drivers/net/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, write to the Free Software
  16  * Foundation, Inc., 59 Temple Place - Suite 330, Boston, MA 02111-1307, USA.
  17  *
  18  * Copyright (C) IBM Corporation, 2003, 2010
  19  *
  20  * Authors: Dave Larson <larson1@us.ibm.com>
  21  *          Santiago Leon <santil@linux.vnet.ibm.com>
  22  *          Brian King <brking@linux.vnet.ibm.com>
  23  *          Robert Jennings <rcj@linux.vnet.ibm.com>
  24  *          Anton Blanchard <anton@au.ibm.com>
  25  */
  26
  27 #include <linux/module.h>
  28 #include <linux/moduleparam.h>
  29 #include <linux/types.h>
  30 #include <linux/errno.h>
  31 #include <linux/dma-mapping.h>
  32 #include <linux/kernel.h>
  33 #include <linux/netdevice.h>
  34 #include <linux/etherdevice.h>
  35 #include <linux/skbuff.h>
  36 #include <linux/init.h>
  37 #include <linux/interrupt.h>
  38 #include <linux/mm.h>
  39 #include <linux/pm.h>
  40 #include <linux/ethtool.h>
  41 #include <linux/in.h>
  42 #include <linux/ip.h>
  43 #include <linux/ipv6.h>
  44 #include <linux/slab.h>
  45 #include <asm/hvcall.h>
  46 #include <linux/atomic.h>
  47 #include <asm/vio.h>
  48 #include <asm/iommu.h>
  49 #include <asm/firmware.h>
  50
  51 #include "ibmveth.h"
  52
  53 static irqreturn_t ibmveth_interrupt(int irq, void *dev_instance);
  54 static void ibmveth_rxq_harvest_buffer(struct ibmveth_adapter *adapter);
  55 static unsigned long ibmveth_get_desired_dma(struct vio_dev *vdev);
  56
  57 static struct kobj_type ktype_veth_pool;
  58
  59
  60 static const char ibmveth_driver_name[] = "ibmveth";
  61 static const char ibmveth_driver_string[] = "IBM Power Virtual Ethernet Driver";
  62 #define ibmveth_driver_version "1.04"
  63
  64 MODULE_AUTHOR("Santiago Leon <santil@linux.vnet.ibm.com>");
  65 MODULE_DESCRIPTION("IBM Power Virtual Ethernet Driver");
  66 MODULE_LICENSE("GPL");
  67 MODULE_VERSION(ibmveth_driver_version);
  68
  69 static unsigned int tx_copybreak __read_mostly = 128;
  70 module_param(tx_copybreak, uint, 0644);
  71 MODULE_PARM_DESC(tx_copybreak,
  72         "Maximum size of packet that is copied to a new buffer on transmit");
  73
  74 static unsigned int rx_copybreak __read_mostly = 128;
  75 module_param(rx_copybreak, uint, 0644);
  76 MODULE_PARM_DESC(rx_copybreak,
  77         "Maximum size of packet that is copied to a new buffer on receive");
  78
  79 static unsigned int rx_flush __read_mostly = 0;
  80 module_param(rx_flush, uint, 0644);
  81 MODULE_PARM_DESC(rx_flush, "Flush receive buffers before use");
  82
  83 struct ibmveth_stat {
  84         char name[ETH_GSTRING_LEN];
  85         int offset;
  86 };
  87
  88 #define IBMVETH_STAT_OFF(stat) offsetof(struct ibmveth_adapter, stat)
  89 #define IBMVETH_GET_STAT(a, off) *((u64 *)(((unsigned long)(a)) + off))
  90
  91 struct ibmveth_stat ibmveth_stats[] = {
  92         { "replenish_task_cycles", IBMVETH_STAT_OFF(replenish_task_cycles) },
  93         { "replenish_no_mem", IBMVETH_STAT_OFF(replenish_no_mem) },
  94         { "replenish_add_buff_failure",
  95                         IBMVETH_STAT_OFF(replenish_add_buff_failure) },
  96         { "replenish_add_buff_success",
  97                         IBMVETH_STAT_OFF(replenish_add_buff_success) },
  98         { "rx_invalid_buffer", IBMVETH_STAT_OFF(rx_invalid_buffer) },
  99         { "rx_no_buffer", IBMVETH_STAT_OFF(rx_no_buffer) },
 100         { "tx_map_failed", IBMVETH_STAT_OFF(tx_map_failed) },
 101         { "tx_send_failed", IBMVETH_STAT_OFF(tx_send_failed) },
 102         { "fw_enabled_ipv4_csum", IBMVETH_STAT_OFF(fw_ipv4_csum_support) },
 103         { "fw_enabled_ipv6_csum", IBMVETH_STAT_OFF(fw_ipv6_csum_support) },
 104 };
 105
 106 /* simple methods of getting data from the current rxq entry */
 107 static inline u32 ibmveth_rxq_flags(struct ibmveth_adapter *adapter)
 108 {
 109         return adapter->rx_queue.queue_addr[adapter->rx_queue.index].flags_off;
 110 }
 111
 112 static inline int ibmveth_rxq_toggle(struct ibmveth_adapter *adapter)
 113 {
 114         return (ibmveth_rxq_flags(adapter) & IBMVETH_RXQ_TOGGLE) >>
 115                         IBMVETH_RXQ_TOGGLE_SHIFT;
 116 }
 117
 118 static inline int ibmveth_rxq_pending_buffer(struct ibmveth_adapter *adapter)
 119 {
 120         return ibmveth_rxq_toggle(adapter) == adapter->rx_queue.toggle;
 121 }
 122
 123 static inline int ibmveth_rxq_buffer_valid(struct ibmveth_adapter *adapter)
 124 {
 125         return ibmveth_rxq_flags(adapter) & IBMVETH_RXQ_VALID;
 126 }
 127
 128 static inline int ibmveth_rxq_frame_offset(struct ibmveth_adapter *adapter)
 129 {
 130         return ibmveth_rxq_flags(adapter) & IBMVETH_RXQ_OFF_MASK;
 131 }
 132
 133 static inline int ibmveth_rxq_frame_length(struct ibmveth_adapter *adapter)
 134 {
 135         return adapter->rx_queue.queue_addr[adapter->rx_queue.index].length;
 136 }
 137
 138 static inline int ibmveth_rxq_csum_good(struct ibmveth_adapter *adapter)
 139 {
 140         return ibmveth_rxq_flags(adapter) & IBMVETH_RXQ_CSUM_GOOD;
 141 }
 142
 143 /* setup the initial settings for a buffer pool */
 144 static void ibmveth_init_buffer_pool(struct ibmveth_buff_pool *pool,
 145                                      u32 pool_index, u32 pool_size,
 146                                      u32 buff_size, u32 pool_active)
 147 {
 148         pool->size = pool_size;
 149         pool->index = pool_index;
 150         pool->buff_size = buff_size;
 151         pool->threshold = pool_size * 7 / 8;
 152         pool->active = pool_active;
 153 }
 154
 155 /* allocate and setup an buffer pool - called during open */
 156 static int ibmveth_alloc_buffer_pool(struct ibmveth_buff_pool *pool)
 157 {
 158         int i;
 159
 160         pool->free_map = kmalloc(sizeof(u16) * pool->size, GFP_KERNEL);
 161
 162         if (!pool->free_map)
 163                 return -1;
 164
 165         pool->dma_addr = kmalloc(sizeof(dma_addr_t) * pool->size, GFP_KERNEL);
 166         if (!pool->dma_addr) {
 167                 kfree(pool->free_map);
 168                 pool->free_map = NULL;
 169                 return -1;
 170         }
 171
 172         pool->skbuff = kcalloc(pool->size, sizeof(void *), GFP_KERNEL);
 173
 174         if (!pool->skbuff) {
 175                 kfree(pool->dma_addr);
 176                 pool->dma_addr = NULL;
 177
 178                 kfree(pool->free_map);
 179                 pool->free_map = NULL;
 180                 return -1;
 181         }
 182
 183         memset(pool->dma_addr, 0, sizeof(dma_addr_t) * pool->size);
 184
 185         for (i = 0; i < pool->size; ++i)
 186                 pool->free_map[i] = i;
 187
 188         atomic_set(&pool->available, 0);
 189         pool->producer_index = 0;
 190         pool->consumer_index = 0;
 191
 192         return 0;
 193 }
 194
 195 static inline void ibmveth_flush_buffer(void *addr, unsigned long length)
 196 {
 197         unsigned long offset;
 198
 199         for (offset = 0; offset < length; offset += SMP_CACHE_BYTES)
 200                 asm("dcbfl %0,%1" :: "b" (addr), "r" (offset));
 201 }
 202
 203 /* replenish the buffers for a pool.  note that we don't need to
 204  * skb_reserve these since they are used for incoming...
