drivers/net/ethernet/ibm/ibmveth.c

   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
  52 static irqreturn_t ibmveth_interrupt(int irq, void *dev_instance);
  53 static void ibmveth_rxq_harvest_buffer(struct ibmveth_adapter *adapter);
  54 static unsigned long ibmveth_get_desired_dma(struct vio_dev *vdev);
  55
  56 static struct kobj_type ktype_veth_pool;
  57
  58
  59 static const char ibmveth_driver_name[] = "ibmveth";
  60 static const char ibmveth_driver_string[] = "IBM Power Virtual Ethernet Driver";
  61 #define ibmveth_driver_version "1.04"
  62
  63 MODULE_AUTHOR("Santiago Leon <santil@linux.vnet.ibm.com>");
  64 MODULE_DESCRIPTION("IBM Power Virtual Ethernet Driver");
  65 MODULE_LICENSE("GPL");
  66 MODULE_VERSION(ibmveth_driver_version);
  67
  68 static unsigned int tx_copybreak __read_mostly = 128;
  69 module_param(tx_copybreak, uint, 0644);
  70 MODULE_PARM_DESC(tx_copybreak,
  71         "Maximum size of packet that is copied to a new buffer on transmit");
  72
  73 static unsigned int rx_copybreak __read_mostly = 128;
  74 module_param(rx_copybreak, uint, 0644);
  75 MODULE_PARM_DESC(rx_copybreak,
  76         "Maximum size of packet that is copied to a new buffer on receive");
  77
  78 static unsigned int rx_flush __read_mostly = 0;
  79 module_param(rx_flush, uint, 0644);
  80 MODULE_PARM_DESC(rx_flush, "Flush receive buffers before use");
  81
  82 struct ibmveth_stat {
  83         char name[ETH_GSTRING_LEN];
  84         int offset;
  85 };
  86
  87 #define IBMVETH_STAT_OFF(stat) offsetof(struct ibmveth_adapter, stat)
  88 #define IBMVETH_GET_STAT(a, off) *((u64 *)(((unsigned long)(a)) + off))
  89
  90 struct ibmveth_stat ibmveth_stats[] = {
  91         { "replenish_task_cycles", IBMVETH_STAT_OFF(replenish_task_cycles) },
  92         { "replenish_no_mem", IBMVETH_STAT_OFF(replenish_no_mem) },
  93         { "replenish_add_buff_failure",
  94                         IBMVETH_STAT_OFF(replenish_add_buff_failure) },
  95         { "replenish_add_buff_success",
  96                         IBMVETH_STAT_OFF(replenish_add_buff_success) },
  97         { "rx_invalid_buffer", IBMVETH_STAT_OFF(rx_invalid_buffer) },
  98         { "rx_no_buffer", IBMVETH_STAT_OFF(rx_no_buffer) },
  99         { "tx_map_failed", IBMVETH_STAT_OFF(tx_map_failed) },
 100         { "tx_send_failed", IBMVETH_STAT_OFF(tx_send_failed) },
 101         { "fw_enabled_ipv4_csum", IBMVETH_STAT_OFF(fw_ipv4_csum_support) },
 102         { "fw_enabled_ipv6_csum", IBMVETH_STAT_OFF(fw_ipv6_csum_support) },
 103 };
 104
 105 /* simple methods of getting data from the current rxq entry */
 106 static inline u32 ibmveth_rxq_flags(struct ibmveth_adapter *adapter)
 107 {
 108         return be32_to_cpu(adapter->rx_queue.queue_addr[adapter->rx_queue.index].flags_off);
 109 }
 110
 111 static inline int ibmveth_rxq_toggle(struct ibmveth_adapter *adapter)
 112 {
 113         return (ibmveth_rxq_flags(adapter) & IBMVETH_RXQ_TOGGLE) >>
 114                         IBMVETH_RXQ_TOGGLE_SHIFT;
 115 }
 116
 117 static inline int ibmveth_rxq_pending_buffer(struct ibmveth_adapter *adapter)
 118 {
 119         return ibmveth_rxq_toggle(adapter) == adapter->rx_queue.toggle;
 120 }
 121
 122 static inline int ibmveth_rxq_buffer_valid(struct ibmveth_adapter *adapter)
 123 {
 124         return ibmveth_rxq_flags(adapter) & IBMVETH_RXQ_VALID;
 125 }
 126
 127 static inline int ibmveth_rxq_frame_offset(struct ibmveth_adapter *adapter)
 128 {
 129         return ibmveth_rxq_flags(adapter) & IBMVETH_RXQ_OFF_MASK;
 130 }
 131
 132 static inline int ibmveth_rxq_frame_length(struct ibmveth_adapter *adapter)
 133 {
 134         return be32_to_cpu(adapter->rx_queue.queue_addr[adapter->rx_queue.index].length);
 135 }
 136
 137 static inline int ibmveth_rxq_csum_good(struct ibmveth_adapter *adapter)
 138 {
 139         return ibmveth_rxq_flags(adapter) & IBMVETH_RXQ_CSUM_GOOD;
 140 }
 141
 142 /* setup the initial settings for a buffer pool */
 143 static void ibmveth_init_buffer_pool(struct ibmveth_buff_pool *pool,
 144                                      u32 pool_index, u32 pool_size,
 145                                      u32 buff_size, u32 pool_active)
 146 {
 147         pool->size = pool_size;
 148         pool->index = pool_index;
 149         pool->buff_size = buff_size;
 150         pool->threshold = pool_size * 7 / 8;
 151         pool->active = pool_active;
 152 }
 153
 154 /* allocate and setup an buffer pool - called during open */
 155 static int ibmveth_alloc_buffer_pool(struct ibmveth_buff_pool *pool)
 156 {
 157         int i;
 158
 159         pool->free_map = kmalloc(sizeof(u16) * pool->size, GFP_KERNEL);
 160
 161         if (!pool->free_map)
 162                 return -1;
 163
 164         pool->dma_addr = kmalloc(sizeof(dma_addr_t) * pool->size, GFP_KERNEL);
 165         if (!pool->dma_addr) {
 166                 kfree(pool->free_map);
 167                 pool->free_map = NULL;
 168                 return -1;
 169         }
 170
 171         pool->skbuff = kcalloc(pool->size, sizeof(void *), GFP_KERNEL);
 172
 173         if (!pool->skbuff) {
 174                 kfree(pool->dma_addr);
 175                 pool->dma_addr = NULL;
 176
 177                 kfree(pool->free_map);
 178                 pool->free_map = NULL;
 179                 return -1;
 180         }
 181
 182         memset(pool->dma_addr, 0, sizeof(dma_addr_t) * pool->size);
 183
 184         for (i = 0; i < pool->size; ++i)
 185                 pool->free_map[i] = i;
 186
 187         atomic_set(&pool->available, 0);
 188         pool->producer_index = 0;
 189         pool->consumer_index = 0;
 190
 191         return 0;
 192 }
 193
 194 static inline void ibmveth_flush_buffer(void *addr, unsigned long length)
 195 {
 196         unsigned long offset;
 197
 198         for (offset = 0; offset < length; offset += SMP_CACHE_BYTES)
 199                 asm("dcbfl %0,%1" :: "b" (addr), "r" (offset));
 200 }
 201
 202 /* replenish the buffers for a pool.  note that we don't need to
 203  * skb_reserve these since they are used for incoming...
