SAFE public projects git trees. - safe/jmp/linux-2.6/blob - 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   - add support for sysfs
  32   - possibly remove procfs support
  33 */
  34
  35 #include <linux/module.h>
  36 #include <linux/moduleparam.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/hvcall.h>
  54 #include <asm/atomic.h>
  55 #include <asm/vio.h>
  56 #include <asm/iommu.h>
  57 #include <asm/uaccess.h>
  58 #include <asm/firmware.h>
  59 #include <linux/seq_file.h>
  60
  61 #include "ibmveth.h"
  62
  63 #undef DEBUG
  64
  65 #define ibmveth_printk(fmt, args...) \
  66   printk(KERN_DEBUG "%s: " fmt, __FILE__, ## args)
  67
  68 #define ibmveth_error_printk(fmt, args...) \
  69   printk(KERN_ERR "(%s:%3.3d ua:%x) ERROR: " fmt, __FILE__, __LINE__ , adapter->vdev->unit_address, ## args)
  70
  71 #ifdef DEBUG
  72 #define ibmveth_debug_printk_no_adapter(fmt, args...) \
  73   printk(KERN_DEBUG "(%s:%3.3d): " fmt, __FILE__, __LINE__ , ## args)
  74 #define ibmveth_debug_printk(fmt, args...) \
  75   printk(KERN_DEBUG "(%s:%3.3d ua:%x): " fmt, __FILE__, __LINE__ , adapter->vdev->unit_address, ## args)
  76 #define ibmveth_assert(expr) \
  77   if(!(expr)) {                                   \
  78     printk(KERN_DEBUG "assertion failed (%s:%3.3d ua:%x): %s\n", __FILE__, __LINE__, adapter->vdev->unit_address, #expr); \
  79     BUG(); \
  80   }
  81 #else
  82 #define ibmveth_debug_printk_no_adapter(fmt, args...)
  83 #define ibmveth_debug_printk(fmt, args...)
  84 #define ibmveth_assert(expr)
  85 #endif
  86
  87 static int ibmveth_open(struct net_device *dev);
  88 static int ibmveth_close(struct net_device *dev);
  89 static int ibmveth_ioctl(struct net_device *dev, struct ifreq *ifr, int cmd);
  90 static int ibmveth_poll(struct napi_struct *napi, int budget);
  91 static int ibmveth_start_xmit(struct sk_buff *skb, 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 unsigned long ibmveth_get_desired_dma(struct vio_dev *vdev);
 101 static struct kobj_type ktype_veth_pool;
 102
 103
 104 #ifdef CONFIG_PROC_FS
 105 #define IBMVETH_PROC_DIR "ibmveth"
 106 static struct proc_dir_entry *ibmveth_proc_dir;
 107 #endif
 108
 109 static const char ibmveth_driver_name[] = "ibmveth";
 110 static const char ibmveth_driver_string[] = "IBM i/pSeries Virtual Ethernet Driver";
 111 #define ibmveth_driver_version "1.03"
 112
 113 MODULE_AUTHOR("Santiago Leon <santil@us.ibm.com>");
 114 MODULE_DESCRIPTION("IBM i/pSeries Virtual Ethernet Driver");
 115 MODULE_LICENSE("GPL");
 116 MODULE_VERSION(ibmveth_driver_version);
 117
 118 struct ibmveth_stat {
 119         char name[ETH_GSTRING_LEN];
 120         int offset;
 121 };
 122
 123 #define IBMVETH_STAT_OFF(stat) offsetof(struct ibmveth_adapter, stat)
 124 #define IBMVETH_GET_STAT(a, off) *((u64 *)(((unsigned long)(a)) + off))
 125
 126 struct ibmveth_stat ibmveth_stats[] = {
 127         { "replenish_task_cycles", IBMVETH_STAT_OFF(replenish_task_cycles) },
 128         { "replenish_no_mem", IBMVETH_STAT_OFF(replenish_no_mem) },
 129         { "replenish_add_buff_failure", IBMVETH_STAT_OFF(replenish_add_buff_failure) },
 130         { "replenish_add_buff_success", IBMVETH_STAT_OFF(replenish_add_buff_success) },
 131         { "rx_invalid_buffer", IBMVETH_STAT_OFF(rx_invalid_buffer) },
 132         { "rx_no_buffer", IBMVETH_STAT_OFF(rx_no_buffer) },
 133         { "tx_map_failed", IBMVETH_STAT_OFF(tx_map_failed) },
 134         { "tx_send_failed", IBMVETH_STAT_OFF(tx_send_failed) },
 135 };
 136
 137 /* simple methods of getting data from the current rxq entry */
 138 static inline u32 ibmveth_rxq_flags(struct ibmveth_adapter *adapter)
 139 {
 140         return adapter->rx_queue.queue_addr[adapter->rx_queue.index].flags_off;
 141 }
 142
 143 static inline int ibmveth_rxq_toggle(struct ibmveth_adapter *adapter)
 144 {
 145         return (ibmveth_rxq_flags(adapter) & IBMVETH_RXQ_TOGGLE) >> IBMVETH_RXQ_TOGGLE_SHIFT;
 146 }
 147
 148 static inline int ibmveth_rxq_pending_buffer(struct ibmveth_adapter *adapter)
 149 {
 150         return (ibmveth_rxq_toggle(adapter) == adapter->rx_queue.toggle);
 151 }
 152
 153 static inline int ibmveth_rxq_buffer_valid(struct ibmveth_adapter *adapter)
 154 {
 155         return (ibmveth_rxq_flags(adapter) & IBMVETH_RXQ_VALID);
 156 }
 157
 158 static inline int ibmveth_rxq_frame_offset(struct ibmveth_adapter *adapter)
 159 {
 160         return (ibmveth_rxq_flags(adapter) & IBMVETH_RXQ_OFF_MASK);
 161 }
 162
 163 static inline int ibmveth_rxq_frame_length(struct ibmveth_adapter *adapter)
 164 {
 165         return (adapter->rx_queue.queue_addr[adapter->rx_queue.index].length);
 166 }
 167
 168 static inline int ibmveth_rxq_csum_good(struct ibmveth_adapter *adapter)
 169 {
 170         return (ibmveth_rxq_flags(adapter) & IBMVETH_RXQ_CSUM_GOOD);
 171 }
 172
 173 /* setup the initial settings for a buffer pool */
 174 static void ibmveth_init_buffer_pool(struct ibmveth_buff_pool *pool, u32 pool_index, u32 pool_size, u32 buff_size, u32 pool_active)
 175 {
 176         pool->size = pool_size;
 177         pool->index = pool_index;
 178         pool->buff_size = buff_size;
 179         pool->threshold = pool_size / 2;
 180         pool->active = pool_active;
 181 }
 182
 183 /* allocate and setup an buffer pool - called during open */
 184 static int ibmveth_alloc_buffer_pool(struct ibmveth_buff_pool *pool)
 185 {
 186         int i;
 187
 188         pool->free_map = kmalloc(sizeof(u16) * pool->size, GFP_KERNEL);
 189
 190         if(!pool->free_map) {
 191                 return -1;
 192         }
 193
 194         pool->dma_addr = kmalloc(sizeof(dma_addr_t) * pool->size, GFP_KERNEL);
 195         if(!pool->dma_addr) {
 196                 kfree(pool->free_map);
 197                 pool->free_map = NULL;
 198                 return -1;
 199         }
 200
 201         pool->skbuff = kmalloc(sizeof(void*) * pool->size, GFP_KERNEL);
 202
 203         if(!pool->skbuff) {
 204                 kfree(pool->dma_addr);
 205                 pool->dma_addr = NULL;
 206
 207                 kfree(pool->free_map);
 208                 pool->free_map = NULL;
 209                 return -1;
 210         }
 211
 212         memset(pool->skbuff, 0, sizeof(void*) * pool->size);
 213         memset(pool->dma_addr, 0, sizeof(dma_addr_t) * pool->size);
 214
 215         for(i = 0; i < pool->size; ++i) {
 216                 pool->free_map[i] = i;
 217         }
 218
 219         atomic_set(&pool->available, 0);
 220         pool->producer_index = 0;
 221         pool->consumer_index = 0;
 222
 223         return 0;
 224 }
 225
 226 /* replenish the buffers for a pool.  note that we don't need to
 227  * skb_reserve these since they are used for incoming...