 205  */
 206 static void ibmveth_replenish_buffer_pool(struct ibmveth_adapter *adapter,
 207                                           struct ibmveth_buff_pool *pool)
 208 {
 209         u32 i;
 210         u32 count = pool->size - atomic_read(&pool->available);
 211         u32 buffers_added = 0;
 212         struct sk_buff *skb;
 213         unsigned int free_index, index;
 214         u64 correlator;
 215         unsigned long lpar_rc;
 216         dma_addr_t dma_addr;
 217
 218         mb();
 219
 220         for (i = 0; i < count; ++i) {
 221                 union ibmveth_buf_desc desc;
 222
 223                 skb = netdev_alloc_skb(adapter->netdev, pool->buff_size);
 224
 225                 if (!skb) {
 226                         netdev_dbg(adapter->netdev,
 227                                    "replenish: unable to allocate skb\n");
 228                         adapter->replenish_no_mem++;
 229                         break;
 230                 }
 231
 232                 free_index = pool->consumer_index;
 233                 pool->consumer_index++;
 234                 if (pool->consumer_index >= pool->size)
 235                         pool->consumer_index = 0;
 236                 index = pool->free_map[free_index];
 237
 238                 BUG_ON(index == IBM_VETH_INVALID_MAP);
 239                 BUG_ON(pool->skbuff[index] != NULL);
 240
 241                 dma_addr = dma_map_single(&adapter->vdev->dev, skb->data,
 242                                 pool->buff_size, DMA_FROM_DEVICE);
 243
 244                 if (dma_mapping_error(&adapter->vdev->dev, dma_addr))
 245                         goto failure;
 246
 247                 pool->free_map[free_index] = IBM_VETH_INVALID_MAP;
 248                 pool->dma_addr[index] = dma_addr;
 249                 pool->skbuff[index] = skb;
 250
 251                 correlator = ((u64)pool->index << 32) | index;
 252                 *(u64 *)skb->data = correlator;
 253
 254                 desc.fields.flags_len = IBMVETH_BUF_VALID | pool->buff_size;
 255                 desc.fields.address = dma_addr;
 256
 257                 if (rx_flush) {
 258                         unsigned int len = min(pool->buff_size,
 259                                                 adapter->netdev->mtu +
 260                                                 IBMVETH_BUFF_OH);
 261                         ibmveth_flush_buffer(skb->data, len);
 262                 }
 263                 lpar_rc = h_add_logical_lan_buffer(adapter->vdev->unit_address,
 264                                                    desc.desc);
 265
 266                 if (lpar_rc != H_SUCCESS) {
 267                         goto failure;
 268                 } else {
 269                         buffers_added++;
 270                         adapter->replenish_add_buff_success++;
 271                 }
 272         }
 273
 274         mb();
 275         atomic_add(buffers_added, &(pool->available));
 276         return;
 277
 278 failure:
 279         pool->free_map[free_index] = index;
 280         pool->skbuff[index] = NULL;
 281         if (pool->consumer_index == 0)
 282                 pool->consumer_index = pool->size - 1;
 283         else
 284                 pool->consumer_index--;
 285         if (!dma_mapping_error(&adapter->vdev->dev, dma_addr))
 286                 dma_unmap_single(&adapter->vdev->dev,
 287                                  pool->dma_addr[index], pool->buff_size,
 288                                  DMA_FROM_DEVICE);
 289         dev_kfree_skb_any(skb);
 290         adapter->replenish_add_buff_failure++;
 291
 292         mb();
 293         atomic_add(buffers_added, &(pool->available));
 294 }
 295
 296 /* replenish routine */
 297 static void ibmveth_replenish_task(struct ibmveth_adapter *adapter)
 298 {
 299         int i;
 300
 301         adapter->replenish_task_cycles++;
 302
 303         for (i = (IBMVETH_NUM_BUFF_POOLS - 1); i >= 0; i--) {
 304                 struct ibmveth_buff_pool *pool = &adapter->rx_buff_pool[i];
 305
 306                 if (pool->active &&
 307                     (atomic_read(&pool->available) < pool->threshold))
 308                         ibmveth_replenish_buffer_pool(adapter, pool);
 309         }
 310
 311         adapter->rx_no_buffer = *(u64 *)(((char*)adapter->buffer_list_addr) +
 312                                                 4096 - 8);
 313 }
 314
 315 /* empty and free ana buffer pool - also used to do cleanup in error paths */
 316 static void ibmveth_free_buffer_pool(struct ibmveth_adapter *adapter,
 317                                      struct ibmveth_buff_pool *pool)
 318 {
 319         int i;
 320
 321         kfree(pool->free_map);
 322         pool->free_map = NULL;
 323
 324         if (pool->skbuff && pool->dma_addr) {
 325                 for (i = 0; i < pool->size; ++i) {
 326                         struct sk_buff *skb = pool->skbuff[i];
 327                         if (skb) {
 328                                 dma_unmap_single(&adapter->vdev->dev,
 329                                                  pool->dma_addr[i],
 330                                                  pool->buff_size,
 331                                                  DMA_FROM_DEVICE);
 332                                 dev_kfree_skb_any(skb);
 333                                 pool->skbuff[i] = NULL;
 334                         }
 335                 }
 336         }
 337
 338         if (pool->dma_addr) {
 339                 kfree(pool->dma_addr);
 340                 pool->dma_addr = NULL;
 341         }
 342
 343         if (pool->skbuff) {
 344                 kfree(pool->skbuff);
 345                 pool->skbuff = NULL;
 346         }
 347 }
 348
 349 /* remove a buffer from a pool */
 350 static void ibmveth_remove_buffer_from_pool(struct ibmveth_adapter *adapter,
 351                                             u64 correlator)
 352 {
 353         unsigned int pool  = correlator >> 32;
 354         unsigned int index = correlator & 0xffffffffUL;
 355         unsigned int free_index;
 356         struct sk_buff *skb;
 357
 358         BUG_ON(pool >= IBMVETH_NUM_BUFF_POOLS);
 359         BUG_ON(index >= adapter->rx_buff_pool[pool].size);
 360
 361         skb = adapter->rx_buff_pool[pool].skbuff[index];
 362
 363         BUG_ON(skb == NULL);
 364
 365         adapter->rx_buff_pool[pool].skbuff[index] = NULL;
 366
 367         dma_unmap_single(&adapter->vdev->dev,
 368                          adapter->rx_buff_pool[pool].dma_addr[index],
 369                          adapter->rx_buff_pool[pool].buff_size,
 370                          DMA_FROM_DEVICE);
 371
 372         free_index = adapter->rx_buff_pool[pool].producer_index;
 373         adapter->rx_buff_pool[pool].producer_index++;
 374         if (adapter->rx_buff_pool[pool].producer_index >=
 375             adapter->rx_buff_pool[pool].size)
 376                 adapter->rx_buff_pool[pool].producer_index = 0;
 377         adapter->rx_buff_pool[pool].free_map[free_index] = index;
 378
 379         mb();
 380
 381         atomic_dec(&(adapter->rx_buff_pool[pool].available));
 382 }
 383
 384 /* get the current buffer on the rx queue */
 385 static inline struct sk_buff *ibmveth_rxq_get_buffer(struct ibmveth_adapter *adapter)
 386 {
 387         u64 correlator = adapter->rx_queue.queue_addr[adapter->rx_queue.index].correlator;
 388         unsigned int pool = correlator >> 32;
 389         unsigned int index = correlator & 0xffffffffUL;
 390
 391         BUG_ON(pool >= IBMVETH_NUM_BUFF_POOLS);
 392         BUG_ON(index >= adapter->rx_buff_pool[pool].size);
 393
 394         return adapter->rx_buff_pool[pool].skbuff[index];
 395 }
 396
 397 /* recycle the current buffer on the rx queue */
 398 static int ibmveth_rxq_recycle_buffer(struct ibmveth_adapter *adapter)
 399 {
 400         u32 q_index = adapter->rx_queue.index;
 401         u64 correlator = adapter->rx_queue.queue_addr[q_index].correlator;
 402         unsigned int pool = correlator >> 32;
 403         unsigned int index = correlator & 0xffffffffUL;
 404         union ibmveth_buf_desc desc;
 405         unsigned long lpar_rc;
 406         int ret = 1;
 407
 408         BUG_ON(pool >= IBMVETH_NUM_BUFF_POOLS);
 409         BUG_ON(index >= adapter->rx_buff_pool[pool].size);
 410
 411         if (!adapter->rx_buff_pool[pool].active) {
 412                 ibmveth_rxq_harvest_buffer(adapter);
 413                 ibmveth_free_buffer_pool(adapter, &adapter->rx_buff_pool[pool]);
 414                 goto out;
 415         }
 416
 417         desc.fields.flags_len = IBMVETH_BUF_VALID |
 418                 adapter->rx_buff_pool[pool].buff_size;
