drivers/net/ethernet/ibm/ibmveth.c

   1 /*
   2  * IBM Power Virtual Ethernet Device Driver
   3  *
   4  * This program is free software; you can redistribute it and/or modify
   5  * it under the terms of the GNU General Public License as published by
   6  * the Free Software Foundation; either version 2 of the License, or
   7  * (at your option) any later version.
   8  *
   9  * This program is distributed in the hope that it will be useful,
  10  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  11  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
  12  * GNU General Public License for more details.
  13  *
  14  * You should have received a copy of the GNU General Public License
  15  * along with this program; if not, 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                 dma_free_coherent(dev, adapter->rx_queue.queue_len,
 476                                   adapter->rx_queue.queue_addr,
 477                                   adapter->rx_queue.queue_dma);
 478                 adapter->rx_queue.queue_addr = NULL;
 479         }
 480
 481         for (i = 0; i < IBMVETH_NUM_BUFF_POOLS; i++)
 482                 if (adapter->rx_buff_pool[i].active)
 483                         ibmveth_free_buffer_pool(adapter,
 484                                                  &adapter->rx_buff_pool[i]);
 485
 486         if (adapter->bounce_buffer != NULL) {
 487                 if (!dma_mapping_error(dev, adapter->bounce_buffer_dma)) {
 488                         dma_unmap_single(&adapter->vdev->dev,
 489                                         adapter->bounce_buffer_dma,
 490                                         adapter->netdev->mtu + IBMVETH_BUFF_OH,
 491                                         DMA_BIDIRECTIONAL);
 492                         adapter->bounce_buffer_dma = DMA_ERROR_CODE;
 493                 }
 494                 kfree(adapter->bounce_buffer);
 495                 adapter->bounce_buffer = NULL;
 496         }
 497 }
 498
 499 static int ibmveth_register_logical_lan(struct ibmveth_adapter *adapter,
 500         union ibmveth_buf_desc rxq_desc, u64 mac_address)
 501 {
 502         int rc, try_again = 1;
 503
 504         /*
 505          * After a kexec the adapter will still be open, so our attempt to
 506          * open it will fail. So if we get a failure we free the adapter and
 507          * try again, but only once.
 508          */
 509 retry:
 510         rc = h_register_logical_lan(adapter->vdev->unit_address,
 511                                     adapter->buffer_list_dma, rxq_desc.desc,
 512                                     adapter->filter_list_dma, mac_address);
 513
 514         if (rc != H_SUCCESS && try_again) {
 515                 do {
 516                         rc = h_free_logical_lan(adapter->vdev->unit_address);
 517                 } while (H_IS_LONG_BUSY(rc) || (rc == H_BUSY));
 518
 519                 try_again = 0;
 520                 goto retry;
 521         }
 522
 523         return rc;
 524 }
 525
 526 static int ibmveth_open(struct net_device *netdev)
 527 {
 528         struct ibmveth_adapter *adapter = netdev_priv(netdev);
 529         u64 mac_address = 0;
 530         int rxq_entries = 1;
 531         unsigned long lpar_rc;
 532         int rc;
 533         union ibmveth_buf_desc rxq_desc;
 534         int i;
 535         struct device *dev;
 536
 537         netdev_dbg(netdev, "open starting\n");
 538
 539         napi_enable(&adapter->napi);
 540
 541         for(i = 0; i < IBMVETH_NUM_BUFF_POOLS; i++)
 542                 rxq_entries += adapter->rx_buff_pool[i].size;
 543
 544         adapter->buffer_list_addr = (void*) get_zeroed_page(GFP_KERNEL);
 545         adapter->filter_list_addr = (void*) get_zeroed_page(GFP_KERNEL);
 546
 547         if (!adapter->buffer_list_addr || !adapter->filter_list_addr) {
 548                 netdev_err(netdev, "unable to allocate filter or buffer list "
 549                            "pages\n");
 550                 rc = -ENOMEM;
 551                 goto err_out;
 552         }
 553
 554         dev = &adapter->vdev->dev;
 555
 556         adapter->rx_queue.queue_len = sizeof(struct ibmveth_rx_q_entry) *
 557                                                 rxq_entries;
 558         adapter->rx_queue.queue_addr =
 559             dma_alloc_coherent(dev, adapter->rx_queue.queue_len,
 560                                &adapter->rx_queue.queue_dma, GFP_KERNEL);
 561
 562         if (!adapter->rx_queue.queue_addr) {
 563                 netdev_err(netdev, "unable to allocate rx queue pages\n");
 564                 rc = -ENOMEM;
 565                 goto err_out;
 566         }
 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
 573         if ((dma_mapping_error(dev, adapter->buffer_list_dma)) ||
 574             (dma_mapping_error(dev, adapter->filter_list_dma))) {
 575                 netdev_err(netdev, "unable to map filter or buffer list "
 576                            "pages\n");
 577                 rc = -ENOMEM;
 578                 goto err_out;
 579         }
 580
 581         adapter->rx_queue.index = 0;
 582         adapter->rx_queue.num_slots = rxq_entries;
 583         adapter->rx_queue.toggle = 1;
 584
 585         memcpy(&mac_address, netdev->dev_addr, netdev->addr_len);
 586         mac_address = mac_address >> 16;
 587
 588         rxq_desc.fields.flags_len = IBMVETH_BUF_VALID |
 589                                         adapter->rx_queue.queue_len;
 590         rxq_desc.fields.address = adapter->rx_queue.queue_dma;
 591
 592         netdev_dbg(netdev, "buffer list @ 0x%p\n", adapter->buffer_list_addr);
 593         netdev_dbg(netdev, "filter list @ 0x%p\n", adapter->filter_list_addr);
 594         netdev_dbg(netdev, "receive q   @ 0x%p\n", adapter->rx_queue.queue_addr);
 595