 228  */
 229 static void ibmveth_replenish_buffer_pool(struct ibmveth_adapter *adapter, struct ibmveth_buff_pool *pool)
 230 {
 231         u32 i;
 232         u32 count = pool->size - atomic_read(&pool->available);
 233         u32 buffers_added = 0;
 234         struct sk_buff *skb;
 235         unsigned int free_index, index;
 236         u64 correlator;
 237         unsigned long lpar_rc;
 238         dma_addr_t dma_addr;
 239
 240         mb();
 241
 242         for(i = 0; i < count; ++i) {
 243                 union ibmveth_buf_desc desc;
 244
 245                 skb = alloc_skb(pool->buff_size, GFP_ATOMIC);
 246
 247                 if(!skb) {
 248                         ibmveth_debug_printk("replenish: unable to allocate skb\n");
 249                         adapter->replenish_no_mem++;
 250                         break;
 251                 }
 252
 253                 free_index = pool->consumer_index;
 254                 pool->consumer_index = (pool->consumer_index + 1) % pool->size;
 255                 index = pool->free_map[free_index];
 256
 257                 ibmveth_assert(index != IBM_VETH_INVALID_MAP);
 258                 ibmveth_assert(pool->skbuff[index] == NULL);
 259
 260                 dma_addr = dma_map_single(&adapter->vdev->dev, skb->data,
 261                                 pool->buff_size, DMA_FROM_DEVICE);
 262
 263                 if (dma_mapping_error(&adapter->vdev->dev, dma_addr))
 264                         goto failure;
 265
 266                 pool->free_map[free_index] = IBM_VETH_INVALID_MAP;
 267                 pool->dma_addr[index] = dma_addr;
 268                 pool->skbuff[index] = skb;
 269
 270                 correlator = ((u64)pool->index << 32) | index;
 271                 *(u64*)skb->data = correlator;
 272
 273                 desc.fields.flags_len = IBMVETH_BUF_VALID | pool->buff_size;
 274                 desc.fields.address = dma_addr;
 275
 276                 lpar_rc = h_add_logical_lan_buffer(adapter->vdev->unit_address, desc.desc);
 277
 278                 if (lpar_rc != H_SUCCESS)
 279                         goto failure;
 280                 else {
 281                         buffers_added++;
 282                         adapter->replenish_add_buff_success++;
 283                 }
 284         }
 285
 286         mb();
 287         atomic_add(buffers_added, &(pool->available));
 288         return;
 289
 290 failure:
 291         pool->free_map[free_index] = index;
 292         pool->skbuff[index] = NULL;
 293         if (pool->consumer_index == 0)
 294                 pool->consumer_index = pool->size - 1;
 295         else
 296                 pool->consumer_index--;
 297         if (!dma_mapping_error(&adapter->vdev->dev, dma_addr))
 298                 dma_unmap_single(&adapter->vdev->dev,
 299                                  pool->dma_addr[index], pool->buff_size,
 300                                  DMA_FROM_DEVICE);
 301         dev_kfree_skb_any(skb);
 302         adapter->replenish_add_buff_failure++;
 303
 304         mb();
 305         atomic_add(buffers_added, &(pool->available));
 306 }
 307
 308 /* replenish routine */
 309 static void ibmveth_replenish_task(struct ibmveth_adapter *adapter)
 310 {
 311         int i;
 312
 313         adapter->replenish_task_cycles++;
 314
 315         for (i = (IbmVethNumBufferPools - 1); i >= 0; i--)
 316                 if(adapter->rx_buff_pool[i].active)
 317                         ibmveth_replenish_buffer_pool(adapter,
 318                                                      &adapter->rx_buff_pool[i]);
 319
 320         adapter->rx_no_buffer = *(u64*)(((char*)adapter->buffer_list_addr) + 4096 - 8);
 321 }
 322
 323 /* empty and free ana buffer pool - also used to do cleanup in error paths */
 324 static void ibmveth_free_buffer_pool(struct ibmveth_adapter *adapter, struct ibmveth_buff_pool *pool)
 325 {
 326         int i;
 327
 328         kfree(pool->free_map);
 329         pool->free_map = NULL;
 330
 331         if(pool->skbuff && pool->dma_addr) {
 332                 for(i = 0; i < pool->size; ++i) {
 333                         struct sk_buff *skb = pool->skbuff[i];
 334                         if(skb) {
 335                                 dma_unmap_single(&adapter->vdev->dev,
 336                                                  pool->dma_addr[i],
 337                                                  pool->buff_size,
 338                                                  DMA_FROM_DEVICE);
 339                                 dev_kfree_skb_any(skb);
 340                                 pool->skbuff[i] = NULL;
 341                         }
 342                 }
 343         }
 344
 345         if(pool->dma_addr) {
 346                 kfree(pool->dma_addr);
 347                 pool->dma_addr = NULL;
 348         }
 349
 350         if(pool->skbuff) {
 351                 kfree(pool->skbuff);
 352                 pool->skbuff = NULL;
 353         }
 354 }
 355
 356 /* remove a buffer from a pool */
 357 static void ibmveth_remove_buffer_from_pool(struct ibmveth_adapter *adapter, u64 correlator)
 358 {
 359         unsigned int pool  = correlator >> 32;
 360         unsigned int index = correlator & 0xffffffffUL;
 361         unsigned int free_index;
 362         struct sk_buff *skb;
 363
 364         ibmveth_assert(pool < IbmVethNumBufferPools);
 365         ibmveth_assert(index < adapter->rx_buff_pool[pool].size);
 366
 367         skb = adapter->rx_buff_pool[pool].skbuff[index];
 368
 369         ibmveth_assert(skb != NULL);
 370
 371         adapter->rx_buff_pool[pool].skbuff[index] = NULL;
 372
 373         dma_unmap_single(&adapter->vdev->dev,
 374                          adapter->rx_buff_pool[pool].dma_addr[index],
 375                          adapter->rx_buff_pool[pool].buff_size,
 376                          DMA_FROM_DEVICE);
 377
 378         free_index = adapter->rx_buff_pool[pool].producer_index;
