drivers/net/ibmveth.c

   1 /**************************************************************************/
   2 /*                                                                        */
   3 /* IBM eServer i/pSeries Virtual Ethernet Device Driver                   */
   4 /* Copyright (C) 2003 IBM Corp.                                           */
   5 /*  Originally written by Dave Larson (larson1@us.ibm.com)                */
   6 /*  Maintained by Santiago Leon (santil@us.ibm.com)                       */
   7 /*                                                                        */
   8 /*  This program is free software; you can redistribute it and/or modify  */
   9 /*  it under the terms of the GNU General Public License as published by  */
  10 /*  the Free Software Foundation; either version 2 of the License, or     */
  11 /*  (at your option) any later version.                                   */
  12 /*                                                                        */
  13 /*  This program is distributed in the hope that it will be useful,       */
  14 /*  but WITHOUT ANY WARRANTY; without even the implied warranty of        */
  15 /*  MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the         */
  16 /*  GNU General Public License for more details.                          */
  17 /*                                                                        */
  18 /*  You should have received a copy of the GNU General Public License     */
  19 /*  along with this program; if not, write to the Free Software           */
  20 /*  Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA  02111-1307  */
  21 /*                                                                   USA  */
  22 /*                                                                        */
  23 /* This module contains the implementation of a virtual ethernet device   */
  24 /* for use with IBM i/pSeries LPAR Linux.  It utilizes the logical LAN    */
  25 /* option of the RS/6000 Platform Architechture to interface with virtual */
  26 /* ethernet NICs that are presented to the partition by the hypervisor.   */
  27 /*                                                                        */
  28 /**************************************************************************/
  29 /*
  30   TODO:
  31   - remove frag processing code - no longer needed
  32   - add support for sysfs
  33   - possibly remove procfs support
  34 */
  35
  36 #include <linux/module.h>
  37 #include <linux/types.h>
  38 #include <linux/errno.h>
  39 #include <linux/ioport.h>
  40 #include <linux/dma-mapping.h>
  41 #include <linux/kernel.h>
  42 #include <linux/netdevice.h>
  43 #include <linux/etherdevice.h>
  44 #include <linux/skbuff.h>
  45 #include <linux/init.h>
  46 #include <linux/delay.h>
  47 #include <linux/mm.h>
  48 #include <linux/ethtool.h>
  49 #include <linux/proc_fs.h>
  50 #include <linux/in.h>
  51 #include <linux/ip.h>
  52 #include <net/net_namespace.h>
  53 #include <asm/semaphore.h>
  54 #include <asm/hvcall.h>
  55 #include <asm/atomic.h>
  56 #include <asm/vio.h>
  57 #include <asm/uaccess.h>
  58 #include <linux/seq_file.h>
  59
  60 #include "ibmveth.h"
  61
  62 #undef DEBUG
  63
  64 #define ibmveth_printk(fmt, args...) \
  65   printk(KERN_DEBUG "%s: " fmt, __FILE__, ## args)
  66
  67 #define ibmveth_error_printk(fmt, args...) \
  68   printk(KERN_ERR "(%s:%3.3d ua:%x) ERROR: " fmt, __FILE__, __LINE__ , adapter->vdev->unit_address, ## args)
  69
  70 #ifdef DEBUG
  71 #define ibmveth_debug_printk_no_adapter(fmt, args...) \
  72   printk(KERN_DEBUG "(%s:%3.3d): " fmt, __FILE__, __LINE__ , ## args)
  73 #define ibmveth_debug_printk(fmt, args...) \
  74   printk(KERN_DEBUG "(%s:%3.3d ua:%x): " fmt, __FILE__, __LINE__ , adapter->vdev->unit_address, ## args)
  75 #define ibmveth_assert(expr) \
  76   if(!(expr)) {                                   \
  77     printk(KERN_DEBUG "assertion failed (%s:%3.3d ua:%x): %s\n", __FILE__, __LINE__, adapter->vdev->unit_address, #expr); \
  78     BUG(); \
  79   }
  80 #else
  81 #define ibmveth_debug_printk_no_adapter(fmt, args...)
  82 #define ibmveth_debug_printk(fmt, args...)
  83 #define ibmveth_assert(expr)
  84 #endif
  85
  86 static int ibmveth_open(struct net_device *dev);
  87 static int ibmveth_close(struct net_device *dev);
  88 static int ibmveth_ioctl(struct net_device *dev, struct ifreq *ifr, int cmd);
  89 static int ibmveth_poll(struct napi_struct *napi, int budget);
  90 static int ibmveth_start_xmit(struct sk_buff *skb, struct net_device *dev);
  91 static struct net_device_stats *ibmveth_get_stats(struct net_device *dev);
  92 static void ibmveth_set_multicast_list(struct net_device *dev);
  93 static int ibmveth_change_mtu(struct net_device *dev, int new_mtu);
  94 static void ibmveth_proc_register_driver(void);
  95 static void ibmveth_proc_unregister_driver(void);
  96 static void ibmveth_proc_register_adapter(struct ibmveth_adapter *adapter);
  97 static void ibmveth_proc_unregister_adapter(struct ibmveth_adapter *adapter);
  98 static irqreturn_t ibmveth_interrupt(int irq, void *dev_instance);
  99 static void ibmveth_rxq_harvest_buffer(struct ibmveth_adapter *adapter);
 100 static struct kobj_type ktype_veth_pool;
 101
 102 #ifdef CONFIG_PROC_FS
 103 #define IBMVETH_PROC_DIR "ibmveth"
 104 static struct proc_dir_entry *ibmveth_proc_dir;
 105 #endif
 106
 107 static const char ibmveth_driver_name[] = "ibmveth";
 108 static const char ibmveth_driver_string[] = "IBM i/pSeries Virtual Ethernet Driver";
 109 #define ibmveth_driver_version "1.03"
 110
 111 MODULE_AUTHOR("Santiago Leon <santil@us.ibm.com>");
 112 MODULE_DESCRIPTION("IBM i/pSeries Virtual Ethernet Driver");
 113 MODULE_LICENSE("GPL");
 114 MODULE_VERSION(ibmveth_driver_version);
 115
 116 /* simple methods of getting data from the current rxq entry */
 117 static inline int ibmveth_rxq_pending_buffer(struct ibmveth_adapter *adapter)
 118 {
 119         return (adapter->rx_queue.queue_addr[adapter->rx_queue.index].toggle == adapter->rx_queue.toggle);
 120 }
 121
 122 static inline int ibmveth_rxq_buffer_valid(struct ibmveth_adapter *adapter)
 123 {
 124         return (adapter->rx_queue.queue_addr[adapter->rx_queue.index].valid);
 125 }
 126
 127 static inline int ibmveth_rxq_frame_offset(struct ibmveth_adapter *adapter)
 128 {
 129         return (adapter->rx_queue.queue_addr[adapter->rx_queue.index].offset);
 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 (adapter->rx_queue.queue_addr[adapter->rx_queue.index].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, u32 pool_index, u32 pool_size, u32 buff_size, u32 pool_active)
 144 {
 145         pool->size = pool_size;
 146         pool->index = pool_index;
 147         pool->buff_size = buff_size;
 148         pool->threshold = pool_size / 2;
 149         pool->active = pool_active;
 150 }
 151
 152 /* allocate and setup an buffer pool - called during open */
 153 static int ibmveth_alloc_buffer_pool(struct ibmveth_buff_pool *pool)
 154 {
 155         int i;
 156
 157         pool->free_map = kmalloc(sizeof(u16) * pool->size, GFP_KERNEL);
 158
 159         if(!pool->free_map) {
 160                 return -1;
 161         }
 162
 163         pool->dma_addr = kmalloc(sizeof(dma_addr_t) * pool->size, GFP_KERNEL);
 164         if(!pool->dma_addr) {
 165                 kfree(pool->free_map);
 166                 pool->free_map = NULL;
 167                 return -1;
 168         }
 169
 170         pool->skbuff = kmalloc(sizeof(void*) * pool->size, GFP_KERNEL);
 171
 172         if(!pool->skbuff) {
 173                 kfree(pool->dma_addr);
 174                 pool->dma_addr = NULL;
 175
 176                 kfree(pool->free_map);
 177                 pool->free_map = NULL;
 178                 return -1;
 179         }
 180
 181         memset(pool->skbuff, 0, sizeof(void*) * pool->size);
 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
 188         atomic_set(&pool->available, 0);
 189         pool->producer_index = 0;
 190         pool->consumer_index = 0;
 191
 192         return 0;
 193 }
 194
 195 /* replenish the buffers for a pool.  note that we don't need to
 196  * skb_reserve these since they are used for incoming...
