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