 379         adapter->rx_buff_pool[pool].producer_index
 380                 = (adapter->rx_buff_pool[pool].producer_index + 1)
 381                 % adapter->rx_buff_pool[pool].size;
 382         adapter->rx_buff_pool[pool].free_map[free_index] = index;
 383
 384         mb();
 385
 386         atomic_dec(&(adapter->rx_buff_pool[pool].available));
 387 }
 388
 389 /* get the current buffer on the rx queue */
 390 static inline struct sk_buff *ibmveth_rxq_get_buffer(struct ibmveth_adapter *adapter)
 391 {
 392         u64 correlator = adapter->rx_queue.queue_addr[adapter->rx_queue.index].correlator;
 393         unsigned int pool = correlator >> 32;
 394         unsigned int index = correlator & 0xffffffffUL;
 395
 396         ibmveth_assert(pool < IbmVethNumBufferPools);
 397         ibmveth_assert(index < adapter->rx_buff_pool[pool].size);
 398
 399         return adapter->rx_buff_pool[pool].skbuff[index];
 400 }
 401
 402 /* recycle the current buffer on the rx queue */
 403 static void ibmveth_rxq_recycle_buffer(struct ibmveth_adapter *adapter)
 404 {
 405         u32 q_index = adapter->rx_queue.index;
 406         u64 correlator = adapter->rx_queue.queue_addr[q_index].correlator;
 407         unsigned int pool = correlator >> 32;
 408         unsigned int index = correlator & 0xffffffffUL;
 409         union ibmveth_buf_desc desc;
 410         unsigned long lpar_rc;
 411
 412         ibmveth_assert(pool < IbmVethNumBufferPools);
 413         ibmveth_assert(index < adapter->rx_buff_pool[pool].size);
 414
 415         if(!adapter->rx_buff_pool[pool].active) {
 416                 ibmveth_rxq_harvest_buffer(adapter);
 417                 ibmveth_free_buffer_pool(adapter, &adapter->rx_buff_pool[pool]);
 418                 return;
 419         }
 420
 421         desc.fields.flags_len = IBMVETH_BUF_VALID |
 422                 adapter->rx_buff_pool[pool].buff_size;
 423         desc.fields.address = adapter->rx_buff_pool[pool].dma_addr[index];
 424
 425         lpar_rc = h_add_logical_lan_buffer(adapter->vdev->unit_address, desc.desc);
 426
 427         if(lpar_rc != H_SUCCESS) {
 428                 ibmveth_debug_printk("h_add_logical_lan_buffer failed during recycle rc=%ld", lpar_rc);
 429                 ibmveth_remove_buffer_from_pool(adapter, adapter->rx_queue.queue_addr[adapter->rx_queue.index].correlator);
 430         }
 431
 432         if(++adapter->rx_queue.index == adapter->rx_queue.num_slots) {
 433                 adapter->rx_queue.index = 0;
 434                 adapter->rx_queue.toggle = !adapter->rx_queue.toggle;
 435         }
 436 }
 437
 438 static void ibmveth_rxq_harvest_buffer(struct ibmveth_adapter *adapter)
 439 {
 440         ibmveth_remove_buffer_from_pool(adapter, adapter->rx_queue.queue_addr[adapter->rx_queue.index].correlator);
 441
 442         if(++adapter->rx_queue.index == adapter->rx_queue.num_slots) {
 443                 adapter->rx_queue.index = 0;
 444                 adapter->rx_queue.toggle = !adapter->rx_queue.toggle;
 445         }
 446 }
 447
 448 static void ibmveth_cleanup(struct ibmveth_adapter *adapter)
 449 {
 450         int i;
 451         struct device *dev = &adapter->vdev->dev;
 452
 453         if(adapter->buffer_list_addr != NULL) {
 454                 if (!dma_mapping_error(dev, adapter->buffer_list_dma)) {
 455                         dma_unmap_single(dev, adapter->buffer_list_dma, 4096,
 456                                         DMA_BIDIRECTIONAL);
 457                         adapter->buffer_list_dma = DMA_ERROR_CODE;
 458                 }
 459                 free_page((unsigned long)adapter->buffer_list_addr);
 460                 adapter->buffer_list_addr = NULL;
 461         }
 462
 463         if(adapter->filter_list_addr != NULL) {
 464                 if (!dma_mapping_error(dev, adapter->filter_list_dma)) {
 465                         dma_unmap_single(dev, adapter->filter_list_dma, 4096,
 466                                         DMA_BIDIRECTIONAL);
 467                         adapter->filter_list_dma = DMA_ERROR_CODE;
 468                 }
 469                 free_page((unsigned long)adapter->filter_list_addr);
 470                 adapter->filter_list_addr = NULL;
 471         }
 472
 473         if(adapter->rx_queue.queue_addr != NULL) {
 474                 if (!dma_mapping_error(dev, adapter->rx_queue.queue_dma)) {
 475                         dma_unmap_single(dev,
 476                                         adapter->rx_queue.queue_dma,
 477                                         adapter->rx_queue.queue_len,
 478                                         DMA_BIDIRECTIONAL);
 479                         adapter->rx_queue.queue_dma = DMA_ERROR_CODE;
 480                 }
 481                 kfree(adapter->rx_queue.queue_addr);
 482                 adapter->rx_queue.queue_addr = NULL;
 483         }
 484
 485         for(i = 0; i<IbmVethNumBufferPools; i++)
 486                 if (adapter->rx_buff_pool[i].active)
 487                         ibmveth_free_buffer_pool(adapter,
 488                                                  &adapter->rx_buff_pool[i]);
 489
 490         if (adapter->bounce_buffer != NULL) {
 491                 if (!dma_mapping_error(dev, adapter->bounce_buffer_dma)) {
 492                         dma_unmap_single(&adapter->vdev->dev,
 493                                         adapter->bounce_buffer_dma,
 494                                         adapter->netdev->mtu + IBMVETH_BUFF_OH,
 495                                         DMA_BIDIRECTIONAL);
 496                         adapter->bounce_buffer_dma = DMA_ERROR_CODE;
 497                 }
 498                 kfree(adapter->bounce_buffer);
 499                 adapter->bounce_buffer = NULL;
 500         }
 501 }
 502
 503 static int ibmveth_register_logical_lan(struct ibmveth_adapter *adapter,
 504         union ibmveth_buf_desc rxq_desc, u64 mac_address)
 505 {
 506         int rc, try_again = 1;