 197  */
 198 static void ibmveth_replenish_buffer_pool(struct ibmveth_adapter *adapter, struct ibmveth_buff_pool *pool)
 199 {
 200         u32 i;
 201         u32 count = pool->size - atomic_read(&pool->available);
 202         u32 buffers_added = 0;
 203
 204         mb();
 205
 206         for(i = 0; i < count; ++i) {
 207                 struct sk_buff *skb;
 208                 unsigned int free_index, index;
 209                 u64 correlator;
 210                 union ibmveth_buf_desc desc;
 211                 unsigned long lpar_rc;
 212                 dma_addr_t dma_addr;
 213
 214                 skb = alloc_skb(pool->buff_size, GFP_ATOMIC);
 215
 216                 if(!skb) {
 217                         ibmveth_debug_printk("replenish: unable to allocate skb\n");
 218                         adapter->replenish_no_mem++;
 219                         break;
 220                 }
 221
 222                 free_index = pool->consumer_index;
 223                 pool->consumer_index = (pool->consumer_index + 1) % pool->size;
 224                 index = pool->free_map[free_index];
 225
 226                 ibmveth_assert(index != IBM_VETH_INVALID_MAP);
 227                 ibmveth_assert(pool->skbuff[index] == NULL);
 228
 229                 dma_addr = dma_map_single(&adapter->vdev->dev, skb->data,
 230                                 pool->buff_size, DMA_FROM_DEVICE);
 231
 232                 pool->free_map[free_index] = IBM_VETH_INVALID_MAP;
 233                 pool->dma_addr[index] = dma_addr;
 234                 pool->skbuff[index] = skb;
 235
 236                 correlator = ((u64)pool->index << 32) | index;
 237                 *(u64*)skb->data = correlator;
 238
 239                 desc.desc = 0;
 240                 desc.fields.valid = 1;
 241                 desc.fields.length = pool->buff_size;
 242                 desc.fields.address = dma_addr;
 243
 244                 lpar_rc = h_add_logical_lan_buffer(adapter->vdev->unit_address, desc.desc);
 245
 246                 if(lpar_rc != H_SUCCESS) {
 247                         pool->free_map[free_index] = index;
 248                         pool->skbuff[index] = NULL;
 249                         if (pool->consumer_index == 0)
 250                                 pool->consumer_index = pool->size - 1;
 251                         else
 252                                 pool->consumer_index--;
 253                         dma_unmap_single(&adapter->vdev->dev,
 254                                         pool->dma_addr[index], pool->buff_size,
 255                                         DMA_FROM_DEVICE);
 256                         dev_kfree_skb_any(skb);
 257                         adapter->replenish_add_buff_failure++;
 258                         break;
 259                 } else {
 260                         buffers_added++;
 261                         adapter->replenish_add_buff_success++;
 262                 }
 263         }
 264
 265         mb();
 266         atomic_add(buffers_added, &(pool->available));
 267 }
 268
 269 /* replenish routine */
 270 static void ibmveth_replenish_task(struct ibmveth_adapter *adapter)
 271 {
 272         int i;
 273
 274         adapter->replenish_task_cycles++;
 275
 276         for(i = 0; i < IbmVethNumBufferPools; i++)
 277                 if(adapter->rx_buff_pool[i].active)
 278                         ibmveth_replenish_buffer_pool(adapter,
 279                                                      &adapter->rx_buff_pool[i]);
 280
 281         adapter->rx_no_buffer = *(u64*)(((char*)adapter->buffer_list_addr) + 4096 - 8);
 282 }
 283
 284 /* empty and free ana buffer pool - also used to do cleanup in error paths */
 285 static void ibmveth_free_buffer_pool(struct ibmveth_adapter *adapter, struct ibmveth_buff_pool *pool)
 286 {
 287         int i;
 288
 289         kfree(pool->free_map);
 290         pool->free_map = NULL;
 291
 292         if(pool->skbuff && pool->dma_addr) {
 293                 for(i = 0; i < pool->size; ++i) {
 294                         struct sk_buff *skb = pool->skbuff[i];
 295                         if(skb) {
 296                                 dma_unmap_single(&adapter->vdev->dev,
 297                                                  pool->dma_addr[i],
 298                                                  pool->buff_size,
 299                                                  DMA_FROM_DEVICE);
 300                                 dev_kfree_skb_any(skb);
 301                                 pool->skbuff[i] = NULL;
 302                         }
 303                 }
 304         }
 305
 306         if(pool->dma_addr) {
 307                 kfree(pool->dma_addr);
 308                 pool->dma_addr = NULL;
 309         }
 310
 311         if(pool->skbuff) {
 312                 kfree(pool->skbuff);
 313                 pool->skbuff = NULL;
 314         }
 315 }
 316
 317 /* remove a buffer from a pool */
 318 static void ibmveth_remove_buffer_from_pool(struct ibmveth_adapter *adapter, u64 correlator)
 319 {
 320         unsigned int pool  = correlator >> 32;
 321         unsigned int index = correlator & 0xffffffffUL;
 322         unsigned int free_index;
 323         struct sk_buff *skb;
 324
 325         ibmveth_assert(pool < IbmVethNumBufferPools);
 326         ibmveth_assert(index < adapter->rx_buff_pool[pool].size);
 327
 328         skb = adapter->rx_buff_pool[pool].skbuff[index];
 329
 330         ibmveth_assert(skb != NULL);
 331
 332         adapter->rx_buff_pool[pool].skbuff[index] = NULL;
 333
 334         dma_unmap_single(&adapter->vdev->dev,
 335                          adapter->rx_buff_pool[pool].dma_addr[index],
 336                          adapter->rx_buff_pool[pool].buff_size,
 337                          DMA_FROM_DEVICE);
 338
 339         free_index = adapter->rx_buff_pool[pool].producer_index;
 340         adapter->rx_buff_pool[pool].producer_index
 341                 = (adapter->rx_buff_pool[pool].producer_index + 1)
 342                 % adapter->rx_buff_pool[pool].size;
 343         adapter->rx_buff_pool[pool].free_map[free_index] = index;
 344
 345         mb();
 346
 347         atomic_dec(&(adapter->rx_buff_pool[pool].available));
 348 }
 349
 350 /* get the current buffer on the rx queue */
 351 static inline struct sk_buff *ibmveth_rxq_get_buffer(struct ibmveth_adapter *adapter)
 352 {
 353         u64 correlator = adapter->rx_queue.queue_addr[adapter->rx_queue.index].correlator;
 354         unsigned int pool = correlator >> 32;
 355         unsigned int index = correlator & 0xffffffffUL;
 356
 357         ibmveth_assert(pool < IbmVethNumBufferPools);
 358         ibmveth_assert(index < adapter->rx_buff_pool[pool].size);
 359
 360         return adapter->rx_buff_pool[pool].skbuff[index];
 361 }
 362
 363 /* recycle the current buffer on the rx queue */
 364 static void ibmveth_rxq_recycle_buffer(struct ibmveth_adapter *adapter)
 365 {
 366         u32 q_index = adapter->rx_queue.index;
 367         u64 correlator = adapter->rx_queue.queue_addr[q_index].correlator;
 368         unsigned int pool = correlator >> 32;
 369         unsigned int index = correlator & 0xffffffffUL;
 370         union ibmveth_buf_desc desc;
 371         unsigned long lpar_rc;
 372
 373         ibmveth_assert(pool < IbmVethNumBufferPools);
 374         ibmveth_assert(index < adapter->rx_buff_pool[pool].size);
 375
 376         if(!adapter->rx_buff_pool[pool].active) {
 377                 ibmveth_rxq_harvest_buffer(adapter);
 378                 ibmveth_free_buffer_pool(adapter, &adapter->rx_buff_pool[pool]);
