *      Number of bytes of IO data the driver will need to perform well.
1408 */
1409static unsigned long ibmveth_get_desired_dma(struct vio_dev *vdev)
1410{
1411        struct net_device *netdev = dev_get_drvdata(&vdev->dev);
1412        struct ibmveth_adapter *adapter;
1413        struct iommu_table *tbl;
1414        unsigned long ret;
1415        int i;
1416        int rxqentries = 1;
1417
1418        tbl = get_iommu_table_base(&vdev->dev);
1419
1420        /* netdev inits at probe time along with the structures we need below*/
1421        if (netdev == NULL)
1422                return IOMMU_PAGE_ALIGN(IBMVETH_IO_ENTITLEMENT_DEFAULT, tbl);
1423
1424        adapter = netdev_priv(netdev);
1425
1426        ret = IBMVETH_BUFF_LIST_SIZE + IBMVETH_FILT_LIST_SIZE;
1427        ret += IOMMU_PAGE_ALIGN(netdev->mtu, tbl);
1428
1429        for (i = 0; i < IBMVETH_NUM_BUFF_POOLS; i++) {
1430                /* add the size of the active receive buffers */
1431                if (adapter->rx_buff_pool[i].active)
1432                        ret +=
1433                            adapter->rx_buff_pool[i].size *
1434                            IOMMU_PAGE_ALIGN(adapter->rx_buff_pool[i].
1435                                             buff_size, tbl);
1436                rxqentries += adapter->rx_buff_pool[i].size;
1437        }
1438        /* add the size of the receive queue entries */
1439        ret += IOMMU_PAGE_ALIGN(
1440                rxqentries * sizeof(struct ibmveth_rx_q_entry), tbl);
1441
1442        return ret;
1443}
1444
1445static int ibmveth_set_mac_addr(struct net_device *dev, void *p)
1446{
1447        struct ibmveth_adapter *adapter = netdev_priv(dev);
1448        struct sockaddr *addr = p;
1449        u64 mac_address;
1450        int rc;
1451
1452        if (!is_valid_ether_addr(addr->sa_data))
1453                return -EADDRNOTAVAIL;
1454
1455        mac_address = ibmveth_encode_mac_addr(addr->sa_data);
1456        rc = h_change_logical_lan_mac(adapter->vdev->unit_address, mac_address);
1457        if (rc) {
1458                netdev_err(adapter->netdev, "h_change_logical_lan_mac failed with rc=%d\n", rc);
1459                return rc;
1460        }
1461
1462        ether_addr_copy(dev->dev_addr, addr->sa_data);
1463
1464        return 0;
1465}
1466
1467static const struct net_device_ops ibmveth_netdev_ops = {
1468        .ndo_open               = ibmveth_open,
1469        .ndo_stop               = ibmveth_close,
1470        .ndo_start_xmit         = ibmveth_start_xmit,
1471        .ndo_set_rx_mode        = ibmveth_set_multicast_list,
1472        .ndo_do_ioctl           = ibmveth_ioctl,
1473        .ndo_change_mtu         = ibmveth_change_mtu,
1474        .ndo_fix_features       = ibmveth_fix_features,
1475        .ndo_set_features       = ibmveth_set_features,
1476        .ndo_validate_addr      = eth_validate_addr,
1477        .ndo_set_mac_address    = ibmveth_set_mac_addr,
1478#ifdef CONFIG_NET_POLL_CONTROLLER
1479        .ndo_poll_controller    = ibmveth_poll_controller,
1480#endif
1481};
1482
1483static int ibmveth_probe(struct vio_dev *dev, const struct vio_device_id *id)
1484{
1485        int rc, i, mac_len;
1486        struct net_device *netdev;
1487        struct ibmveth_adapter *adapter;
1488        unsigned char *mac_addr_p;
1489        unsigned int *mcastFilterSize_p;
1490        long ret;
1491        unsigned long ret_attr;
1492
1493        dev_dbg(&dev->dev, "entering ibmveth_probe for UA 0x%x\n",
1494                dev->unit_address);
1495
1496        mac_addr_p = (unsigned char *)vio_get_attribute(dev, VETH_MAC_ADDR,
1497                                                        &mac_len);
1498        if (!mac_addr_p) {
1499                dev_err(&dev->dev, "Can't find VETH_MAC_ADDR attribute\n");
1500                return -EINVAL;
1501        }
1502        /* Workaround for old/broken pHyp */
1503        if (mac_len == 8)
1504                mac_addr_p += 2;