 596         h_vio_signal(adapter->vdev->unit_address, VIO_IRQ_DISABLE);
 597
 598         lpar_rc = ibmveth_register_logical_lan(adapter, rxq_desc, mac_address);
 599
 600         if (lpar_rc != H_SUCCESS) {
 601                 netdev_err(netdev, "h_register_logical_lan failed with %ld\n",
 602                            lpar_rc);
 603                 netdev_err(netdev, "buffer TCE:0x%llx filter TCE:0x%llx rxq "
 604                            "desc:0x%llx MAC:0x%llx\n",
 605                                      adapter->buffer_list_dma,
 606                                      adapter->filter_list_dma,
 607                                      rxq_desc.desc,
 608                                      mac_address);
 609                 rc = -ENONET;
 610                 goto err_out;
 611         }
 612
 613         for (i = 0; i < IBMVETH_NUM_BUFF_POOLS; i++) {
 614                 if (!adapter->rx_buff_pool[i].active)
 615                         continue;
 616                 if (ibmveth_alloc_buffer_pool(&adapter->rx_buff_pool[i])) {
 617                         netdev_err(netdev, "unable to alloc pool\n");
 618                         adapter->rx_buff_pool[i].active = 0;
 619                         rc = -ENOMEM;
 620                         goto err_out;
 621                 }
 622         }
 623
 624         netdev_dbg(netdev, "registering irq 0x%x\n", netdev->irq);
 625         rc = request_irq(netdev->irq, ibmveth_interrupt, 0, netdev->name,
 626                          netdev);
 627         if (rc != 0) {
 628                 netdev_err(netdev, "unable to request irq 0x%x, rc %d\n",
 629                            netdev->irq, rc);
 630                 do {
 631                         lpar_rc = h_free_logical_lan(adapter->vdev->unit_address);
 632                 } while (H_IS_LONG_BUSY(lpar_rc) || (lpar_rc == H_BUSY));
 633
 634                 goto err_out;
 635         }
 636
 637         adapter->bounce_buffer =
 638             kmalloc(netdev->mtu + IBMVETH_BUFF_OH, GFP_KERNEL);
 639         if (!adapter->bounce_buffer) {
 640                 netdev_err(netdev, "unable to allocate bounce buffer\n");
 641                 rc = -ENOMEM;
 642                 goto err_out_free_irq;
 643         }
 644         adapter->bounce_buffer_dma =
 645             dma_map_single(&adapter->vdev->dev, adapter->bounce_buffer,
 646                            netdev->mtu + IBMVETH_BUFF_OH, DMA_BIDIRECTIONAL);
 647         if (dma_mapping_error(dev, adapter->bounce_buffer_dma)) {
 648                 netdev_err(netdev, "unable to map bounce buffer\n");
 649                 rc = -ENOMEM;
 650                 goto err_out_free_irq;
 651         }
 652
 653         netdev_dbg(netdev, "initial replenish cycle\n");
 654         ibmveth_interrupt(netdev->irq, netdev);
 655
 656         netif_start_queue(netdev);
 657
 658         netdev_dbg(netdev, "open complete\n");
 659
 660         return 0;
 661
 662 err_out_free_irq:
 663         free_irq(netdev->irq, netdev);
 664 err_out:
 665         ibmveth_cleanup(adapter);
 666         napi_disable(&adapter->napi);
 667         return rc;
 668 }
 669
 670 static int ibmveth_close(struct net_device *netdev)
 671 {
 672         struct ibmveth_adapter *adapter = netdev_priv(netdev);
 673         long lpar_rc;
 674
 675         netdev_dbg(netdev, "close starting\n");
 676
 677         napi_disable(&adapter->napi);
 678
 679         if (!adapter->pool_config)
 680                 netif_stop_queue(netdev);
 681
 682         h_vio_signal(adapter->vdev->unit_address, VIO_IRQ_DISABLE);
 683
 684         do {
 685                 lpar_rc = h_free_logical_lan(adapter->vdev->unit_address);
 686         } while (H_IS_LONG_BUSY(lpar_rc) || (lpar_rc == H_BUSY));
 687
 688         if (lpar_rc != H_SUCCESS) {
 689                 netdev_err(netdev, "h_free_logical_lan failed with %lx, "
 690                            "continuing with close\n", lpar_rc);
 691         }
 692
 693         free_irq(netdev->irq, netdev);
 694
 695         adapter->rx_no_buffer = *(u64 *)(((char *)adapter->buffer_list_addr) +
 696                                                 4096 - 8);
 697
 698         ibmveth_cleanup(adapter);
 699
 700         netdev_dbg(netdev, "close complete\n");
 701
 702         return 0;
 703 }
 704
 705 static int netdev_get_settings(struct net_device *dev, struct ethtool_cmd *cmd)
 706 {
 707         cmd->supported = (SUPPORTED_1000baseT_Full | SUPPORTED_Autoneg |
 708                                 SUPPORTED_FIBRE);
 709         cmd->advertising = (ADVERTISED_1000baseT_Full | ADVERTISED_Autoneg |
 710                                 ADVERTISED_FIBRE);
 711         ethtool_cmd_speed_set(cmd, SPEED_1000);
 712         cmd->duplex = DUPLEX_FULL;
 713         cmd->port = PORT_FIBRE;
 714         cmd->phy_address = 0;
 715         cmd->transceiver = XCVR_INTERNAL;
 716         cmd->autoneg = AUTONEG_ENABLE;
 717         cmd->maxtxpkt = 0;
 718         cmd->maxrxpkt = 1;
 719         return 0;
 720 }
 721
 722 static void netdev_get_drvinfo(struct net_device *dev,
 723                                struct ethtool_drvinfo *info)
 724 {
 725         strncpy(info->driver, ibmveth_driver_name, sizeof(info->driver) - 1);
 726         strncpy(info->version, ibmveth_driver_version,
 727                 sizeof(info->version) - 1);
 728 }
 729
 730 static u32 ibmveth_fix_features(struct net_device *dev, u32 features)
 731 {
 732         /*
 733          * Since the ibmveth firmware interface does not have the
 734          * concept of separate tx/rx checksum offload enable, if rx
 735          * checksum is disabled we also have to disable tx checksum
 736          * offload. Once we disable rx checksum offload, we are no
 737          * longer allowed to send tx buffers that are not properly
 738          * checksummed.
