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/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 <asm/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 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 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 void 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
 406         BUG_ON(pool >= IBMVETH_NUM_BUFF_POOLS);
 407         BUG_ON(index >= adapter->rx_buff_pool[pool].size);
 408
 409         if (!adapter->rx_buff_pool[pool].active) {
 410                 ibmveth_rxq_harvest_buffer(adapter);
 411                 ibmveth_free_buffer_pool(adapter, &adapter->rx_buff_pool[pool]);
 412                 return;
 413         }
 414
 415         desc.fields.flags_len = IBMVETH_BUF_VALID |
 416                 adapter->rx_buff_pool[pool].buff_size;
 417         desc.fields.address = adapter->rx_buff_pool[pool].dma_addr[index];
 418
 419         lpar_rc = h_add_logical_lan_buffer(adapter->vdev->unit_address, desc.desc);
 420
 421         if (lpar_rc != H_SUCCESS) {
 422                 netdev_dbg(adapter->netdev, "h_add_logical_lan_buffer failed "
 423                            "during recycle rc=%ld", lpar_rc);
 424                 ibmveth_remove_buffer_from_pool(adapter, adapter->rx_queue.queue_addr[adapter->rx_queue.index].correlator);
 425         }
 426
 427         if (++adapter->rx_queue.index == adapter->rx_queue.num_slots) {
 428                 adapter->rx_queue.index = 0;
 429                 adapter->rx_queue.toggle = !adapter->rx_queue.toggle;
 430         }
 431 }
 432
 433 static void ibmveth_rxq_harvest_buffer(struct ibmveth_adapter *adapter)
 434 {
 435         ibmveth_remove_buffer_from_pool(adapter, adapter->rx_queue.queue_addr[adapter->rx_queue.index].correlator);
 436
 437         if (++adapter->rx_queue.index == adapter->rx_queue.num_slots) {
 438                 adapter->rx_queue.index = 0;
 439                 adapter->rx_queue.toggle = !adapter->rx_queue.toggle;
 440         }
 441 }
 442
 443 static void ibmveth_cleanup(struct ibmveth_adapter *adapter)
 444 {
 445         int i;
 446         struct device *dev = &adapter->vdev->dev;
 447
 448         if (adapter->buffer_list_addr != NULL) {
 449                 if (!dma_mapping_error(dev, adapter->buffer_list_dma)) {
 450                         dma_unmap_single(dev, adapter->buffer_list_dma, 4096,
 451                                         DMA_BIDIRECTIONAL);
 452                         adapter->buffer_list_dma = DMA_ERROR_CODE;
 453                 }
 454                 free_page((unsigned long)adapter->buffer_list_addr);
 455                 adapter->buffer_list_addr = NULL;
 456         }
 457
 458         if (adapter->filter_list_addr != NULL) {
 459                 if (!dma_mapping_error(dev, adapter->filter_list_dma)) {
 460                         dma_unmap_single(dev, adapter->filter_list_dma, 4096,
 461                                         DMA_BIDIRECTIONAL);
 462                         adapter->filter_list_dma = DMA_ERROR_CODE;
 463                 }
 464                 free_page((unsigned long)adapter->filter_list_addr);
 465                 adapter->filter_list_addr = NULL;
 466         }
 467
 468         if (adapter->rx_queue.queue_addr != NULL) {
 469                 if (!dma_mapping_error(dev, adapter->rx_queue.queue_dma)) {
 470                         dma_unmap_single(dev,
 471                                         adapter->rx_queue.queue_dma,
 472                                         adapter->rx_queue.queue_len,
 473                                         DMA_BIDIRECTIONAL);
 474                         adapter->rx_queue.queue_dma = DMA_ERROR_CODE;
 475                 }
 476                 kfree(adapter->rx_queue.queue_addr);
 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                 ibmveth_cleanup(adapter);
 550                 napi_disable(&adapter->napi);
 551                 return -ENOMEM;
 552         }
 553
 554         adapter->rx_queue.queue_len = sizeof(struct ibmveth_rx_q_entry) *
 555                                                 rxq_entries;
 556         adapter->rx_queue.queue_addr = kmalloc(adapter->rx_queue.queue_len,
 557                                                 GFP_KERNEL);
 558
 559         if (!adapter->rx_queue.queue_addr) {
 560                 netdev_err(netdev, "unable to allocate rx queue pages\n");
 561                 ibmveth_cleanup(adapter);
 562                 napi_disable(&adapter->napi);
 563                 return -ENOMEM;
 564         }
 565
 566         dev = &adapter->vdev->dev;
 567
 568         adapter->buffer_list_dma = dma_map_single(dev,
 569                         adapter->buffer_list_addr, 4096, DMA_BIDIRECTIONAL);
 570         adapter->filter_list_dma = dma_map_single(dev,
 571                         adapter->filter_list_addr, 4096, DMA_BIDIRECTIONAL);
 572         adapter->rx_queue.queue_dma = dma_map_single(dev,
 573                         adapter->rx_queue.queue_addr,
 574                         adapter->rx_queue.queue_len, DMA_BIDIRECTIONAL);
 575
 576         if ((dma_mapping_error(dev, adapter->buffer_list_dma)) ||
 577             (dma_mapping_error(dev, adapter->filter_list_dma)) ||
 578             (dma_mapping_error(dev, adapter->rx_queue.queue_dma))) {
 579                 netdev_err(netdev, "unable to map filter or buffer list "
 580                            "pages\n");
 581                 ibmveth_cleanup(adapter);
 582                 napi_disable(&adapter->napi);
 583                 return -ENOMEM;
 584         }
 585
 586         adapter->rx_queue.index = 0;
 587         adapter->rx_queue.num_slots = rxq_entries;
 588         adapter->rx_queue.toggle = 1;
 589
 590         memcpy(&mac_address, netdev->dev_addr, netdev->addr_len);
 591         mac_address = mac_address >> 16;
 592
 593         rxq_desc.fields.flags_len = IBMVETH_BUF_VALID |
 594                                         adapter->rx_queue.queue_len;
 595         rxq_desc.fields.address = adapter->rx_queue.queue_dma;
 596
 597         netdev_dbg(netdev, "buffer list @ 0x%p\n", adapter->buffer_list_addr);
 598         netdev_dbg(netdev, "filter list @ 0x%p\n", adapter->filter_list_addr);
 599         netdev_dbg(netdev, "receive q   @ 0x%p\n", adapter->rx_queue.queue_addr);