 419         desc.fields.address = adapter->rx_buff_pool[pool].dma_addr[index];
 420
 421         lpar_rc = h_add_logical_lan_buffer(adapter->vdev->unit_address, desc.desc);
 422
 423         if (lpar_rc != H_SUCCESS) {
 424                 netdev_dbg(adapter->netdev, "h_add_logical_lan_buffer failed "
 425                            "during recycle rc=%ld", lpar_rc);
 426                 ibmveth_remove_buffer_from_pool(adapter, adapter->rx_queue.queue_addr[adapter->rx_queue.index].correlator);
 427                 ret = 0;
 428         }
 429
 430         if (++adapter->rx_queue.index == adapter->rx_queue.num_slots) {
 431                 adapter->rx_queue.index = 0;
 432                 adapter->rx_queue.toggle = !adapter->rx_queue.toggle;
 433         }
 434
 435 out:
 436         return ret;
 437 }
 438
 439 static void ibmveth_rxq_harvest_buffer(struct ibmveth_adapter *adapter)
 440 {
 441         ibmveth_remove_buffer_from_pool(adapter, adapter->rx_queue.queue_addr[adapter->rx_queue.index].correlator);
 442
 443         if (++adapter->rx_queue.index == adapter->rx_queue.num_slots) {
 444                 adapter->rx_queue.index = 0;
 445                 adapter->rx_queue.toggle = !adapter->rx_queue.toggle;
 446         }
 447 }
 448
 449 static void ibmveth_cleanup(struct ibmveth_adapter *adapter)
 450 {
 451         int i;
 452         struct device *dev = &adapter->vdev->dev;
 453
 454         if (adapter->buffer_list_addr != NULL) {
 455                 if (!dma_mapping_error(dev, adapter->buffer_list_dma)) {
 456                         dma_unmap_single(dev, adapter->buffer_list_dma, 4096,
 457                                         DMA_BIDIRECTIONAL);
 458                         adapter->buffer_list_dma = DMA_ERROR_CODE;
 459                 }
 460                 free_page((unsigned long)adapter->buffer_list_addr);
 461                 adapter->buffer_list_addr = NULL;
 462         }
 463
 464         if (adapter->filter_list_addr != NULL) {
 465                 if (!dma_mapping_error(dev, adapter->filter_list_dma)) {
 466                         dma_unmap_single(dev, adapter->filter_list_dma, 4096,
 467                                         DMA_BIDIRECTIONAL);
 468                         adapter->filter_list_dma = DMA_ERROR_CODE;
 469                 }
 470                 free_page((unsigned long)adapter->filter_list_addr);
 471                 adapter->filter_list_addr = NULL;
 472         }
 473
 474         if (adapter->rx_queue.queue_addr != NULL) {
 475                 if (!dma_mapping_error(dev, adapter->rx_queue.queue_dma)) {
 476                         dma_unmap_single(dev,
 477                                         adapter->rx_queue.queue_dma,
 478                                         adapter->rx_queue.queue_len,
 479                                         DMA_BIDIRECTIONAL);
 480                         adapter->rx_queue.queue_dma = DMA_ERROR_CODE;
 481                 }
 482                 kfree(adapter->rx_queue.queue_addr);
 483                 adapter->rx_queue.queue_addr = NULL;
 484         }
 485
 486         for (i = 0; i < IBMVETH_NUM_BUFF_POOLS; i++)
 487                 if (adapter->rx_buff_pool[i].active)
 488                         ibmveth_free_buffer_pool(adapter,
 489                                                  &adapter->rx_buff_pool[i]);
 490
 491         if (adapter->bounce_buffer != NULL) {
 492                 if (!dma_mapping_error(dev, adapter->bounce_buffer_dma)) {
 493                         dma_unmap_single(&adapter->vdev->dev,
 494                                         adapter->bounce_buffer_dma,
 495                                         adapter->netdev->mtu + IBMVETH_BUFF_OH,
 496                                         DMA_BIDIRECTIONAL);
 497                         adapter->bounce_buffer_dma = DMA_ERROR_CODE;
 498                 }
 499                 kfree(adapter->bounce_buffer);
 500                 adapter->bounce_buffer = NULL;
 501         }
 502 }
 503
 504 static int ibmveth_register_logical_lan(struct ibmveth_adapter *adapter,
 505         union ibmveth_buf_desc rxq_desc, u64 mac_address)
 506 {
 507         int rc, try_again = 1;
 508
 509         /*
 510          * After a kexec the adapter will still be open, so our attempt to
 511          * open it will fail. So if we get a failure we free the adapter and
 512          * try again, but only once.
 513          */
 514 retry:
 515         rc = h_register_logical_lan(adapter->vdev->unit_address,
 516                                     adapter->buffer_list_dma, rxq_desc.desc,
 517                                     adapter->filter_list_dma, mac_address);
 518
 519         if (rc != H_SUCCESS && try_again) {
 520                 do {
 521                         rc = h_free_logical_lan(adapter->vdev->unit_address);
 522                 } while (H_IS_LONG_BUSY(rc) || (rc == H_BUSY));
 523
 524                 try_again = 0;
 525                 goto retry;
 526         }
 527
 528         return rc;
 529 }
 530
 531 static int ibmveth_open(struct net_device *netdev)
 532 {
 533         struct ibmveth_adapter *adapter = netdev_priv(netdev);
 534         u64 mac_address = 0;
 535         int rxq_entries = 1;
 536         unsigned long lpar_rc;
 537         int rc;
 538         union ibmveth_buf_desc rxq_desc;
 539         int i;
 540         struct device *dev;
 541
 542         netdev_dbg(netdev, "open starting\n");
 543
 544         napi_enable(&adapter->napi);
 545
 546         for(i = 0; i < IBMVETH_NUM_BUFF_POOLS; i++)
 547                 rxq_entries += adapter->rx_buff_pool[i].size;
 548
 549         adapter->buffer_list_addr = (void*) get_zeroed_page(GFP_KERNEL);
 550         adapter->filter_list_addr = (void*) get_zeroed_page(GFP_KERNEL);
 551
 552         if (!adapter->buffer_list_addr || !adapter->filter_list_addr) {
 553                 netdev_err(netdev, "unable to allocate filter or buffer list "
 554                            "pages\n");
 555                 rc = -ENOMEM;
 556                 goto err_out;
 557         }
 558
 559         adapter->rx_queue.queue_len = sizeof(struct ibmveth_rx_q_entry) *
 560                                                 rxq_entries;
 561         adapter->rx_queue.queue_addr = kmalloc(adapter->rx_queue.queue_len,
 562                                                 GFP_KERNEL);
 563
 564         if (!adapter->rx_queue.queue_addr) {
 565                 netdev_err(netdev, "unable to allocate rx queue pages\n");
 566                 rc = -ENOMEM;
 567                 goto err_out;
 568         }
 569
 570         dev = &adapter->vdev->dev;
 571
 572         adapter->buffer_list_dma = dma_map_single(dev,
 573                         adapter->buffer_list_addr, 4096, DMA_BIDIRECTIONAL);
 574         adapter->filter_list_dma = dma_map_single(dev,
 575                         adapter->filter_list_addr, 4096, DMA_BIDIRECTIONAL);
 576         adapter->rx_queue.queue_dma = dma_map_single(dev,
 577                         adapter->rx_queue.queue_addr,
 578                         adapter->rx_queue.queue_len, DMA_BIDIRECTIONAL);
 579
 580         if ((dma_mapping_error(dev, adapter->buffer_list_dma)) ||
 581             (dma_mapping_error(dev, adapter->filter_list_dma)) ||
 582             (dma_mapping_error(dev, adapter->rx_queue.queue_dma))) {
 583                 netdev_err(netdev, "unable to map filter or buffer list "
 584                            "pages\n");
 585                 rc = -ENOMEM;
 586                 goto err_out;
 587         }
 588
 589         adapter->rx_queue.index = 0;
 590         adapter->rx_queue.num_slots = rxq_entries;
 591         adapter->rx_queue.toggle = 1;
 592
 593         memcpy(&mac_address, netdev->dev_addr, netdev->addr_len);
 594         mac_address = mac_address >> 16;
 595
 596         rxq_desc.fields.flags_len = IBMVETH_BUF_VALID |
 597                                         adapter->rx_queue.queue_len;
 598         rxq_desc.fields.address = adapter->rx_queue.queue_dma;
 599
 600         netdev_dbg(netdev, "buffer list @ 0x%p\n", adapter->buffer_list_addr);
 601         netdev_dbg(netdev, "filter list @ 0x%p\n", adapter->filter_list_addr);
 602         netdev_dbg(netdev, "receive q   @ 0x%p\n", adapter->rx_queue.queue_addr);
 603
 604         h_vio_signal(adapter->vdev->unit_address, VIO_IRQ_DISABLE);
 605
 606         lpar_rc = ibmveth_register_logical_lan(adapter, rxq_desc, mac_address);
 607
 608         if (lpar_rc != H_SUCCESS) {
 609                 netdev_err(netdev, "h_register_logical_lan failed with %ld\n",
 610                            lpar_rc);