 507
 508         /* After a kexec the adapter will still be open, so our attempt to
 509         * open it will fail. So if we get a failure we free the adapter and
 510         * try again, but only once. */
 511 retry:
 512         rc = h_register_logical_lan(adapter->vdev->unit_address,
 513                                     adapter->buffer_list_dma, rxq_desc.desc,
 514                                     adapter->filter_list_dma, mac_address);
 515
 516         if (rc != H_SUCCESS && try_again) {
 517                 do {
 518                         rc = h_free_logical_lan(adapter->vdev->unit_address);
 519                 } while (H_IS_LONG_BUSY(rc) || (rc == H_BUSY));
 520
 521                 try_again = 0;
 522                 goto retry;
 523         }
 524
 525         return rc;
 526 }
 527
 528 static int ibmveth_open(struct net_device *netdev)
 529 {
 530         struct ibmveth_adapter *adapter = netdev_priv(netdev);
 531         u64 mac_address = 0;
 532         int rxq_entries = 1;
 533         unsigned long lpar_rc;
 534         int rc;
 535         union ibmveth_buf_desc rxq_desc;
 536         int i;
 537         struct device *dev;
 538
 539         ibmveth_debug_printk("open starting\n");
 540
 541         napi_enable(&adapter->napi);
 542
 543         for(i = 0; i<IbmVethNumBufferPools; i++)
 544                 rxq_entries += adapter->rx_buff_pool[i].size;
 545
 546         adapter->buffer_list_addr = (void*) get_zeroed_page(GFP_KERNEL);
 547         adapter->filter_list_addr = (void*) get_zeroed_page(GFP_KERNEL);
 548
 549         if(!adapter->buffer_list_addr || !adapter->filter_list_addr) {
 550                 ibmveth_error_printk("unable to allocate filter or buffer list pages\n");
 551                 ibmveth_cleanup(adapter);
 552                 napi_disable(&adapter->napi);
 553                 return -ENOMEM;
 554         }
 555
 556         adapter->rx_queue.queue_len = sizeof(struct ibmveth_rx_q_entry) * rxq_entries;
 557         adapter->rx_queue.queue_addr = kmalloc(adapter->rx_queue.queue_len, GFP_KERNEL);
 558
 559         if(!adapter->rx_queue.queue_addr) {
 560                 ibmveth_error_printk("unable to allocate rx queue pages\n");
 561                 ibmveth_cleanup(adapter);
 562                 napi_disable(&adapter->napi);
 563                 return -ENOMEM;
 564         }
 565
 566         dev = &adapter->vdev->dev;
 567
 568         adapter->buffer_list_dma = dma_map_single(dev,
 569                         adapter->buffer_list_addr, 4096, DMA_BIDIRECTIONAL);
 570         adapter->filter_list_dma = dma_map_single(dev,
 571                         adapter->filter_list_addr, 4096, DMA_BIDIRECTIONAL);
 572         adapter->rx_queue.queue_dma = dma_map_single(dev,
 573                         adapter->rx_queue.queue_addr,
 574                         adapter->rx_queue.queue_len, DMA_BIDIRECTIONAL);
 575
 576         if ((dma_mapping_error(dev, adapter->buffer_list_dma)) ||
 577             (dma_mapping_error(dev, adapter->filter_list_dma)) ||
 578             (dma_mapping_error(dev, adapter->rx_queue.queue_dma))) {
 579                 ibmveth_error_printk("unable to map filter or buffer list pages\n");
 580                 ibmveth_cleanup(adapter);
 581                 napi_disable(&adapter->napi);
 582                 return -ENOMEM;
 583         }
 584
 585         adapter->rx_queue.index = 0;
 586         adapter->rx_queue.num_slots = rxq_entries;
 587         adapter->rx_queue.toggle = 1;
 588
 589         memcpy(&mac_address, netdev->dev_addr, netdev->addr_len);
 590         mac_address = mac_address >> 16;
 591
 592         rxq_desc.fields.flags_len = IBMVETH_BUF_VALID | adapter->rx_queue.queue_len;
 593         rxq_desc.fields.address = adapter->rx_queue.queue_dma;
 594
 595         ibmveth_debug_printk("buffer list @ 0x%p\n", adapter->buffer_list_addr);
 596         ibmveth_debug_printk("filter list @ 0x%p\n", adapter->filter_list_addr);
 597         ibmveth_debug_printk("receive q   @ 0x%p\n", adapter->rx_queue.queue_addr);
 598
 599         h_vio_signal(adapter->vdev->unit_address, VIO_IRQ_DISABLE);
 600
 601         lpar_rc = ibmveth_register_logical_lan(adapter, rxq_desc, mac_address);
 602
 603         if(lpar_rc != H_SUCCESS) {
 604                 ibmveth_error_printk("h_register_logical_lan failed with %ld\n", lpar_rc);
 605                 ibmveth_error_printk("buffer TCE:0x%llx filter TCE:0x%llx rxq desc:0x%llx MAC:0x%llx\n",
 606                                      adapter->buffer_list_dma,
 607                                      adapter->filter_list_dma,
 608                                      rxq_desc.desc,
 609                                      mac_address);
 610                 ibmveth_cleanup(adapter);
 611                 napi_disable(&adapter->napi);
 612                 return -ENONET;
 613         }
 614
 615         for(i = 0; i<IbmVethNumBufferPools; i++) {
 616                 if(!adapter->rx_buff_pool[i].active)
 617                         continue;
 618                 if (ibmveth_alloc_buffer_pool(&adapter->rx_buff_pool[i])) {
 619                         ibmveth_error_printk("unable to alloc pool\n");
 620                         adapter->rx_buff_pool[i].active = 0;
 621                         ibmveth_cleanup(adapter);
 622                         napi_disable(&adapter->napi);
 623                         return -ENOMEM ;
 624                 }
 625         }
 626
 627         ibmveth_debug_printk("registering irq 0x%x\n", netdev->irq);
 628         if((rc = request_irq(netdev->irq, ibmveth_interrupt, 0, netdev->name, netdev)) != 0) {
 629                 ibmveth_error_printk("unable to request irq 0x%x, rc %d\n", netdev->irq, rc);
 630                 do {
 631                         rc = h_free_logical_lan(adapter->vdev->unit_address);
 632                 } while (H_IS_LONG_BUSY(rc) || (rc == H_BUSY));
 633
 634                 ibmveth_cleanup(adapter);