 600
 601         h_vio_signal(adapter->vdev->unit_address, VIO_IRQ_DISABLE);
 602
 603         lpar_rc = ibmveth_register_logical_lan(adapter, rxq_desc, mac_address);
 604
 605         if (lpar_rc != H_SUCCESS) {
 606                 netdev_err(netdev, "h_register_logical_lan failed with %ld\n",
 607                            lpar_rc);
 608                 netdev_err(netdev, "buffer TCE:0x%llx filter TCE:0x%llx rxq "
 609                            "desc:0x%llx MAC:0x%llx\n",
 610                                      adapter->buffer_list_dma,
 611                                      adapter->filter_list_dma,
 612                                      rxq_desc.desc,
 613                                      mac_address);
 614                 ibmveth_cleanup(adapter);
 615                 napi_disable(&adapter->napi);
 616                 return -ENONET;
 617         }
 618
 619         for (i = 0; i < IBMVETH_NUM_BUFF_POOLS; i++) {
 620                 if (!adapter->rx_buff_pool[i].active)
 621                         continue;
 622                 if (ibmveth_alloc_buffer_pool(&adapter->rx_buff_pool[i])) {
 623                         netdev_err(netdev, "unable to alloc pool\n");
 624                         adapter->rx_buff_pool[i].active = 0;
 625                         ibmveth_cleanup(adapter);
 626                         napi_disable(&adapter->napi);
 627                         return -ENOMEM ;
 628                 }
 629         }
 630
 631         netdev_dbg(netdev, "registering irq 0x%x\n", netdev->irq);
 632         rc = request_irq(netdev->irq, ibmveth_interrupt, 0, netdev->name,
 633                          netdev);
 634         if (rc != 0) {
 635                 netdev_err(netdev, "unable to request irq 0x%x, rc %d\n",
 636                            netdev->irq, rc);
 637                 do {
 638                         rc = h_free_logical_lan(adapter->vdev->unit_address);
 639                 } while (H_IS_LONG_BUSY(rc) || (rc == H_BUSY));
 640
 641                 ibmveth_cleanup(adapter);
 642                 napi_disable(&adapter->napi);
 643                 return rc;
 644         }
 645
 646         adapter->bounce_buffer =
 647             kmalloc(netdev->mtu + IBMVETH_BUFF_OH, GFP_KERNEL);
 648         if (!adapter->bounce_buffer) {
 649                 netdev_err(netdev, "unable to allocate bounce buffer\n");
 650                 ibmveth_cleanup(adapter);
 651                 napi_disable(&adapter->napi);
 652                 return -ENOMEM;
 653         }
 654         adapter->bounce_buffer_dma =
 655             dma_map_single(&adapter->vdev->dev, adapter->bounce_buffer,
 656                            netdev->mtu + IBMVETH_BUFF_OH, DMA_BIDIRECTIONAL);
 657         if (dma_mapping_error(dev, adapter->bounce_buffer_dma)) {
 658                 netdev_err(netdev, "unable to map bounce buffer\n");
 659                 ibmveth_cleanup(adapter);
 660                 napi_disable(&adapter->napi);
 661                 return -ENOMEM;
 662         }
 663
 664         netdev_dbg(netdev, "initial replenish cycle\n");
 665         ibmveth_interrupt(netdev->irq, netdev);
 666
 667         netif_start_queue(netdev);
 668
 669         netdev_dbg(netdev, "open complete\n");
 670
 671         return 0;
 672 }
 673
 674 static int ibmveth_close(struct net_device *netdev)
 675 {
 676         struct ibmveth_adapter *adapter = netdev_priv(netdev);
 677         long lpar_rc;
 678
 679         netdev_dbg(netdev, "close starting\n");
 680
 681         napi_disable(&adapter->napi);
 682
 683         if (!adapter->pool_config)
 684                 netif_stop_queue(netdev);
 685
 686         h_vio_signal(adapter->vdev->unit_address, VIO_IRQ_DISABLE);
 687
 688         do {
 689                 lpar_rc = h_free_logical_lan(adapter->vdev->unit_address);
 690         } while (H_IS_LONG_BUSY(lpar_rc) || (lpar_rc == H_BUSY));
 691
 692         if (lpar_rc != H_SUCCESS) {
 693                 netdev_err(netdev, "h_free_logical_lan failed with %lx, "
 694                            "continuing with close\n", lpar_rc);
 695         }
 696
 697         free_irq(netdev->irq, netdev);
 698
 699         adapter->rx_no_buffer = *(u64 *)(((char *)adapter->buffer_list_addr) +
 700                                                 4096 - 8);
 701
 702         ibmveth_cleanup(adapter);
 703
 704         netdev_dbg(netdev, "close complete\n");
 705
 706         return 0;
 707 }
 708
 709 static int netdev_get_settings(struct net_device *dev, struct ethtool_cmd *cmd)
 710 {
 711         cmd->supported = (SUPPORTED_1000baseT_Full | SUPPORTED_Autoneg |
 712                                 SUPPORTED_FIBRE);
 713         cmd->advertising = (ADVERTISED_1000baseT_Full | ADVERTISED_Autoneg |
 714                                 ADVERTISED_FIBRE);
 715         cmd->speed = SPEED_1000;
 716         cmd->duplex = DUPLEX_FULL;
 717         cmd->port = PORT_FIBRE;
 718         cmd->phy_address = 0;
 719         cmd->transceiver = XCVR_INTERNAL;
 720         cmd->autoneg = AUTONEG_ENABLE;
 721         cmd->maxtxpkt = 0;
 722         cmd->maxrxpkt = 1;
 723         return 0;
 724 }
 725
 726 static void netdev_get_drvinfo(struct net_device *dev,
 727                                struct ethtool_drvinfo *info)
 728 {
 729         strncpy(info->driver, ibmveth_driver_name, sizeof(info->driver) - 1);
 730         strncpy(info->version, ibmveth_driver_version,
 731                 sizeof(info->version) - 1);
 732 }
 733
 734 static u32 netdev_get_link(struct net_device *dev)
 735 {
 736         return 1;
 737 }
 738
 739 static void ibmveth_set_rx_csum_flags(struct net_device *dev, u32 data)
 740 {
 741         struct ibmveth_adapter *adapter = netdev_priv(dev);
 742
 743         if (data) {
 744                 adapter->rx_csum = 1;
 745         } else {
 746                 /*
 747                  * Since the ibmveth firmware interface does not have the
 748                  * concept of separate tx/rx checksum offload enable, if rx
 749                  * checksum is disabled we also have to disable tx checksum
 750                  * offload. Once we disable rx checksum offload, we are no
 751                  * longer allowed to send tx buffers that are not properly
 752                  * checksummed.