 611                 netdev_err(netdev, "buffer TCE:0x%llx filter TCE:0x%llx rxq "
 612                            "desc:0x%llx MAC:0x%llx\n",
 613                                      adapter->buffer_list_dma,
 614                                      adapter->filter_list_dma,
 615                                      rxq_desc.desc,
 616                                      mac_address);
 617                 rc = -ENONET;
 618                 goto err_out;
 619         }
 620
 621         for (i = 0; i < IBMVETH_NUM_BUFF_POOLS; i++) {
 622                 if (!adapter->rx_buff_pool[i].active)
 623                         continue;
 624                 if (ibmveth_alloc_buffer_pool(&adapter->rx_buff_pool[i])) {
 625                         netdev_err(netdev, "unable to alloc pool\n");
 626                         adapter->rx_buff_pool[i].active = 0;
 627                         rc = -ENOMEM;
 628                         goto err_out;
 629                 }
 630         }
 631
 632         netdev_dbg(netdev, "registering irq 0x%x\n", netdev->irq);
 633         rc = request_irq(netdev->irq, ibmveth_interrupt, 0, netdev->name,
 634                          netdev);
 635         if (rc != 0) {
 636                 netdev_err(netdev, "unable to request irq 0x%x, rc %d\n",
 637                            netdev->irq, rc);
 638                 do {
 639                         rc = h_free_logical_lan(adapter->vdev->unit_address);
 640                 } while (H_IS_LONG_BUSY(rc) || (rc == H_BUSY));
 641
 642                 goto err_out;
 643         }
 644
 645         adapter->bounce_buffer =
 646             kmalloc(netdev->mtu + IBMVETH_BUFF_OH, GFP_KERNEL);
 647         if (!adapter->bounce_buffer) {
 648                 netdev_err(netdev, "unable to allocate bounce buffer\n");
 649                 rc = -ENOMEM;
 650                 goto err_out_free_irq;
 651         }
 652         adapter->bounce_buffer_dma =
 653             dma_map_single(&adapter->vdev->dev, adapter->bounce_buffer,
 654                            netdev->mtu + IBMVETH_BUFF_OH, DMA_BIDIRECTIONAL);
 655         if (dma_mapping_error(dev, adapter->bounce_buffer_dma)) {
 656                 netdev_err(netdev, "unable to map bounce buffer\n");
 657                 rc = -ENOMEM;
 658                 goto err_out_free_irq;
 659         }
 660
 661         netdev_dbg(netdev, "initial replenish cycle\n");
 662         ibmveth_interrupt(netdev->irq, netdev);
 663
 664         netif_start_queue(netdev);
 665
 666         netdev_dbg(netdev, "open complete\n");
 667
 668         return 0;
 669
 670 err_out_free_irq:
 671         free_irq(netdev->irq, netdev);
 672 err_out:
 673         ibmveth_cleanup(adapter);
 674         napi_disable(&adapter->napi);
 675         return rc;
 676 }
 677
 678 static int ibmveth_close(struct net_device *netdev)
 679 {
 680         struct ibmveth_adapter *adapter = netdev_priv(netdev);
 681         long lpar_rc;
 682
 683         netdev_dbg(netdev, "close starting\n");
 684
 685         napi_disable(&adapter->napi);
 686
 687         if (!adapter->pool_config)
 688                 netif_stop_queue(netdev);
 689
 690         h_vio_signal(adapter->vdev->unit_address, VIO_IRQ_DISABLE);
 691
 692         do {
 693                 lpar_rc = h_free_logical_lan(adapter->vdev->unit_address);
 694         } while (H_IS_LONG_BUSY(lpar_rc) || (lpar_rc == H_BUSY));
 695
 696         if (lpar_rc != H_SUCCESS) {
 697                 netdev_err(netdev, "h_free_logical_lan failed with %lx, "
 698                            "continuing with close\n", lpar_rc);
 699         }
 700
 701         free_irq(netdev->irq, netdev);
 702
 703         adapter->rx_no_buffer = *(u64 *)(((char *)adapter->buffer_list_addr) +
 704                                                 4096 - 8);
 705
 706         ibmveth_cleanup(adapter);
 707
 708         netdev_dbg(netdev, "close complete\n");
 709
 710         return 0;
 711 }
 712
 713 static int netdev_get_settings(struct net_device *dev, struct ethtool_cmd *cmd)
 714 {
 715         cmd->supported = (SUPPORTED_1000baseT_Full | SUPPORTED_Autoneg |
 716                                 SUPPORTED_FIBRE);
 717         cmd->advertising = (ADVERTISED_1000baseT_Full | ADVERTISED_Autoneg |
 718                                 ADVERTISED_FIBRE);
 719         ethtool_cmd_speed_set(cmd, SPEED_1000);
 720         cmd->duplex = DUPLEX_FULL;
 721         cmd->port = PORT_FIBRE;
 722         cmd->phy_address = 0;
 723         cmd->transceiver = XCVR_INTERNAL;
 724         cmd->autoneg = AUTONEG_ENABLE;
 725         cmd->maxtxpkt = 0;
 726         cmd->maxrxpkt = 1;
 727         return 0;
 728 }
 729
 730 static void netdev_get_drvinfo(struct net_device *dev,
 731                                struct ethtool_drvinfo *info)
 732 {
 733         strncpy(info->driver, ibmveth_driver_name, sizeof(info->driver) - 1);
 734         strncpy(info->version, ibmveth_driver_version,
 735                 sizeof(info->version) - 1);
 736 }
 737
 738 static u32 ibmveth_fix_features(struct net_device *dev, u32 features)
 739 {
 740         /*
 741          * Since the ibmveth firmware interface does not have the
 742          * concept of separate tx/rx checksum offload enable, if rx
 743          * checksum is disabled we also have to disable tx checksum
 744          * offload. Once we disable rx checksum offload, we are no
 745          * longer allowed to send tx buffers that are not properly
 746          * checksummed.
 747          */
 748
 749         if (!(features & NETIF_F_RXCSUM))
 750                 features &= ~NETIF_F_ALL_CSUM;
 751
 752         return features;
 753 }
 754
 755 static int ibmveth_set_csum_offload(struct net_device *dev, u32 data)
 756 {
 757         struct ibmveth_adapter *adapter = netdev_priv(dev);
 758         unsigned long set_attr, clr_attr, ret_attr;
 759         unsigned long set_attr6, clr_attr6;
 760         long ret, ret6;
 761         int rc1 = 0, rc2 = 0;
 762         int restart = 0;
 763
 764         if (netif_running(dev)) {
 765                 restart = 1;
 766                 adapter->pool_config = 1;
 767                 ibmveth_close(dev);
 768                 adapter->pool_config = 0;
 769         }
 770
 771         set_attr = 0;
 772         clr_attr = 0;
 773
 774         if (data) {
 775                 set_attr = IBMVETH_ILLAN_IPV4_TCP_CSUM;
 776                 set_attr6 = IBMVETH_ILLAN_IPV6_TCP_CSUM;
 777         } else {
 778                 clr_attr = IBMVETH_ILLAN_IPV4_TCP_CSUM;
 779                 clr_attr6 = IBMVETH_ILLAN_IPV6_TCP_CSUM;
 780         }
 781
 782         ret = h_illan_attributes(adapter->vdev->unit_address, 0, 0, &ret_attr);
 783
 784         if (ret == H_SUCCESS && !(ret_attr & IBMVETH_ILLAN_ACTIVE_TRUNK) &&
 785             !(ret_attr & IBMVETH_ILLAN_TRUNK_PRI_MASK) &&
 786             (ret_attr & IBMVETH_ILLAN_PADDED_PKT_CSUM)) {
 787                 ret = h_illan_attributes(adapter->vdev->unit_address, clr_attr,
 788                                          set_attr, &ret_attr);
 789
 790                 if (ret != H_SUCCESS) {
 791                         netdev_err(dev, "unable to change IPv4 checksum "
 792                                         "offload settings. %d rc=%ld\n",
 793                                         data, ret);
 794
 795                         ret = h_illan_attributes(adapter->vdev->unit_address,
 796                                                  set_attr, clr_attr, &ret_attr);
 797                 } else {
 798                         adapter->fw_ipv4_csum_support = data;
 799                 }
 800
 801                 ret6 = h_illan_attributes(adapter->vdev->unit_address,
 802                                          clr_attr6, set_attr6, &ret_attr);
 803
 804                 if (ret6 != H_SUCCESS) {
 805                         netdev_err(dev, "unable to change IPv6 checksum "
 806                                         "offload settings. %d rc=%ld\n",
 807                                         data, ret);
 808
 809                         ret = h_illan_attributes(adapter->vdev->unit_address,
 810                                                  set_attr6, clr_attr6,
 811                                                  &ret_attr);
 812                 } else
 813                         adapter->fw_ipv6_csum_support = data;
 814
 815                 if (ret != H_SUCCESS || ret6 != H_SUCCESS)
 816                         adapter->rx_csum = data;
 817                 else
 818                         rc1 = -EIO;
 819         } else {
 820                 rc1 = -EIO;
 821                 netdev_err(dev, "unable to change checksum offload settings."