 379                 return;
 380         }
 381
 382         desc.desc = 0;
 383         desc.fields.valid = 1;
 384         desc.fields.length = adapter->rx_buff_pool[pool].buff_size;
 385         desc.fields.address = adapter->rx_buff_pool[pool].dma_addr[index];
 386
 387         lpar_rc = h_add_logical_lan_buffer(adapter->vdev->unit_address, desc.desc);
 388
 389         if(lpar_rc != H_SUCCESS) {
 390                 ibmveth_debug_printk("h_add_logical_lan_buffer failed during recycle rc=%ld", lpar_rc);
 391                 ibmveth_remove_buffer_from_pool(adapter, adapter->rx_queue.queue_addr[adapter->rx_queue.index].correlator);
 392         }
 393
 394         if(++adapter->rx_queue.index == adapter->rx_queue.num_slots) {
 395                 adapter->rx_queue.index = 0;
 396                 adapter->rx_queue.toggle = !adapter->rx_queue.toggle;
 397         }
 398 }
 399
 400 static void ibmveth_rxq_harvest_buffer(struct ibmveth_adapter *adapter)
 401 {
 402         ibmveth_remove_buffer_from_pool(adapter, adapter->rx_queue.queue_addr[adapter->rx_queue.index].correlator);
 403
 404         if(++adapter->rx_queue.index == adapter->rx_queue.num_slots) {
 405                 adapter->rx_queue.index = 0;
 406                 adapter->rx_queue.toggle = !adapter->rx_queue.toggle;
 407         }
 408 }
 409
 410 static void ibmveth_cleanup(struct ibmveth_adapter *adapter)
 411 {
 412         int i;
 413
 414         if(adapter->buffer_list_addr != NULL) {
 415                 if(!dma_mapping_error(adapter->buffer_list_dma)) {
 416                         dma_unmap_single(&adapter->vdev->dev,
 417                                         adapter->buffer_list_dma, 4096,
 418                                         DMA_BIDIRECTIONAL);
 419                         adapter->buffer_list_dma = DMA_ERROR_CODE;
 420                 }
 421                 free_page((unsigned long)adapter->buffer_list_addr);
 422                 adapter->buffer_list_addr = NULL;
 423         }
 424
 425         if(adapter->filter_list_addr != NULL) {
 426                 if(!dma_mapping_error(adapter->filter_list_dma)) {
 427                         dma_unmap_single(&adapter->vdev->dev,
 428                                         adapter->filter_list_dma, 4096,
 429                                         DMA_BIDIRECTIONAL);
 430                         adapter->filter_list_dma = DMA_ERROR_CODE;
 431                 }
 432                 free_page((unsigned long)adapter->filter_list_addr);
 433                 adapter->filter_list_addr = NULL;
 434         }
 435
 436         if(adapter->rx_queue.queue_addr != NULL) {
 437                 if(!dma_mapping_error(adapter->rx_queue.queue_dma)) {
 438                         dma_unmap_single(&adapter->vdev->dev,
 439                                         adapter->rx_queue.queue_dma,
 440                                         adapter->rx_queue.queue_len,
 441                                         DMA_BIDIRECTIONAL);
 442                         adapter->rx_queue.queue_dma = DMA_ERROR_CODE;
 443                 }
 444                 kfree(adapter->rx_queue.queue_addr);
 445                 adapter->rx_queue.queue_addr = NULL;
 446         }
 447
 448         for(i = 0; i<IbmVethNumBufferPools; i++)
 449                 if (adapter->rx_buff_pool[i].active)
 450                         ibmveth_free_buffer_pool(adapter,
 451                                                  &adapter->rx_buff_pool[i]);
 452 }
 453
 454 static int ibmveth_register_logical_lan(struct ibmveth_adapter *adapter,
 455         union ibmveth_buf_desc rxq_desc, u64 mac_address)
 456 {
 457         int rc, try_again = 1;
 458
 459         /* After a kexec the adapter will still be open, so our attempt to
 460         * open it will fail. So if we get a failure we free the adapter and
 461         * try again, but only once. */
 462 retry:
 463         rc = h_register_logical_lan(adapter->vdev->unit_address,
 464                                     adapter->buffer_list_dma, rxq_desc.desc,
 465                                     adapter->filter_list_dma, mac_address);
 466
 467         if (rc != H_SUCCESS && try_again) {
 468                 do {
 469                         rc = h_free_logical_lan(adapter->vdev->unit_address);
 470                 } while (H_IS_LONG_BUSY(rc) || (rc == H_BUSY));
 471
 472                 try_again = 0;
 473                 goto retry;
 474         }
 475
 476         return rc;
 477 }
 478
 479 static int ibmveth_open(struct net_device *netdev)
 480 {
 481         struct ibmveth_adapter *adapter = netdev->priv;
 482         u64 mac_address = 0;
 483         int rxq_entries = 1;
 484         unsigned long lpar_rc;
 485         int rc;
 486         union ibmveth_buf_desc rxq_desc;
 487         int i;
 488
 489         ibmveth_debug_printk("open starting\n");
 490
 491         napi_enable(&adapter->napi);
 492
 493         for(i = 0; i<IbmVethNumBufferPools; i++)
 494                 rxq_entries += adapter->rx_buff_pool[i].size;
 495
 496         adapter->buffer_list_addr = (void*) get_zeroed_page(GFP_KERNEL);
 497         adapter->filter_list_addr = (void*) get_zeroed_page(GFP_KERNEL);
 498
 499         if(!adapter->buffer_list_addr || !adapter->filter_list_addr) {
 500                 ibmveth_error_printk("unable to allocate filter or buffer list pages\n");
 501                 ibmveth_cleanup(adapter);
 502                 napi_disable(&adapter->napi);
 503                 return -ENOMEM;
 504         }
 505
 506         adapter->rx_queue.queue_len = sizeof(struct ibmveth_rx_q_entry) * rxq_entries;
 507         adapter->rx_queue.queue_addr = kmalloc(adapter->rx_queue.queue_len, GFP_KERNEL);
 508
 509         if(!adapter->rx_queue.queue_addr) {
 510                 ibmveth_error_printk("unable to allocate rx queue pages\n");
 511                 ibmveth_cleanup(adapter);
 512                 napi_disable(&adapter->napi);
 513                 return -ENOMEM;
 514         }
 515
 516         adapter->buffer_list_dma = dma_map_single(&adapter->vdev->dev,
 517                         adapter->buffer_list_addr, 4096, DMA_BIDIRECTIONAL);
 518         adapter->filter_list_dma = dma_map_single(&adapter->vdev->dev,
 519                         adapter->filter_list_addr, 4096, DMA_BIDIRECTIONAL);
 520         adapter->rx_queue.queue_dma = dma_map_single(&adapter->vdev->dev,
 521                         adapter->rx_queue.queue_addr,
 522                         adapter->rx_queue.queue_len, DMA_BIDIRECTIONAL);
 523
 524         if((dma_mapping_error(adapter->buffer_list_dma) ) ||
 525            (dma_mapping_error(adapter->filter_list_dma)) ||
 526            (dma_mapping_error(adapter->rx_queue.queue_dma))) {
 527                 ibmveth_error_printk("unable to map filter or buffer list pages\n");
 528                 ibmveth_cleanup(adapter);
 529                 napi_disable(&adapter->napi);
 530                 return -ENOMEM;
 531         }
 532
 533         adapter->rx_queue.index = 0;
 534         adapter->rx_queue.num_slots = rxq_entries;
 535         adapter->rx_queue.toggle = 1;
 536
 537         memcpy(&mac_address, netdev->dev_addr, netdev->addr_len);
 538         mac_address = mac_address >> 16;
 539
 540         rxq_desc.desc = 0;
 541         rxq_desc.fields.valid = 1;
 542         rxq_desc.fields.length = adapter->rx_queue.queue_len;
 543         rxq_desc.fields.address = adapter->rx_queue.queue_dma;
 544
 545         ibmveth_debug_printk("buffer list @ 0x%p\n", adapter->buffer_list_addr);