 204  */
 205 static void ibmveth_replenish_buffer_pool(struct ibmveth_adapter *adapter,
 206                                           struct ibmveth_buff_pool *pool)
 207 {
 208         u32 i;
 209         u32 count = pool->size - atomic_read(&pool->available);
 210         u32 buffers_added = 0;
 211         struct sk_buff *skb;
 212         unsigned int free_index, index;
 213         u64 correlator;
 214         unsigned long lpar_rc;
 215         dma_addr_t dma_addr;
 216
 217         mb();
 218
 219         for (i = 0; i < count; ++i) {
 220                 union ibmveth_buf_desc desc;
 221
 222                 skb = netdev_alloc_skb(adapter->netdev, pool->buff_size);
 223
 224                 if (!skb) {
 225                         netdev_dbg(adapter->netdev,
 226                                    "replenish: unable to allocate skb\n");
 227                         adapter->replenish_no_mem++;
 228                         break;
 229                 }
 230
 231                 free_index = pool->consumer_index;
 232                 pool->consumer_index++;
 233                 if (pool->consumer_index >= pool->size)
 234                         pool->consumer_index = 0;
 235                 index = pool->free_map[free_index];
 236
 237                 BUG_ON(index == IBM_VETH_INVALID_MAP);
 238                 BUG_ON(pool->skbuff[index] != NULL);
 239
 240                 dma_addr = dma_map_single(&adapter->vdev->dev, skb->data,
 241                                 pool->buff_size, DMA_FROM_DEVICE);
 242
 243                 if (dma_mapping_error(&adapter->vdev->dev, dma_addr))
 244                         goto failure;
 245
 246                 pool->free_map[free_index] = IBM_VETH_INVALID_MAP;
 247                 pool->dma_addr[index] = dma_addr;
 248                 pool->skbuff[index] = skb;
 249
 250                 correlator = ((u64)pool->index << 32) | index;
 251                 *(u64 *)skb->data = correlator;
 252
 253                 desc.fields.flags_len = IBMVETH_BUF_VALID | pool->buff_size;
 254                 desc.fields.address = dma_addr;
 255
 256                 if (rx_flush) {
 257                         unsigned int len = min(pool->buff_size,
 258                                                 adapter->netdev->mtu +
 259                                                 IBMVETH_BUFF_OH);
 260                         ibmveth_flush_buffer(skb->data, len);
 261                 }
 262                 lpar_rc = h_add_logical_lan_buffer(adapter->vdev->unit_address,
 263                                                    desc.desc);
 264
 265                 if (lpar_rc != H_SUCCESS) {
 266                         goto failure;
 267                 } else {
 268                         buffers_added++;
 269                         adapter->replenish_add_buff_success++;
 270                 }
 271         }
 272
 273         mb();
 274         atomic_add(buffers_added, &(pool->available));
 275         return;
 276
 277 failure:
 278         pool->free_map[free_index] = index;
 279         pool->skbuff[index] = NULL;
 280         if (pool->consumer_index == 0)
 281                 pool->consumer_index = pool->size - 1;
 282         else
 283                 pool->consumer_index--;
 284         if (!dma_mapping_error(&adapter->vdev->dev, dma_addr))
 285                 dma_unmap_single(&adapter->vdev->dev,
 286                                  pool->dma_addr[index], pool->buff_size,
 287                                  DMA_FROM_DEVICE);
 288         dev_kfree_skb_any(skb);
 289         adapter->replenish_add_buff_failure++;
 290
 291         mb();
 292         atomic_add(buffers_added, &(pool->available));
 293 }
 294
 295 /* replenish routine */
 296 static void ibmveth_replenish_task(struct ibmveth_adapter *adapter)
 297 {
 298         int i;
 299
 300         adapter->replenish_task_cycles++;
 301
 302         for (i = (IBMVETH_NUM_BUFF_POOLS - 1); i >= 0; i--) {
 303                 struct ibmveth_buff_pool *pool = &adapter->rx_buff_pool[i];
 304
 305                 if (pool->active &&
 306                     (atomic_read(&pool->available) < pool->threshold))
 307                         ibmveth_replenish_buffer_pool(adapter, pool);
 308         }
 309
 310         adapter->rx_no_buffer = *(u64 *)(((char*)adapter->buffer_list_addr) +
 311                                                 4096 - 8);
 312 }
 313
 314 /* empty and free ana buffer pool - also used to do cleanup in error paths */
 315 static void ibmveth_free_buffer_pool(struct ibmveth_adapter *adapter,
 316                                      struct ibmveth_buff_pool *pool)
 317 {
 318         int i;
 319
 320         kfree(pool->free_map);
 321         pool->free_map = NULL;
 322
 323         if (pool->skbuff && pool->dma_addr) {
 324                 for (i = 0; i < pool->size; ++i) {
 325                         struct sk_buff *skb = pool->skbuff[i];
 326                         if (skb) {
 327                                 dma_unmap_single(&adapter->vdev->dev,
 328                                                  pool->dma_addr[i],
 329                                                  pool->buff_size,
 330                                                  DMA_FROM_DEVICE);
 331                                 dev_kfree_skb_any(skb);
 332                                 pool->skbuff[i] = NULL;
 333                         }
 334                 }
 335         }
 336
 337         if (pool->dma_addr) {
 338                 kfree(pool->dma_addr);
 339                 pool->dma_addr = NULL;
 340         }
 341
 342         if (pool->skbuff) {
 343                 kfree(pool->skbuff);
 344                 pool->skbuff = NULL;
 345         }
 346 }
 347
 348 /* remove a buffer from a pool */
 349 static void ibmveth_remove_buffer_from_pool(struct ibmveth_adapter *adapter,
 350                                             u64 correlator)
 351 {
 352         unsigned int pool  = correlator >> 32;
 353         unsigned int index = correlator & 0xffffffffUL;
 354         unsigned int free_index;
 355         struct sk_buff *skb;
 356
 357         BUG_ON(pool >= IBMVETH_NUM_BUFF_POOLS);
 358         BUG_ON(index >= adapter->rx_buff_pool[pool].size);
 359
 360         skb = adapter->rx_buff_pool[pool].skbuff[index];
 361
 362         BUG_ON(skb == NULL);
 363
 364         adapter->rx_buff_pool[pool].skbuff[index] = NULL;
 365
 366         dma_unmap_single(&adapter->vdev->dev,
 367                          adapter->rx_buff_pool[pool].dma_addr[index],
 368                          adapter->rx_buff_pool[pool].buff_size,
 369                          DMA_FROM_DEVICE);
 370
 371         free_index = adapter->rx_buff_pool[pool].producer_index;
 372         adapter->rx_buff_pool[pool].producer_index++;
 373         if (adapter->rx_buff_pool[pool].producer_index >=
 374             adapter->rx_buff_pool[pool].size)
 375                 adapter->rx_buff_pool[pool].producer_index = 0;
 376         adapter->rx_buff_pool[pool].free_map[free_index] = index;
 377
 378         mb();
 379
 380         atomic_dec(&(adapter->rx_buff_pool[pool].available));
 381 }
 382
 383 /* get the current buffer on the rx queue */
 384 static inline struct sk_buff *ibmveth_rxq_get_buffer(struct ibmveth_adapter *adapter)
 385 {
 386         u64 correlator = adapter->rx_queue.queue_addr[adapter->rx_queue.index].correlator;
 387         unsigned int pool = correlator >> 32;
 388         unsigned int index = correlator & 0xffffffffUL;
 389
 390         BUG_ON(pool >= IBMVETH_NUM_BUFF_POOLS);
 391         BUG_ON(index >= adapter->rx_buff_pool[pool].size);
 392
 393         return adapter->rx_buff_pool[pool].skbuff[index];
 394 }
 395
 396 /* recycle the current buffer on the rx queue */
 397 static int ibmveth_rxq_recycle_buffer(struct ibmveth_adapter *adapter)
 398 {
 399         u32 q_index = adapter->rx_queue.index;
 400         u64 correlator = adapter->rx_queue.queue_addr[q_index].correlator;
 401         unsigned int pool = correlator >> 32;
 402         unsigned int index = correlator & 0xffffffffUL;
 403         union ibmveth_buf_desc desc;
 404         unsigned long lpar_rc;
 405         int ret = 1;
 406
 407         BUG_ON(pool >= IBMVETH_NUM_BUFF_POOLS);
 408         BUG_ON(index >= adapter->rx_buff_pool[pool].size);
 409
 410         if (!adapter->rx_buff_pool[pool].active) {
 411                 ibmveth_rxq_harvest_buffer(adapter);
 412                 ibmveth_free_buffer_pool(adapter, &adapter->rx_buff_pool[pool]);