 753                  */
 754                 adapter->rx_csum = 0;
 755                 dev->features &= ~NETIF_F_IP_CSUM;
 756                 dev->features &= ~NETIF_F_IPV6_CSUM;
 757         }
 758 }
 759
 760 static void ibmveth_set_tx_csum_flags(struct net_device *dev, u32 data)
 761 {
 762         struct ibmveth_adapter *adapter = netdev_priv(dev);
 763
 764         if (data) {
 765                 if (adapter->fw_ipv4_csum_support)
 766                         dev->features |= NETIF_F_IP_CSUM;
 767                 if (adapter->fw_ipv6_csum_support)
 768                         dev->features |= NETIF_F_IPV6_CSUM;
 769                 adapter->rx_csum = 1;
 770         } else {
 771                 dev->features &= ~NETIF_F_IP_CSUM;
 772                 dev->features &= ~NETIF_F_IPV6_CSUM;
 773         }
 774 }
 775
 776 static int ibmveth_set_csum_offload(struct net_device *dev, u32 data,
 777                                     void (*done) (struct net_device *, u32))
 778 {
 779         struct ibmveth_adapter *adapter = netdev_priv(dev);
 780         unsigned long set_attr, clr_attr, ret_attr;
 781         unsigned long set_attr6, clr_attr6;
 782         long ret, ret6;
 783         int rc1 = 0, rc2 = 0;
 784         int restart = 0;
 785
 786         if (netif_running(dev)) {
 787                 restart = 1;
 788                 adapter->pool_config = 1;
 789                 ibmveth_close(dev);
 790                 adapter->pool_config = 0;
 791         }
 792
 793         set_attr = 0;
 794         clr_attr = 0;
 795
 796         if (data) {
 797                 set_attr = IBMVETH_ILLAN_IPV4_TCP_CSUM;
 798                 set_attr6 = IBMVETH_ILLAN_IPV6_TCP_CSUM;
 799         } else {
 800                 clr_attr = IBMVETH_ILLAN_IPV4_TCP_CSUM;
 801                 clr_attr6 = IBMVETH_ILLAN_IPV6_TCP_CSUM;
 802         }
 803
 804         ret = h_illan_attributes(adapter->vdev->unit_address, 0, 0, &ret_attr);
 805
 806         if (ret == H_SUCCESS && !(ret_attr & IBMVETH_ILLAN_ACTIVE_TRUNK) &&
 807             !(ret_attr & IBMVETH_ILLAN_TRUNK_PRI_MASK) &&
 808             (ret_attr & IBMVETH_ILLAN_PADDED_PKT_CSUM)) {
 809                 ret = h_illan_attributes(adapter->vdev->unit_address, clr_attr,
 810                                          set_attr, &ret_attr);
 811
 812                 if (ret != H_SUCCESS) {
 813                         netdev_err(dev, "unable to change IPv4 checksum "
 814                                         "offload settings. %d rc=%ld\n",
 815                                         data, ret);
 816
 817                         ret = h_illan_attributes(adapter->vdev->unit_address,
 818                                                  set_attr, clr_attr, &ret_attr);
 819                 } else {
 820                         adapter->fw_ipv4_csum_support = data;
 821                 }
 822
 823                 ret6 = h_illan_attributes(adapter->vdev->unit_address,
 824                                          clr_attr6, set_attr6, &ret_attr);
 825
 826                 if (ret6 != H_SUCCESS) {
 827                         netdev_err(dev, "unable to change IPv6 checksum "
 828                                         "offload settings. %d rc=%ld\n",
 829                                         data, ret);
 830
 831                         ret = h_illan_attributes(adapter->vdev->unit_address,
 832                                                  set_attr6, clr_attr6,
 833                                                  &ret_attr);
 834                 } else
 835                         adapter->fw_ipv6_csum_support = data;
 836
 837                 if (ret == H_SUCCESS || ret6 == H_SUCCESS)
 838                         done(dev, data);
 839                 else
 840                         rc1 = -EIO;
 841         } else {
 842                 rc1 = -EIO;
 843                 netdev_err(dev, "unable to change checksum offload settings."
 844                                      " %d rc=%ld ret_attr=%lx\n", data, ret,
 845                                      ret_attr);
 846         }
 847
 848         if (restart)
 849                 rc2 = ibmveth_open(dev);
 850
 851         return rc1 ? rc1 : rc2;
 852 }
 853
 854 static int ibmveth_set_rx_csum(struct net_device *dev, u32 data)
 855 {
 856         struct ibmveth_adapter *adapter = netdev_priv(dev);
 857
 858         if ((data && adapter->rx_csum) || (!data && !adapter->rx_csum))
 859                 return 0;
 860
 861         return ibmveth_set_csum_offload(dev, data, ibmveth_set_rx_csum_flags);
 862 }
 863
 864 static int ibmveth_set_tx_csum(struct net_device *dev, u32 data)
 865 {
 866         struct ibmveth_adapter *adapter = netdev_priv(dev);
 867         int rc = 0;
 868
 869         if (data && (dev->features & (NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM)))
 870                 return 0;
 871         if (!data && !(dev->features & (NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM)))
 872                 return 0;
 873
 874         if (data && !adapter->rx_csum)
 875                 rc = ibmveth_set_csum_offload(dev, data,
 876                                               ibmveth_set_tx_csum_flags);
 877         else
 878                 ibmveth_set_tx_csum_flags(dev, data);
 879
 880         return rc;
 881 }
 882