 822                                      " %d rc=%ld ret_attr=%lx\n", data, ret,
 823                                      ret_attr);
 824         }
 825
 826         if (restart)
 827                 rc2 = ibmveth_open(dev);
 828
 829         return rc1 ? rc1 : rc2;
 830 }
 831
 832 static int ibmveth_set_features(struct net_device *dev, u32 features)
 833 {
 834         struct ibmveth_adapter *adapter = netdev_priv(dev);
 835         int rx_csum = !!(features & NETIF_F_RXCSUM);
 836         int rc;
 837
 838         if (rx_csum == adapter->rx_csum)
 839                 return 0;
 840
 841         rc = ibmveth_set_csum_offload(dev, rx_csum);
 842         if (rc && !adapter->rx_csum)
 843                 dev->features = features & ~(NETIF_F_ALL_CSUM | NETIF_F_RXCSUM);
 844
 845         return rc;
 846 }
 847
 848 static void ibmveth_get_strings(struct net_device *dev, u32 stringset, u8 *data)
 849 {
 850         int i;
 851
 852         if (stringset != ETH_SS_STATS)
 853                 return;
 854
 855         for (i = 0; i < ARRAY_SIZE(ibmveth_stats); i++, data += ETH_GSTRING_LEN)
 856                 memcpy(data, ibmveth_stats[i].name, ETH_GSTRING_LEN);
 857 }
 858
 859 static int ibmveth_get_sset_count(struct net_device *dev, int sset)
 860 {
 861         switch (sset) {
 862         case ETH_SS_STATS:
 863                 return ARRAY_SIZE(ibmveth_stats);
 864         default:
 865                 return -EOPNOTSUPP;
 866         }
 867 }
 868
 869 static void ibmveth_get_ethtool_stats(struct net_device *dev,
 870                                       struct ethtool_stats *stats, u64 *data)
 871 {
 872         int i;
 873         struct ibmveth_adapter *adapter = netdev_priv(dev);
 874
 875         for (i = 0; i < ARRAY_SIZE(ibmveth_stats); i++)
 876                 data[i] = IBMVETH_GET_STAT(adapter, ibmveth_stats[i].offset);
 877 }
 878
 879 static const struct ethtool_ops netdev_ethtool_ops = {
 880         .get_drvinfo            = netdev_get_drvinfo,
 881         .get_settings           = netdev_get_settings,
 882         .get_link               = ethtool_op_get_link,
 883         .get_strings            = ibmveth_get_strings,
 884         .get_sset_count         = ibmveth_get_sset_count,
 885         .get_ethtool_stats      = ibmveth_get_ethtool_stats,
 886 };
 887
 888 static int ibmveth_ioctl(struct net_device *dev, struct ifreq *ifr, int cmd)
 889 {
 890         return -EOPNOTSUPP;
 891 }
 892
 893 #define page_offset(v) ((unsigned long)(v) & ((1 << 12) - 1))
 894
 895 static int ibmveth_send(struct ibmveth_adapter *adapter,
 896                         union ibmveth_buf_desc *descs)
 897 {
 898         unsigned long correlator;
 899         unsigned int retry_count;
 900         unsigned long ret;
 901
 902         /*
 903          * The retry count sets a maximum for the number of broadcast and
 904          * multicast destinations within the system.
 905          */
 906         retry_count = 1024;
 907         correlator = 0;
 908         do {
 909                 ret = h_send_logical_lan(adapter->vdev->unit_address,
 910                                              descs[0].desc, descs[1].desc,
 911                                              descs[2].desc, descs[3].desc,
 912                                              descs[4].desc, descs[5].desc,
 913                                              correlator, &correlator);
 914         } while ((ret == H_BUSY) && (retry_count--));
 915
 916         if (ret != H_SUCCESS && ret != H_DROPPED) {
 917                 netdev_err(adapter->netdev, "tx: h_send_logical_lan failed "
 918                            "with rc=%ld\n", ret);
 919                 return 1;
 920         }
 921
 922         return 0;
 923 }
 924
 925 static netdev_tx_t ibmveth_start_xmit(struct sk_buff *skb,
 926                                       struct net_device *netdev)
 927 {
 928         struct ibmveth_adapter *adapter = netdev_priv(netdev);
 929         unsigned int desc_flags;
 930         union ibmveth_buf_desc descs[6];
 931         int last, i;
 932         int force_bounce = 0;
 933
 934         /*
 935          * veth handles a maximum of 6 segments including the header, so
 936          * we have to linearize the skb if there are more than this.
 937          */
 938         if (skb_shinfo(skb)->nr_frags > 5 && __skb_linearize(skb)) {
 939                 netdev->stats.tx_dropped++;
 940                 goto out;
 941         }
 942
 943         /* veth can't checksum offload UDP */
 944         if (skb->ip_summed == CHECKSUM_PARTIAL &&
 945             ((skb->protocol == htons(ETH_P_IP) &&
 946               ip_hdr(skb)->protocol != IPPROTO_TCP) ||
 947              (skb->protocol == htons(ETH_P_IPV6) &&
 948               ipv6_hdr(skb)->nexthdr != IPPROTO_TCP)) &&
 949             skb_checksum_help(skb)) {
 950
 951                 netdev_err(netdev, "tx: failed to checksum packet\n");
 952                 netdev->stats.tx_dropped++;
 953                 goto out;
 954         }
 955
 956         desc_flags = IBMVETH_BUF_VALID;
 957
 958         if (skb->ip_summed == CHECKSUM_PARTIAL) {
 959                 unsigned char *buf = skb_transport_header(skb) +
 960                                                 skb->csum_offset;
 961
 962                 desc_flags |= (IBMVETH_BUF_NO_CSUM | IBMVETH_BUF_CSUM_GOOD);
 963
 964                 /* Need to zero out the checksum */
 965                 buf[0] = 0;
 966                 buf[1] = 0;
 967         }
 968
 969 retry_bounce:
 970         memset(descs, 0, sizeof(descs));
 971
 972         /*
 973          * If a linear packet is below the rx threshold then
 974          * copy it into the static bounce buffer. This avoids the
 975          * cost of a TCE insert and remove.