 413                 goto out;
 414         }
 415
 416         desc.fields.flags_len = IBMVETH_BUF_VALID |
 417                 adapter->rx_buff_pool[pool].buff_size;
 418         desc.fields.address = adapter->rx_buff_pool[pool].dma_addr[index];
 419
 420         lpar_rc = h_add_logical_lan_buffer(adapter->vdev->unit_address, desc.desc);
 421
 422         if (lpar_rc != H_SUCCESS) {
 423                 netdev_dbg(adapter->netdev, "h_add_logical_lan_buffer failed "
 424                            "during recycle rc=%ld", lpar_rc);
 425                 ibmveth_remove_buffer_from_pool(adapter, adapter->rx_queue.queue_addr[adapter->rx_queue.index].correlator);
 426                 ret = 0;
 427         }
 428
 429         if (++adapter->rx_queue.index == adapter->rx_queue.num_slots) {
 430                 adapter->rx_queue.index = 0;
 431                 adapter->rx_queue.toggle = !adapter->rx_queue.toggle;
 432         }
 433
 434 out:
 435         return ret;
 436 }
 437
 438 static void ibmveth_rxq_harvest_buffer(struct ibmveth_adapter *adapter)
 439 {
 440         ibmveth_remove_buffer_from_pool(adapter, adapter->rx_queue.queue_addr[adapter->rx_queue.index].correlator);
 441
 442         if (++adapter->rx_queue.index == adapter->rx_queue.num_slots) {
 443                 adapter->rx_queue.index = 0;
 444                 adapter->rx_queue.toggle = !adapter->rx_queue.toggle;
 445         }
 446 }
 447
 448 static void ibmveth_cleanup(struct ibmveth_adapter *adapter)
 449 {
 450         int i;
 451         struct device *dev = &adapter->vdev->dev;
 452
 453         if (adapter->buffer_list_addr != NULL) {
 454                 if (!dma_mapping_error(dev, adapter->buffer_list_dma)) {
 455                         dma_unmap_single(dev, adapter->buffer_list_dma, 4096,
 456                                         DMA_BIDIRECTIONAL);
 457                         adapter->buffer_list_dma = DMA_ERROR_CODE;
 458                 }
 459                 free_page((unsigned long)adapter->buffer_list_addr);
 460                 adapter->buffer_list_addr = NULL;
 461         }
 462
 463         if (adapter->filter_list_addr != NULL) {
 464                 if (!dma_mapping_error(dev, adapter->filter_list_dma)) {
 465                         dma_unmap_single(dev, adapter->filter_list_dma, 4096,
 466                                         DMA_BIDIRECTIONAL);
 467                         adapter->filter_list_dma = DMA_ERROR_CODE;
 468                 }
 469                 free_page((unsigned long)adapter->filter_list_addr);
 470                 adapter->filter_list_addr = NULL;
 471         }
 472
 473         if (adapter->rx_queue.queue_addr != NULL) {
 474                 dma_free_coherent(dev, adapter->rx_queue.queue_len,
 475                                   adapter->rx_queue.queue_addr,
 476                                   adapter->rx_queue.queue_dma);
 477                 adapter->rx_queue.queue_addr = NULL;
 478         }
 479
 480         for (i = 0; i < IBMVETH_NUM_BUFF_POOLS; i++)
 481                 if (adapter->rx_buff_pool[i].active)
 482                         ibmveth_free_buffer_pool(adapter,
 483                                                  &adapter->rx_buff_pool[i]);
 484
 485         if (adapter->bounce_buffer != NULL) {
 486                 if (!dma_mapping_error(dev, adapter->bounce_buffer_dma)) {
 487                         dma_unmap_single(&adapter->vdev->dev,
 488                                         adapter->bounce_buffer_dma,
 489                                         adapter->netdev->mtu + IBMVETH_BUFF_OH,
 490                                         DMA_BIDIRECTIONAL);
 491                         adapter->bounce_buffer_dma = DMA_ERROR_CODE;
 492                 }
 493                 kfree(adapter->bounce_buffer);
 494                 adapter->bounce_buffer = NULL;
 495         }
 496 }
 497
 498 static int ibmveth_register_logical_lan(struct ibmveth_adapter *adapter,
 499         union ibmveth_buf_desc rxq_desc, u64 mac_address)
 500 {
 501         int rc, try_again = 1;
 502
 503         /*
 504          * After a kexec the adapter will still be open, so our attempt to
 505          * open it will fail. So if we get a failure we free the adapter and
 506          * try again, but only once.
 507          */
 508 retry:
 509         rc = h_register_logical_lan(adapter->vdev->unit_address,
 510                                     adapter->buffer_list_dma, rxq_desc.desc,
 511                                     adapter->filter_list_dma, mac_address);
 512
 513         if (rc != H_SUCCESS && try_again) {
 514                 do {
 515                         rc = h_free_logical_lan(adapter->vdev->unit_address);
 516                 } while (H_IS_LONG_BUSY(rc) || (rc == H_BUSY));
 517
 518                 try_again = 0;
 519                 goto retry;
 520         }
 521
 522         return rc;
 523 }
 524
 525 static int ibmveth_open(struct net_device *netdev)
 526 {
 527         struct ibmveth_adapter *adapter = netdev_priv(netdev);
 528         u64 mac_address = 0;
 529         int rxq_entries = 1;
 530         unsigned long lpar_rc;
 531         int rc;
 532         union ibmveth_buf_desc rxq_desc;
 533         int i;
 534         struct device *dev;
 535
 536         netdev_dbg(netdev, "open starting\n");
 537
 538         napi_enable(&adapter->napi);
 539
 540         for(i = 0; i < IBMVETH_NUM_BUFF_POOLS; i++)
 541                 rxq_entries += adapter->rx_buff_pool[i].size;
 542
 543         adapter->buffer_list_addr = (void*) get_zeroed_page(GFP_KERNEL);
 544         adapter->filter_list_addr = (void*) get_zeroed_page(GFP_KERNEL);
 545
 546         if (!adapter->buffer_list_addr || !adapter->filter_list_addr) {
 547                 netdev_err(netdev, "unable to allocate filter or buffer list "
 548                            "pages\n");
 549                 rc = -ENOMEM;
 550                 goto err_out;
 551         }
 552
 553         dev = &adapter->vdev->dev;
 554
 555         adapter->rx_queue.queue_len = sizeof(struct ibmveth_rx_q_entry) *
 556                                                 rxq_entries;
 557         adapter->rx_queue.queue_addr =
 558                 dma_alloc_coherent(dev, adapter->rx_queue.queue_len,
 559                                    &adapter->rx_queue.queue_dma, GFP_KERNEL);
 560         if (!adapter->rx_queue.queue_addr) {
 561                 rc = -ENOMEM;
 562                 goto err_out;
 563         }
 564
 565         adapter->buffer_list_dma = dma_map_single(dev,
 566                         adapter->buffer_list_addr, 4096, DMA_BIDIRECTIONAL);
 567         adapter->filter_list_dma = dma_map_single(dev,
 568                         adapter->filter_list_addr, 4096, DMA_BIDIRECTIONAL);
 569
 570         if ((dma_mapping_error(dev, adapter->buffer_list_dma)) ||
 571             (dma_mapping_error(dev, adapter->filter_list_dma))) {
 572                 netdev_err(netdev, "unable to map filter or buffer list "
 573                            "pages\n");
 574                 rc = -ENOMEM;
 575                 goto err_out;
 576         }
 577
 578         adapter->rx_queue.index = 0;
 579         adapter->rx_queue.num_slots = rxq_entries;
 580         adapter->rx_queue.toggle = 1;
 581
 582         memcpy(&mac_address, netdev->dev_addr, netdev->addr_len);
 583         mac_address = mac_address >> 16;
 584
 585         rxq_desc.fields.flags_len = IBMVETH_BUF_VALID |
 586                                         adapter->rx_queue.queue_len;
 587         rxq_desc.fields.address = adapter->rx_queue.queue_dma;
 588
 589         netdev_dbg(netdev, "buffer list @ 0x%p\n", adapter->buffer_list_addr);
 590         netdev_dbg(netdev, "filter list @ 0x%p\n", adapter->filter_list_addr);
 591         netdev_dbg(netdev, "receive q   @ 0x%p\n", adapter->rx_queue.queue_addr);
 592
 593         h_vio_signal(adapter->vdev->unit_address, VIO_IRQ_DISABLE);
 594
 595         lpar_rc = ibmveth_register_logical_lan(adapter, rxq_desc, mac_address);
 596
 597         if (lpar_rc != H_SUCCESS) {
 598                 netdev_err(netdev, "h_register_logical_lan failed with %ld\n",
 599                            lpar_rc);
 600                 netdev_err(netdev, "buffer TCE:0x%llx filter TCE:0x%llx rxq "
 601                            "desc:0x%llx MAC:0x%llx\n",
 602                                      adapter->buffer_list_dma,
 603                                      adapter->filter_list_dma,
 604                                      rxq_desc.desc,