 739          */
 740
 741         if (!(features & NETIF_F_RXCSUM))
 742                 features &= ~NETIF_F_ALL_CSUM;
 743
 744         return features;
 745 }
 746
 747 static int ibmveth_set_csum_offload(struct net_device *dev, u32 data)
 748 {
 749         struct ibmveth_adapter *adapter = netdev_priv(dev);
 750         unsigned long set_attr, clr_attr, ret_attr;
 751         unsigned long set_attr6, clr_attr6;
 752         long ret, ret4, ret6;
 753         int rc1 = 0, rc2 = 0;
 754         int restart = 0;
 755
 756         if (netif_running(dev)) {
 757                 restart = 1;
 758                 adapter->pool_config = 1;
 759                 ibmveth_close(dev);
 760                 adapter->pool_config = 0;
 761         }
 762
 763         set_attr = 0;
 764         clr_attr = 0;
 765         set_attr6 = 0;
 766         clr_attr6 = 0;
 767
 768         if (data) {
 769                 set_attr = IBMVETH_ILLAN_IPV4_TCP_CSUM;
 770                 set_attr6 = IBMVETH_ILLAN_IPV6_TCP_CSUM;
 771         } else {
 772                 clr_attr = IBMVETH_ILLAN_IPV4_TCP_CSUM;
 773                 clr_attr6 = IBMVETH_ILLAN_IPV6_TCP_CSUM;
 774         }
 775
 776         ret = h_illan_attributes(adapter->vdev->unit_address, 0, 0, &ret_attr);
 777
 778         if (ret == H_SUCCESS && !(ret_attr & IBMVETH_ILLAN_ACTIVE_TRUNK) &&
 779             !(ret_attr & IBMVETH_ILLAN_TRUNK_PRI_MASK) &&
 780             (ret_attr & IBMVETH_ILLAN_PADDED_PKT_CSUM)) {
 781                 ret4 = h_illan_attributes(adapter->vdev->unit_address, clr_attr,
 782                                          set_attr, &ret_attr);
 783
 784                 if (ret4 != H_SUCCESS) {
 785                         netdev_err(dev, "unable to change IPv4 checksum "
 786                                         "offload settings. %d rc=%ld\n",
 787                                         data, ret4);
 788
 789                         h_illan_attributes(adapter->vdev->unit_address,
 790                                            set_attr, clr_attr, &ret_attr);
 791
 792                         if (data == 1)
 793                                 dev->features &= ~NETIF_F_IP_CSUM;
 794
 795                 } else {
 796                         adapter->fw_ipv4_csum_support = data;
 797                 }
 798
 799                 ret6 = h_illan_attributes(adapter->vdev->unit_address,
 800                                          clr_attr6, set_attr6, &ret_attr);
 801
 802                 if (ret6 != H_SUCCESS) {
 803                         netdev_err(dev, "unable to change IPv6 checksum "
 804                                         "offload settings. %d rc=%ld\n",
 805                                         data, ret6);
 806
 807                         h_illan_attributes(adapter->vdev->unit_address,
 808                                            set_attr6, clr_attr6, &ret_attr);
 809
 810                         if (data == 1)
 811                                 dev->features &= ~NETIF_F_IPV6_CSUM;
 812
 813                 } else
 814                         adapter->fw_ipv6_csum_support = data;
 815
 816                 if (ret4 == H_SUCCESS || ret6 == H_SUCCESS)
 817                         adapter->rx_csum = data;
 818                 else
 819                         rc1 = -EIO;
 820         } else {
 821                 rc1 = -EIO;
 822                 netdev_err(dev, "unable to change checksum offload settings."
 823                                      " %d rc=%ld ret_attr=%lx\n", data, ret,
 824                                      ret_attr);
 825         }
 826
 827         if (restart)
 828                 rc2 = ibmveth_open(dev);
 829
 830         return rc1 ? rc1 : rc2;
 831 }
 832
 833 static int ibmveth_set_features(struct net_device *dev, u32 features)
 834 {
 835         struct ibmveth_adapter *adapter = netdev_priv(dev);
 836         int rx_csum = !!(features & NETIF_F_RXCSUM);
 837         int rc;
 838
 839         if (rx_csum == adapter->rx_csum)
 840                 return 0;
 841
 842         rc = ibmveth_set_csum_offload(dev, rx_csum);
 843         if (rc && !adapter->rx_csum)
 844                 dev->features = features & ~(NETIF_F_ALL_CSUM | NETIF_F_RXCSUM);
 845
 846         return rc;
 847 }
 848
 849 static void ibmveth_get_strings(struct net_device *dev, u32 stringset, u8 *data)
 850 {
 851         int i;
 852
 853         if (stringset != ETH_SS_STATS)
 854                 return;
 855
 856         for (i = 0; i < ARRAY_SIZE(ibmveth_stats); i++, data += ETH_GSTRING_LEN)
 857                 memcpy(data, ibmveth_stats[i].name, ETH_GSTRING_LEN);
 858 }
 859
 860 static int ibmveth_get_sset_count(struct net_device *dev, int sset)
 861 {
 862         switch (sset) {
 863         case ETH_SS_STATS:
 864                 return ARRAY_SIZE(ibmveth_stats);
 865         default:
 866                 return -EOPNOTSUPP;
 867         }
 868 }
 869
 870 static void ibmveth_get_ethtool_stats(struct net_device *dev,
 871                                       struct ethtool_stats *stats, u64 *data)
 872 {
 873         int i;
 874         struct ibmveth_adapter *adapter = netdev_priv(dev);
 875
 876         for (i = 0; i < ARRAY_SIZE(ibmveth_stats); i++)
 877                 data[i] = IBMVETH_GET_STAT(adapter, ibmveth_stats[i].offset);
 878 }
 879
 880 static const struct ethtool_ops netdev_ethtool_ops = {
 881         .get_drvinfo            = netdev_get_drvinfo,
 882         .get_settings           = netdev_get_settings,
 883         .get_link               = ethtool_op_get_link,
 884         .get_strings            = ibmveth_get_strings,
 885         .get_sset_count         = ibmveth_get_sset_count,
 886         .get_ethtool_stats      = ibmveth_get_ethtool_stats,
 887 };
 888
 889 static int ibmveth_ioctl(struct net_device *dev, struct ifreq *ifr, int cmd)
 890 {
 891         return -EOPNOTSUPP;
 892 }
 893
 894 #define page_offset(v) ((unsigned long)(v) & ((1 << 12) - 1))
 895
 896 static int ibmveth_send(struct ibmveth_adapter *adapter,
 897                         union ibmveth_buf_desc *descs)
 898 {
 899         unsigned long correlator;
 900         unsigned int retry_count;
 901         unsigned long ret;
 902
 903         /*
 904          * The retry count sets a maximum for the number of broadcast and
 905          * multicast destinations within the system.