 546         ibmveth_debug_printk("filter list @ 0x%p\n", adapter->filter_list_addr);
 547         ibmveth_debug_printk("receive q   @ 0x%p\n", adapter->rx_queue.queue_addr);
 548
 549         h_vio_signal(adapter->vdev->unit_address, VIO_IRQ_DISABLE);
 550
 551         lpar_rc = ibmveth_register_logical_lan(adapter, rxq_desc, mac_address);
 552
 553         if(lpar_rc != H_SUCCESS) {
 554                 ibmveth_error_printk("h_register_logical_lan failed with %ld\n", lpar_rc);
 555                 ibmveth_error_printk("buffer TCE:0x%lx filter TCE:0x%lx rxq desc:0x%lx MAC:0x%lx\n",
 556                                      adapter->buffer_list_dma,
 557                                      adapter->filter_list_dma,
 558                                      rxq_desc.desc,
 559                                      mac_address);
 560                 ibmveth_cleanup(adapter);
 561                 napi_disable(&adapter->napi);
 562                 return -ENONET;
 563         }
 564
 565         for(i = 0; i<IbmVethNumBufferPools; i++) {
 566                 if(!adapter->rx_buff_pool[i].active)
 567                         continue;
 568                 if (ibmveth_alloc_buffer_pool(&adapter->rx_buff_pool[i])) {
 569                         ibmveth_error_printk("unable to alloc pool\n");
 570                         adapter->rx_buff_pool[i].active = 0;
 571                         ibmveth_cleanup(adapter);
 572                         napi_disable(&adapter->napi);
 573                         return -ENOMEM ;
 574                 }
 575         }
 576
 577         ibmveth_debug_printk("registering irq 0x%x\n", netdev->irq);
 578         if((rc = request_irq(netdev->irq, &ibmveth_interrupt, 0, netdev->name, netdev)) != 0) {
 579                 ibmveth_error_printk("unable to request irq 0x%x, rc %d\n", netdev->irq, rc);
 580                 do {
 581                         rc = h_free_logical_lan(adapter->vdev->unit_address);
 582                 } while (H_IS_LONG_BUSY(rc) || (rc == H_BUSY));
 583
 584                 ibmveth_cleanup(adapter);
 585                 napi_disable(&adapter->napi);
 586                 return rc;
 587         }
 588
 589         ibmveth_debug_printk("initial replenish cycle\n");
 590         ibmveth_interrupt(netdev->irq, netdev);
 591
 592         netif_start_queue(netdev);
 593
 594         ibmveth_debug_printk("open complete\n");
 595
 596         return 0;
 597 }
 598
 599 static int ibmveth_close(struct net_device *netdev)
 600 {
 601         struct ibmveth_adapter *adapter = netdev->priv;
 602         long lpar_rc;
 603
 604         ibmveth_debug_printk("close starting\n");
 605
 606         napi_disable(&adapter->napi);
 607
 608         if (!adapter->pool_config)
 609                 netif_stop_queue(netdev);
 610
 611         free_irq(netdev->irq, netdev);
 612
 613         do {
 614                 lpar_rc = h_free_logical_lan(adapter->vdev->unit_address);
 615         } while (H_IS_LONG_BUSY(lpar_rc) || (lpar_rc == H_BUSY));
 616
 617         if(lpar_rc != H_SUCCESS)
 618         {
 619                 ibmveth_error_printk("h_free_logical_lan failed with %lx, continuing with close\n",
 620                                      lpar_rc);
 621         }
 622
 623         adapter->rx_no_buffer = *(u64*)(((char*)adapter->buffer_list_addr) + 4096 - 8);
 624
 625         ibmveth_cleanup(adapter);
 626
 627         ibmveth_debug_printk("close complete\n");
 628
 629         return 0;
 630 }
 631
 632 static int netdev_get_settings(struct net_device *dev, struct ethtool_cmd *cmd) {
 633         cmd->supported = (SUPPORTED_1000baseT_Full | SUPPORTED_Autoneg | SUPPORTED_FIBRE);
 634         cmd->advertising = (ADVERTISED_1000baseT_Full | ADVERTISED_Autoneg | ADVERTISED_FIBRE);
 635         cmd->speed = SPEED_1000;
 636         cmd->duplex = DUPLEX_FULL;
 637         cmd->port = PORT_FIBRE;
 638         cmd->phy_address = 0;
 639         cmd->transceiver = XCVR_INTERNAL;
 640         cmd->autoneg = AUTONEG_ENABLE;
 641         cmd->maxtxpkt = 0;
 642         cmd->maxrxpkt = 1;
 643         return 0;
 644 }
 645
 646 static void netdev_get_drvinfo (struct net_device *dev, struct ethtool_drvinfo *info) {
 647         strncpy(info->driver, ibmveth_driver_name, sizeof(info->driver) - 1);
 648         strncpy(info->version, ibmveth_driver_version, sizeof(info->version) - 1);
 649 }
 650
 651 static u32 netdev_get_link(struct net_device *dev) {
 652         return 1;
 653 }
 654
 655 static void ibmveth_set_rx_csum_flags(struct net_device *dev, u32 data)
 656 {
 657         struct ibmveth_adapter *adapter = dev->priv;
 658
 659         if (data)
 660                 adapter->rx_csum = 1;
 661         else {
 662                 /*
 663                  * Since the ibmveth firmware interface does not have the concept of
 664                  * separate tx/rx checksum offload enable, if rx checksum is disabled
 665                  * we also have to disable tx checksum offload. Once we disable rx
 666                  * checksum offload, we are no longer allowed to send tx buffers that
 667                  * are not properly checksummed.
 668                  */
 669                 adapter->rx_csum = 0;
 670                 dev->features &= ~NETIF_F_IP_CSUM;
 671         }
 672 }
 673
 674 static void ibmveth_set_tx_csum_flags(struct net_device *dev, u32 data)
 675 {
 676         struct ibmveth_adapter *adapter = dev->priv;
 677
 678         if (data) {
 679                 dev->features |= NETIF_F_IP_CSUM;
 680                 adapter->rx_csum = 1;
 681         } else
 682                 dev->features &= ~NETIF_F_IP_CSUM;
 683 }
 684
 685 static int ibmveth_set_csum_offload(struct net_device *dev, u32 data,
 686                                     void (*done) (struct net_device *, u32))
 687 {
 688         struct ibmveth_adapter *adapter = dev->priv;
 689         union ibmveth_illan_attributes set_attr, clr_attr, ret_attr;
 690         long ret;
 691         int rc1 = 0, rc2 = 0;
 692         int restart = 0;
 693
 694         if (netif_running(dev)) {
 695                 restart = 1;
 696                 adapter->pool_config = 1;
 697                 ibmveth_close(dev);
 698                 adapter->pool_config = 0;
 699         }
 700
 701         set_attr.desc = 0;
 702         clr_attr.desc = 0;
 703
 704         if (data)
 705                 set_attr.fields.tcp_csum_offload_ipv4 = 1;
 706         else
 707                 clr_attr.fields.tcp_csum_offload_ipv4 = 1;
 708
 709         ret = h_illan_attributes(adapter->vdev->unit_address, 0, 0, &ret_attr.desc);
 710
 711         if (ret == H_SUCCESS && !ret_attr.fields.active_trunk &&
 712             !ret_attr.fields.trunk_priority &&
 713             ret_attr.fields.csum_offload_padded_pkt_support) {
 714                 ret = h_illan_attributes(adapter->vdev->unit_address, clr_attr.desc,
 715                                          set_attr.desc, &ret_attr.desc);
 716
 717                 if (ret != H_SUCCESS) {
 718                         rc1 = -EIO;
 719                         ibmveth_error_printk("unable to change checksum offload settings."
 720                                              " %d rc=%ld\n", data, ret);
 721
 722                         ret = h_illan_attributes(adapter->vdev->unit_address,
 723                                                  set_attr.desc, clr_attr.desc, &ret_attr.desc);
 724                 } else
 725                         done(dev, data);
 726         } else {
 727                 rc1 = -EIO;
 728                 ibmveth_error_printk("unable to change checksum offload settings."