 605                                      mac_address);
 606                 rc = -ENONET;
 607                 goto err_out;
 608         }
 609
 610         for (i = 0; i < IBMVETH_NUM_BUFF_POOLS; i++) {
 611                 if (!adapter->rx_buff_pool[i].active)
 612                         continue;
 613                 if (ibmveth_alloc_buffer_pool(&adapter->rx_buff_pool[i])) {
 614                         netdev_err(netdev, "unable to alloc pool\n");
 615                         adapter->rx_buff_pool[i].active = 0;
 616                         rc = -ENOMEM;
 617                         goto err_out;
 618                 }
 619         }
 620
 621         netdev_dbg(netdev, "registering irq 0x%x\n", netdev->irq);
 622         rc = request_irq(netdev->irq, ibmveth_interrupt, 0, netdev->name,
 623                          netdev);
 624         if (rc != 0) {
 625                 netdev_err(netdev, "unable to request irq 0x%x, rc %d\n",
 626                            netdev->irq, rc);
 627                 do {
 628                         lpar_rc = h_free_logical_lan(adapter->vdev->unit_address);
 629                 } while (H_IS_LONG_BUSY(lpar_rc) || (lpar_rc == H_BUSY));
 630
 631                 goto err_out;
 632         }
 633
 634         adapter->bounce_buffer =
 635             kmalloc(netdev->mtu + IBMVETH_BUFF_OH, GFP_KERNEL);
 636         if (!adapter->bounce_buffer) {
 637                 rc = -ENOMEM;
 638                 goto err_out_free_irq;
 639         }
 640         adapter->bounce_buffer_dma =
 641             dma_map_single(&adapter->vdev->dev, adapter->bounce_buffer,
 642                            netdev->mtu + IBMVETH_BUFF_OH, DMA_BIDIRECTIONAL);
 643         if (dma_mapping_error(dev, adapter->bounce_buffer_dma)) {
 644                 netdev_err(netdev, "unable to map bounce buffer\n");
 645                 rc = -ENOMEM;
 646                 goto err_out_free_irq;
 647         }
 648
 649         netdev_dbg(netdev, "initial replenish cycle\n");
 650         ibmveth_interrupt(netdev->irq, netdev);
 651
 652         netif_start_queue(netdev);
 653
 654         netdev_dbg(netdev, "open complete\n");
 655
 656         return 0;
 657
 658 err_out_free_irq:
 659         free_irq(netdev->irq, netdev);
 660 err_out:
 661         ibmveth_cleanup(adapter);
 662         napi_disable(&adapter->napi);
 663         return rc;
 664 }
 665
 666 static int ibmveth_close(struct net_device *netdev)
 667 {
 668         struct ibmveth_adapter *adapter = netdev_priv(netdev);
 669         long lpar_rc;
 670
 671         netdev_dbg(netdev, "close starting\n");
 672
 673         napi_disable(&adapter->napi);
 674
 675         if (!adapter->pool_config)
 676                 netif_stop_queue(netdev);
 677
 678         h_vio_signal(adapter->vdev->unit_address, VIO_IRQ_DISABLE);
 679
 680         do {
 681                 lpar_rc = h_free_logical_lan(adapter->vdev->unit_address);
 682         } while (H_IS_LONG_BUSY(lpar_rc) || (lpar_rc == H_BUSY));
 683
 684         if (lpar_rc != H_SUCCESS) {
 685                 netdev_err(netdev, "h_free_logical_lan failed with %lx, "
 686                            "continuing with close\n", lpar_rc);
 687         }
 688
 689         free_irq(netdev->irq, netdev);
 690
 691         adapter->rx_no_buffer = *(u64 *)(((char *)adapter->buffer_list_addr) +
 692                                                 4096 - 8);
 693
 694         ibmveth_cleanup(adapter);
 695
 696         netdev_dbg(netdev, "close complete\n");
 697
 698         return 0;
 699 }
 700
 701 static int netdev_get_settings(struct net_device *dev, struct ethtool_cmd *cmd)
 702 {
 703         cmd->supported = (SUPPORTED_1000baseT_Full | SUPPORTED_Autoneg |
 704                                 SUPPORTED_FIBRE);
 705         cmd->advertising = (ADVERTISED_1000baseT_Full | ADVERTISED_Autoneg |
 706                                 ADVERTISED_FIBRE);
 707         ethtool_cmd_speed_set(cmd, SPEED_1000);
 708         cmd->duplex = DUPLEX_FULL;
 709         cmd->port = PORT_FIBRE;
 710         cmd->phy_address = 0;
 711         cmd->transceiver = XCVR_INTERNAL;
 712         cmd->autoneg = AUTONEG_ENABLE;
 713         cmd->maxtxpkt = 0;
 714         cmd->maxrxpkt = 1;
 715         return 0;
 716 }
 717
 718 static void netdev_get_drvinfo(struct net_device *dev,
 719                                struct ethtool_drvinfo *info)
 720 {
 721         strlcpy(info->driver, ibmveth_driver_name, sizeof(info->driver));
 722         strlcpy(info->version, ibmveth_driver_version, sizeof(info->version));
 723 }
 724
 725 static netdev_features_t ibmveth_fix_features(struct net_device *dev,
 726         netdev_features_t features)
 727 {
 728         /*
 729          * Since the ibmveth firmware interface does not have the
 730          * concept of separate tx/rx checksum offload enable, if rx
 731          * checksum is disabled we also have to disable tx checksum
 732          * offload. Once we disable rx checksum offload, we are no
 733          * longer allowed to send tx buffers that are not properly
 734          * checksummed.
 735          */
 736
 737         if (!(features & NETIF_F_RXCSUM))
 738                 features &= ~NETIF_F_ALL_CSUM;
 739
 740         return features;
 741 }
 742
 743 static int ibmveth_set_csum_offload(struct net_device *dev, u32 data)
 744 {
 745         struct ibmveth_adapter *adapter = netdev_priv(dev);
 746         unsigned long set_attr, clr_attr, ret_attr;
 747         unsigned long set_attr6, clr_attr6;
 748         long ret, ret4, ret6;
 749         int rc1 = 0, rc2 = 0;
 750         int restart = 0;
 751
 752         if (netif_running(dev)) {
 753                 restart = 1;
 754                 adapter->pool_config = 1;
 755                 ibmveth_close(dev);
 756                 adapter->pool_config = 0;
 757         }
 758
 759         set_attr = 0;
 760         clr_attr = 0;
 761         set_attr6 = 0;
 762         clr_attr6 = 0;
 763
 764         if (data) {
 765                 set_attr = IBMVETH_ILLAN_IPV4_TCP_CSUM;
 766                 set_attr6 = IBMVETH_ILLAN_IPV6_TCP_CSUM;
 767         } else {
 768                 clr_attr = IBMVETH_ILLAN_IPV4_TCP_CSUM;
 769                 clr_attr6 = IBMVETH_ILLAN_IPV6_TCP_CSUM;
 770         }
 771
 772         ret = h_illan_attributes(adapter->vdev->unit_address, 0, 0, &ret_attr);
 773
 774         if (ret == H_SUCCESS && !(ret_attr & IBMVETH_ILLAN_ACTIVE_TRUNK) &&
 775             !(ret_attr & IBMVETH_ILLAN_TRUNK_PRI_MASK) &&
 776             (ret_attr & IBMVETH_ILLAN_PADDED_PKT_CSUM)) {
 777                 ret4 = h_illan_attributes(adapter->vdev->unit_address, clr_attr,
 778                                          set_attr, &ret_attr);
 779
 780                 if (ret4 != H_SUCCESS) {
 781                         netdev_err(dev, "unable to change IPv4 checksum "
 782                                         "offload settings. %d rc=%ld\n",
 783                                         data, ret4);
 784
 785                         h_illan_attributes(adapter->vdev->unit_address,
 786                                            set_attr, clr_attr, &ret_attr);
 787
 788                         if (data == 1)
 789                                 dev->features &= ~NETIF_F_IP_CSUM;
 790
 791                 } else {
 792                         adapter->fw_ipv4_csum_support = data;
 793                 }
 794
 795                 ret6 = h_illan_attributes(adapter->vdev->unit_address,
 796                                          clr_attr6, set_attr6, &ret_attr);
 797
 798                 if (ret6 != H_SUCCESS) {
 799                         netdev_err(dev, "unable to change IPv6 checksum "
 800                                         "offload settings. %d rc=%ld\n",
 801                                         data, ret6);
 802
 803                         h_illan_attributes(adapter->vdev->unit_address,
 804                                            set_attr6, clr_attr6, &ret_attr);
 805
 806                         if (data == 1)
 807                                 dev->features &= ~NETIF_F_IPV6_CSUM;
 808
 809                 } else
 810                         adapter->fw_ipv6_csum_support = data;
 811
 812                 if (ret4 == H_SUCCESS || ret6 == H_SUCCESS)
 813                         adapter->rx_csum = data;
 814                 else
 815                         rc1 = -EIO;
 816         } else {
 817                 rc1 = -EIO;
 818                 netdev_err(dev, "unable to change checksum offload settings."