 635                 napi_disable(&adapter->napi);
 636                 return rc;
 637         }
 638
 639         adapter->bounce_buffer =
 640             kmalloc(netdev->mtu + IBMVETH_BUFF_OH, GFP_KERNEL);
 641         if (!adapter->bounce_buffer) {
 642                 ibmveth_error_printk("unable to allocate bounce buffer\n");
 643                 ibmveth_cleanup(adapter);
 644                 napi_disable(&adapter->napi);
 645                 return -ENOMEM;
 646         }
 647         adapter->bounce_buffer_dma =
 648             dma_map_single(&adapter->vdev->dev, adapter->bounce_buffer,
 649                            netdev->mtu + IBMVETH_BUFF_OH, DMA_BIDIRECTIONAL);
 650         if (dma_mapping_error(dev, adapter->bounce_buffer_dma)) {
 651                 ibmveth_error_printk("unable to map bounce buffer\n");
 652                 ibmveth_cleanup(adapter);
 653                 napi_disable(&adapter->napi);
 654                 return -ENOMEM;
 655         }
 656
 657         ibmveth_debug_printk("initial replenish cycle\n");
 658         ibmveth_interrupt(netdev->irq, netdev);
 659
 660         netif_start_queue(netdev);
 661
 662         ibmveth_debug_printk("open complete\n");
 663
 664         return 0;
 665 }
 666
 667 static int ibmveth_close(struct net_device *netdev)
 668 {
 669         struct ibmveth_adapter *adapter = netdev_priv(netdev);
 670         long lpar_rc;
 671
 672         ibmveth_debug_printk("close starting\n");
 673
 674         napi_disable(&adapter->napi);
 675
 676         if (!adapter->pool_config)
 677                 netif_stop_queue(netdev);
 678
 679         free_irq(netdev->irq, netdev);
 680
 681         do {
 682                 lpar_rc = h_free_logical_lan(adapter->vdev->unit_address);
 683         } while (H_IS_LONG_BUSY(lpar_rc) || (lpar_rc == H_BUSY));
 684
 685         if(lpar_rc != H_SUCCESS)
 686         {
 687                 ibmveth_error_printk("h_free_logical_lan failed with %lx, continuing with close\n",
 688                                      lpar_rc);
 689         }
 690
 691         adapter->rx_no_buffer = *(u64*)(((char*)adapter->buffer_list_addr) + 4096 - 8);
 692
 693         ibmveth_cleanup(adapter);
 694
 695         ibmveth_debug_printk("close complete\n");
 696
 697         return 0;
 698 }
 699
 700 static int netdev_get_settings(struct net_device *dev, struct ethtool_cmd *cmd) {
 701         cmd->supported = (SUPPORTED_1000baseT_Full | SUPPORTED_Autoneg | SUPPORTED_FIBRE);
 702         cmd->advertising = (ADVERTISED_1000baseT_Full | ADVERTISED_Autoneg | ADVERTISED_FIBRE);
 703         cmd->speed = SPEED_1000;
 704         cmd->duplex = DUPLEX_FULL;
 705         cmd->port = PORT_FIBRE;
 706         cmd->phy_address = 0;
 707         cmd->transceiver = XCVR_INTERNAL;
 708         cmd->autoneg = AUTONEG_ENABLE;
 709         cmd->maxtxpkt = 0;
 710         cmd->maxrxpkt = 1;
 711         return 0;
 712 }
 713
 714 static void netdev_get_drvinfo (struct net_device *dev, struct ethtool_drvinfo *info) {
 715         strncpy(info->driver, ibmveth_driver_name, sizeof(info->driver) - 1);
 716         strncpy(info->version, ibmveth_driver_version, sizeof(info->version) - 1);
 717 }
 718
 719 static u32 netdev_get_link(struct net_device *dev) {
 720         return 1;
 721 }
 722
 723 static void ibmveth_set_rx_csum_flags(struct net_device *dev, u32 data)
 724 {
 725         struct ibmveth_adapter *adapter = netdev_priv(dev);
 726
 727         if (data)
 728                 adapter->rx_csum = 1;
 729         else {
 730                 /*
 731                  * Since the ibmveth firmware interface does not have the concept of
 732                  * separate tx/rx checksum offload enable, if rx checksum is disabled
 733                  * we also have to disable tx checksum offload. Once we disable rx
 734                  * checksum offload, we are no longer allowed to send tx buffers that
 735                  * are not properly checksummed.
 736                  */
 737                 adapter->rx_csum = 0;
 738                 dev->features &= ~NETIF_F_IP_CSUM;
 739         }
 740 }
 741
 742 static void ibmveth_set_tx_csum_flags(struct net_device *dev, u32 data)
 743 {
 744         struct ibmveth_adapter *adapter = netdev_priv(dev);
 745
 746         if (data) {
 747                 dev->features |= NETIF_F_IP_CSUM;
 748                 adapter->rx_csum = 1;
 749         } else
 750                 dev->features &= ~NETIF_F_IP_CSUM;
 751 }
 752
 753 static int ibmveth_set_csum_offload(struct net_device *dev, u32 data,
 754                                     void (*done) (struct net_device *, u32))
 755 {
 756         struct ibmveth_adapter *adapter = netdev_priv(dev);
 757         unsigned long set_attr, clr_attr, ret_attr;
 758         long ret;
 759         int rc1 = 0, rc2 = 0;
 760         int restart = 0;
 761
 762         if (netif_running(dev)) {
 763                 restart = 1;
 764                 adapter->pool_config = 1;
 765                 ibmveth_close(dev);
 766                 adapter->pool_config = 0;
 767         }
 768
 769         set_attr = 0;
 770         clr_attr = 0;
 771
 772         if (data)
 773                 set_attr = IBMVETH_ILLAN_IPV4_TCP_CSUM;
 774         else
 775                 clr_attr = IBMVETH_ILLAN_IPV4_TCP_CSUM;
 776
 777         ret = h_illan_attributes(adapter->vdev->unit_address, 0, 0, &ret_attr);
 778
 779         if (ret == H_SUCCESS && !(ret_attr & IBMVETH_ILLAN_ACTIVE_TRUNK) &&
 780             !(ret_attr & IBMVETH_ILLAN_TRUNK_PRI_MASK) &&
 781             (ret_attr & IBMVETH_ILLAN_PADDED_PKT_CSUM)) {
 782                 ret = h_illan_attributes(adapter->vdev->unit_address, clr_attr,
 783                                          set_attr, &ret_attr);
 784
 785                 if (ret != H_SUCCESS) {
 786                         rc1 = -EIO;
 787                         ibmveth_error_printk("unable to change checksum offload settings."