 729                                      " %d rc=%ld ret_attr=%lx\n", data, ret, ret_attr.desc);
 730         }
 731
 732         if (restart)
 733                 rc2 = ibmveth_open(dev);
 734
 735         return rc1 ? rc1 : rc2;
 736 }
 737
 738 static int ibmveth_set_rx_csum(struct net_device *dev, u32 data)
 739 {
 740         struct ibmveth_adapter *adapter = dev->priv;
 741
 742         if ((data && adapter->rx_csum) || (!data && !adapter->rx_csum))
 743                 return 0;
 744
 745         return ibmveth_set_csum_offload(dev, data, ibmveth_set_rx_csum_flags);
 746 }
 747
 748 static int ibmveth_set_tx_csum(struct net_device *dev, u32 data)
 749 {
 750         struct ibmveth_adapter *adapter = dev->priv;
 751         int rc = 0;
 752
 753         if (data && (dev->features & NETIF_F_IP_CSUM))
 754                 return 0;
 755         if (!data && !(dev->features & NETIF_F_IP_CSUM))
 756                 return 0;
 757
 758         if (data && !adapter->rx_csum)
 759                 rc = ibmveth_set_csum_offload(dev, data, ibmveth_set_tx_csum_flags);
 760         else
 761                 ibmveth_set_tx_csum_flags(dev, data);
 762
 763         return rc;
 764 }
 765
 766 static u32 ibmveth_get_rx_csum(struct net_device *dev)
 767 {
 768         struct ibmveth_adapter *adapter = dev->priv;
 769         return adapter->rx_csum;
 770 }
 771
 772 static const struct ethtool_ops netdev_ethtool_ops = {
 773         .get_drvinfo            = netdev_get_drvinfo,
 774         .get_settings           = netdev_get_settings,
 775         .get_link               = netdev_get_link,
 776         .get_sg                 = ethtool_op_get_sg,
 777         .get_tx_csum            = ethtool_op_get_tx_csum,
 778         .set_tx_csum            = ibmveth_set_tx_csum,
 779         .get_rx_csum            = ibmveth_get_rx_csum,
 780         .set_rx_csum            = ibmveth_set_rx_csum,
 781 };
 782
 783 static int ibmveth_ioctl(struct net_device *dev, struct ifreq *ifr, int cmd)
 784 {
 785         return -EOPNOTSUPP;
 786 }
 787
 788 #define page_offset(v) ((unsigned long)(v) & ((1 << 12) - 1))
 789
 790 static int ibmveth_start_xmit(struct sk_buff *skb, struct net_device *netdev)
 791 {
 792         struct ibmveth_adapter *adapter = netdev->priv;
 793         union ibmveth_buf_desc desc[IbmVethMaxSendFrags];
 794         unsigned long lpar_rc;
 795         int nfrags = 0, curfrag;
 796         unsigned long correlator;
 797         unsigned long flags;
 798         unsigned int retry_count;
 799         unsigned int tx_dropped = 0;
 800         unsigned int tx_bytes = 0;
 801         unsigned int tx_packets = 0;
 802         unsigned int tx_send_failed = 0;
 803         unsigned int tx_map_failed = 0;
 804
 805
 806         if ((skb_shinfo(skb)->nr_frags + 1) > IbmVethMaxSendFrags) {
 807                 tx_dropped++;
 808                 goto out;
 809         }
 810
 811         memset(&desc, 0, sizeof(desc));
 812
 813         /* nfrags = number of frags after the initial fragment */
 814         nfrags = skb_shinfo(skb)->nr_frags;
 815
 816         if(nfrags)
 817                 adapter->tx_multidesc_send++;
 818
 819         /* map the initial fragment */
 820         desc[0].fields.length  = nfrags ? skb->len - skb->data_len : skb->len;
 821         desc[0].fields.address = dma_map_single(&adapter->vdev->dev, skb->data,
 822                                         desc[0].fields.length, DMA_TO_DEVICE);
 823         desc[0].fields.valid   = 1;
 824
 825         if (skb->ip_summed == CHECKSUM_PARTIAL &&
 826             ip_hdr(skb)->protocol != IPPROTO_TCP && skb_checksum_help(skb)) {
 827                 ibmveth_error_printk("tx: failed to checksum packet\n");
 828                 tx_dropped++;
 829                 goto out;
 830         }
 831
 832         if (skb->ip_summed == CHECKSUM_PARTIAL) {
 833                 unsigned char *buf = skb_transport_header(skb) + skb->csum_offset;
 834
 835                 desc[0].fields.no_csum = 1;
 836                 desc[0].fields.csum_good = 1;
 837
 838                 /* Need to zero out the checksum */
 839                 buf[0] = 0;
 840                 buf[1] = 0;
 841         }
 842
 843         if(dma_mapping_error(desc[0].fields.address)) {
 844                 ibmveth_error_printk("tx: unable to map initial fragment\n");
 845                 tx_map_failed++;
 846                 tx_dropped++;
 847                 goto out;
 848         }
 849
 850         curfrag = nfrags;
 851
 852         /* map fragments past the initial portion if there are any */
 853         while(curfrag--) {
 854                 skb_frag_t *frag = &skb_shinfo(skb)->frags[curfrag];
 855                 desc[curfrag+1].fields.address
 856                         = dma_map_single(&adapter->vdev->dev,
 857                                 page_address(frag->page) + frag->page_offset,
 858                                 frag->size, DMA_TO_DEVICE);
 859                 desc[curfrag+1].fields.length = frag->size;
 860                 desc[curfrag+1].fields.valid  = 1;
 861                 if (skb->ip_summed == CHECKSUM_PARTIAL) {
 862                         desc[curfrag+1].fields.no_csum = 1;
 863                         desc[curfrag+1].fields.csum_good = 1;
 864                 }
 865
 866                 if(dma_mapping_error(desc[curfrag+1].fields.address)) {
 867                         ibmveth_error_printk("tx: unable to map fragment %d\n", curfrag);
 868                         tx_map_failed++;
 869                         tx_dropped++;
 870                         /* Free all the mappings we just created */
 871                         while(curfrag < nfrags) {
 872                                 dma_unmap_single(&adapter->vdev->dev,
 873                                                  desc[curfrag+1].fields.address,
 874                                                  desc[curfrag+1].fields.length,
 875                                                  DMA_TO_DEVICE);
 876                                 curfrag++;
 877                         }
 878                         goto out;
 879                 }
 880         }
 881
 882         /* send the frame. Arbitrarily set retrycount to 1024 */
 883         correlator = 0;
 884         retry_count = 1024;
 885         do {
 886                 lpar_rc = h_send_logical_lan(adapter->vdev->unit_address,
 887                                              desc[0].desc,
 888                                              desc[1].desc,
 889                                              desc[2].desc,
 890                                              desc[3].desc,
 891                                              desc[4].desc,
 892                                              desc[5].desc,
 893                                              correlator,
 894                                              &correlator);
 895         } while ((lpar_rc == H_BUSY) && (retry_count--));
 896
 897         if(lpar_rc != H_SUCCESS && lpar_rc != H_DROPPED) {
 898                 int i;
 899                 ibmveth_error_printk("tx: h_send_logical_lan failed with rc=%ld\n", lpar_rc);
 900                 for(i = 0; i < 6; i++) {
 901                         ibmveth_error_printk("tx: desc[%i] valid=%d, len=%d, address=0x%d\n", i,
 902                                              desc[i].fields.valid, desc[i].fields.length, desc[i].fields.address);
 903                 }
 904                 tx_send_failed++;
 905                 tx_dropped++;
 906         } else {
 907                 tx_packets++;
 908                 tx_bytes += skb->len;
 909                 netdev->trans_start = jiffies;
 910         }
 911
 912         do {
 913                 dma_unmap_single(&adapter->vdev->dev,
 914                                 desc[nfrags].fields.address,
 915                                 desc[nfrags].fields.length, DMA_TO_DEVICE);
 916         } while(--nfrags >= 0);
 917
 918 out:    spin_lock_irqsave(&adapter->stats_lock, flags);
 919         adapter->stats.tx_dropped += tx_dropped;
 920         adapter->stats.tx_bytes += tx_bytes;
 921         adapter->stats.tx_packets += tx_packets;
 922         adapter->tx_send_failed += tx_send_failed;
 923         adapter->tx_map_failed += tx_map_failed;
 924         spin_unlock_irqrestore(&adapter->stats_lock, flags);
 925
 926         dev_kfree_skb(skb);
 927         return 0;
 928 }
 929
 930 static int ibmveth_poll(struct napi_struct *napi, int budget)
 931 {
 932         struct ibmveth_adapter *adapter = container_of(napi, struct ibmveth_adapter, napi);
 933         struct net_device *netdev = adapter->netdev;
 934         int frames_processed = 0;
 935         unsigned long lpar_rc;
 936
 937  restart_poll:
 938         do {
 939                 struct sk_buff *skb;
 940
 941                 if (!ibmveth_rxq_pending_buffer(adapter))
 942                         break;
 943
 944                 rmb();
 945                 if (!ibmveth_rxq_buffer_valid(adapter)) {
 946                         wmb(); /* suggested by larson1 */
 947                         adapter->rx_invalid_buffer++;
 948                         ibmveth_debug_printk("recycling invalid buffer\n");
 949                         ibmveth_rxq_recycle_buffer(adapter);
 950                 } else {
 951                         int length = ibmveth_rxq_frame_length(adapter);
 952                         int offset = ibmveth_rxq_frame_offset(adapter);
 953                         int csum_good = ibmveth_rxq_csum_good(adapter);
 954
 955                         skb = ibmveth_rxq_get_buffer(adapter);
 956                         if (csum_good)
 957                                 skb->ip_summed = CHECKSUM_UNNECESSARY;
 958
 959                         ibmveth_rxq_harvest_buffer(adapter);
 960
 961                         skb_reserve(skb, offset);
 962                         skb_put(skb, length);
 963                         skb->protocol = eth_type_trans(skb, netdev);
 964
 965                         netif_receive_skb(skb); /* send it up */
 966
 967                         adapter->stats.rx_packets++;
 968                         adapter->stats.rx_bytes += length;
 969                         frames_processed++;
 970                         netdev->last_rx = jiffies;
 971                 }
 972         } while (frames_processed < budget);
 973
 974         ibmveth_replenish_task(adapter);
 975
 976         if (frames_processed < budget) {
 977                 /* We think we are done - reenable interrupts,
 978                  * then check once more to make sure we are done.