 883 static u32 ibmveth_get_rx_csum(struct net_device *dev)
 884 {
 885         struct ibmveth_adapter *adapter = netdev_priv(dev);
 886         return adapter->rx_csum;
 887 }
 888
 889 static void ibmveth_get_strings(struct net_device *dev, u32 stringset, u8 *data)
 890 {
 891         int i;
 892
 893         if (stringset != ETH_SS_STATS)
 894                 return;
 895
 896         for (i = 0; i < ARRAY_SIZE(ibmveth_stats); i++, data += ETH_GSTRING_LEN)
 897                 memcpy(data, ibmveth_stats[i].name, ETH_GSTRING_LEN);
 898 }
 899
 900 static int ibmveth_get_sset_count(struct net_device *dev, int sset)
 901 {
 902         switch (sset) {
 903         case ETH_SS_STATS:
 904                 return ARRAY_SIZE(ibmveth_stats);
 905         default:
 906                 return -EOPNOTSUPP;
 907         }
 908 }
 909
 910 static void ibmveth_get_ethtool_stats(struct net_device *dev,
 911                                       struct ethtool_stats *stats, u64 *data)
 912 {
 913         int i;
 914         struct ibmveth_adapter *adapter = netdev_priv(dev);
 915
 916         for (i = 0; i < ARRAY_SIZE(ibmveth_stats); i++)
 917                 data[i] = IBMVETH_GET_STAT(adapter, ibmveth_stats[i].offset);
 918 }
 919
 920 static const struct ethtool_ops netdev_ethtool_ops = {
 921         .get_drvinfo            = netdev_get_drvinfo,
 922         .get_settings           = netdev_get_settings,
 923         .get_link               = netdev_get_link,
 924         .set_tx_csum            = ibmveth_set_tx_csum,
 925         .get_rx_csum            = ibmveth_get_rx_csum,
 926         .set_rx_csum            = ibmveth_set_rx_csum,
 927         .get_strings            = ibmveth_get_strings,
 928         .get_sset_count         = ibmveth_get_sset_count,
 929         .get_ethtool_stats      = ibmveth_get_ethtool_stats,
 930         .set_sg                 = ethtool_op_set_sg,
 931 };
 932
 933 static int ibmveth_ioctl(struct net_device *dev, struct ifreq *ifr, int cmd)
 934 {
 935         return -EOPNOTSUPP;
 936 }
 937
 938 #define page_offset(v) ((unsigned long)(v) & ((1 << 12) - 1))
 939
 940 static int ibmveth_send(struct ibmveth_adapter *adapter,
 941                         union ibmveth_buf_desc *descs)
 942 {
 943         unsigned long correlator;
 944         unsigned int retry_count;
 945         unsigned long ret;
 946
 947         /*
 948          * The retry count sets a maximum for the number of broadcast and
 949          * multicast destinations within the system.
 950          */
 951         retry_count = 1024;
 952         correlator = 0;
 953         do {
 954                 ret = h_send_logical_lan(adapter->vdev->unit_address,
 955                                              descs[0].desc, descs[1].desc,
 956                                              descs[2].desc, descs[3].desc,
 957                                              descs[4].desc, descs[5].desc,
 958                                              correlator, &correlator);
 959         } while ((ret == H_BUSY) && (retry_count--));
 960
 961         if (ret != H_SUCCESS && ret != H_DROPPED) {
 962                 netdev_err(adapter->netdev, "tx: h_send_logical_lan failed "
 963                            "with rc=%ld\n", ret);
 964                 return 1;
 965         }
 966
 967         return 0;
 968 }
 969
 970 static netdev_tx_t ibmveth_start_xmit(struct sk_buff *skb,
 971                                       struct net_device *netdev)
 972 {
 973         struct ibmveth_adapter *adapter = netdev_priv(netdev);
 974         unsigned int desc_flags;
 975         union ibmveth_buf_desc descs[6];
 976         int last, i;
 977         int force_bounce = 0;
 978
 979         /*
 980          * veth handles a maximum of 6 segments including the header, so
 981          * we have to linearize the skb if there are more than this.
 982          */
 983         if (skb_shinfo(skb)->nr_frags > 5 && __skb_linearize(skb)) {
 984                 netdev->stats.tx_dropped++;
 985                 goto out;
 986         }
 987
 988         /* veth can't checksum offload UDP */
 989         if (skb->ip_summed == CHECKSUM_PARTIAL &&
 990             ((skb->protocol == htons(ETH_P_IP) &&
 991               ip_hdr(skb)->protocol != IPPROTO_TCP) ||
 992              (skb->protocol == htons(ETH_P_IPV6) &&
 993               ipv6_hdr(skb)->nexthdr != IPPROTO_TCP)) &&
 994             skb_checksum_help(skb)) {
 995
 996                 netdev_err(netdev, "tx: failed to checksum packet\n");
 997                 netdev->stats.tx_dropped++;
 998                 goto out;
 999         }
1000
1001         desc_flags = IBMVETH_BUF_VALID;
1002
1003         if (skb->ip_summed == CHECKSUM_PARTIAL) {
1004                 unsigned char *buf = skb_transport_header(skb) +
1005                                                 skb->csum_offset;
1006
1007                 desc_flags |= (IBMVETH_BUF_NO_CSUM | IBMVETH_BUF_CSUM_GOOD);
1008
1009                 /* Need to zero out the checksum */
1010                 buf[0] = 0;
1011                 buf[1] = 0;
1012         }
1013
1014 retry_bounce:
1015         memset(descs, 0, sizeof(descs));
1016
1017         /*
1018          * If a linear packet is below the rx threshold then
1019          * copy it into the static bounce buffer. This avoids the
1020          * cost of a TCE insert and remove.