 976          */
 977         if (force_bounce || (!skb_is_nonlinear(skb) &&
 978                                 (skb->len < tx_copybreak))) {
 979                 skb_copy_from_linear_data(skb, adapter->bounce_buffer,
 980                                           skb->len);
 981
 982                 descs[0].fields.flags_len = desc_flags | skb->len;
 983                 descs[0].fields.address = adapter->bounce_buffer_dma;
 984
 985                 if (ibmveth_send(adapter, descs)) {
 986                         adapter->tx_send_failed++;
 987                         netdev->stats.tx_dropped++;
 988                 } else {
 989                         netdev->stats.tx_packets++;
 990                         netdev->stats.tx_bytes += skb->len;
 991                 }
 992
 993                 goto out;
 994         }
 995
 996         /* Map the header */
 997         descs[0].fields.address = dma_map_single(&adapter->vdev->dev, skb->data,
 998                                                  skb_headlen(skb),
 999                                                  DMA_TO_DEVICE);
1000         if (dma_mapping_error(&adapter->vdev->dev, descs[0].fields.address))
1001                 goto map_failed;
1002
1003         descs[0].fields.flags_len = desc_flags | skb_headlen(skb);
1004
1005         /* Map the frags */
1006         for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
1007                 unsigned long dma_addr;
1008                 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
1009
1010                 dma_addr = dma_map_page(&adapter->vdev->dev, frag->page,
1011                                         frag->page_offset, frag->size,
1012                                         DMA_TO_DEVICE);
1013
1014                 if (dma_mapping_error(&adapter->vdev->dev, dma_addr))
1015                         goto map_failed_frags;
1016
1017                 descs[i+1].fields.flags_len = desc_flags | frag->size;
1018                 descs[i+1].fields.address = dma_addr;
1019         }
1020
1021         if (ibmveth_send(adapter, descs)) {
1022                 adapter->tx_send_failed++;
1023                 netdev->stats.tx_dropped++;
1024         } else {
1025                 netdev->stats.tx_packets++;
1026                 netdev->stats.tx_bytes += skb->len;
1027         }
1028
1029         for (i = 0; i < skb_shinfo(skb)->nr_frags + 1; i++)
1030                 dma_unmap_page(&adapter->vdev->dev, descs[i].fields.address,
1031                                descs[i].fields.flags_len & IBMVETH_BUF_LEN_MASK,
1032                                DMA_TO_DEVICE);
1033
1034 out:
1035         dev_kfree_skb(skb);
1036         return NETDEV_TX_OK;
1037
1038 map_failed_frags:
1039         last = i+1;
1040         for (i = 0; i < last; i++)
1041                 dma_unmap_page(&adapter->vdev->dev, descs[i].fields.address,
1042                                descs[i].fields.flags_len & IBMVETH_BUF_LEN_MASK,
1043                                DMA_TO_DEVICE);
1044
1045 map_failed:
1046         if (!firmware_has_feature(FW_FEATURE_CMO))
1047                 netdev_err(netdev, "tx: unable to map xmit buffer\n");
1048         adapter->tx_map_failed++;
1049         skb_linearize(skb);
1050         force_bounce = 1;
1051         goto retry_bounce;
1052 }
1053
1054 static int ibmveth_poll(struct napi_struct *napi, int budget)
1055 {
1056         struct ibmveth_adapter *adapter =
1057                         container_of(napi, struct ibmveth_adapter, napi);
1058         struct net_device *netdev = adapter->netdev;
1059         int frames_processed = 0;
1060         unsigned long lpar_rc;
1061
1062 restart_poll:
1063         do {
1064                 if (!ibmveth_rxq_pending_buffer(adapter))
1065                         break;
1066
1067                 smp_rmb();
1068                 if (!ibmveth_rxq_buffer_valid(adapter)) {
1069                         wmb(); /* suggested by larson1 */
1070                         adapter->rx_invalid_buffer++;
1071                         netdev_dbg(netdev, "recycling invalid buffer\n");
1072                         ibmveth_rxq_recycle_buffer(adapter);
1073                 } else {
1074                         struct sk_buff *skb, *new_skb;
1075                         int length = ibmveth_rxq_frame_length(adapter);
1076                         int offset = ibmveth_rxq_frame_offset(adapter);
1077                         int csum_good = ibmveth_rxq_csum_good(adapter);
1078
1079                         skb = ibmveth_rxq_get_buffer(adapter);
1080
1081                         new_skb = NULL;
1082                         if (length < rx_copybreak)
1083                                 new_skb = netdev_alloc_skb(netdev, length);
1084
1085                         if (new_skb) {
1086                                 skb_copy_to_linear_data(new_skb,
1087                                                         skb->data + offset,
1088                                                         length);
1089                                 if (rx_flush)
1090                                         ibmveth_flush_buffer(skb->data,
1091                                                 length + offset);
1092                                 if (!ibmveth_rxq_recycle_buffer(adapter))
1093                                         kfree_skb(skb);
1094                                 skb = new_skb;
1095                         } else {
1096                                 ibmveth_rxq_harvest_buffer(adapter);
1097                                 skb_reserve(skb, offset);
1098                         }
1099
1100                         skb_put(skb, length);
1101                         skb->protocol = eth_type_trans(skb, netdev);
1102
1103                         if (csum_good)
1104                                 skb->ip_summed = CHECKSUM_UNNECESSARY;
1105
1106                         netif_receive_skb(skb); /* send it up */
1107
1108                         netdev->stats.rx_packets++;
1109                         netdev->stats.rx_bytes += length;
1110                         frames_processed++;
1111                 }
1112         } while (frames_processed < budget);
1113
1114         ibmveth_replenish_task(adapter);
1115
1116         if (frames_processed < budget) {
1117                 /* We think we are done - reenable interrupts,
1118                  * then check once more to make sure we are done.
1119                  */
1120                 lpar_rc = h_vio_signal(adapter->vdev->unit_address,
1121                                        VIO_IRQ_ENABLE);
1122
1123                 BUG_ON(lpar_rc != H_SUCCESS);
1124
1125                 napi_complete(napi);
1126
1127                 if (ibmveth_rxq_pending_buffer(adapter) &&
1128                     napi_reschedule(napi)) {
1129                         lpar_rc = h_vio_signal(adapter->vdev->unit_address,
1130                                                VIO_IRQ_DISABLE);
1131                         goto restart_poll;
1132                 }
1133         }
1134
1135         return frames_processed;
1136 }
1137
1138 static irqreturn_t ibmveth_interrupt(int irq, void *dev_instance)
1139 {
1140         struct net_device *netdev = dev_instance;
1141         struct ibmveth_adapter *adapter = netdev_priv(netdev);
1142         unsigned long lpar_rc;
1143
1144         if (napi_schedule_prep(&adapter->napi)) {
1145                 lpar_rc = h_vio_signal(adapter->vdev->unit_address,
1146                                        VIO_IRQ_DISABLE);
1147                 BUG_ON(lpar_rc != H_SUCCESS);
1148                 __napi_schedule(&adapter->napi);
1149         }
1150         return IRQ_HANDLED;
1151 }
1152
1153 static void ibmveth_set_multicast_list(struct net_device *netdev)
1154 {
1155         struct ibmveth_adapter *adapter = netdev_priv(netdev);
1156         unsigned long lpar_rc;
1157
1158         if ((netdev->flags & IFF_PROMISC) ||
1159             (netdev_mc_count(netdev) > adapter->mcastFilterSize)) {
1160                 lpar_rc = h_multicast_ctrl(adapter->vdev->unit_address,
1161                                            IbmVethMcastEnableRecv |
1162                                            IbmVethMcastDisableFiltering,
1163                                            0);
1164                 if (lpar_rc != H_SUCCESS) {
1165                         netdev_err(netdev, "h_multicast_ctrl rc=%ld when "
1166                                    "entering promisc mode\n", lpar_rc);
1167                 }
1168         } else {
1169                 struct netdev_hw_addr *ha;
1170                 /* clear the filter table & disable filtering */
1171                 lpar_rc = h_multicast_ctrl(adapter->vdev->unit_address,
1172                                            IbmVethMcastEnableRecv |
1173                                            IbmVethMcastDisableFiltering |
1174                                            IbmVethMcastClearFilterTable,
1175                                            0);
1176                 if (lpar_rc != H_SUCCESS) {
1177                         netdev_err(netdev, "h_multicast_ctrl rc=%ld when "
1178                                    "attempting to clear filter table\n",
1179                                    lpar_rc);
1180                 }
1181                 /* add the addresses to the filter table */
1182                 netdev_for_each_mc_addr(ha, netdev) {
1183                         /* add the multicast address to the filter table */
1184                         unsigned long mcast_addr = 0;
1185                         memcpy(((char *)&mcast_addr)+2, ha->addr, 6);
1186                         lpar_rc = h_multicast_ctrl(adapter->vdev->unit_address,
1187                                                    IbmVethMcastAddFilter,
1188                                                    mcast_addr);
1189                         if (lpar_rc != H_SUCCESS) {
1190                                 netdev_err(netdev, "h_multicast_ctrl rc=%ld "
1191                                            "when adding an entry to the filter "
1192                                            "table\n", lpar_rc);
1193                         }
1194                 }
1195
1196                 /* re-enable filtering */
1197                 lpar_rc = h_multicast_ctrl(adapter->vdev->unit_address,
1198                                            IbmVethMcastEnableFiltering,
1199                                            0);
1200                 if (lpar_rc != H_SUCCESS) {
1201                         netdev_err(netdev, "h_multicast_ctrl rc=%ld when "
1202                                    "enabling filtering\n", lpar_rc);
1203                 }
1204         }
1205 }
1206
1207 static int ibmveth_change_mtu(struct net_device *dev, int new_mtu)
1208 {
1209         struct ibmveth_adapter *adapter = netdev_priv(dev);
1210         struct vio_dev *viodev = adapter->vdev;
1211         int new_mtu_oh = new_mtu + IBMVETH_BUFF_OH;
1212         int i, rc;
1213         int need_restart = 0;
1214
1215         if (new_mtu < IBMVETH_MIN_MTU)
1216                 return -EINVAL;
1217
1218         for (i = 0; i < IBMVETH_NUM_BUFF_POOLS; i++)
1219                 if (new_mtu_oh < adapter->rx_buff_pool[i].buff_size)
1220                         break;
1221
1222         if (i == IBMVETH_NUM_BUFF_POOLS)
1223                 return -EINVAL;
1224
1225         /* Deactivate all the buffer pools so that the next loop can activate
1226            only the buffer pools necessary to hold the new MTU */
1227         if (netif_running(adapter->netdev)) {
1228                 need_restart = 1;
1229                 adapter->pool_config = 1;
1230                 ibmveth_close(adapter->netdev);
1231                 adapter->pool_config = 0;
1232         }
1233
1234         /* Look for an active buffer pool that can hold the new MTU */
1235         for (i = 0; i < IBMVETH_NUM_BUFF_POOLS; i++) {
1236                 adapter->rx_buff_pool[i].active = 1;
1237
1238                 if (new_mtu_oh < adapter->rx_buff_pool[i].buff_size) {
1239                         dev->mtu = new_mtu;
1240                         vio_cmo_set_dev_desired(viodev,
1241                                                 ibmveth_get_desired_dma
1242                                                 (viodev));
1243                         if (need_restart) {
1244                                 return ibmveth_open(adapter->netdev);
1245                         }
1246                         return 0;
1247                 }
1248         }
1249
1250         if (need_restart && (rc = ibmveth_open(adapter->netdev)))
1251                 return rc;
1252
1253         return -EINVAL;
1254 }
1255
1256 #ifdef CONFIG_NET_POLL_CONTROLLER
1257 static void ibmveth_poll_controller(struct net_device *dev)
1258 {
1259         ibmveth_replenish_task(netdev_priv(dev));
1260         ibmveth_interrupt(dev->irq, dev);
1261 }
1262 #endif
1263
1264 /**
1265  * ibmveth_get_desired_dma - Calculate IO memory desired by the driver
1266  *
1267  * @vdev: struct vio_dev for the device whose desired IO mem is to be returned
1268  *
1269  * Return value:
1270  *      Number of bytes of IO data the driver will need to perform well.