 979                  */
 980                 lpar_rc = h_vio_signal(adapter->vdev->unit_address,
 981                                        VIO_IRQ_ENABLE);
 982
 983                 ibmveth_assert(lpar_rc == H_SUCCESS);
 984
 985                 netif_rx_complete(netdev, napi);
 986
 987                 if (ibmveth_rxq_pending_buffer(adapter) &&
 988                     netif_rx_reschedule(netdev, napi)) {
 989                         lpar_rc = h_vio_signal(adapter->vdev->unit_address,
 990                                                VIO_IRQ_DISABLE);
 991                         goto restart_poll;
 992                 }
 993         }
 994
 995         return frames_processed;
 996 }
 997
 998 static irqreturn_t ibmveth_interrupt(int irq, void *dev_instance)
 999 {
1000         struct net_device *netdev = dev_instance;
1001         struct ibmveth_adapter *adapter = netdev->priv;
1002         unsigned long lpar_rc;
1003
1004         if (netif_rx_schedule_prep(netdev, &adapter->napi)) {
1005                 lpar_rc = h_vio_signal(adapter->vdev->unit_address,
1006                                        VIO_IRQ_DISABLE);
1007                 ibmveth_assert(lpar_rc == H_SUCCESS);
1008                 __netif_rx_schedule(netdev, &adapter->napi);
1009         }
1010         return IRQ_HANDLED;
1011 }
1012
1013 static struct net_device_stats *ibmveth_get_stats(struct net_device *dev)
1014 {
1015         struct ibmveth_adapter *adapter = dev->priv;
1016         return &adapter->stats;
1017 }
1018
1019 static void ibmveth_set_multicast_list(struct net_device *netdev)
1020 {
1021         struct ibmveth_adapter *adapter = netdev->priv;
1022         unsigned long lpar_rc;
1023
1024         if((netdev->flags & IFF_PROMISC) || (netdev->mc_count > adapter->mcastFilterSize)) {
1025                 lpar_rc = h_multicast_ctrl(adapter->vdev->unit_address,
1026                                            IbmVethMcastEnableRecv |
1027                                            IbmVethMcastDisableFiltering,
1028                                            0);
1029                 if(lpar_rc != H_SUCCESS) {
1030                         ibmveth_error_printk("h_multicast_ctrl rc=%ld when entering promisc mode\n", lpar_rc);
1031                 }
1032         } else {
1033                 struct dev_mc_list *mclist = netdev->mc_list;
1034                 int i;
1035                 /* clear the filter table & disable filtering */
1036                 lpar_rc = h_multicast_ctrl(adapter->vdev->unit_address,
1037                                            IbmVethMcastEnableRecv |
1038                                            IbmVethMcastDisableFiltering |
1039                                            IbmVethMcastClearFilterTable,
1040                                            0);
1041                 if(lpar_rc != H_SUCCESS) {
1042                         ibmveth_error_printk("h_multicast_ctrl rc=%ld when attempting to clear filter table\n", lpar_rc);
1043                 }
1044                 /* add the addresses to the filter table */
1045                 for(i = 0; i < netdev->mc_count; ++i, mclist = mclist->next) {
1046                         // add the multicast address to the filter table
1047                         unsigned long mcast_addr = 0;
1048                         memcpy(((char *)&mcast_addr)+2, mclist->dmi_addr, 6);
1049                         lpar_rc = h_multicast_ctrl(adapter->vdev->unit_address,
1050                                                    IbmVethMcastAddFilter,
1051                                                    mcast_addr);
1052                         if(lpar_rc != H_SUCCESS) {
1053                                 ibmveth_error_printk("h_multicast_ctrl rc=%ld when adding an entry to the filter table\n", lpar_rc);
1054                         }
1055                 }
1056
1057                 /* re-enable filtering */
1058                 lpar_rc = h_multicast_ctrl(adapter->vdev->unit_address,
1059                                            IbmVethMcastEnableFiltering,
1060                                            0);
1061                 if(lpar_rc != H_SUCCESS) {
1062                         ibmveth_error_printk("h_multicast_ctrl rc=%ld when enabling filtering\n", lpar_rc);
1063                 }
1064         }
1065 }
1066
1067 static int ibmveth_change_mtu(struct net_device *dev, int new_mtu)
1068 {
1069         struct ibmveth_adapter *adapter = dev->priv;
1070         int new_mtu_oh = new_mtu + IBMVETH_BUFF_OH;
1071         int reinit = 0;
1072         int i, rc;
1073
1074         if (new_mtu < IBMVETH_MAX_MTU)
1075                 return -EINVAL;
1076
1077         for (i = 0; i < IbmVethNumBufferPools; i++)
1078                 if (new_mtu_oh < adapter->rx_buff_pool[i].buff_size)
1079                         break;
1080
1081         if (i == IbmVethNumBufferPools)
1082                 return -EINVAL;
1083
1084         /* Look for an active buffer pool that can hold the new MTU */
1085         for(i = 0; i<IbmVethNumBufferPools; i++) {
1086                 if (!adapter->rx_buff_pool[i].active) {
1087                         adapter->rx_buff_pool[i].active = 1;
1088                         reinit = 1;
1089                 }
1090
1091                 if (new_mtu_oh < adapter->rx_buff_pool[i].buff_size) {
1092                         if (reinit && netif_running(adapter->netdev)) {
1093                                 adapter->pool_config = 1;
1094                                 ibmveth_close(adapter->netdev);
1095                                 adapter->pool_config = 0;
1096                                 dev->mtu = new_mtu;
1097                                 if ((rc = ibmveth_open(adapter->netdev)))
1098                                         return rc;
1099                         } else
1100                                 dev->mtu = new_mtu;
1101                         return 0;
1102                 }
1103         }
1104         return -EINVAL;
1105 }
1106
1107 #ifdef CONFIG_NET_POLL_CONTROLLER
1108 static void ibmveth_poll_controller(struct net_device *dev)
1109 {
1110         ibmveth_replenish_task(dev->priv);
1111         ibmveth_interrupt(dev->irq, dev);
1112 }
1113 #endif
1114
1115 static int __devinit ibmveth_probe(struct vio_dev *dev, const struct vio_device_id *id)
1116 {
1117         int rc, i;
1118         long ret;
1119         struct net_device *netdev;
1120         struct ibmveth_adapter *adapter;
1121         union ibmveth_illan_attributes set_attr, ret_attr;
1122
1123         unsigned char *mac_addr_p;
1124         unsigned int *mcastFilterSize_p;
1125
1126
1127         ibmveth_debug_printk_no_adapter("entering ibmveth_probe for UA 0x%x\n",
1128                                         dev->unit_address);
1129
1130         mac_addr_p = (unsigned char *) vio_get_attribute(dev,
1131                                                 VETH_MAC_ADDR, NULL);
1132         if(!mac_addr_p) {
1133                 printk(KERN_ERR "(%s:%3.3d) ERROR: Can't find VETH_MAC_ADDR "
1134                                 "attribute\n", __FILE__, __LINE__);
1135                 return 0;
1136         }
1137
1138         mcastFilterSize_p = (unsigned int *) vio_get_attribute(dev,
1139                                                 VETH_MCAST_FILTER_SIZE, NULL);
1140         if(!mcastFilterSize_p) {
1141                 printk(KERN_ERR "(%s:%3.3d) ERROR: Can't find "
1142                                 "VETH_MCAST_FILTER_SIZE attribute\n",
1143                                 __FILE__, __LINE__);
1144                 return 0;
1145         }
1146
1147         netdev = alloc_etherdev(sizeof(struct ibmveth_adapter));
1148
1149         if(!netdev)
1150                 return -ENOMEM;
1151
1152         SET_MODULE_OWNER(netdev);
1153
1154         adapter = netdev->priv;
1155         dev->dev.driver_data = netdev;
1156
1157         adapter->vdev = dev;
1158         adapter->netdev = netdev;
1159         adapter->mcastFilterSize= *mcastFilterSize_p;
1160         adapter->pool_config = 0;
1161
1162         netif_napi_add(netdev, &adapter->napi, ibmveth_poll, 16);
1163
1164         /*      Some older boxes running PHYP non-natively have an OF that
1165                 returns a 8-byte local-mac-address field (and the first
1166                 2 bytes have to be ignored) while newer boxes' OF return
1167                 a 6-byte field. Note that IEEE 1275 specifies that
1168                 local-mac-address must be a 6-byte field.