1021          */
1022         if (force_bounce || (!skb_is_nonlinear(skb) &&
1023                                 (skb->len < tx_copybreak))) {
1024                 skb_copy_from_linear_data(skb, adapter->bounce_buffer,
1025                                           skb->len);
1026
1027                 descs[0].fields.flags_len = desc_flags | skb->len;
1028                 descs[0].fields.address = adapter->bounce_buffer_dma;
1029
1030                 if (ibmveth_send(adapter, descs)) {
1031                         adapter->tx_send_failed++;
1032                         netdev->stats.tx_dropped++;
1033                 } else {
1034                         netdev->stats.tx_packets++;
1035                         netdev->stats.tx_bytes += skb->len;
1036                 }
1037
1038                 goto out;
1039         }
1040
1041         /* Map the header */
1042         descs[0].fields.address = dma_map_single(&adapter->vdev->dev, skb->data,
1043                                                  skb_headlen(skb),
1044                                                  DMA_TO_DEVICE);
1045         if (dma_mapping_error(&adapter->vdev->dev, descs[0].fields.address))
1046                 goto map_failed;
1047
1048         descs[0].fields.flags_len = desc_flags | skb_headlen(skb);
1049
1050         /* Map the frags */
1051         for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
1052                 unsigned long dma_addr;
1053                 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
1054
1055                 dma_addr = dma_map_page(&adapter->vdev->dev, frag->page,
1056                                         frag->page_offset, frag->size,
1057                                         DMA_TO_DEVICE);
1058
1059                 if (dma_mapping_error(&adapter->vdev->dev, dma_addr))
1060                         goto map_failed_frags;
1061
1062                 descs[i+1].fields.flags_len = desc_flags | frag->size;
1063                 descs[i+1].fields.address = dma_addr;
1064         }
1065
1066         if (ibmveth_send(adapter, descs)) {
1067                 adapter->tx_send_failed++;
1068                 netdev->stats.tx_dropped++;
1069         } else {
1070                 netdev->stats.tx_packets++;
1071                 netdev->stats.tx_bytes += skb->len;
1072         }
1073
1074         for (i = 0; i < skb_shinfo(skb)->nr_frags + 1; i++)
1075                 dma_unmap_page(&adapter->vdev->dev, descs[i].fields.address,
1076                                descs[i].fields.flags_len & IBMVETH_BUF_LEN_MASK,
1077                                DMA_TO_DEVICE);
1078
1079 out:
1080         dev_kfree_skb(skb);
1081         return NETDEV_TX_OK;
1082
1083 map_failed_frags:
1084         last = i+1;
1085         for (i = 0; i < last; i++)
1086                 dma_unmap_page(&adapter->vdev->dev, descs[i].fields.address,
1087                                descs[i].fields.flags_len & IBMVETH_BUF_LEN_MASK,
1088                                DMA_TO_DEVICE);
1089
1090 map_failed:
1091         if (!firmware_has_feature(FW_FEATURE_CMO))
1092                 netdev_err(netdev, "tx: unable to map xmit buffer\n");
1093         adapter->tx_map_failed++;
1094         skb_linearize(skb);
1095         force_bounce = 1;
1096         goto retry_bounce;
1097 }
1098
1099 static int ibmveth_poll(struct napi_struct *napi, int budget)
1100 {
1101         struct ibmveth_adapter *adapter =
1102                         container_of(napi, struct ibmveth_adapter, napi);
1103         struct net_device *netdev = adapter->netdev;
1104         int frames_processed = 0;
1105         unsigned long lpar_rc;
1106
1107 restart_poll:
1108         do {
1109                 if (!ibmveth_rxq_pending_buffer(adapter))
1110                         break;
1111
1112                 smp_rmb();
1113                 if (!ibmveth_rxq_buffer_valid(adapter)) {
1114                         wmb(); /* suggested by larson1 */
1115                         adapter->rx_invalid_buffer++;
1116                         netdev_dbg(netdev, "recycling invalid buffer\n");
1117                         ibmveth_rxq_recycle_buffer(adapter);
1118                 } else {
1119                         struct sk_buff *skb, *new_skb;
1120                         int length = ibmveth_rxq_frame_length(adapter);
1121                         int offset = ibmveth_rxq_frame_offset(adapter);
1122                         int csum_good = ibmveth_rxq_csum_good(adapter);
1123
1124                         skb = ibmveth_rxq_get_buffer(adapter);
1125
1126                         new_skb = NULL;
1127                         if (length < rx_copybreak)
1128                                 new_skb = netdev_alloc_skb(netdev, length);
1129
1130                         if (new_skb) {
1131                                 skb_copy_to_linear_data(new_skb,
1132                                                         skb->data + offset,
1133                                                         length);
1134                                 if (rx_flush)
1135                                         ibmveth_flush_buffer(skb->data,
1136                                                 length + offset);
1137                                 skb = new_skb;
1138                                 ibmveth_rxq_recycle_buffer(adapter);
1139                         } else {
1140                                 ibmveth_rxq_harvest_buffer(adapter);
1141                                 skb_reserve(skb, offset);
1142                         }
1143
1144                         skb_put(skb, length);
1145                         skb->protocol = eth_type_trans(skb, netdev);
1146
1147                         if (csum_good)
1148                                 skb->ip_summed = CHECKSUM_UNNECESSARY;
1149
1150                         netif_receive_skb(skb); /* send it up */
1151
1152                         netdev->stats.rx_packets++;
1153                         netdev->stats.rx_bytes += length;
1154                         frames_processed++;
1155                 }
1156         } while (frames_processed < budget);
1157
1158         ibmveth_replenish_task(adapter);
1159
1160         if (frames_processed < budget) {
1161                 /* We think we are done - reenable interrupts,
1162                  * then check once more to make sure we are done.
1163                  */
1164                 lpar_rc = h_vio_signal(adapter->vdev->unit_address,
1165                                        VIO_IRQ_ENABLE);
1166
1167                 BUG_ON(lpar_rc != H_SUCCESS);
1168
1169                 napi_complete(napi);
1170
1171                 if (ibmveth_rxq_pending_buffer(adapter) &&
1172                     napi_reschedule(napi)) {
1173                         lpar_rc = h_vio_signal(adapter->vdev->unit_address,
1174                                                VIO_IRQ_DISABLE);
1175                         goto restart_poll;
1176                 }
1177         }
1178
1179         return frames_processed;