 819                                      " %d rc=%ld ret_attr=%lx\n", data, ret,
 820                                      ret_attr);
 821         }
 822
 823         if (restart)
 824                 rc2 = ibmveth_open(dev);
 825
 826         return rc1 ? rc1 : rc2;
 827 }
 828
 829 static int ibmveth_set_features(struct net_device *dev,
 830         netdev_features_t features)
 831 {
 832         struct ibmveth_adapter *adapter = netdev_priv(dev);
 833         int rx_csum = !!(features & NETIF_F_RXCSUM);
 834         int rc;
 835
 836         if (rx_csum == adapter->rx_csum)
 837                 return 0;
 838
 839         rc = ibmveth_set_csum_offload(dev, rx_csum);
 840         if (rc && !adapter->rx_csum)
 841                 dev->features = features & ~(NETIF_F_ALL_CSUM | NETIF_F_RXCSUM);
 842
 843         return rc;
 844 }
 845
 846 static void ibmveth_get_strings(struct net_device *dev, u32 stringset, u8 *data)
 847 {
 848         int i;
 849
 850         if (stringset != ETH_SS_STATS)
 851                 return;
 852
 853         for (i = 0; i < ARRAY_SIZE(ibmveth_stats); i++, data += ETH_GSTRING_LEN)
 854                 memcpy(data, ibmveth_stats[i].name, ETH_GSTRING_LEN);
 855 }
 856
 857 static int ibmveth_get_sset_count(struct net_device *dev, int sset)
 858 {
 859         switch (sset) {
 860         case ETH_SS_STATS:
 861                 return ARRAY_SIZE(ibmveth_stats);
 862         default:
 863                 return -EOPNOTSUPP;
 864         }
 865 }
 866
 867 static void ibmveth_get_ethtool_stats(struct net_device *dev,
 868                                       struct ethtool_stats *stats, u64 *data)
 869 {
 870         int i;
 871         struct ibmveth_adapter *adapter = netdev_priv(dev);
 872
 873         for (i = 0; i < ARRAY_SIZE(ibmveth_stats); i++)
 874                 data[i] = IBMVETH_GET_STAT(adapter, ibmveth_stats[i].offset);
 875 }
 876
 877 static const struct ethtool_ops netdev_ethtool_ops = {
 878         .get_drvinfo            = netdev_get_drvinfo,
 879         .get_settings           = netdev_get_settings,
 880         .get_link               = ethtool_op_get_link,
 881         .get_strings            = ibmveth_get_strings,
 882         .get_sset_count         = ibmveth_get_sset_count,
 883         .get_ethtool_stats      = ibmveth_get_ethtool_stats,
 884 };
 885
 886 static int ibmveth_ioctl(struct net_device *dev, struct ifreq *ifr, int cmd)
 887 {
 888         return -EOPNOTSUPP;
 889 }
 890
 891 #define page_offset(v) ((unsigned long)(v) & ((1 << 12) - 1))
 892
 893 static int ibmveth_send(struct ibmveth_adapter *adapter,
 894                         union ibmveth_buf_desc *descs)
 895 {
 896         unsigned long correlator;
 897         unsigned int retry_count;
 898         unsigned long ret;
 899
 900         /*
 901          * The retry count sets a maximum for the number of broadcast and
 902          * multicast destinations within the system.
 903          */
 904         retry_count = 1024;
 905         correlator = 0;
 906         do {
 907                 ret = h_send_logical_lan(adapter->vdev->unit_address,
 908                                              descs[0].desc, descs[1].desc,
 909                                              descs[2].desc, descs[3].desc,
 910                                              descs[4].desc, descs[5].desc,
 911                                              correlator, &correlator);
 912         } while ((ret == H_BUSY) && (retry_count--));
 913
 914         if (ret != H_SUCCESS && ret != H_DROPPED) {
 915                 netdev_err(adapter->netdev, "tx: h_send_logical_lan failed "
 916                            "with rc=%ld\n", ret);
 917                 return 1;
 918         }
 919
 920         return 0;
 921 }
 922
 923 static netdev_tx_t ibmveth_start_xmit(struct sk_buff *skb,
 924                                       struct net_device *netdev)
 925 {
 926         struct ibmveth_adapter *adapter = netdev_priv(netdev);
 927         unsigned int desc_flags;
 928         union ibmveth_buf_desc descs[6];
 929         int last, i;
 930         int force_bounce = 0;
 931         dma_addr_t dma_addr;
 932
 933         /*
 934          * veth handles a maximum of 6 segments including the header, so
 935          * we have to linearize the skb if there are more than this.
 936          */
 937         if (skb_shinfo(skb)->nr_frags > 5 && __skb_linearize(skb)) {
 938                 netdev->stats.tx_dropped++;
 939                 goto out;
 940         }
 941
 942         /* veth can't checksum offload UDP */
 943         if (skb->ip_summed == CHECKSUM_PARTIAL &&
 944             ((skb->protocol == htons(ETH_P_IP) &&
 945               ip_hdr(skb)->protocol != IPPROTO_TCP) ||
 946              (skb->protocol == htons(ETH_P_IPV6) &&
 947               ipv6_hdr(skb)->nexthdr != IPPROTO_TCP)) &&
 948             skb_checksum_help(skb)) {
 949
 950                 netdev_err(netdev, "tx: failed to checksum packet\n");
 951                 netdev->stats.tx_dropped++;
 952                 goto out;
 953         }
 954
 955         desc_flags = IBMVETH_BUF_VALID;
 956
 957         if (skb->ip_summed == CHECKSUM_PARTIAL) {
 958                 unsigned char *buf = skb_transport_header(skb) +
 959                                                 skb->csum_offset;
 960
 961                 desc_flags |= (IBMVETH_BUF_NO_CSUM | IBMVETH_BUF_CSUM_GOOD);
 962
 963                 /* Need to zero out the checksum */
 964                 buf[0] = 0;
 965                 buf[1] = 0;
 966         }
 967
 968 retry_bounce:
 969         memset(descs, 0, sizeof(descs));
 970
 971         /*
 972          * If a linear packet is below the rx threshold then
 973          * copy it into the static bounce buffer. This avoids the
 974          * cost of a TCE insert and remove.