1169                 The RPA doc specifies that the first byte must be 10b, so
1170                 we'll just look for it to solve this 8 vs. 6 byte field issue */
1171
1172         if ((*mac_addr_p & 0x3) != 0x02)
1173                 mac_addr_p += 2;
1174
1175         adapter->mac_addr = 0;
1176         memcpy(&adapter->mac_addr, mac_addr_p, 6);
1177
1178         netdev->irq = dev->irq;
1179         netdev->open               = ibmveth_open;
1180         netdev->stop               = ibmveth_close;
1181         netdev->hard_start_xmit    = ibmveth_start_xmit;
1182         netdev->get_stats          = ibmveth_get_stats;
1183         netdev->set_multicast_list = ibmveth_set_multicast_list;
1184         netdev->do_ioctl           = ibmveth_ioctl;
1185         netdev->ethtool_ops           = &netdev_ethtool_ops;
1186         netdev->change_mtu         = ibmveth_change_mtu;
1187         SET_NETDEV_DEV(netdev, &dev->dev);
1188 #ifdef CONFIG_NET_POLL_CONTROLLER
1189         netdev->poll_controller = ibmveth_poll_controller;
1190 #endif
1191         netdev->features |= NETIF_F_LLTX;
1192         spin_lock_init(&adapter->stats_lock);
1193
1194         memcpy(&netdev->dev_addr, &adapter->mac_addr, netdev->addr_len);
1195
1196         for(i = 0; i<IbmVethNumBufferPools; i++) {
1197                 struct kobject *kobj = &adapter->rx_buff_pool[i].kobj;
1198                 ibmveth_init_buffer_pool(&adapter->rx_buff_pool[i], i,
1199                                          pool_count[i], pool_size[i],
1200                                          pool_active[i]);
1201                 kobj->parent = &dev->dev.kobj;
1202                 sprintf(kobj->name, "pool%d", i);
1203                 kobj->ktype = &ktype_veth_pool;
1204                 kobject_register(kobj);
1205         }
1206
1207         ibmveth_debug_printk("adapter @ 0x%p\n", adapter);
1208
1209         adapter->buffer_list_dma = DMA_ERROR_CODE;
1210         adapter->filter_list_dma = DMA_ERROR_CODE;
1211         adapter->rx_queue.queue_dma = DMA_ERROR_CODE;
1212
1213         ibmveth_debug_printk("registering netdev...\n");
1214
1215         ret = h_illan_attributes(dev->unit_address, 0, 0, &ret_attr.desc);
1216
1217         if (ret == H_SUCCESS && !ret_attr.fields.active_trunk &&
1218             !ret_attr.fields.trunk_priority &&
1219             ret_attr.fields.csum_offload_padded_pkt_support) {
1220                 set_attr.desc = 0;
1221                 set_attr.fields.tcp_csum_offload_ipv4 = 1;
1222
1223                 ret = h_illan_attributes(dev->unit_address, 0, set_attr.desc,
1224                                          &ret_attr.desc);
1225
1226                 if (ret == H_SUCCESS) {
1227                         adapter->rx_csum = 1;
1228                         netdev->features |= NETIF_F_IP_CSUM;
1229                 } else
1230                         ret = h_illan_attributes(dev->unit_address, set_attr.desc,
1231                                                  0, &ret_attr.desc);
1232         }
1233
1234         rc = register_netdev(netdev);
1235
1236         if(rc) {
1237                 ibmveth_debug_printk("failed to register netdev rc=%d\n", rc);
1238                 free_netdev(netdev);
1239                 return rc;
1240         }
1241
1242         ibmveth_debug_printk("registered\n");
1243
1244         ibmveth_proc_register_adapter(adapter);
1245
1246         return 0;
1247 }
1248
1249 static int __devexit ibmveth_remove(struct vio_dev *dev)
1250 {
1251         struct net_device *netdev = dev->dev.driver_data;
1252         struct ibmveth_adapter *adapter = netdev->priv;
1253         int i;
1254
1255         for(i = 0; i<IbmVethNumBufferPools; i++)
1256                 kobject_unregister(&adapter->rx_buff_pool[i].kobj);
1257
1258         unregister_netdev(netdev);
1259
1260         ibmveth_proc_unregister_adapter(adapter);
1261
1262         free_netdev(netdev);
1263         return 0;
1264 }
1265
1266 #ifdef CONFIG_PROC_FS
1267 static void ibmveth_proc_register_driver(void)
1268 {
1269         ibmveth_proc_dir = proc_mkdir(IBMVETH_PROC_DIR, init_net.proc_net);
1270         if (ibmveth_proc_dir) {
1271                 SET_MODULE_OWNER(ibmveth_proc_dir);
1272         }
1273 }
1274
1275 static void ibmveth_proc_unregister_driver(void)
1276 {
1277         remove_proc_entry(IBMVETH_PROC_DIR, init_net.proc_net);
1278 }
1279
1280 static void *ibmveth_seq_start(struct seq_file *seq, loff_t *pos)
1281 {
1282         if (*pos == 0) {
1283                 return (void *)1;
1284         } else {
1285                 return NULL;
1286         }
1287 }
1288
1289 static void *ibmveth_seq_next(struct seq_file *seq, void *v, loff_t *pos)
1290 {
1291         ++*pos;
1292         return NULL;
1293 }
1294
1295 static void ibmveth_seq_stop(struct seq_file *seq, void *v)
1296 {
1297 }
1298
1299 static int ibmveth_seq_show(struct seq_file *seq, void *v)
1300 {
1301         struct ibmveth_adapter *adapter = seq->private;
1302         char *current_mac = ((char*) &adapter->netdev->dev_addr);
1303         char *firmware_mac = ((char*) &adapter->mac_addr) ;
1304
1305         seq_printf(seq, "%s %s\n\n", ibmveth_driver_string, ibmveth_driver_version);
1306
1307         seq_printf(seq, "Unit Address:    0x%x\n", adapter->vdev->unit_address);
1308         seq_printf(seq, "Current MAC:     %02X:%02X:%02X:%02X:%02X:%02X\n",
1309                    current_mac[0], current_mac[1], current_mac[2],
1310                    current_mac[3], current_mac[4], current_mac[5]);
1311         seq_printf(seq, "Firmware MAC:    %02X:%02X:%02X:%02X:%02X:%02X\n",
1312                    firmware_mac[0], firmware_mac[1], firmware_mac[2],
1313                    firmware_mac[3], firmware_mac[4], firmware_mac[5]);
1314
1315         seq_printf(seq, "\nAdapter Statistics:\n");
1316         seq_printf(seq, "  TX:  skbuffs linearized:          %ld\n", adapter->tx_linearized);
1317         seq_printf(seq, "       multi-descriptor sends:      %ld\n", adapter->tx_multidesc_send);
1318         seq_printf(seq, "       skb_linearize failures:      %ld\n", adapter->tx_linearize_failed);
1319         seq_printf(seq, "       vio_map_single failres:      %ld\n", adapter->tx_map_failed);
1320         seq_printf(seq, "       send failures:               %ld\n", adapter->tx_send_failed);
1321         seq_printf(seq, "  RX:  replenish task cycles:       %ld\n", adapter->replenish_task_cycles);
1322         seq_printf(seq, "       alloc_skb_failures:          %ld\n", adapter->replenish_no_mem);
1323         seq_printf(seq, "       add buffer failures:         %ld\n", adapter->replenish_add_buff_failure);
1324         seq_printf(seq, "       invalid buffers:             %ld\n", adapter->rx_invalid_buffer);
1325         seq_printf(seq, "       no buffers:                  %ld\n", adapter->rx_no_buffer);
1326
1327         return 0;
1328 }
1329 static struct seq_operations ibmveth_seq_ops = {
1330         .start = ibmveth_seq_start,
1331         .next  = ibmveth_seq_next,
1332         .stop  = ibmveth_seq_stop,
1333         .show  = ibmveth_seq_show,
1334 };
1335
1336 static int ibmveth_proc_open(struct inode *inode, struct file *file)
1337 {
1338         struct seq_file *seq;
1339         struct proc_dir_entry *proc;
1340         int rc;
1341
1342         rc = seq_open(file, &ibmveth_seq_ops);
1343         if (!rc) {
1344                 /* recover the pointer buried in proc_dir_entry data */
1345                 seq = file->private_data;
1346                 proc = PDE(inode);
1347                 seq->private = proc->data;
1348         }
1349         return rc;
1350 }
1351
1352 static const struct file_operations ibmveth_proc_fops = {
1353         .owner   = THIS_MODULE,
1354         .open    = ibmveth_proc_open,
1355         .read    = seq_read,
1356         .llseek  = seq_lseek,
1357         .release = seq_release,
1358 };
1359
1360 static void ibmveth_proc_register_adapter(struct ibmveth_adapter *adapter)
1361 {
1362         struct proc_dir_entry *entry;
1363         if (ibmveth_proc_dir) {
1364                 char u_addr[10];
1365                 sprintf(u_addr, "%x", adapter->vdev->unit_address);
1366                 entry = create_proc_entry(u_addr, S_IFREG, ibmveth_proc_dir);
1367                 if (!entry) {
1368                         ibmveth_error_printk("Cannot create adapter proc entry");
1369                 } else {
1370                         entry->data = (void *) adapter;