 906          */
 907         retry_count = 1024;
 908         correlator = 0;
 909         do {
 910                 ret = h_send_logical_lan(adapter->vdev->unit_address,
 911                                              descs[0].desc, descs[1].desc,
 912                                              descs[2].desc, descs[3].desc,
 913                                              descs[4].desc, descs[5].desc,
 914                                              correlator, &correlator);
 915         } while ((ret == H_BUSY) && (retry_count--));
 916
 917         if (ret != H_SUCCESS && ret != H_DROPPED) {
 918                 netdev_err(adapter->netdev, "tx: h_send_logical_lan failed "
 919                            "with rc=%ld\n", ret);
 920                 return 1;
 921         }
 922
 923         return 0;
 924 }
 925
 926 static netdev_tx_t ibmveth_start_xmit(struct sk_buff *skb,
 927                                       struct net_device *netdev)
 928 {
 929         struct ibmveth_adapter *adapter = netdev_priv(netdev);
 930         unsigned int desc_flags;
 931         union ibmveth_buf_desc descs[6];
 932         int last, i;
 933         int force_bounce = 0;
 934         dma_addr_t dma_addr;
 935
 936         /*
 937          * veth handles a maximum of 6 segments including the header, so
 938          * we have to linearize the skb if there are more than this.
 939          */
 940         if (skb_shinfo(skb)->nr_frags > 5 && __skb_linearize(skb)) {
 941                 netdev->stats.tx_dropped++;
 942                 goto out;
 943         }
 944
 945         /* veth can't checksum offload UDP */
 946         if (skb->ip_summed == CHECKSUM_PARTIAL &&
 947             ((skb->protocol == htons(ETH_P_IP) &&
 948               ip_hdr(skb)->protocol != IPPROTO_TCP) ||
 949              (skb->protocol == htons(ETH_P_IPV6) &&
 950               ipv6_hdr(skb)->nexthdr != IPPROTO_TCP)) &&
 951             skb_checksum_help(skb)) {
 952
 953                 netdev_err(netdev, "tx: failed to checksum packet\n");
 954                 netdev->stats.tx_dropped++;
 955                 goto out;
 956         }
 957
 958         desc_flags = IBMVETH_BUF_VALID;
 959
 960         if (skb->ip_summed == CHECKSUM_PARTIAL) {
 961                 unsigned char *buf = skb_transport_header(skb) +
 962                                                 skb->csum_offset;
 963
 964                 desc_flags |= (IBMVETH_BUF_NO_CSUM | IBMVETH_BUF_CSUM_GOOD);
 965
 966                 /* Need to zero out the checksum */
 967                 buf[0] = 0;
 968                 buf[1] = 0;
 969         }
 970
 971 retry_bounce:
 972         memset(descs, 0, sizeof(descs));
 973
 974         /*
 975          * If a linear packet is below the rx threshold then
 976          * copy it into the static bounce buffer. This avoids the
 977          * cost of a TCE insert and remove.
 978          */
 979         if (force_bounce || (!skb_is_nonlinear(skb) &&
 980                                 (skb->len < tx_copybreak))) {
 981                 skb_copy_from_linear_data(skb, adapter->bounce_buffer,
 982                                           skb->len);
 983
 984                 descs[0].fields.flags_len = desc_flags | skb->len;
 985                 descs[0].fields.address = adapter->bounce_buffer_dma;
 986
 987                 if (ibmveth_send(adapter, descs)) {
 988                         adapter->tx_send_failed++;
 989                         netdev->stats.tx_dropped++;
 990                 } else {
 991                         netdev->stats.tx_packets++;
 992                         netdev->stats.tx_bytes += skb->len;
 993                 }
 994
 995                 goto out;
 996         }
 997
 998         /* Map the header */
 999         dma_addr = dma_map_single(&adapter->vdev->dev, skb->data,
1000                                   skb_headlen(skb), DMA_TO_DEVICE);
1001         if (dma_mapping_error(&adapter->vdev->dev, dma_addr))
1002                 goto map_failed;
1003
1004         descs[0].fields.flags_len = desc_flags | skb_headlen(skb);
1005         descs[0].fields.address = dma_addr;
1006
1007         /* Map the frags */
1008         for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
1009                 const skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
1010
1011                 dma_addr = skb_frag_dma_map(&adapter->vdev->dev, frag, 0,
1012                                             skb_frag_size(frag), 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 | skb_frag_size(frag);
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         dma_unmap_single(&adapter->vdev->dev,
1030                          descs[0].fields.address,
1031                          descs[0].fields.flags_len & IBMVETH_BUF_LEN_MASK,
1032                          DMA_TO_DEVICE);
1033
1034         for (i = 1; i < skb_shinfo(skb)->nr_frags + 1; i++)
1035                 dma_unmap_page(&adapter->vdev->dev, descs[i].fields.address,
1036                                descs[i].fields.flags_len & IBMVETH_BUF_LEN_MASK,
1037                                DMA_TO_DEVICE);
1038
1039 out:
1040         dev_kfree_skb(skb);
1041         return NETDEV_TX_OK;
1042
1043 map_failed_frags:
1044         last = i+1;
1045         for (i = 0; i < last; i++)
1046                 dma_unmap_page(&adapter->vdev->dev, descs[i].fields.address,
1047                                descs[i].fields.flags_len & IBMVETH_BUF_LEN_MASK,
1048                                DMA_TO_DEVICE);