1505        else if (mac_len != 6) {
1506                dev_err(&dev->dev, "VETH_MAC_ADDR attribute wrong len %d\n",
1507                        mac_len);
1508                return -EINVAL;
1509        }
1510
1511        mcastFilterSize_p = (unsigned int *)vio_get_attribute(dev,
1512                                                VETH_MCAST_FILTER_SIZE, NULL);
1513        if (!mcastFilterSize_p) {
1514                dev_err(&dev->dev, "Can't find VETH_MCAST_FILTER_SIZE "
1515                        "attribute\n");
1516                return -EINVAL;
1517        }
1518
1519        netdev = alloc_etherdev(sizeof(struct ibmveth_adapter));
1520
1521        if (!netdev)
1522                return -ENOMEM;
1523
1524        adapter = netdev_priv(netdev);
1525        dev_set_drvdata(&dev->dev, netdev);
1526
1527        adapter->vdev = dev;
1528        adapter->netdev = netdev;
1529        adapter->mcastFilterSize = *mcastFilterSize_p;
1530        adapter->pool_config = 0;
1531
1532        netif_napi_add(netdev, &adapter->napi, ibmveth_poll, 16);
1533
1534        netdev->irq = dev->irq;
1535        netdev->netdev_ops = &ibmveth_netdev_ops;
1536        netdev->ethtool_ops = &netdev_ethtool_ops;
1537        SET_NETDEV_DEV(netdev, &dev->dev);
1538        netdev->hw_features = NETIF_F_SG | NETIF_F_RXCSUM |
1539                NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM;
1540
1541        netdev->features |= netdev->hw_features;
1542
1543        ret = h_illan_attributes(adapter->vdev->unit_address, 0, 0, &ret_attr);
1544
1545        /* If running older firmware, TSO should not be enabled by default */
1546        if (ret == H_SUCCESS && (ret_attr & IBMVETH_ILLAN_LRG_SND_SUPPORT) &&
1547            !old_large_send) {
1548                netdev->hw_features |= NETIF_F_TSO | NETIF_F_TSO6;
1549                netdev->features |= netdev->hw_features;
1550        } else {
1551                netdev->hw_features |= NETIF_F_TSO;
1552        }
1553
1554        memcpy(netdev->dev_addr, mac_addr_p, ETH_ALEN);
1555
1556        if (firmware_has_feature(FW_FEATURE_CMO))
1557                memcpy(pool_count, pool_count_cmo, sizeof(pool_count));
1558
1559        for (i = 0; i < IBMVETH_NUM_BUFF_POOLS; i++) {
1560                struct kobject *kobj = &adapter->rx_buff_pool[i].kobj;
1561                int error;
1562
1563                ibmveth_init_buffer_pool(&adapter->rx_buff_pool[i], i,
1564                                         pool_count[i], pool_size[i],
1565                                         pool_active[i]);
1566                error = kobject_init_and_add(kobj, &ktype_veth_pool,
1567                                             &dev->dev.kobj, "pool%d", i);
1568                if (!error)
1569                        kobject_uevent(kobj, KOBJ_ADD);
1570        }
1571
1572        netdev_dbg(netdev, "adapter @ 0x%p\n", adapter);
1573
1574        adapter->buffer_list_dma = DMA_ERROR_CODE;
1575        adapter->filter_list_dma = DMA_ERROR_CODE;
1576        adapter->rx_queue.queue_dma = DMA_ERROR_CODE;
1577
1578        netdev_dbg(netdev, "registering netdev...\n");
1579
1580        ibmveth_set_features(netdev, netdev->features);
1581
1582        rc = register_netdev(netdev);
1583
1584        if (rc) {
1585                netdev_dbg(netdev, "failed to register netdev rc=%d\n", rc);
1586                free_netdev(netdev);
1587                return rc;
1588        }
1589
1590        netdev_dbg(netdev, "registered\n");
1591
1592        return 0;
1593}
1594
1595static int ibmveth_remove(struct vio_dev *dev)
1596{
1597        struct net_device *netdev = dev_get_drvdata(&dev->dev);
1598        struct ibmveth_adapter *adapter = netdev_priv(netdev);
1599        int i;
1600
1601        for (i = 0; i < IBMVETH_NUM_BUFF_POOLS; i++)
1602                kobject_put(&adapter->rx_buff_pool[i].kobj);
1603
1604        unregister_netdev(netdev);
1605
1606        free_netdev(netdev);