 975          */
 976         if (force_bounce || (!skb_is_nonlinear(skb) &&
 977                                 (skb->len < tx_copybreak))) {
 978                 skb_copy_from_linear_data(skb, adapter->bounce_buffer,
 979                                           skb->len);
 980
 981                 descs[0].fields.flags_len = desc_flags | skb->len;
 982                 descs[0].fields.address = adapter->bounce_buffer_dma;
 983
 984                 if (ibmveth_send(adapter, descs)) {
 985                         adapter->tx_send_failed++;
 986                         netdev->stats.tx_dropped++;
 987                 } else {
 988                         netdev->stats.tx_packets++;
 989                         netdev->stats.tx_bytes += skb->len;
 990                 }
 991
 992                 goto out;
 993         }
 994
 995         /* Map the header */
 996         dma_addr = dma_map_single(&adapter->vdev->dev, skb->data,
 997                                   skb_headlen(skb), DMA_TO_DEVICE);
 998         if (dma_mapping_error(&adapter->vdev->dev, dma_addr))
 999                 goto map_failed;
1000
1001         descs[0].fields.flags_len = desc_flags | skb_headlen(skb);
1002         descs[0].fields.address = dma_addr;
1003
1004         /* Map the frags */
1005         for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
1006                 const skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
1007
1008                 dma_addr = skb_frag_dma_map(&adapter->vdev->dev, frag, 0,
1009                                             skb_frag_size(frag), DMA_TO_DEVICE);
1010
1011                 if (dma_mapping_error(&adapter->vdev->dev, dma_addr))
1012                         goto map_failed_frags;
1013
1014                 descs[i+1].fields.flags_len = desc_flags | skb_frag_size(frag);
1015                 descs[i+1].fields.address = dma_addr;
1016         }
1017
1018         if (ibmveth_send(adapter, descs)) {
1019                 adapter->tx_send_failed++;
1020                 netdev->stats.tx_dropped++;
1021         } else {
1022                 netdev->stats.tx_packets++;
1023                 netdev->stats.tx_bytes += skb->len;
1024         }
1025
1026         dma_unmap_single(&adapter->vdev->dev,
1027                          descs[0].fields.address,
1028                          descs[0].fields.flags_len & IBMVETH_BUF_LEN_MASK,
1029                          DMA_TO_DEVICE);
1030
1031         for (i = 1; i < skb_shinfo(skb)->nr_frags + 1; i++)
1032                 dma_unmap_page(&adapter->vdev->dev, descs[i].fields.address,
1033                                descs[i].fields.flags_len & IBMVETH_BUF_LEN_MASK,
1034                                DMA_TO_DEVICE);
1035
1036 out:
1037         dev_kfree_skb(skb);
1038         return NETDEV_TX_OK;
1039
1040 map_failed_frags:
1041         last = i+1;
1042         for (i = 0; i < last; i++)
1043                 dma_unmap_page(&adapter->vdev->dev, descs[i].fields.address,
1044                                descs[i].fields.flags_len & IBMVETH_BUF_LEN_MASK,
1045                                DMA_TO_DEVICE);
1046
1047 map_failed:
1048         if (!firmware_has_feature(FW_FEATURE_CMO))
1049                 netdev_err(netdev, "tx: unable to map xmit buffer\n");
1050         adapter->tx_map_failed++;
1051         skb_linearize(skb);
1052         force_bounce = 1;
1053         goto retry_bounce;
1054 }
1055
1056 static int ibmveth_poll(struct napi_struct *napi, int budget)
1057 {
1058         struct ibmveth_adapter *adapter =
1059                         container_of(napi, struct ibmveth_adapter, napi);
1060         struct net_device *netdev = adapter->netdev;
1061         int frames_processed = 0;
1062         unsigned long lpar_rc;
1063
1064 restart_poll:
1065         do {
1066                 if (!ibmveth_rxq_pending_buffer(adapter))
1067                         break;
1068
1069                 smp_rmb();
1070                 if (!ibmveth_rxq_buffer_valid(adapter)) {
1071                         wmb(); /* suggested by larson1 */
1072                         adapter->rx_invalid_buffer++;
1073                         netdev_dbg(netdev, "recycling invalid buffer\n");
1074                         ibmveth_rxq_recycle_buffer(adapter);
1075                 } else {
1076                         struct sk_buff *skb, *new_skb;
1077                         int length = ibmveth_rxq_frame_length(adapter);
1078                         int offset = ibmveth_rxq_frame_offset(adapter);
1079                         int csum_good = ibmveth_rxq_csum_good(adapter);
1080
1081                         skb = ibmveth_rxq_get_buffer(adapter);
1082
1083                         new_skb = NULL;
1084                         if (length < rx_copybreak)
1085                                 new_skb = netdev_alloc_skb(netdev, length);
1086
1087                         if (new_skb) {
1088                                 skb_copy_to_linear_data(new_skb,
1089                                                         skb->data + offset,
1090                                                         length);
1091                                 if (rx_flush)
1092                                         ibmveth_flush_buffer(skb->data,
1093                                                 length + offset);
1094                                 if (!ibmveth_rxq_recycle_buffer(adapter))
1095                                         kfree_skb(skb);
1096                                 skb = new_skb;
1097                         } else {
1098                                 ibmveth_rxq_harvest_buffer(adapter);
1099                                 skb_reserve(skb, offset);
1100                         }
1101
1102                         skb_put(skb, length);
1103                         skb->protocol = eth_type_trans(skb, netdev);
1104
1105                         if (csum_good)
1106                                 skb->ip_summed = CHECKSUM_UNNECESSARY;
1107
1108                         netif_receive_skb(skb); /* send it up */
1109
1110                         netdev->stats.rx_packets++;
1111                         netdev->stats.rx_bytes += length;
1112                         frames_processed++;
1113                 }
1114         } while (frames_processed < budget);
1115
1116         ibmveth_replenish_task(adapter);
1117
1118         if (frames_processed < budget) {
1119                 /* We think we are done - reenable interrupts,
1120                  * then check once more to make sure we are done.
1121                  */
1122                 lpar_rc = h_vio_signal(adapter->vdev->unit_address,
1123                                        VIO_IRQ_ENABLE);
1124
1125                 BUG_ON(lpar_rc != H_SUCCESS);
1126
1127                 napi_complete(napi);
1128
1129                 if (ibmveth_rxq_pending_buffer(adapter) &&
1130                     napi_reschedule(napi)) {
1131                         lpar_rc = h_vio_signal(adapter->vdev->unit_address,
1132                                                VIO_IRQ_DISABLE);
1133                         goto restart_poll;
1134                 }
1135         }
1136
1137         return frames_processed;
1138 }
1139
1140 static irqreturn_t ibmveth_interrupt(int irq, void *dev_instance)
1141 {
1142         struct net_device *netdev = dev_instance;
1143         struct ibmveth_adapter *adapter = netdev_priv(netdev);
1144         unsigned long lpar_rc;
1145
1146         if (napi_schedule_prep(&adapter->napi)) {
1147                 lpar_rc = h_vio_signal(adapter->vdev->unit_address,
1148                                        VIO_IRQ_DISABLE);
1149                 BUG_ON(lpar_rc != H_SUCCESS);
1150                 __napi_schedule(&adapter->napi);
1151         }
1152         return IRQ_HANDLED;
1153 }
1154
1155 static void ibmveth_set_multicast_list(struct net_device *netdev)
1156 {
1157         struct ibmveth_adapter *adapter = netdev_priv(netdev);
1158         unsigned long lpar_rc;
1159
1160         if ((netdev->flags & IFF_PROMISC) ||
1161             (netdev_mc_count(netdev) > adapter->mcastFilterSize)) {
1162                 lpar_rc = h_multicast_ctrl(adapter->vdev->unit_address,
1163                                            IbmVethMcastEnableRecv |
1164                                            IbmVethMcastDisableFiltering,
1165                                            0);
1166                 if (lpar_rc != H_SUCCESS) {
1167                         netdev_err(netdev, "h_multicast_ctrl rc=%ld when "
1168                                    "entering promisc mode\n", lpar_rc);
1169                 }
1170         } else {
1171                 struct netdev_hw_addr *ha;
1172                 /* clear the filter table & disable filtering */
1173                 lpar_rc = h_multicast_ctrl(adapter->vdev->unit_address,
1174                                            IbmVethMcastEnableRecv |
1175                                            IbmVethMcastDisableFiltering |
1176                                            IbmVethMcastClearFilterTable,
1177                                            0);
1178                 if (lpar_rc != H_SUCCESS) {
1179                         netdev_err(netdev, "h_multicast_ctrl rc=%ld when "
1180                                    "attempting to clear filter table\n",
1181                                    lpar_rc);
1182                 }
1183                 /* add the addresses to the filter table */
1184                 netdev_for_each_mc_addr(ha, netdev) {
1185                         /* add the multicast address to the filter table */
1186                         unsigned long mcast_addr = 0;
1187                         memcpy(((char *)&mcast_addr)+2, ha->addr, ETH_ALEN);
1188                         lpar_rc = h_multicast_ctrl(adapter->vdev->unit_address,
1189                                                    IbmVethMcastAddFilter,
1190                                                    mcast_addr);
1191                         if (lpar_rc != H_SUCCESS) {
1192                                 netdev_err(netdev, "h_multicast_ctrl rc=%ld "
1193                                            "when adding an entry to the filter "
1194                                            "table\n", lpar_rc);
1195                         }
1196                 }
1197
1198                 /* re-enable filtering */
1199                 lpar_rc = h_multicast_ctrl(adapter->vdev->unit_address,
1200                                            IbmVethMcastEnableFiltering,
1201                                            0);
1202                 if (lpar_rc != H_SUCCESS) {
1203                         netdev_err(netdev, "h_multicast_ctrl rc=%ld when "
1204                                    "enabling filtering\n", lpar_rc);
1205                 }
1206         }
1207 }
1208
1209 static int ibmveth_change_mtu(struct net_device *dev, int new_mtu)
1210 {
1211         struct ibmveth_adapter *adapter = netdev_priv(dev);
1212         struct vio_dev *viodev = adapter->vdev;
1213         int new_mtu_oh = new_mtu + IBMVETH_BUFF_OH;
1214         int i, rc;
1215         int need_restart = 0;
1216
1217         if (new_mtu < IBMVETH_MIN_MTU)
1218                 return -EINVAL;
1219
1220         for (i = 0; i < IBMVETH_NUM_BUFF_POOLS; i++)
1221                 if (new_mtu_oh < adapter->rx_buff_pool[i].buff_size)
1222                         break;
1223
1224         if (i == IBMVETH_NUM_BUFF_POOLS)
1225                 return -EINVAL;
1226
1227         /* Deactivate all the buffer pools so that the next loop can activate
1228            only the buffer pools necessary to hold the new MTU */
1229         if (netif_running(adapter->netdev)) {
1230                 need_restart = 1;
1231                 adapter->pool_config = 1;
1232                 ibmveth_close(adapter->netdev);
1233                 adapter->pool_config = 0;
1234         }
1235
1236         /* Look for an active buffer pool that can hold the new MTU */
1237         for (i = 0; i < IBMVETH_NUM_BUFF_POOLS; i++) {
1238                 adapter->rx_buff_pool[i].active = 1;
1239
1240                 if (new_mtu_oh < adapter->rx_buff_pool[i].buff_size) {
1241                         dev->mtu = new_mtu;
1242                         vio_cmo_set_dev_desired(viodev,
1243                                                 ibmveth_get_desired_dma
1244                                                 (viodev));
1245                         if (need_restart) {
1246                                 return ibmveth_open(adapter->netdev);
1247                         }
1248                         return 0;
1249                 }
1250         }
1251
1252         if (need_restart && (rc = ibmveth_open(adapter->netdev)))
1253                 return rc;
1254
1255         return -EINVAL;
1256 }
1257
1258 #ifdef CONFIG_NET_POLL_CONTROLLER
1259 static void ibmveth_poll_controller(struct net_device *dev)
1260 {
1261         ibmveth_replenish_task(netdev_priv(dev));
1262         ibmveth_interrupt(dev->irq, dev);
1263 }
1264 #endif
1265
1266 /**
1267  * ibmveth_get_desired_dma - Calculate IO memory desired by the driver
1268  *
1269  * @vdev: struct vio_dev for the device whose desired IO mem is to be returned
1270  *
1271  * Return value:
1272  *      Number of bytes of IO data the driver will need to perform well.