1049
1050 map_failed:
1051         if (!firmware_has_feature(FW_FEATURE_CMO))
1052                 netdev_err(netdev, "tx: unable to map xmit buffer\n");
1053         adapter->tx_map_failed++;
1054         skb_linearize(skb);
1055         force_bounce = 1;
1056         goto retry_bounce;
1057 }
1058
1059 static int ibmveth_poll(struct napi_struct *napi, int budget)
1060 {
1061         struct ibmveth_adapter *adapter =
1062                         container_of(napi, struct ibmveth_adapter, napi);
1063         struct net_device *netdev = adapter->netdev;
1064         int frames_processed = 0;
1065         unsigned long lpar_rc;
1066
1067 restart_poll:
1068         do {
1069                 if (!ibmveth_rxq_pending_buffer(adapter))
1070                         break;
1071
1072                 smp_rmb();
1073                 if (!ibmveth_rxq_buffer_valid(adapter)) {
1074                         wmb(); /* suggested by larson1 */
1075                         adapter->rx_invalid_buffer++;
1076                         netdev_dbg(netdev, "recycling invalid buffer\n");
1077                         ibmveth_rxq_recycle_buffer(adapter);
1078                 } else {
1079                         struct sk_buff *skb, *new_skb;
1080                         int length = ibmveth_rxq_frame_length(adapter);
1081                         int offset = ibmveth_rxq_frame_offset(adapter);
1082                         int csum_good = ibmveth_rxq_csum_good(adapter);
1083
1084                         skb = ibmveth_rxq_get_buffer(adapter);
1085
1086                         new_skb = NULL;
1087                         if (length < rx_copybreak)
1088                                 new_skb = netdev_alloc_skb(netdev, length);
1089
1090                         if (new_skb) {
1091                                 skb_copy_to_linear_data(new_skb,
1092                                                         skb->data + offset,
1093                                                         length);
1094                                 if (rx_flush)
1095                                         ibmveth_flush_buffer(skb->data,
1096                                                 length + offset);
1097                                 if (!ibmveth_rxq_recycle_buffer(adapter))
1098                                         kfree_skb(skb);
1099                                 skb = new_skb;
1100                         } else {
1101                                 ibmveth_rxq_harvest_buffer(adapter);
1102                                 skb_reserve(skb, offset);
1103                         }
1104
1105                         skb_put(skb, length);
1106                         skb->protocol = eth_type_trans(skb, netdev);
1107
1108                         if (csum_good)
1109                                 skb->ip_summed = CHECKSUM_UNNECESSARY;
1110
1111                         netif_receive_skb(skb); /* send it up */
1112
1113                         netdev->stats.rx_packets++;
1114                         netdev->stats.rx_bytes += length;
1115                         frames_processed++;
1116                 }
1117         } while (frames_processed < budget);
1118
1119         ibmveth_replenish_task(adapter);
1120
1121         if (frames_processed < budget) {
1122                 /* We think we are done - reenable interrupts,
1123                  * then check once more to make sure we are done.
1124                  */
1125                 lpar_rc = h_vio_signal(adapter->vdev->unit_address,
1126                                        VIO_IRQ_ENABLE);
1127
1128                 BUG_ON(lpar_rc != H_SUCCESS);
1129
1130                 napi_complete(napi);
1131
1132                 if (ibmveth_rxq_pending_buffer(adapter) &&
1133                     napi_reschedule(napi)) {
1134                         lpar_rc = h_vio_signal(adapter->vdev->unit_address,
1135                                                VIO_IRQ_DISABLE);
1136                         goto restart_poll;
1137                 }
1138         }
1139
1140         return frames_processed;
1141 }
1142
1143 static irqreturn_t ibmveth_interrupt(int irq, void *dev_instance)
1144 {
1145         struct net_device *netdev = dev_instance;
1146         struct ibmveth_adapter *adapter = netdev_priv(netdev);
1147         unsigned long lpar_rc;
1148
1149         if (napi_schedule_prep(&adapter->napi)) {
1150                 lpar_rc = h_vio_signal(adapter->vdev->unit_address,
1151                                        VIO_IRQ_DISABLE);
1152                 BUG_ON(lpar_rc != H_SUCCESS);
1153                 __napi_schedule(&adapter->napi);
1154         }
1155         return IRQ_HANDLED;
1156 }
1157
1158 static void ibmveth_set_multicast_list(struct net_device *netdev)
1159 {
1160         struct ibmveth_adapter *adapter = netdev_priv(netdev);
1161         unsigned long lpar_rc;
1162
1163         if ((netdev->flags & IFF_PROMISC) ||
1164             (netdev_mc_count(netdev) > adapter->mcastFilterSize)) {
1165                 lpar_rc = h_multicast_ctrl(adapter->vdev->unit_address,
1166                                            IbmVethMcastEnableRecv |
1167                                            IbmVethMcastDisableFiltering,
1168                                            0);
1169                 if (lpar_rc != H_SUCCESS) {
1170                         netdev_err(netdev, "h_multicast_ctrl rc=%ld when "
1171                                    "entering promisc mode\n", lpar_rc);
1172                 }
1173         } else {