1607        dev_set_drvdata(&dev->dev, NULL);
1608
1609        return 0;
1610}
1611
1612static struct attribute veth_active_attr;
1613static struct attribute veth_num_attr;
1614static struct attribute veth_size_attr;
1615
1616static ssize_t veth_pool_show(struct kobject *kobj,
1617                              struct attribute *attr, char *buf)
1618{
1619        struct ibmveth_buff_pool *pool = container_of(kobj,
1620                                                      struct ibmveth_buff_pool,
1621                                                      kobj);
1622
1623        if (attr == &veth_active_attr)
1624                return sprintf(buf, "%d\n", pool->active);
1625        else if (attr == &veth_num_attr)
1626                return sprintf(buf, "%d\n", pool->size);
1627        else if (attr == &veth_size_attr)
1628                return sprintf(buf, "%d\n", pool->buff_size);
1629        return 0;
1630}
1631
1632static ssize_t veth_pool_store(struct kobject *kobj, struct attribute *attr,
1633                               const char *buf, size_t count)
1634{
1635        struct ibmveth_buff_pool *pool = container_of(kobj,
1636                                                      struct ibmveth_buff_pool,
1637                                                      kobj);
1638        struct net_device *netdev = dev_get_drvdata(
1639            container_of(kobj->parent, struct device, kobj));
1640        struct ibmveth_adapter *adapter = netdev_priv(netdev);
1641        long value = simple_strtol(buf, NULL, 10);
1642        long rc;
1643
1644        if (attr == &veth_active_attr) {
1645                if (value && !pool->active) {
1646                        if (netif_running(netdev)) {
1647                                if (ibmveth_alloc_buffer_pool(pool)) {
1648                                        netdev_err(netdev,
1649                                                   "unable to alloc pool\n");
1650                                        return -ENOMEM;
1651                                }
1652                                pool->active = 1;
1653                                adapter->pool_config = 1;
1654                                ibmveth_close(netdev);
1655                                adapter->pool_config = 0;
1656                                if ((rc = ibmveth_open(netdev)))
1657                                        return rc;
1658                        } else {
1659                                pool->active = 1;
1660                        }
1661                } else if (!value && pool->active) {
1662                        int mtu = netdev->mtu + IBMVETH_BUFF_OH;
1663                        int i;
1664                        /* Make sure there is a buffer pool with buffers that
1665                           can hold a packet of the size of the MTU */
1666                        for (i = 0; i < IBMVETH_NUM_BUFF_POOLS; i++) {
1667                                if (pool == &adapter->rx_buff_pool[i])
1668                                        continue;
1669                                if (!adapter->rx_buff_pool[i].active)
1670                                        continue;
1671                                if (mtu <= adapter->rx_buff_pool[i].buff_size)
1672                                        break;
1673                        }
1674
1675                        if (i == IBMVETH_NUM_BUFF_POOLS) {
1676                                netdev_err(netdev, "no active pool >= MTU\n");
1677                                return -EPERM;
1678                        }
1679
1680                        if (netif_running(netdev)) {
1681                                adapter->pool_config = 1;
1682                                ibmveth_close(netdev);
1683                                pool->active = 0;
1684                                adapter->pool_config = 0;
1685                                if ((rc = ibmveth_open(netdev)))
1686                                        return rc;