1271  */
1272 static unsigned long ibmveth_get_desired_dma(struct vio_dev *vdev)
1273 {
1274         struct net_device *netdev = dev_get_drvdata(&vdev->dev);
1275         struct ibmveth_adapter *adapter;
1276         unsigned long ret;
1277         int i;
1278         int rxqentries = 1;
1279
1280         /* netdev inits at probe time along with the structures we need below*/
1281         if (netdev == NULL)
1282                 return IOMMU_PAGE_ALIGN(IBMVETH_IO_ENTITLEMENT_DEFAULT);
1283
1284         adapter = netdev_priv(netdev);
1285
1286         ret = IBMVETH_BUFF_LIST_SIZE + IBMVETH_FILT_LIST_SIZE;
1287         ret += IOMMU_PAGE_ALIGN(netdev->mtu);
1288
1289         for (i = 0; i < IBMVETH_NUM_BUFF_POOLS; i++) {
1290                 /* add the size of the active receive buffers */
1291                 if (adapter->rx_buff_pool[i].active)
1292                         ret +=
1293                             adapter->rx_buff_pool[i].size *
1294                             IOMMU_PAGE_ALIGN(adapter->rx_buff_pool[i].
1295                                     buff_size);
1296                 rxqentries += adapter->rx_buff_pool[i].size;
1297         }
1298         /* add the size of the receive queue entries */
1299         ret += IOMMU_PAGE_ALIGN(rxqentries * sizeof(struct ibmveth_rx_q_entry));
1300
1301         return ret;
1302 }
1303
1304 static const struct net_device_ops ibmveth_netdev_ops = {
1305         .ndo_open               = ibmveth_open,
1306         .ndo_stop               = ibmveth_close,
1307         .ndo_start_xmit         = ibmveth_start_xmit,
1308         .ndo_set_multicast_list = ibmveth_set_multicast_list,
1309         .ndo_do_ioctl           = ibmveth_ioctl,
1310         .ndo_change_mtu         = ibmveth_change_mtu,
1311         .ndo_fix_features       = ibmveth_fix_features,
1312         .ndo_set_features       = ibmveth_set_features,
1313         .ndo_validate_addr      = eth_validate_addr,
1314         .ndo_set_mac_address    = eth_mac_addr,
1315 #ifdef CONFIG_NET_POLL_CONTROLLER
1316         .ndo_poll_controller    = ibmveth_poll_controller,
1317 #endif
1318 };
1319
1320 static int __devinit ibmveth_probe(struct vio_dev *dev,
1321                                    const struct vio_device_id *id)
1322 {
1323         int rc, i;
1324         struct net_device *netdev;
1325         struct ibmveth_adapter *adapter;
1326         unsigned char *mac_addr_p;
1327         unsigned int *mcastFilterSize_p;
1328
1329         dev_dbg(&dev->dev, "entering ibmveth_probe for UA 0x%x\n",
1330                 dev->unit_address);
1331
1332         mac_addr_p = (unsigned char *)vio_get_attribute(dev, VETH_MAC_ADDR,
1333                                                         NULL);
1334         if (!mac_addr_p) {
1335                 dev_err(&dev->dev, "Can't find VETH_MAC_ADDR attribute\n");
1336                 return -EINVAL;
1337         }
1338
1339         mcastFilterSize_p = (unsigned int *)vio_get_attribute(dev,
1340                                                 VETH_MCAST_FILTER_SIZE, NULL);
1341         if (!mcastFilterSize_p) {
1342                 dev_err(&dev->dev, "Can't find VETH_MCAST_FILTER_SIZE "
1343                         "attribute\n");
1344                 return -EINVAL;
1345         }
1346
1347         netdev = alloc_etherdev(sizeof(struct ibmveth_adapter));
1348
1349         if (!netdev)
1350                 return -ENOMEM;
1351
1352         adapter = netdev_priv(netdev);
1353         dev_set_drvdata(&dev->dev, netdev);
1354
1355         adapter->vdev = dev;
1356         adapter->netdev = netdev;
1357         adapter->mcastFilterSize = *mcastFilterSize_p;
1358         adapter->pool_config = 0;
1359
1360         netif_napi_add(netdev, &adapter->napi, ibmveth_poll, 16);
1361
1362         /*
1363          * Some older boxes running PHYP non-natively have an OF that returns
1364          * a 8-byte local-mac-address field (and the first 2 bytes have to be
1365          * ignored) while newer boxes' OF return a 6-byte field. Note that
1366          * IEEE 1275 specifies that local-mac-address must be a 6-byte field.