1174                 struct netdev_hw_addr *ha;
1175                 /* clear the filter table & disable filtering */
1176                 lpar_rc = h_multicast_ctrl(adapter->vdev->unit_address,
1177                                            IbmVethMcastEnableRecv |
1178                                            IbmVethMcastDisableFiltering |
1179                                            IbmVethMcastClearFilterTable,
1180                                            0);
1181                 if (lpar_rc != H_SUCCESS) {
1182                         netdev_err(netdev, "h_multicast_ctrl rc=%ld when "
1183                                    "attempting to clear filter table\n",
1184                                    lpar_rc);
1185                 }
1186                 /* add the addresses to the filter table */
1187                 netdev_for_each_mc_addr(ha, netdev) {
1188                         /* add the multicast address to the filter table */
1189                         unsigned long mcast_addr = 0;
1190                         memcpy(((char *)&mcast_addr)+2, ha->addr, 6);
1191                         lpar_rc = h_multicast_ctrl(adapter->vdev->unit_address,
1192                                                    IbmVethMcastAddFilter,
1193                                                    mcast_addr);
1194                         if (lpar_rc != H_SUCCESS) {
1195                                 netdev_err(netdev, "h_multicast_ctrl rc=%ld "
1196                                            "when adding an entry to the filter "
1197                                            "table\n", lpar_rc);
1198                         }
1199                 }
1200
1201                 /* re-enable filtering */
1202                 lpar_rc = h_multicast_ctrl(adapter->vdev->unit_address,
1203                                            IbmVethMcastEnableFiltering,
1204                                            0);
1205                 if (lpar_rc != H_SUCCESS) {
1206                         netdev_err(netdev, "h_multicast_ctrl rc=%ld when "
1207                                    "enabling filtering\n", lpar_rc);
1208                 }
1209         }
1210 }
1211
1212 static int ibmveth_change_mtu(struct net_device *dev, int new_mtu)
1213 {
1214         struct ibmveth_adapter *adapter = netdev_priv(dev);
1215         struct vio_dev *viodev = adapter->vdev;
1216         int new_mtu_oh = new_mtu + IBMVETH_BUFF_OH;
1217         int i, rc;
1218         int need_restart = 0;
1219
1220         if (new_mtu < IBMVETH_MIN_MTU)
1221                 return -EINVAL;
1222
1223         for (i = 0; i < IBMVETH_NUM_BUFF_POOLS; i++)
1224                 if (new_mtu_oh < adapter->rx_buff_pool[i].buff_size)
1225                         break;
1226
1227         if (i == IBMVETH_NUM_BUFF_POOLS)
1228                 return -EINVAL;
1229
1230         /* Deactivate all the buffer pools so that the next loop can activate
1231            only the buffer pools necessary to hold the new MTU */
1232         if (netif_running(adapter->netdev)) {
1233                 need_restart = 1;
1234                 adapter->pool_config = 1;
1235                 ibmveth_close(adapter->netdev);
1236                 adapter->pool_config = 0;
1237         }
1238
1239         /* Look for an active buffer pool that can hold the new MTU */
1240         for (i = 0; i < IBMVETH_NUM_BUFF_POOLS; i++) {
1241                 adapter->rx_buff_pool[i].active = 1;
1242
1243                 if (new_mtu_oh < adapter->rx_buff_pool[i].buff_size) {
1244                         dev->mtu = new_mtu;
1245                         vio_cmo_set_dev_desired(viodev,
1246                                                 ibmveth_get_desired_dma
1247                                                 (viodev));
1248                         if (need_restart) {
1249                                 return ibmveth_open(adapter->netdev);
1250                         }
1251                         return 0;
1252                 }
1253         }
1254
1255         if (need_restart && (rc = ibmveth_open(adapter->netdev)))
1256                 return rc;
1257
1258         return -EINVAL;
1259 }
1260
1261 #ifdef CONFIG_NET_POLL_CONTROLLER
1262 static void ibmveth_poll_controller(struct net_device *dev)
1263 {
1264         ibmveth_replenish_task(netdev_priv(dev));
1265         ibmveth_interrupt(dev->irq, dev);
1266 }
1267 #endif
1268
1269 /**
1270  * ibmveth_get_desired_dma - Calculate IO memory desired by the driver
1271  *
1272  * @vdev: struct vio_dev for the device whose desired IO mem is to be returned
1273  *
1274  * Return value:
1275  *      Number of bytes of IO data the driver will need to perform well.
1276  */
1277 static unsigned long ibmveth_get_desired_dma(struct vio_dev *vdev)
1278 {
1279         struct net_device *netdev = dev_get_drvdata(&vdev->dev);
1280         struct ibmveth_adapter *adapter;
1281         unsigned long ret;
1282         int i;
1283         int rxqentries = 1;
1284
1285         /* netdev inits at probe time along with the structures we need below*/
1286         if (netdev == NULL)
1287                 return IOMMU_PAGE_ALIGN(IBMVETH_IO_ENTITLEMENT_DEFAULT);
1288
1289         adapter = netdev_priv(netdev);
1290
1291         ret = IBMVETH_BUFF_LIST_SIZE + IBMVETH_FILT_LIST_SIZE;
1292         ret += IOMMU_PAGE_ALIGN(netdev->mtu);
1293
1294         for (i = 0; i < IBMVETH_NUM_BUFF_POOLS; i++) {
1295                 /* add the size of the active receive buffers */
1296                 if (adapter->rx_buff_pool[i].active)