 788                                              " %d rc=%ld\n", data, ret);
 789
 790                         ret = h_illan_attributes(adapter->vdev->unit_address,
 791                                                  set_attr, clr_attr, &ret_attr);
 792                 } else
 793                         done(dev, data);
 794         } else {
 795                 rc1 = -EIO;
 796                 ibmveth_error_printk("unable to change checksum offload settings."
 797                                      " %d rc=%ld ret_attr=%lx\n", data, ret, ret_attr);
 798         }
 799
 800         if (restart)
 801                 rc2 = ibmveth_open(dev);
 802
 803         return rc1 ? rc1 : rc2;
 804 }
 805
 806 static int ibmveth_set_rx_csum(struct net_device *dev, u32 data)
 807 {
 808         struct ibmveth_adapter *adapter = netdev_priv(dev);
 809
 810         if ((data && adapter->rx_csum) || (!data && !adapter->rx_csum))
 811                 return 0;
 812
 813         return ibmveth_set_csum_offload(dev, data, ibmveth_set_rx_csum_flags);
 814 }
 815
 816 static int ibmveth_set_tx_csum(struct net_device *dev, u32 data)
 817 {
 818         struct ibmveth_adapter *adapter = netdev_priv(dev);
 819         int rc = 0;
 820
 821         if (data && (dev->features & NETIF_F_IP_CSUM))
 822                 return 0;
 823         if (!data && !(dev->features & NETIF_F_IP_CSUM))
 824                 return 0;
 825
 826         if (data && !adapter->rx_csum)
 827                 rc = ibmveth_set_csum_offload(dev, data, ibmveth_set_tx_csum_flags);
 828         else
 829                 ibmveth_set_tx_csum_flags(dev, data);
 830
 831         return rc;
 832 }
 833
 834 static u32 ibmveth_get_rx_csum(struct net_device *dev)
 835 {
 836         struct ibmveth_adapter *adapter = netdev_priv(dev);
 837         return adapter->rx_csum;
 838 }
 839
 840 static void ibmveth_get_strings(struct net_device *dev, u32 stringset, u8 *data)
 841 {
 842         int i;
 843
 844         if (stringset != ETH_SS_STATS)
 845                 return;
 846
 847         for (i = 0; i < ARRAY_SIZE(ibmveth_stats); i++, data += ETH_GSTRING_LEN)
 848                 memcpy(data, ibmveth_stats[i].name, ETH_GSTRING_LEN);
 849 }
 850
 851 static int ibmveth_get_sset_count(struct net_device *dev, int sset)
 852 {
 853         switch (sset) {
 854         case ETH_SS_STATS:
 855                 return ARRAY_SIZE(ibmveth_stats);
 856         default:
 857                 return -EOPNOTSUPP;
 858         }
 859 }
 860
 861 static void ibmveth_get_ethtool_stats(struct net_device *dev,
 862                                       struct ethtool_stats *stats, u64 *data)
 863 {
 864         int i;
 865         struct ibmveth_adapter *adapter = netdev_priv(dev);
 866
 867         for (i = 0; i < ARRAY_SIZE(ibmveth_stats); i++)
 868                 data[i] = IBMVETH_GET_STAT(adapter, ibmveth_stats[i].offset);
 869 }
 870
 871 static const struct ethtool_ops netdev_ethtool_ops = {
 872         .get_drvinfo            = netdev_get_drvinfo,
 873         .get_settings           = netdev_get_settings,
 874         .get_link               = netdev_get_link,
 875         .set_tx_csum            = ibmveth_set_tx_csum,
 876         .get_rx_csum            = ibmveth_get_rx_csum,
 877         .set_rx_csum            = ibmveth_set_rx_csum,
 878         .get_strings            = ibmveth_get_strings,
 879         .get_sset_count         = ibmveth_get_sset_count,
 880         .get_ethtool_stats      = ibmveth_get_ethtool_stats,
 881 };
 882
 883 static int ibmveth_ioctl(struct net_device *dev, struct ifreq *ifr, int cmd)
 884 {
 885         return -EOPNOTSUPP;
 886 }
 887
 888 #define page_offset(v) ((unsigned long)(v) & ((1 << 12) - 1))
 889
 890 static netdev_tx_t ibmveth_start_xmit(struct sk_buff *skb,
 891                                       struct net_device *netdev)
 892 {
 893         struct ibmveth_adapter *adapter = netdev_priv(netdev);
 894         union ibmveth_buf_desc desc;
 895         unsigned long lpar_rc;
 896         unsigned long correlator;
 897         unsigned long flags;
 898         unsigned int retry_count;
 899         unsigned int tx_dropped = 0;
 900         unsigned int tx_bytes = 0;
 901         unsigned int tx_packets = 0;
 902         unsigned int tx_send_failed = 0;
 903         unsigned int tx_map_failed = 0;
 904         int used_bounce = 0;
 905         unsigned long data_dma_addr;
 906
 907         desc.fields.flags_len = IBMVETH_BUF_VALID | skb->len;
 908
 909         if (skb->ip_summed == CHECKSUM_PARTIAL &&
 910             ip_hdr(skb)->protocol != IPPROTO_TCP && skb_checksum_help(skb)) {
 911                 ibmveth_error_printk("tx: failed to checksum packet\n");
 912                 tx_dropped++;
 913                 goto out;
 914         }
 915
 916         if (skb->ip_summed == CHECKSUM_PARTIAL) {
 917                 unsigned char *buf = skb_transport_header(skb) + skb->csum_offset;
 918
 919                 desc.fields.flags_len |= (IBMVETH_BUF_NO_CSUM | IBMVETH_BUF_CSUM_GOOD);
 920
 921                 /* Need to zero out the checksum */
 922                 buf[0] = 0;
 923                 buf[1] = 0;
 924         }
 925
 926         data_dma_addr = dma_map_single(&adapter->vdev->dev, skb->data,
 927                                        skb->len, DMA_TO_DEVICE);
 928         if (dma_mapping_error(&adapter->vdev->dev, data_dma_addr)) {
 929                 if (!firmware_has_feature(FW_FEATURE_CMO))
 930                         ibmveth_error_printk("tx: unable to map xmit buffer\n");
 931                 skb_copy_from_linear_data(skb, adapter->bounce_buffer,
 932                                           skb->len);
 933                 desc.fields.address = adapter->bounce_buffer_dma;
 934                 tx_map_failed++;
 935                 used_bounce = 1;
 936                 wmb();
 937         } else
 938                 desc.fields.address = data_dma_addr;
 939
 940         /* send the frame. Arbitrarily set retrycount to 1024 */
 941         correlator = 0;
 942         retry_count = 1024;
 943         do {
 944                 lpar_rc = h_send_logical_lan(adapter->vdev->unit_address,
 945                                              desc.desc, 0, 0, 0, 0, 0,
 946                                              correlator, &correlator);
 947         } while ((lpar_rc == H_BUSY) && (retry_count--));