1273  */
1274 static unsigned long ibmveth_get_desired_dma(struct vio_dev *vdev)
1275 {
1276         struct net_device *netdev = dev_get_drvdata(&vdev->dev);
1277         struct ibmveth_adapter *adapter;
1278         unsigned long ret;
1279         int i;
1280         int rxqentries = 1;
1281
1282         /* netdev inits at probe time along with the structures we need below*/
1283         if (netdev == NULL)
1284                 return IOMMU_PAGE_ALIGN(IBMVETH_IO_ENTITLEMENT_DEFAULT);
1285
1286         adapter = netdev_priv(netdev);
1287
1288         ret = IBMVETH_BUFF_LIST_SIZE + IBMVETH_FILT_LIST_SIZE;
1289         ret += IOMMU_PAGE_ALIGN(netdev->mtu);
1290
1291         for (i = 0; i < IBMVETH_NUM_BUFF_POOLS; i++) {
1292                 /* add the size of the active receive buffers */
1293                 if (adapter->rx_buff_pool[i].active)
1294                         ret +=
1295                             adapter->rx_buff_pool[i].size *
1296                             IOMMU_PAGE_ALIGN(adapter->rx_buff_pool[i].
1297                                     buff_size);
1298                 rxqentries += adapter->rx_buff_pool[i].size;
1299         }
1300         /* add the size of the receive queue entries */
1301         ret += IOMMU_PAGE_ALIGN(rxqentries * sizeof(struct ibmveth_rx_q_entry));
1302
1303         return ret;
1304 }
1305
1306 static const struct net_device_ops ibmveth_netdev_ops = {
1307         .ndo_open               = ibmveth_open,
1308         .ndo_stop               = ibmveth_close,
1309         .ndo_start_xmit         = ibmveth_start_xmit,
1310         .ndo_set_rx_mode        = ibmveth_set_multicast_list,
1311         .ndo_do_ioctl           = ibmveth_ioctl,
1312         .ndo_change_mtu         = ibmveth_change_mtu,
1313         .ndo_fix_features       = ibmveth_fix_features,
1314         .ndo_set_features       = ibmveth_set_features,
1315         .ndo_validate_addr      = eth_validate_addr,
1316         .ndo_set_mac_address    = eth_mac_addr,
1317 #ifdef CONFIG_NET_POLL_CONTROLLER
1318         .ndo_poll_controller    = ibmveth_poll_controller,
1319 #endif
1320 };
1321
1322 static int ibmveth_probe(struct vio_dev *dev, const struct vio_device_id *id)
1323 {
1324         int rc, i, mac_len;
1325         struct net_device *netdev;
1326         struct ibmveth_adapter *adapter;
1327         unsigned char *mac_addr_p;
1328         unsigned int *mcastFilterSize_p;
1329
1330         dev_dbg(&dev->dev, "entering ibmveth_probe for UA 0x%x\n",
1331                 dev->unit_address);
1332
1333         mac_addr_p = (unsigned char *)vio_get_attribute(dev, VETH_MAC_ADDR,
1334                                                         &mac_len);
1335         if (!mac_addr_p) {
1336                 dev_err(&dev->dev, "Can't find VETH_MAC_ADDR attribute\n");
1337                 return -EINVAL;
1338         }
1339         /* Workaround for old/broken pHyp */
1340         if (mac_len == 8)
1341                 mac_addr_p += 2;
1342         else if (mac_len != 6) {
1343                 dev_err(&dev->dev, "VETH_MAC_ADDR attribute wrong len %d\n",
1344                         mac_len);
1345                 return -EINVAL;
1346         }
1347
1348         mcastFilterSize_p = (unsigned int *)vio_get_attribute(dev,
1349                                                 VETH_MCAST_FILTER_SIZE, NULL);
1350         if (!mcastFilterSize_p) {
1351                 dev_err(&dev->dev, "Can't find VETH_MCAST_FILTER_SIZE "
1352                         "attribute\n");
1353                 return -EINVAL;
1354         }
1355
1356         netdev = alloc_etherdev(sizeof(struct ibmveth_adapter));
1357
1358         if (!netdev)
1359                 return -ENOMEM;
1360
1361         adapter = netdev_priv(netdev);
1362         dev_set_drvdata(&dev->dev, netdev);
1363
1364         adapter->vdev = dev;
1365         adapter->netdev = netdev;
1366         adapter->mcastFilterSize = *mcastFilterSize_p;
1367         adapter->pool_config = 0;
1368
1369         netif_napi_add(netdev, &adapter->napi, ibmveth_poll, 16);
1370
1371         adapter->mac_addr = 0;
1372         memcpy(&adapter->mac_addr, mac_addr_p, ETH_ALEN);
1373
1374         netdev->irq = dev->irq;
1375         netdev->netdev_ops = &ibmveth_netdev_ops;
1376         netdev->ethtool_ops = &netdev_ethtool_ops;
1377         SET_NETDEV_DEV(netdev, &dev->dev);
1378         netdev->hw_features = NETIF_F_SG | NETIF_F_RXCSUM |
1379                 NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM;
1380         netdev->features |= netdev->hw_features;
1381
1382         memcpy(netdev->dev_addr, &adapter->mac_addr, netdev->addr_len);
1383
1384         for (i = 0; i < IBMVETH_NUM_BUFF_POOLS; i++) {
1385                 struct kobject *kobj = &adapter->rx_buff_pool[i].kobj;
1386                 int error;
1387
1388                 ibmveth_init_buffer_pool(&adapter->rx_buff_pool[i], i,
1389                                          pool_count[i], pool_size[i],
1390                                          pool_active[i]);
1391                 error = kobject_init_and_add(kobj, &ktype_veth_pool,
1392                                              &dev->dev.kobj, "pool%d", i);
1393                 if (!error)
1394                         kobject_uevent(kobj, KOBJ_ADD);
1395         }
1396
1397         netdev_dbg(netdev, "adapter @ 0x%p\n", adapter);
1398
1399         adapter->buffer_list_dma = DMA_ERROR_CODE;
1400         adapter->filter_list_dma = DMA_ERROR_CODE;
1401         adapter->rx_queue.queue_dma = DMA_ERROR_CODE;
1402
1403         netdev_dbg(netdev, "registering netdev...\n");
1404
1405         ibmveth_set_features(netdev, netdev->features);
1406
1407         rc = register_netdev(netdev);
1408
1409         if (rc) {
1410                 netdev_dbg(netdev, "failed to register netdev rc=%d\n", rc);
1411                 free_netdev(netdev);
1412                 return rc;
1413         }
1414
1415         netdev_dbg(netdev, "registered\n");
1416
1417         return 0;
1418 }
1419
1420 static int ibmveth_remove(struct vio_dev *dev)
1421 {
1422         struct net_device *netdev = dev_get_drvdata(&dev->dev);
1423         struct ibmveth_adapter *adapter = netdev_priv(netdev);
1424         int i;
1425
1426         for (i = 0; i < IBMVETH_NUM_BUFF_POOLS; i++)
1427                 kobject_put(&adapter->rx_buff_pool[i].kobj);
1428
1429         unregister_netdev(netdev);
1430
1431         free_netdev(netdev);
1432         dev_set_drvdata(&dev->dev, NULL);
1433
1434         return 0;
1435 }
1436
1437 static struct attribute veth_active_attr;
1438 static struct attribute veth_num_attr;
1439 static struct attribute veth_size_attr;
1440
1441 static ssize_t veth_pool_show(struct kobject *kobj,
1442                               struct attribute *attr, char *buf)
1443 {
1444         struct ibmveth_buff_pool *pool = container_of(kobj,
1445                                                       struct ibmveth_buff_pool,
1446                                                       kobj);
1447
1448         if (attr == &veth_active_attr)
1449                 return sprintf(buf, "%d\n", pool->active);
1450         else if (attr == &veth_num_attr)
1451                 return sprintf(buf, "%d\n", pool->size);