1297                         ret +=
1298                             adapter->rx_buff_pool[i].size *
1299                             IOMMU_PAGE_ALIGN(adapter->rx_buff_pool[i].
1300                                     buff_size);
1301                 rxqentries += adapter->rx_buff_pool[i].size;
1302         }
1303         /* add the size of the receive queue entries */
1304         ret += IOMMU_PAGE_ALIGN(rxqentries * sizeof(struct ibmveth_rx_q_entry));
1305
1306         return ret;
1307 }
1308
1309 static const struct net_device_ops ibmveth_netdev_ops = {
1310         .ndo_open               = ibmveth_open,
1311         .ndo_stop               = ibmveth_close,
1312         .ndo_start_xmit         = ibmveth_start_xmit,
1313         .ndo_set_rx_mode        = ibmveth_set_multicast_list,
1314         .ndo_do_ioctl           = ibmveth_ioctl,
1315         .ndo_change_mtu         = ibmveth_change_mtu,
1316         .ndo_fix_features       = ibmveth_fix_features,
1317         .ndo_set_features       = ibmveth_set_features,
1318         .ndo_validate_addr      = eth_validate_addr,
1319         .ndo_set_mac_address    = eth_mac_addr,
1320 #ifdef CONFIG_NET_POLL_CONTROLLER
1321         .ndo_poll_controller    = ibmveth_poll_controller,
1322 #endif
1323 };
1324
1325 static int __devinit ibmveth_probe(struct vio_dev *dev,
1326                                    const struct vio_device_id *id)
1327 {
1328         int rc, i;
1329         struct net_device *netdev;
1330         struct ibmveth_adapter *adapter;
1331         unsigned char *mac_addr_p;
1332         unsigned int *mcastFilterSize_p;
1333
1334         dev_dbg(&dev->dev, "entering ibmveth_probe for UA 0x%x\n",
1335                 dev->unit_address);
1336
1337         mac_addr_p = (unsigned char *)vio_get_attribute(dev, VETH_MAC_ADDR,
1338                                                         NULL);
1339         if (!mac_addr_p) {
1340                 dev_err(&dev->dev, "Can't find VETH_MAC_ADDR attribute\n");
1341                 return -EINVAL;
1342         }
1343
1344         mcastFilterSize_p = (unsigned int *)vio_get_attribute(dev,
1345                                                 VETH_MCAST_FILTER_SIZE, NULL);
1346         if (!mcastFilterSize_p) {
1347                 dev_err(&dev->dev, "Can't find VETH_MCAST_FILTER_SIZE "
1348                         "attribute\n");
1349                 return -EINVAL;
1350         }
1351
1352         netdev = alloc_etherdev(sizeof(struct ibmveth_adapter));
1353
1354         if (!netdev)
1355                 return -ENOMEM;
1356
1357         adapter = netdev_priv(netdev);
1358         dev_set_drvdata(&dev->dev, netdev);
1359
1360         adapter->vdev = dev;
1361         adapter->netdev = netdev;
1362         adapter->mcastFilterSize = *mcastFilterSize_p;
1363         adapter->pool_config = 0;
1364
1365         netif_napi_add(netdev, &adapter->napi, ibmveth_poll, 16);
1366
1367         /*
1368          * Some older boxes running PHYP non-natively have an OF that returns
1369          * a 8-byte local-mac-address field (and the first 2 bytes have to be
1370          * ignored) while newer boxes' OF return a 6-byte field. Note that
1371          * IEEE 1275 specifies that local-mac-address must be a 6-byte field.
1372          * The RPA doc specifies that the first byte must be 10b, so we'll
1373          * just look for it to solve this 8 vs. 6 byte field issue
1374          */
1375         if ((*mac_addr_p & 0x3) != 0x02)
1376                 mac_addr_p += 2;
1377
1378         adapter->mac_addr = 0;
1379         memcpy(&adapter->mac_addr, mac_addr_p, 6);
1380
1381         netdev->irq = dev->irq;
1382         netdev->netdev_ops = &ibmveth_netdev_ops;
1383         netdev->ethtool_ops = &netdev_ethtool_ops;
1384         SET_NETDEV_DEV(netdev, &dev->dev);
1385         netdev->hw_features = NETIF_F_SG | NETIF_F_RXCSUM |
1386                 NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM;
1387         netdev->features |= netdev->hw_features;
1388
1389         memcpy(netdev->dev_addr, &adapter->mac_addr, netdev->addr_len);
1390
1391         for (i = 0; i < IBMVETH_NUM_BUFF_POOLS; i++) {
1392                 struct kobject *kobj = &adapter->rx_buff_pool[i].kobj;
1393                 int error;
1394
1395                 ibmveth_init_buffer_pool(&adapter->rx_buff_pool[i], i,
1396                                          pool_count[i], pool_size[i],
1397                                          pool_active[i]);
1398                 error = kobject_init_and_add(kobj, &ktype_veth_pool,
1399                                              &dev->dev.kobj, "pool%d", i);
1400                 if (!error)
1401                         kobject_uevent(kobj, KOBJ_ADD);
1402         }
1403
1404         netdev_dbg(netdev, "adapter @ 0x%p\n", adapter);
1405
1406         adapter->buffer_list_dma = DMA_ERROR_CODE;
1407         adapter->filter_list_dma = DMA_ERROR_CODE;
1408         adapter->rx_queue.queue_dma = DMA_ERROR_CODE;
1409
1410         netdev_dbg(netdev, "registering netdev...\n");
1411
1412         ibmveth_set_features(netdev, netdev->features);
1413
1414         rc = register_netdev(netdev);
1415
1416         if (rc) {
1417                 netdev_dbg(netdev, "failed to register netdev rc=%d\n", rc);
1418                 free_netdev(netdev);
1419                 return rc;
1420         }
1421
1422         netdev_dbg(netdev, "registered\n");
1423
1424         return 0;
1425 }
1426
1427 static int __devexit ibmveth_remove(struct vio_dev *dev)
1428 {
1429         struct net_device *netdev = dev_get_drvdata(&dev->dev);
1430         struct ibmveth_adapter *adapter = netdev_priv(netdev);
1431         int i;
1432
1433         for (i = 0; i < IBMVETH_NUM_BUFF_POOLS; i++)
1434                 kobject_put(&adapter->rx_buff_pool[i].kobj);
1435
1436         unregister_netdev(netdev);
1437
1438         free_netdev(netdev);
1439         dev_set_drvdata(&dev->dev, NULL);
1440
1441         return 0;
1442 }
1443
1444 static struct attribute veth_active_attr;
1445 static struct attribute veth_num_attr;
1446 static struct attribute veth_size_attr;
1447
1448 static ssize_t veth_pool_show(struct kobject *kobj,
1449                               struct attribute *attr, char *buf)
1450 {
1451         struct ibmveth_buff_pool *pool = container_of(kobj,
1452                                                       struct ibmveth_buff_pool,
1453                                                       kobj);
1454
1455         if (attr == &veth_active_attr)
1456                 return sprintf(buf, "%d\n", pool->active);
1457         else if (attr == &veth_num_attr)
1458                 return sprintf(buf, "%d\n", pool->size);
1459         else if (attr == &veth_size_attr)
1460                 return sprintf(buf, "%d\n", pool->buff_size);
1461         return 0;
1462 }
1463
1464 static ssize_t veth_pool_store(struct kobject *kobj, struct attribute *attr,
1465                                const char *buf, size_t count)
1466 {