1180 }
1181
1182 static irqreturn_t ibmveth_interrupt(int irq, void *dev_instance)
1183 {
1184         struct net_device *netdev = dev_instance;
1185         struct ibmveth_adapter *adapter = netdev_priv(netdev);
1186         unsigned long lpar_rc;
1187
1188         if (napi_schedule_prep(&adapter->napi)) {
1189                 lpar_rc = h_vio_signal(adapter->vdev->unit_address,
1190                                        VIO_IRQ_DISABLE);
1191                 BUG_ON(lpar_rc != H_SUCCESS);
1192                 __napi_schedule(&adapter->napi);
1193         }
1194         return IRQ_HANDLED;
1195 }
1196
1197 static void ibmveth_set_multicast_list(struct net_device *netdev)
1198 {
1199         struct ibmveth_adapter *adapter = netdev_priv(netdev);
1200         unsigned long lpar_rc;
1201
1202         if ((netdev->flags & IFF_PROMISC) ||
1203             (netdev_mc_count(netdev) > adapter->mcastFilterSize)) {
1204                 lpar_rc = h_multicast_ctrl(adapter->vdev->unit_address,
1205                                            IbmVethMcastEnableRecv |
1206                                            IbmVethMcastDisableFiltering,
1207                                            0);
1208                 if (lpar_rc != H_SUCCESS) {
1209                         netdev_err(netdev, "h_multicast_ctrl rc=%ld when "
1210                                    "entering promisc mode\n", lpar_rc);
1211                 }
1212         } else {
1213                 struct netdev_hw_addr *ha;
1214                 /* clear the filter table & disable filtering */
1215                 lpar_rc = h_multicast_ctrl(adapter->vdev->unit_address,
1216                                            IbmVethMcastEnableRecv |
1217                                            IbmVethMcastDisableFiltering |
1218                                            IbmVethMcastClearFilterTable,
1219                                            0);
1220                 if (lpar_rc != H_SUCCESS) {
1221                         netdev_err(netdev, "h_multicast_ctrl rc=%ld when "
1222                                    "attempting to clear filter table\n",
1223                                    lpar_rc);
1224                 }
1225                 /* add the addresses to the filter table */
1226                 netdev_for_each_mc_addr(ha, netdev) {
1227                         /* add the multicast address to the filter table */
1228                         unsigned long mcast_addr = 0;
1229                         memcpy(((char *)&mcast_addr)+2, ha->addr, 6);
1230                         lpar_rc = h_multicast_ctrl(adapter->vdev->unit_address,
1231                                                    IbmVethMcastAddFilter,
1232                                                    mcast_addr);
1233                         if (lpar_rc != H_SUCCESS) {
1234                                 netdev_err(netdev, "h_multicast_ctrl rc=%ld "
1235                                            "when adding an entry to the filter "
1236                                            "table\n", lpar_rc);
1237                         }
1238                 }
1239
1240                 /* re-enable filtering */
1241                 lpar_rc = h_multicast_ctrl(adapter->vdev->unit_address,
1242                                            IbmVethMcastEnableFiltering,
1243                                            0);
1244                 if (lpar_rc != H_SUCCESS) {
1245                         netdev_err(netdev, "h_multicast_ctrl rc=%ld when "
1246                                    "enabling filtering\n", lpar_rc);
1247                 }
1248         }
1249 }
1250
1251 static int ibmveth_change_mtu(struct net_device *dev, int new_mtu)
1252 {
1253         struct ibmveth_adapter *adapter = netdev_priv(dev);
1254         struct vio_dev *viodev = adapter->vdev;
1255         int new_mtu_oh = new_mtu + IBMVETH_BUFF_OH;
1256         int i, rc;
1257         int need_restart = 0;
1258
1259         if (new_mtu < IBMVETH_MIN_MTU)
1260                 return -EINVAL;
1261
1262         for (i = 0; i < IBMVETH_NUM_BUFF_POOLS; i++)
1263                 if (new_mtu_oh < adapter->rx_buff_pool[i].buff_size)
1264                         break;
1265
1266         if (i == IBMVETH_NUM_BUFF_POOLS)
1267                 return -EINVAL;
1268
1269         /* Deactivate all the buffer pools so that the next loop can activate
1270            only the buffer pools necessary to hold the new MTU */
1271         if (netif_running(adapter->netdev)) {
1272                 need_restart = 1;
1273                 adapter->pool_config = 1;
1274                 ibmveth_close(adapter->netdev);
1275                 adapter->pool_config = 0;
1276         }
1277
1278         /* Look for an active buffer pool that can hold the new MTU */
1279         for (i = 0; i < IBMVETH_NUM_BUFF_POOLS; i++) {
1280                 adapter->rx_buff_pool[i].active = 1;
1281
1282                 if (new_mtu_oh < adapter->rx_buff_pool[i].buff_size) {
1283                         dev->mtu = new_mtu;
1284                         vio_cmo_set_dev_desired(viodev,
1285                                                 ibmveth_get_desired_dma
1286                                                 (viodev));
1287                         if (need_restart) {
1288                                 return ibmveth_open(adapter->netdev);
1289                         }
1290                         return 0;
1291                 }
1292         }
1293
1294         if (need_restart && (rc = ibmveth_open(adapter->netdev)))
1295                 return rc;
1296
1297         return -EINVAL;
1298 }
1299
1300 #ifdef CONFIG_NET_POLL_CONTROLLER
1301 static void ibmveth_poll_controller(struct net_device *dev)
1302 {
1303         ibmveth_replenish_task(netdev_priv(dev));
1304         ibmveth_interrupt(dev->irq, dev);
1305 }
1306 #endif
1307
1308 /**
1309  * ibmveth_get_desired_dma - Calculate IO memory desired by the driver
1310  *
1311  * @vdev: struct vio_dev for the device whose desired IO mem is to be returned
1312  *
1313  * Return value:
1314  *      Number of bytes of IO data the driver will need to perform well.
1315  */
1316 static unsigned long ibmveth_get_desired_dma(struct vio_dev *vdev)
1317 {
1318         struct net_device *netdev = dev_get_drvdata(&vdev->dev);
1319         struct ibmveth_adapter *adapter;
1320         unsigned long ret;
1321         int i;
1322         int rxqentries = 1;
1323
1324         /* netdev inits at probe time along with the structures we need below*/
1325         if (netdev == NULL)
1326                 return IOMMU_PAGE_ALIGN(IBMVETH_IO_ENTITLEMENT_DEFAULT);
1327
1328         adapter = netdev_priv(netdev);
1329
1330         ret = IBMVETH_BUFF_LIST_SIZE + IBMVETH_FILT_LIST_SIZE;
1331         ret += IOMMU_PAGE_ALIGN(netdev->mtu);
1332
1333         for (i = 0; i < IBMVETH_NUM_BUFF_POOLS; i++) {
1334                 /* add the size of the active receive buffers */
1335                 if (adapter->rx_buff_pool[i].active)