1371                         entry->proc_fops = &ibmveth_proc_fops;
1372                         SET_MODULE_OWNER(entry);
1373                 }
1374         }
1375         return;
1376 }
1377
1378 static void ibmveth_proc_unregister_adapter(struct ibmveth_adapter *adapter)
1379 {
1380         if (ibmveth_proc_dir) {
1381                 char u_addr[10];
1382                 sprintf(u_addr, "%x", adapter->vdev->unit_address);
1383                 remove_proc_entry(u_addr, ibmveth_proc_dir);
1384         }
1385 }
1386
1387 #else /* CONFIG_PROC_FS */
1388 static void ibmveth_proc_register_adapter(struct ibmveth_adapter *adapter)
1389 {
1390 }
1391
1392 static void ibmveth_proc_unregister_adapter(struct ibmveth_adapter *adapter)
1393 {
1394 }
1395 static void ibmveth_proc_register_driver(void)
1396 {
1397 }
1398
1399 static void ibmveth_proc_unregister_driver(void)
1400 {
1401 }
1402 #endif /* CONFIG_PROC_FS */
1403
1404 static struct attribute veth_active_attr;
1405 static struct attribute veth_num_attr;
1406 static struct attribute veth_size_attr;
1407
1408 static ssize_t veth_pool_show(struct kobject * kobj,
1409                               struct attribute * attr, char * buf)
1410 {
1411         struct ibmveth_buff_pool *pool = container_of(kobj,
1412                                                       struct ibmveth_buff_pool,
1413                                                       kobj);
1414
1415         if (attr == &veth_active_attr)
1416                 return sprintf(buf, "%d\n", pool->active);
1417         else if (attr == &veth_num_attr)
1418                 return sprintf(buf, "%d\n", pool->size);
1419         else if (attr == &veth_size_attr)
1420                 return sprintf(buf, "%d\n", pool->buff_size);
1421         return 0;
1422 }
1423
1424 static ssize_t veth_pool_store(struct kobject * kobj, struct attribute * attr,
1425 const char * buf, size_t count)
1426 {
1427         struct ibmveth_buff_pool *pool = container_of(kobj,
1428                                                       struct ibmveth_buff_pool,
1429                                                       kobj);
1430         struct net_device *netdev =
1431             container_of(kobj->parent, struct device, kobj)->driver_data;
1432         struct ibmveth_adapter *adapter = netdev->priv;
1433         long value = simple_strtol(buf, NULL, 10);
1434         long rc;
1435
1436         if (attr == &veth_active_attr) {
1437                 if (value && !pool->active) {
1438                         if (netif_running(netdev)) {
1439                                 if(ibmveth_alloc_buffer_pool(pool)) {
1440                                         ibmveth_error_printk("unable to alloc pool\n");
1441                                         return -ENOMEM;
1442                                 }
1443                                 pool->active = 1;
1444                                 adapter->pool_config = 1;
1445                                 ibmveth_close(netdev);
1446                                 adapter->pool_config = 0;
1447                                 if ((rc = ibmveth_open(netdev)))
1448                                         return rc;
1449                         } else
1450                                 pool->active = 1;
1451                 } else if (!value && pool->active) {
1452                         int mtu = netdev->mtu + IBMVETH_BUFF_OH;
1453                         int i;
1454                         /* Make sure there is a buffer pool with buffers that
1455                            can hold a packet of the size of the MTU */
1456                         for (i = 0; i < IbmVethNumBufferPools; i++) {
1457                                 if (pool == &adapter->rx_buff_pool[i])
1458                                         continue;
1459                                 if (!adapter->rx_buff_pool[i].active)
1460                                         continue;
1461                                 if (mtu <= adapter->rx_buff_pool[i].buff_size)
1462                                         break;
1463                         }
1464
1465                         if (i == IbmVethNumBufferPools) {
1466                                 ibmveth_error_printk("no active pool >= MTU\n");
1467                                 return -EPERM;
1468                         }
1469
1470                         pool->active = 0;
1471                         if (netif_running(netdev)) {
1472                                 adapter->pool_config = 1;
1473                                 ibmveth_close(netdev);
1474                                 adapter->pool_config = 0;
1475                                 if ((rc = ibmveth_open(netdev)))
1476                                         return rc;
1477                         }
1478                 }
1479         } else if (attr == &veth_num_attr) {
1480                 if (value <= 0 || value > IBMVETH_MAX_POOL_COUNT)
1481                         return -EINVAL;
1482                 else {
1483                         if (netif_running(netdev)) {
1484                                 adapter->pool_config = 1;
1485                                 ibmveth_close(netdev);
1486                                 adapter->pool_config = 0;
1487                                 pool->size = value;
1488                                 if ((rc = ibmveth_open(netdev)))
1489                                         return rc;
1490                         } else
1491                                 pool->size = value;
1492                 }
1493         } else if (attr == &veth_size_attr) {
1494                 if (value <= IBMVETH_BUFF_OH || value > IBMVETH_MAX_BUF_SIZE)
1495                         return -EINVAL;
1496                 else {
1497                         if (netif_running(netdev)) {
1498                                 adapter->pool_config = 1;
1499                                 ibmveth_close(netdev);
1500                                 adapter->pool_config = 0;
1501                                 pool->buff_size = value;
1502                                 if ((rc = ibmveth_open(netdev)))
1503                                         return rc;
1504                         } else
1505                                 pool->buff_size = value;
1506                 }
1507         }
1508
1509         /* kick the interrupt handler to allocate/deallocate pools */
1510         ibmveth_interrupt(netdev->irq, netdev);
1511         return count;
1512 }
1513
1514
1515 #define ATTR(_name, _mode)      \
1516         struct attribute veth_##_name##_attr = {               \
1517         .name = __stringify(_name), .mode = _mode, \
1518         };
1519
1520 static ATTR(active, 0644);
1521 static ATTR(num, 0644);
1522 static ATTR(size, 0644);
1523
1524 static struct attribute * veth_pool_attrs[] = {
1525         &veth_active_attr,
1526         &veth_num_attr,
1527         &veth_size_attr,
1528         NULL,
1529 };
1530
1531 static struct sysfs_ops veth_pool_ops = {
1532         .show   = veth_pool_show,
1533         .store  = veth_pool_store,
1534 };
1535
1536 static struct kobj_type ktype_veth_pool = {
1537         .release        = NULL,
1538         .sysfs_ops      = &veth_pool_ops,
1539         .default_attrs  = veth_pool_attrs,
1540 };
1541
1542
1543 static struct vio_device_id ibmveth_device_table[] __devinitdata= {
1544         { "network", "IBM,l-lan"},
1545         { "", "" }
1546 };
1547 MODULE_DEVICE_TABLE(vio, ibmveth_device_table);
1548
1549 static struct vio_driver ibmveth_driver = {
1550         .id_table       = ibmveth_device_table,
1551         .probe          = ibmveth_probe,
1552         .remove         = ibmveth_remove,
1553         .driver         = {
1554                 .name   = ibmveth_driver_name,
1555                 .owner  = THIS_MODULE,
1556         }
1557 };
1558
1559 static int __init ibmveth_module_init(void)
1560 {
1561         ibmveth_printk("%s: %s %s\n", ibmveth_driver_name, ibmveth_driver_string, ibmveth_driver_version);
1562
1563         ibmveth_proc_register_driver();
1564
1565         return vio_register_driver(&ibmveth_driver);
1566 }
1567
1568 static void __exit ibmveth_module_exit(void)
1569 {
1570         vio_unregister_driver(&ibmveth_driver);
1571         ibmveth_proc_unregister_driver();
1572 }
1573
1574 module_init(ibmveth_module_init);
1575 module_exit(ibmveth_module_exit);