 948
 949         if(lpar_rc != H_SUCCESS && lpar_rc != H_DROPPED) {
 950                 ibmveth_error_printk("tx: h_send_logical_lan failed with rc=%ld\n", lpar_rc);
 951                 ibmveth_error_printk("tx: valid=%d, len=%d, address=0x%08x\n",
 952                                      (desc.fields.flags_len & IBMVETH_BUF_VALID) ? 1 : 0,
 953                                      skb->len, desc.fields.address);
 954                 tx_send_failed++;
 955                 tx_dropped++;
 956         } else {
 957                 tx_packets++;
 958                 tx_bytes += skb->len;
 959                 netdev->trans_start = jiffies;
 960         }
 961
 962         if (!used_bounce)
 963                 dma_unmap_single(&adapter->vdev->dev, data_dma_addr,
 964                                  skb->len, DMA_TO_DEVICE);
 965
 966 out:    spin_lock_irqsave(&adapter->stats_lock, flags);
 967         netdev->stats.tx_dropped += tx_dropped;
 968         netdev->stats.tx_bytes += tx_bytes;
 969         netdev->stats.tx_packets += tx_packets;
 970         adapter->tx_send_failed += tx_send_failed;
 971         adapter->tx_map_failed += tx_map_failed;
 972         spin_unlock_irqrestore(&adapter->stats_lock, flags);
 973
 974         dev_kfree_skb(skb);
 975         return NETDEV_TX_OK;
 976 }
 977
 978 static int ibmveth_poll(struct napi_struct *napi, int budget)
 979 {
 980         struct ibmveth_adapter *adapter = container_of(napi, struct ibmveth_adapter, napi);
 981         struct net_device *netdev = adapter->netdev;
 982         int frames_processed = 0;
 983         unsigned long lpar_rc;
 984
 985  restart_poll:
 986         do {
 987                 struct sk_buff *skb;
 988
 989                 if (!ibmveth_rxq_pending_buffer(adapter))
 990                         break;
 991
 992                 rmb();
 993                 if (!ibmveth_rxq_buffer_valid(adapter)) {
 994                         wmb(); /* suggested by larson1 */
 995                         adapter->rx_invalid_buffer++;
 996                         ibmveth_debug_printk("recycling invalid buffer\n");
 997                         ibmveth_rxq_recycle_buffer(adapter);
 998                 } else {
 999                         int length = ibmveth_rxq_frame_length(adapter);
1000                         int offset = ibmveth_rxq_frame_offset(adapter);
1001                         int csum_good = ibmveth_rxq_csum_good(adapter);
1002
1003                         skb = ibmveth_rxq_get_buffer(adapter);
1004                         if (csum_good)
1005                                 skb->ip_summed = CHECKSUM_UNNECESSARY;
1006
1007                         ibmveth_rxq_harvest_buffer(adapter);
1008
1009                         skb_reserve(skb, offset);
1010                         skb_put(skb, length);
1011                         skb->protocol = eth_type_trans(skb, netdev);
1012
1013                         netif_receive_skb(skb); /* send it up */
1014
1015                         netdev->stats.rx_packets++;
1016                         netdev->stats.rx_bytes += length;
1017                         frames_processed++;
1018                 }
1019         } while (frames_processed < budget);
1020
1021         ibmveth_replenish_task(adapter);
1022
1023         if (frames_processed < budget) {
1024                 /* We think we are done - reenable interrupts,
1025                  * then check once more to make sure we are done.
1026                  */
1027                 lpar_rc = h_vio_signal(adapter->vdev->unit_address,
1028                                        VIO_IRQ_ENABLE);
1029
1030                 ibmveth_assert(lpar_rc == H_SUCCESS);
1031
1032                 napi_complete(napi);
1033
1034                 if (ibmveth_rxq_pending_buffer(adapter) &&
1035                     napi_reschedule(napi)) {
1036                         lpar_rc = h_vio_signal(adapter->vdev->unit_address,
1037                                                VIO_IRQ_DISABLE);
1038                         goto restart_poll;
1039                 }
1040         }
1041
1042         return frames_processed;
1043 }
1044
1045 static irqreturn_t ibmveth_interrupt(int irq, void *dev_instance)
1046 {
1047         struct net_device *netdev = dev_instance;
1048         struct ibmveth_adapter *adapter = netdev_priv(netdev);
1049         unsigned long lpar_rc;
1050
1051         if (napi_schedule_prep(&adapter->napi)) {
1052                 lpar_rc = h_vio_signal(adapter->vdev->unit_address,
1053                                        VIO_IRQ_DISABLE);
1054                 ibmveth_assert(lpar_rc == H_SUCCESS);
1055                 __napi_schedule(&adapter->napi);
1056         }
1057         return IRQ_HANDLED;
1058 }
1059
1060 static void ibmveth_set_multicast_list(struct net_device *netdev)
1061 {
1062         struct ibmveth_adapter *adapter = netdev_priv(netdev);
1063         unsigned long lpar_rc;
1064
1065         if((netdev->flags & IFF_PROMISC) || (netdev->mc_count > adapter->mcastFilterSize)) {
1066                 lpar_rc = h_multicast_ctrl(adapter->vdev->unit_address,
1067                                            IbmVethMcastEnableRecv |
1068                                            IbmVethMcastDisableFiltering,
1069                                            0);
1070                 if(lpar_rc != H_SUCCESS) {
1071                         ibmveth_error_printk("h_multicast_ctrl rc=%ld when entering promisc mode\n", lpar_rc);
1072                 }
1073         } else {
1074                 struct dev_mc_list *mclist = netdev->mc_list;
1075                 int i;
1076                 /* clear the filter table & disable filtering */
1077                 lpar_rc = h_multicast_ctrl(adapter->vdev->unit_address,
1078                                            IbmVethMcastEnableRecv |
1079                                            IbmVethMcastDisableFiltering |
1080                                            IbmVethMcastClearFilterTable,
1081                                            0);
1082                 if(lpar_rc != H_SUCCESS) {
1083                         ibmveth_error_printk("h_multicast_ctrl rc=%ld when attempting to clear filter table\n", lpar_rc);
1084                 }
1085                 /* add the addresses to the filter table */
1086                 for(i = 0; i < netdev->mc_count; ++i, mclist = mclist->next) {