1336                         ret +=
1337                             adapter->rx_buff_pool[i].size *
1338                             IOMMU_PAGE_ALIGN(adapter->rx_buff_pool[i].
1339                                     buff_size);
1340                 rxqentries += adapter->rx_buff_pool[i].size;
1341         }
1342         /* add the size of the receive queue entries */
1343         ret += IOMMU_PAGE_ALIGN(rxqentries * sizeof(struct ibmveth_rx_q_entry));
1344
1345         return ret;
1346 }
1347
1348 static const struct net_device_ops ibmveth_netdev_ops = {
1349         .ndo_open               = ibmveth_open,
1350         .ndo_stop               = ibmveth_close,
1351         .ndo_start_xmit         = ibmveth_start_xmit,
1352         .ndo_set_multicast_list = ibmveth_set_multicast_list,
1353         .ndo_do_ioctl           = ibmveth_ioctl,
1354         .ndo_change_mtu         = ibmveth_change_mtu,
1355         .ndo_validate_addr      = eth_validate_addr,
1356         .ndo_set_mac_address    = eth_mac_addr,
1357 #ifdef CONFIG_NET_POLL_CONTROLLER
1358         .ndo_poll_controller    = ibmveth_poll_controller,
1359 #endif
1360 };
1361
1362 static int __devinit ibmveth_probe(struct vio_dev *dev,
1363                                    const struct vio_device_id *id)
1364 {
1365         int rc, i;
1366         struct net_device *netdev;
1367         struct ibmveth_adapter *adapter;
1368         unsigned char *mac_addr_p;
1369         unsigned int *mcastFilterSize_p;
1370
1371         dev_dbg(&dev->dev, "entering ibmveth_probe for UA 0x%x\n",
1372                 dev->unit_address);
1373
1374         mac_addr_p = (unsigned char *)vio_get_attribute(dev, VETH_MAC_ADDR,
1375                                                         NULL);
1376         if (!mac_addr_p) {
1377                 dev_err(&dev->dev, "Can't find VETH_MAC_ADDR attribute\n");
1378                 return -EINVAL;
1379         }
1380
1381         mcastFilterSize_p = (unsigned int *)vio_get_attribute(dev,
1382                                                 VETH_MCAST_FILTER_SIZE, NULL);
1383         if (!mcastFilterSize_p) {
1384                 dev_err(&dev->dev, "Can't find VETH_MCAST_FILTER_SIZE "
1385                         "attribute\n");
1386                 return -EINVAL;
1387         }
1388
1389         netdev = alloc_etherdev(sizeof(struct ibmveth_adapter));
1390
1391         if (!netdev)
1392                 return -ENOMEM;
1393
1394         adapter = netdev_priv(netdev);
1395         dev_set_drvdata(&dev->dev, netdev);
1396
1397         adapter->vdev = dev;
1398         adapter->netdev = netdev;
1399         adapter->mcastFilterSize = *mcastFilterSize_p;
1400         adapter->pool_config = 0;
1401
1402         netif_napi_add(netdev, &adapter->napi, ibmveth_poll, 16);
1403
1404         /*
1405          * Some older boxes running PHYP non-natively have an OF that returns
1406          * a 8-byte local-mac-address field (and the first 2 bytes have to be
1407          * ignored) while newer boxes' OF return a 6-byte field. Note that
1408          * IEEE 1275 specifies that local-mac-address must be a 6-byte field.
1409          * The RPA doc specifies that the first byte must be 10b, so we'll
1410          * just look for it to solve this 8 vs. 6 byte field issue
1411          */
1412         if ((*mac_addr_p & 0x3) != 0x02)
1413                 mac_addr_p += 2;
1414
1415         adapter->mac_addr = 0;
1416         memcpy(&adapter->mac_addr, mac_addr_p, 6);
1417
1418         netdev->irq = dev->irq;
1419         netdev->netdev_ops = &ibmveth_netdev_ops;
1420         netdev->ethtool_ops = &netdev_ethtool_ops;
1421         SET_NETDEV_DEV(netdev, &dev->dev);
1422         netdev->features |= NETIF_F_SG;
1423
1424         memcpy(netdev->dev_addr, &adapter->mac_addr, netdev->addr_len);
1425
1426         for (i = 0; i < IBMVETH_NUM_BUFF_POOLS; i++) {
1427                 struct kobject *kobj = &adapter->rx_buff_pool[i].kobj;
1428                 int error;
1429
1430                 ibmveth_init_buffer_pool(&adapter->rx_buff_pool[i], i,
1431                                          pool_count[i], pool_size[i],
1432                                          pool_active[i]);
1433                 error = kobject_init_and_add(kobj, &ktype_veth_pool,
1434                                              &dev->dev.kobj, "pool%d", i);
1435                 if (!error)
1436                         kobject_uevent(kobj, KOBJ_ADD);
1437         }
1438
1439         netdev_dbg(netdev, "adapter @ 0x%p\n", adapter);
1440
1441         adapter->buffer_list_dma = DMA_ERROR_CODE;
1442         adapter->filter_list_dma = DMA_ERROR_CODE;
1443         adapter->rx_queue.queue_dma = DMA_ERROR_CODE;
1444
1445         netdev_dbg(netdev, "registering netdev...\n");
1446
1447         ibmveth_set_csum_offload(netdev, 1, ibmveth_set_tx_csum_flags);
1448
1449         rc = register_netdev(netdev);
1450
1451         if (rc) {
1452                 netdev_dbg(netdev, "failed to register netdev rc=%d\n", rc);
1453                 free_netdev(netdev);
1454                 return rc;
1455         }
1456
1457         netdev_dbg(netdev, "registered\n");
1458
1459         return 0;
1460 }
1461
1462 static int __devexit ibmveth_remove(struct vio_dev *dev)
1463 {
1464         struct net_device *netdev = dev_get_drvdata(&dev->dev);
1465         struct ibmveth_adapter *adapter = netdev_priv(netdev);
1466         int i;
1467
1468         for (i = 0; i < IBMVETH_NUM_BUFF_POOLS; i++)
1469                 kobject_put(&adapter->rx_buff_pool[i].kobj);
1470
1471         unregister_netdev(netdev);
1472
1473         free_netdev(netdev);
1474         dev_set_drvdata(&dev->dev, NULL);
1475
1476         return 0;
1477 }
1478
1479 static struct attribute veth_active_attr;
1480 static struct attribute veth_num_attr;
1481 static struct attribute veth_size_attr;
1482
1483 static ssize_t veth_pool_show(struct kobject *kobj,
1484                               struct attribute *attr, char *buf)
1485 {
1486         struct ibmveth_buff_pool *pool = container_of(kobj,
1487                                                       struct ibmveth_buff_pool,
1488                                                       kobj);
1489
1490         if (attr == &veth_active_attr)
1491                 return sprintf(buf, "%d\n", pool->active);
1492         else if (attr == &veth_num_attr)
1493                 return sprintf(buf, "%d\n", pool->size);
1494         else if (attr == &veth_size_attr)
1495                 return sprintf(buf, "%d\n", pool->buff_size);
1496         return 0;
1497 }
1498
1499 static ssize_t veth_pool_store(struct kobject *kobj, struct attribute *attr,
1500                                const char *buf, size_t count)
1501 {
1502         struct ibmveth_buff_pool *pool = container_of(kobj,
1503                                                       struct ibmveth_buff_pool,