1367          * The RPA doc specifies that the first byte must be 10b, so we'll
1368          * just look for it to solve this 8 vs. 6 byte field issue
1369          */
1370         if ((*mac_addr_p & 0x3) != 0x02)
1371                 mac_addr_p += 2;
1372
1373         adapter->mac_addr = 0;
1374         memcpy(&adapter->mac_addr, mac_addr_p, 6);
1375
1376         netdev->irq = dev->irq;
1377         netdev->netdev_ops = &ibmveth_netdev_ops;
1378         netdev->ethtool_ops = &netdev_ethtool_ops;
1379         SET_NETDEV_DEV(netdev, &dev->dev);
1380         netdev->hw_features = NETIF_F_SG | NETIF_F_RXCSUM |
1381                 NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM;
1382         netdev->features |= netdev->hw_features;
1383
1384         memcpy(netdev->dev_addr, &adapter->mac_addr, netdev->addr_len);
1385
1386         for (i = 0; i < IBMVETH_NUM_BUFF_POOLS; i++) {
1387                 struct kobject *kobj = &adapter->rx_buff_pool[i].kobj;
1388                 int error;
1389
1390                 ibmveth_init_buffer_pool(&adapter->rx_buff_pool[i], i,
1391                                          pool_count[i], pool_size[i],
1392                                          pool_active[i]);
1393                 error = kobject_init_and_add(kobj, &ktype_veth_pool,
1394                                              &dev->dev.kobj, "pool%d", i);
1395                 if (!error)
1396                         kobject_uevent(kobj, KOBJ_ADD);
1397         }
1398
1399         netdev_dbg(netdev, "adapter @ 0x%p\n", adapter);
1400
1401         adapter->buffer_list_dma = DMA_ERROR_CODE;
1402         adapter->filter_list_dma = DMA_ERROR_CODE;
1403         adapter->rx_queue.queue_dma = DMA_ERROR_CODE;
1404
1405         netdev_dbg(netdev, "registering netdev...\n");
1406
1407         ibmveth_set_features(netdev, netdev->features);
1408
1409         rc = register_netdev(netdev);
1410
1411         if (rc) {
1412                 netdev_dbg(netdev, "failed to register netdev rc=%d\n", rc);
1413                 free_netdev(netdev);
1414                 return rc;
1415         }
1416
1417         netdev_dbg(netdev, "registered\n");
1418
1419         return 0;
1420 }
1421
1422 static int __devexit ibmveth_remove(struct vio_dev *dev)
1423 {
1424         struct net_device *netdev = dev_get_drvdata(&dev->dev);
1425         struct ibmveth_adapter *adapter = netdev_priv(netdev);
1426         int i;
1427
1428         for (i = 0; i < IBMVETH_NUM_BUFF_POOLS; i++)
1429                 kobject_put(&adapter->rx_buff_pool[i].kobj);
1430
1431         unregister_netdev(netdev);
1432
1433         free_netdev(netdev);
1434         dev_set_drvdata(&dev->dev, NULL);
1435
1436         return 0;
1437 }
1438
1439 static struct attribute veth_active_attr;
1440 static struct attribute veth_num_attr;
1441 static struct attribute veth_size_attr;
1442
1443 static ssize_t veth_pool_show(struct kobject *kobj,
1444                               struct attribute *attr, char *buf)
1445 {
1446         struct ibmveth_buff_pool *pool = container_of(kobj,
1447                                                       struct ibmveth_buff_pool,
1448                                                       kobj);
1449
1450         if (attr == &veth_active_attr)
1451                 return sprintf(buf, "%d\n", pool->active);
1452         else if (attr == &veth_num_attr)
1453                 return sprintf(buf, "%d\n", pool->size);
1454         else if (attr == &veth_size_attr)
1455                 return sprintf(buf, "%d\n", pool->buff_size);
1456         return 0;
1457 }
1458
1459 static ssize_t veth_pool_store(struct kobject *kobj, struct attribute *attr,
1460                                const char *buf, size_t count)
1461 {
1462         struct ibmveth_buff_pool *pool = container_of(kobj,
1463                                                       struct ibmveth_buff_pool,
1464                                                       kobj);
1465         struct net_device *netdev = dev_get_drvdata(
1466             container_of(kobj->parent, struct device, kobj));
1467         struct ibmveth_adapter *adapter = netdev_priv(netdev);
1468         long value = simple_strtol(buf, NULL, 10);
1469         long rc;
1470
1471         if (attr == &veth_active_attr) {
1472                 if (value && !pool->active) {
1473                         if (netif_running(netdev)) {
1474                                 if (ibmveth_alloc_buffer_pool(pool)) {
1475                                         netdev_err(netdev,
1476                                                    "unable to alloc pool\n");
1477                                         return -ENOMEM;
1478                                 }
1479                                 pool->active = 1;
1480                                 adapter->pool_config = 1;
1481                                 ibmveth_close(netdev);
1482                                 adapter->pool_config = 0;
1483                                 if ((rc = ibmveth_open(netdev)))
1484                                         return rc;
1485                         } else {
1486                                 pool->active = 1;
1487                         }
1488                 } else if (!value && pool->active) {
1489                         int mtu = netdev->mtu + IBMVETH_BUFF_OH;
1490                         int i;
1491                         /* Make sure there is a buffer pool with buffers that
1492                            can hold a packet of the size of the MTU */
1493                         for (i = 0; i < IBMVETH_NUM_BUFF_POOLS; i++) {
1494                                 if (pool == &adapter->rx_buff_pool[i])
1495                                         continue;
1496                                 if (!adapter->rx_buff_pool[i].active)
1497                                         continue;
1498                                 if (mtu <= adapter->rx_buff_pool[i].buff_size)
1499                                         break;
1500                         }
1501
1502                         if (i == IBMVETH_NUM_BUFF_POOLS) {
1503                                 netdev_err(netdev, "no active pool >= MTU\n");
1504                                 return -EPERM;
1505                         }
1506
1507                         if (netif_running(netdev)) {
1508                                 adapter->pool_config = 1;
1509                                 ibmveth_close(netdev);
1510                                 pool->active = 0;
1511                                 adapter->pool_config = 0;
1512                                 if ((rc = ibmveth_open(netdev)))
1513                                         return rc;
1514                         }
1515                         pool->active = 0;
1516                 }
1517         } else if (attr == &veth_num_attr) {
1518                 if (value <= 0 || value > IBMVETH_MAX_POOL_COUNT) {
1519                         return -EINVAL;
1520                 } else {
1521                         if (netif_running(netdev)) {
1522                                 adapter->pool_config = 1;
1523                                 ibmveth_close(netdev);
1524                                 adapter->pool_config = 0;
1525                                 pool->size = value;
1526                                 if ((rc = ibmveth_open(netdev)))
1527                                         return rc;
1528                         } else {
1529                                 pool->size = value;
1530                         }
1531                 }
1532         } else if (attr == &veth_size_attr) {
1533                 if (value <= IBMVETH_BUFF_OH || value > IBMVETH_MAX_BUF_SIZE) {
1534                         return -EINVAL;
1535                 } else {
1536                         if (netif_running(netdev)) {
1537                                 adapter->pool_config = 1;
1538                                 ibmveth_close(netdev);
1539                                 adapter->pool_config = 0;
1540                                 pool->buff_size = value;
1541                                 if ((rc = ibmveth_open(netdev)))
1542                                         return rc;
1543                         } else {
1544                                 pool->buff_size = value;
1545                         }
1546                 }
1547         }
1548
1549         /* kick the interrupt handler to allocate/deallocate pools */
1550         ibmveth_interrupt(netdev->irq, netdev);
1551         return count;
1552 }
1553
1554
1555 #define ATTR(_name, _mode)                              \
1556         struct attribute veth_##_name##_attr = {        \
1557         .name = __stringify(_name), .mode = _mode,      \
1558         };
1559
1560 static ATTR(active, 0644);
1561 static ATTR(num, 0644);
1562 static ATTR(size, 0644);
1563
1564 static struct attribute *veth_pool_attrs[] = {
1565         &veth_active_attr,
1566         &veth_num_attr,
1567         &veth_size_attr,
1568         NULL,
1569 };
1570
1571 static const struct sysfs_ops veth_pool_ops = {
1572         .show   = veth_pool_show,
1573         .store  = veth_pool_store,
1574 };
1575
1576 static struct kobj_type ktype_veth_pool = {
1577         .release        = NULL,
1578         .sysfs_ops      = &veth_pool_ops,
1579         .default_attrs  = veth_pool_attrs,
1580 };
1581
1582 static int ibmveth_resume(struct device *dev)
1583 {
1584         struct net_device *netdev = dev_get_drvdata(dev);
1585         ibmveth_interrupt(netdev->irq, netdev);
1586         return 0;
1587 }
1588
1589 static struct vio_device_id ibmveth_device_table[] __devinitdata = {
1590         { "network", "IBM,l-lan"},
1591         { "", "" }
1592 };
1593 MODULE_DEVICE_TABLE(vio, ibmveth_device_table);
1594
1595 static struct dev_pm_ops ibmveth_pm_ops = {
1596         .resume = ibmveth_resume
1597 };
1598
1599 static struct vio_driver ibmveth_driver = {
1600         .id_table       = ibmveth_device_table,
1601         .probe          = ibmveth_probe,
1602         .remove         = ibmveth_remove,
1603         .get_desired_dma = ibmveth_get_desired_dma,
1604         .driver         = {
1605                 .name   = ibmveth_driver_name,
1606                 .owner  = THIS_MODULE,
1607                 .pm = &ibmveth_pm_ops,
1608         }
1609 };
1610
1611 static int __init ibmveth_module_init(void)
1612 {
1613         printk(KERN_DEBUG "%s: %s %s\n", ibmveth_driver_name,
1614                ibmveth_driver_string, ibmveth_driver_version);
1615
1616         return vio_register_driver(&ibmveth_driver);
1617 }
1618
1619 static void __exit ibmveth_module_exit(void)
1620 {
1621         vio_unregister_driver(&ibmveth_driver);
1622 }
1623
1624 module_init(ibmveth_module_init);
1625 module_exit(ibmveth_module_exit);