1467         struct ibmveth_buff_pool *pool = container_of(kobj,
1468                                                       struct ibmveth_buff_pool,
1469                                                       kobj);
1470         struct net_device *netdev = dev_get_drvdata(
1471             container_of(kobj->parent, struct device, kobj));
1472         struct ibmveth_adapter *adapter = netdev_priv(netdev);
1473         long value = simple_strtol(buf, NULL, 10);
1474         long rc;
1475
1476         if (attr == &veth_active_attr) {
1477                 if (value && !pool->active) {
1478                         if (netif_running(netdev)) {
1479                                 if (ibmveth_alloc_buffer_pool(pool)) {
1480                                         netdev_err(netdev,
1481                                                    "unable to alloc pool\n");
1482                                         return -ENOMEM;
1483                                 }
1484                                 pool->active = 1;
1485                                 adapter->pool_config = 1;
1486                                 ibmveth_close(netdev);
1487                                 adapter->pool_config = 0;
1488                                 if ((rc = ibmveth_open(netdev)))
1489                                         return rc;
1490                         } else {
1491                                 pool->active = 1;
1492                         }
1493                 } else if (!value && pool->active) {
1494                         int mtu = netdev->mtu + IBMVETH_BUFF_OH;
1495                         int i;
1496                         /* Make sure there is a buffer pool with buffers that
1497                            can hold a packet of the size of the MTU */
1498                         for (i = 0; i < IBMVETH_NUM_BUFF_POOLS; i++) {
1499                                 if (pool == &adapter->rx_buff_pool[i])
1500                                         continue;
1501                                 if (!adapter->rx_buff_pool[i].active)
1502                                         continue;
1503                                 if (mtu <= adapter->rx_buff_pool[i].buff_size)
1504                                         break;
1505                         }
1506
1507                         if (i == IBMVETH_NUM_BUFF_POOLS) {
1508                                 netdev_err(netdev, "no active pool >= MTU\n");
1509                                 return -EPERM;
1510                         }
1511
1512                         if (netif_running(netdev)) {
1513                                 adapter->pool_config = 1;
1514                                 ibmveth_close(netdev);
1515                                 pool->active = 0;
1516                                 adapter->pool_config = 0;
1517                                 if ((rc = ibmveth_open(netdev)))
1518                                         return rc;
1519                         }
1520                         pool->active = 0;
1521                 }
1522         } else if (attr == &veth_num_attr) {
1523                 if (value <= 0 || value > IBMVETH_MAX_POOL_COUNT) {
1524                         return -EINVAL;
1525                 } else {
1526                         if (netif_running(netdev)) {
1527                                 adapter->pool_config = 1;
1528                                 ibmveth_close(netdev);
1529                                 adapter->pool_config = 0;
1530                                 pool->size = value;
1531                                 if ((rc = ibmveth_open(netdev)))
1532                                         return rc;
1533                         } else {
1534                                 pool->size = value;
1535                         }
1536                 }
1537         } else if (attr == &veth_size_attr) {
1538                 if (value <= IBMVETH_BUFF_OH || value > IBMVETH_MAX_BUF_SIZE) {
1539                         return -EINVAL;
1540                 } else {
1541                         if (netif_running(netdev)) {
1542                                 adapter->pool_config = 1;
1543                                 ibmveth_close(netdev);
1544                                 adapter->pool_config = 0;
1545                                 pool->buff_size = value;
1546                                 if ((rc = ibmveth_open(netdev)))
1547                                         return rc;
1548                         } else {
1549                                 pool->buff_size = value;
1550                         }
1551                 }
1552         }
1553
1554         /* kick the interrupt handler to allocate/deallocate pools */
1555         ibmveth_interrupt(netdev->irq, netdev);
1556         return count;
1557 }
1558
1559
1560 #define ATTR(_name, _mode)                              \
1561         struct attribute veth_##_name##_attr = {        \
1562         .name = __stringify(_name), .mode = _mode,      \
1563         };
1564
1565 static ATTR(active, 0644);
1566 static ATTR(num, 0644);
1567 static ATTR(size, 0644);
1568
1569 static struct attribute *veth_pool_attrs[] = {
1570         &veth_active_attr,
1571         &veth_num_attr,
1572         &veth_size_attr,
1573         NULL,
1574 };
1575
1576 static const struct sysfs_ops veth_pool_ops = {
1577         .show   = veth_pool_show,
1578         .store  = veth_pool_store,
1579 };
1580
1581 static struct kobj_type ktype_veth_pool = {
1582         .release        = NULL,
1583         .sysfs_ops      = &veth_pool_ops,
1584         .default_attrs  = veth_pool_attrs,
1585 };
1586
1587 static int ibmveth_resume(struct device *dev)
1588 {
1589         struct net_device *netdev = dev_get_drvdata(dev);
1590         ibmveth_interrupt(netdev->irq, netdev);
1591         return 0;
1592 }
1593
1594 static struct vio_device_id ibmveth_device_table[] __devinitdata = {
1595         { "network", "IBM,l-lan"},
1596         { "", "" }
1597 };
1598 MODULE_DEVICE_TABLE(vio, ibmveth_device_table);
1599
1600 static struct dev_pm_ops ibmveth_pm_ops = {
1601         .resume = ibmveth_resume
1602 };
1603
1604 static struct vio_driver ibmveth_driver = {
1605         .id_table       = ibmveth_device_table,
1606         .probe          = ibmveth_probe,
1607         .remove         = ibmveth_remove,
1608         .get_desired_dma = ibmveth_get_desired_dma,
1609         .driver         = {
1610                 .name   = ibmveth_driver_name,
1611                 .owner  = THIS_MODULE,
1612                 .pm = &ibmveth_pm_ops,
1613         }
1614 };
1615
1616 static int __init ibmveth_module_init(void)
1617 {
1618         printk(KERN_DEBUG "%s: %s %s\n", ibmveth_driver_name,
1619                ibmveth_driver_string, ibmveth_driver_version);
1620
1621         return vio_register_driver(&ibmveth_driver);
1622 }
1623
1624 static void __exit ibmveth_module_exit(void)
1625 {
1626         vio_unregister_driver(&ibmveth_driver);
1627 }
1628
1629 module_init(ibmveth_module_init);
1630 module_exit(ibmveth_module_exit);