1504                                                       kobj);
1505         struct net_device *netdev = dev_get_drvdata(
1506             container_of(kobj->parent, struct device, kobj));
1507         struct ibmveth_adapter *adapter = netdev_priv(netdev);
1508         long value = simple_strtol(buf, NULL, 10);
1509         long rc;
1510
1511         if (attr == &veth_active_attr) {
1512                 if (value && !pool->active) {
1513                         if (netif_running(netdev)) {
1514                                 if (ibmveth_alloc_buffer_pool(pool)) {
1515                                         netdev_err(netdev,
1516                                                    "unable to alloc pool\n");
1517                                         return -ENOMEM;
1518                                 }
1519                                 pool->active = 1;
1520                                 adapter->pool_config = 1;
1521                                 ibmveth_close(netdev);
1522                                 adapter->pool_config = 0;
1523                                 if ((rc = ibmveth_open(netdev)))
1524                                         return rc;
1525                         } else {
1526                                 pool->active = 1;
1527                         }
1528                 } else if (!value && pool->active) {
1529                         int mtu = netdev->mtu + IBMVETH_BUFF_OH;
1530                         int i;
1531                         /* Make sure there is a buffer pool with buffers that
1532                            can hold a packet of the size of the MTU */
1533                         for (i = 0; i < IBMVETH_NUM_BUFF_POOLS; i++) {
1534                                 if (pool == &adapter->rx_buff_pool[i])
1535                                         continue;
1536                                 if (!adapter->rx_buff_pool[i].active)
1537                                         continue;
1538                                 if (mtu <= adapter->rx_buff_pool[i].buff_size)
1539                                         break;
1540                         }
1541
1542                         if (i == IBMVETH_NUM_BUFF_POOLS) {
1543                                 netdev_err(netdev, "no active pool >= MTU\n");
1544                                 return -EPERM;
1545                         }
1546
1547                         if (netif_running(netdev)) {
1548                                 adapter->pool_config = 1;
1549                                 ibmveth_close(netdev);
1550                                 pool->active = 0;
1551                                 adapter->pool_config = 0;
1552                                 if ((rc = ibmveth_open(netdev)))
1553                                         return rc;
1554                         }
1555                         pool->active = 0;
1556                 }
1557         } else if (attr == &veth_num_attr) {
1558                 if (value <= 0 || value > IBMVETH_MAX_POOL_COUNT) {
1559                         return -EINVAL;
1560                 } else {
1561                         if (netif_running(netdev)) {
1562                                 adapter->pool_config = 1;
1563                                 ibmveth_close(netdev);
1564                                 adapter->pool_config = 0;
1565                                 pool->size = value;
1566                                 if ((rc = ibmveth_open(netdev)))
1567                                         return rc;
1568                         } else {
1569                                 pool->size = value;
1570                         }
1571                 }
1572         } else if (attr == &veth_size_attr) {
1573                 if (value <= IBMVETH_BUFF_OH || value > IBMVETH_MAX_BUF_SIZE) {
1574                         return -EINVAL;
1575                 } else {
1576                         if (netif_running(netdev)) {
1577                                 adapter->pool_config = 1;
1578                                 ibmveth_close(netdev);
1579                                 adapter->pool_config = 0;
1580                                 pool->buff_size = value;
1581                                 if ((rc = ibmveth_open(netdev)))
1582                                         return rc;
1583                         } else {
1584                                 pool->buff_size = value;
1585                         }
1586                 }
1587         }
1588
1589         /* kick the interrupt handler to allocate/deallocate pools */
1590         ibmveth_interrupt(netdev->irq, netdev);
1591         return count;
1592 }
1593
1594
1595 #define ATTR(_name, _mode)                              \
1596         struct attribute veth_##_name##_attr = {        \
1597         .name = __stringify(_name), .mode = _mode,      \
1598         };
1599
1600 static ATTR(active, 0644);
1601 static ATTR(num, 0644);
1602 static ATTR(size, 0644);
1603
1604 static struct attribute *veth_pool_attrs[] = {
1605         &veth_active_attr,
1606         &veth_num_attr,
1607         &veth_size_attr,
1608         NULL,
1609 };
1610
1611 static const struct sysfs_ops veth_pool_ops = {
1612         .show   = veth_pool_show,
1613         .store  = veth_pool_store,
1614 };
1615
1616 static struct kobj_type ktype_veth_pool = {
1617         .release        = NULL,
1618         .sysfs_ops      = &veth_pool_ops,
1619         .default_attrs  = veth_pool_attrs,
1620 };
1621
1622 static int ibmveth_resume(struct device *dev)
1623 {
1624         struct net_device *netdev = dev_get_drvdata(dev);
1625         ibmveth_interrupt(netdev->irq, netdev);
1626         return 0;
1627 }
1628
1629 static struct vio_device_id ibmveth_device_table[] __devinitdata = {
1630         { "network", "IBM,l-lan"},
1631         { "", "" }
1632 };
1633 MODULE_DEVICE_TABLE(vio, ibmveth_device_table);
1634
1635 static struct dev_pm_ops ibmveth_pm_ops = {
1636         .resume = ibmveth_resume
1637 };
1638
1639 static struct vio_driver ibmveth_driver = {
1640         .id_table       = ibmveth_device_table,
1641         .probe          = ibmveth_probe,
1642         .remove         = ibmveth_remove,
1643         .get_desired_dma = ibmveth_get_desired_dma,
1644         .driver         = {
1645                 .name   = ibmveth_driver_name,
1646                 .owner  = THIS_MODULE,
1647                 .pm = &ibmveth_pm_ops,
1648         }
1649 };
1650
1651 static int __init ibmveth_module_init(void)
1652 {
1653         printk(KERN_DEBUG "%s: %s %s\n", ibmveth_driver_name,
1654                ibmveth_driver_string, ibmveth_driver_version);
1655
1656         return vio_register_driver(&ibmveth_driver);
1657 }
1658
1659 static void __exit ibmveth_module_exit(void)
1660 {
1661         vio_unregister_driver(&ibmveth_driver);
1662 }
1663
1664 module_init(ibmveth_module_init);
1665 module_exit(ibmveth_module_exit);