1452         else if (attr == &veth_size_attr)
1453                 return sprintf(buf, "%d\n", pool->buff_size);
1454         return 0;
1455 }
1456
1457 static ssize_t veth_pool_store(struct kobject *kobj, struct attribute *attr,
1458                                const char *buf, size_t count)
1459 {
1460         struct ibmveth_buff_pool *pool = container_of(kobj,
1461                                                       struct ibmveth_buff_pool,
1462                                                       kobj);
1463         struct net_device *netdev = dev_get_drvdata(
1464             container_of(kobj->parent, struct device, kobj));
1465         struct ibmveth_adapter *adapter = netdev_priv(netdev);
1466         long value = simple_strtol(buf, NULL, 10);
1467         long rc;
1468
1469         if (attr == &veth_active_attr) {
1470                 if (value && !pool->active) {
1471                         if (netif_running(netdev)) {
1472                                 if (ibmveth_alloc_buffer_pool(pool)) {
1473                                         netdev_err(netdev,
1474                                                    "unable to alloc pool\n");
1475                                         return -ENOMEM;
1476                                 }
1477                                 pool->active = 1;
1478                                 adapter->pool_config = 1;
1479                                 ibmveth_close(netdev);
1480                                 adapter->pool_config = 0;
1481                                 if ((rc = ibmveth_open(netdev)))
1482                                         return rc;
1483                         } else {
1484                                 pool->active = 1;
1485                         }
1486                 } else if (!value && pool->active) {
1487                         int mtu = netdev->mtu + IBMVETH_BUFF_OH;
1488                         int i;
1489                         /* Make sure there is a buffer pool with buffers that
1490                            can hold a packet of the size of the MTU */
1491                         for (i = 0; i < IBMVETH_NUM_BUFF_POOLS; i++) {
1492                                 if (pool == &adapter->rx_buff_pool[i])
1493                                         continue;
1494                                 if (!adapter->rx_buff_pool[i].active)
1495                                         continue;
1496                                 if (mtu <= adapter->rx_buff_pool[i].buff_size)
1497                                         break;
1498                         }
1499
1500                         if (i == IBMVETH_NUM_BUFF_POOLS) {
1501                                 netdev_err(netdev, "no active pool >= MTU\n");
1502                                 return -EPERM;
1503                         }
1504
1505                         if (netif_running(netdev)) {
1506                                 adapter->pool_config = 1;
1507                                 ibmveth_close(netdev);
1508                                 pool->active = 0;
1509                                 adapter->pool_config = 0;
1510                                 if ((rc = ibmveth_open(netdev)))
1511                                         return rc;
1512                         }
1513                         pool->active = 0;
1514                 }
1515         } else if (attr == &veth_num_attr) {
1516                 if (value <= 0 || value > IBMVETH_MAX_POOL_COUNT) {
1517                         return -EINVAL;
1518                 } else {
1519                         if (netif_running(netdev)) {
1520                                 adapter->pool_config = 1;
1521                                 ibmveth_close(netdev);
1522                                 adapter->pool_config = 0;
1523                                 pool->size = value;
1524                                 if ((rc = ibmveth_open(netdev)))
1525                                         return rc;
1526                         } else {
1527                                 pool->size = value;
1528                         }
1529                 }
1530         } else if (attr == &veth_size_attr) {
1531                 if (value <= IBMVETH_BUFF_OH || value > IBMVETH_MAX_BUF_SIZE) {
1532                         return -EINVAL;
1533                 } else {
1534                         if (netif_running(netdev)) {
1535                                 adapter->pool_config = 1;
1536                                 ibmveth_close(netdev);
1537                                 adapter->pool_config = 0;
1538                                 pool->buff_size = value;
1539                                 if ((rc = ibmveth_open(netdev)))
1540                                         return rc;
1541                         } else {
1542                                 pool->buff_size = value;
1543                         }
1544                 }
1545         }
1546
1547         /* kick the interrupt handler to allocate/deallocate pools */
1548         ibmveth_interrupt(netdev->irq, netdev);
1549         return count;
1550 }
1551
1552
1553 #define ATTR(_name, _mode)                              \
1554         struct attribute veth_##_name##_attr = {        \
1555         .name = __stringify(_name), .mode = _mode,      \
1556         };
1557
1558 static ATTR(active, 0644);
1559 static ATTR(num, 0644);
1560 static ATTR(size, 0644);
1561
1562 static struct attribute *veth_pool_attrs[] = {
1563         &veth_active_attr,
1564         &veth_num_attr,
1565         &veth_size_attr,
1566         NULL,
1567 };
1568
1569 static const struct sysfs_ops veth_pool_ops = {
1570         .show   = veth_pool_show,
1571         .store  = veth_pool_store,
1572 };
1573
1574 static struct kobj_type ktype_veth_pool = {
1575         .release        = NULL,
1576         .sysfs_ops      = &veth_pool_ops,
1577         .default_attrs  = veth_pool_attrs,
1578 };
1579
1580 static int ibmveth_resume(struct device *dev)
1581 {
1582         struct net_device *netdev = dev_get_drvdata(dev);
1583         ibmveth_interrupt(netdev->irq, netdev);
1584         return 0;
1585 }
1586
1587 static struct vio_device_id ibmveth_device_table[] = {
1588         { "network", "IBM,l-lan"},
1589         { "", "" }
1590 };
1591 MODULE_DEVICE_TABLE(vio, ibmveth_device_table);
1592
1593 static struct dev_pm_ops ibmveth_pm_ops = {
1594         .resume = ibmveth_resume
1595 };
1596
1597 static struct vio_driver ibmveth_driver = {
1598         .id_table       = ibmveth_device_table,
1599         .probe          = ibmveth_probe,
1600         .remove         = ibmveth_remove,
1601         .get_desired_dma = ibmveth_get_desired_dma,
1602         .name           = ibmveth_driver_name,
1603         .pm             = &ibmveth_pm_ops,
1604 };
1605
1606 static int __init ibmveth_module_init(void)
1607 {
1608         printk(KERN_DEBUG "%s: %s %s\n", ibmveth_driver_name,
1609                ibmveth_driver_string, ibmveth_driver_version);
1610
1611         return vio_register_driver(&ibmveth_driver);
1612 }
1613
1614 static void __exit ibmveth_module_exit(void)
1615 {
1616         vio_unregister_driver(&ibmveth_driver);
1617 }
1618
1619 module_init(ibmveth_module_init);
1620 module_exit(ibmveth_module_exit);