1687                        }
1688                        pool->active = 0;
1689                }
1690        } else if (attr == &veth_num_attr) {
1691                if (value <= 0 || value > IBMVETH_MAX_POOL_COUNT) {
1692                        return -EINVAL;
1693                } else {
1694                        if (netif_running(netdev)) {
1695                                adapter->pool_config = 1;
1696                                ibmveth_close(netdev);
1697                                adapter->pool_config = 0;
1698                                pool->size = value;
1699                                if ((rc = ibmveth_open(netdev)))
1700                                        return rc;
1701                        } else {
1702                                pool->size = value;
1703                        }
1704                }
1705        } else if (attr == &veth_size_attr) {
1706                if (value <= IBMVETH_BUFF_OH || value > IBMVETH_MAX_BUF_SIZE) {
1707                        return -EINVAL;
1708                } else {
1709                        if (netif_running(netdev)) {
1710                                adapter->pool_config = 1;
1711                                ibmveth_close(netdev);
1712                                adapter->pool_config = 0;
1713                                pool->buff_size = value;
1714                                if ((rc = ibmveth_open(netdev)))
1715                                        return rc;
1716                        } else {
1717                                pool->buff_size = value;
1718                        }
1719                }
1720        }
1721
1722        /* kick the interrupt handler to allocate/deallocate pools */
1723        ibmveth_interrupt(netdev->irq, netdev);
1724        return count;
1725}
1726
1727
1728#define ATTR(_name, _mode)                              \
1729        struct attribute veth_##_name##_attr = {        \
1730        .name = __stringify(_name), .mode = _mode,      \
1731        };
1732
1733static ATTR(active, 0644);
1734static ATTR(num, 0644);
1735static ATTR(size, 0644);
1736
1737static struct attribute *veth_pool_attrs[] = {
1738        &veth_active_attr,
1739        &veth_num_attr,
1740        &veth_size_attr,
1741        NULL,
1742};
1743
1744static const struct sysfs_ops veth_pool_ops = {
1745        .show   = veth_pool_show,
1746        .store  = veth_pool_store,
1747};
1748
1749static struct kobj_type ktype_veth_pool = {
1750        .release        = NULL,
1751        .sysfs_ops      = &veth_pool_ops,
1752        .default_attrs  = veth_pool_attrs,
1753};
1754
1755static int ibmveth_resume(struct device *dev)
1756{
1757        struct net_device *netdev = dev_get_drvdata(dev);
1758        ibmveth_interrupt(netdev->irq, netdev);
1759        return 0;
1760}
1761
1762static struct vio_device_id ibmveth_device_table[] = {
1763        { "network", "IBM,l-lan"},
1764        { "", "" }
1765};
1766MODULE_DEVICE_TABLE(vio, ibmveth_device_table);
1767
1768static struct dev_pm_ops ibmveth_pm_ops = {
1769        .resume = ibmveth_resume
1770};
1771
1772static struct vio_driver ibmveth_driver = {
1773        .id_table       = ibmveth_device_table,
1774        .probe          = ibmveth_probe,
1775        .remove         = ibmveth_remove,
1776        .get_desired_dma = ibmveth_get_desired_dma,
1777        .name           = ibmveth_driver_name,
1778        .pm             = &ibmveth_pm_ops,
1779};
1780
1781static int __init ibmveth_module_init(void)
1782{
1783        printk(KERN_DEBUG "%s: %s %s\n", ibmveth_driver_name,
1784               ibmveth_driver_string, ibmveth_driver_version);
1785
1786        return vio_register_driver(&ibmveth_driver);
1787}
1788
1789static void __exit ibmveth_module_exit(void)
1790{
1791        vio_unregister_driver(&ibmveth_driver);
1792}
1793
1794module_init(ibmveth_module_init);
1795module_exit(ibmveth_module_exit);
1796