1087                         // add the multicast address to the filter table
1088                         unsigned long mcast_addr = 0;
1089                         memcpy(((char *)&mcast_addr)+2, mclist->dmi_addr, 6);
1090                         lpar_rc = h_multicast_ctrl(adapter->vdev->unit_address,
1091                                                    IbmVethMcastAddFilter,
1092                                                    mcast_addr);
1093                         if(lpar_rc != H_SUCCESS) {
1094                                 ibmveth_error_printk("h_multicast_ctrl rc=%ld when adding an entry to the filter table\n", lpar_rc);
1095                         }
1096                 }
1097
1098                 /* re-enable filtering */
1099                 lpar_rc = h_multicast_ctrl(adapter->vdev->unit_address,
1100                                            IbmVethMcastEnableFiltering,
1101                                            0);
1102                 if(lpar_rc != H_SUCCESS) {
1103                         ibmveth_error_printk("h_multicast_ctrl rc=%ld when enabling filtering\n", lpar_rc);
1104                 }
1105         }
1106 }
1107
1108 static int ibmveth_change_mtu(struct net_device *dev, int new_mtu)
1109 {
1110         struct ibmveth_adapter *adapter = netdev_priv(dev);
1111         struct vio_dev *viodev = adapter->vdev;
1112         int new_mtu_oh = new_mtu + IBMVETH_BUFF_OH;
1113         int i;
1114
1115         if (new_mtu < IBMVETH_MAX_MTU)
1116                 return -EINVAL;
1117
1118         for (i = 0; i < IbmVethNumBufferPools; i++)
1119                 if (new_mtu_oh < adapter->rx_buff_pool[i].buff_size)
1120                         break;
1121
1122         if (i == IbmVethNumBufferPools)
1123                 return -EINVAL;
1124
1125         /* Deactivate all the buffer pools so that the next loop can activate
1126            only the buffer pools necessary to hold the new MTU */
1127         for (i = 0; i < IbmVethNumBufferPools; i++)
1128                 if (adapter->rx_buff_pool[i].active) {
1129                         ibmveth_free_buffer_pool(adapter,
1130                                                  &adapter->rx_buff_pool[i]);
1131                         adapter->rx_buff_pool[i].active = 0;
1132                 }
1133
1134         /* Look for an active buffer pool that can hold the new MTU */
1135         for(i = 0; i<IbmVethNumBufferPools; i++) {
1136                 adapter->rx_buff_pool[i].active = 1;
1137
1138                 if (new_mtu_oh < adapter->rx_buff_pool[i].buff_size) {
1139                         if (netif_running(adapter->netdev)) {
1140                                 adapter->pool_config = 1;
1141                                 ibmveth_close(adapter->netdev);
1142                                 adapter->pool_config = 0;
1143                                 dev->mtu = new_mtu;
1144                                 vio_cmo_set_dev_desired(viodev,
1145                                                 ibmveth_get_desired_dma
1146                                                 (viodev));
1147                                 return ibmveth_open(adapter->netdev);
1148                         }
1149                         dev->mtu = new_mtu;
1150                         vio_cmo_set_dev_desired(viodev,
1151                                                 ibmveth_get_desired_dma
1152                                                 (viodev));
1153                         return 0;
1154                 }
1155         }
1156         return -EINVAL;
1157 }
1158
1159 #ifdef CONFIG_NET_POLL_CONTROLLER
1160 static void ibmveth_poll_controller(struct net_device *dev)
1161 {
1162         ibmveth_replenish_task(netdev_priv(dev));
1163         ibmveth_interrupt(dev->irq, dev);
1164 }
1165 #endif
1166
1167 /**
1168  * ibmveth_get_desired_dma - Calculate IO memory desired by the driver
1169  *
1170  * @vdev: struct vio_dev for the device whose desired IO mem is to be returned
1171  *
1172  * Return value:
1173  *      Number of bytes of IO data the driver will need to perform well.
1174  */
1175 static unsigned long ibmveth_get_desired_dma(struct vio_dev *vdev)
1176 {
1177         struct net_device *netdev = dev_get_drvdata(&vdev->dev);
1178         struct ibmveth_adapter *adapter;
1179         unsigned long ret;
1180         int i;
1181         int rxqentries = 1;
1182
1183         /* netdev inits at probe time along with the structures we need below*/
1184         if (netdev == NULL)
1185                 return IOMMU_PAGE_ALIGN(IBMVETH_IO_ENTITLEMENT_DEFAULT);
1186
1187         adapter = netdev_priv(netdev);
1188
1189         ret = IBMVETH_BUFF_LIST_SIZE + IBMVETH_FILT_LIST_SIZE;
1190         ret += IOMMU_PAGE_ALIGN(netdev->mtu);
1191
1192         for (i = 0; i < IbmVethNumBufferPools; i++) {
1193                 /* add the size of the active receive buffers */
1194                 if (adapter->rx_buff_pool[i].active)
1195                         ret +=
1196                             adapter->rx_buff_pool[i].size *
1197                             IOMMU_PAGE_ALIGN(adapter->rx_buff_pool[i].
1198                                     buff_size);
1199                 rxqentries += adapter->rx_buff_pool[i].size;
1200         }
1201         /* add the size of the receive queue entries */
1202         ret += IOMMU_PAGE_ALIGN(rxqentries * sizeof(struct ibmveth_rx_q_entry));
1203
1204         return ret;
1205 }
1206
1207 static const struct net_device_ops ibmveth_netdev_ops = {
1208         .ndo_open               = ibmveth_open,
1209         .ndo_stop               = ibmveth_close,
1210         .ndo_start_xmit         = ibmveth_start_xmit,
1211         .ndo_set_multicast_list = ibmveth_set_multicast_list,
1212         .ndo_do_ioctl           = ibmveth_ioctl,
1213         .ndo_change_mtu         = ibmveth_change_mtu,
1214         .ndo_validate_addr      = eth_validate_addr,
1215         .ndo_set_mac_address    = eth_mac_addr,
1216 #ifdef CONFIG_NET_POLL_CONTROLLER
1217         .ndo_poll_controller    = ibmveth_poll_controller,
1218 #endif
1219 };
1220
1221 static int __devinit ibmveth_probe(struct vio_dev *dev, const struct vio_device_id *id)
1222 {
1223         int rc, i;
1224         long ret;
1225         struct net_device *netdev;
1226         struct ibmveth_adapter *adapter;
1227         unsigned long set_attr, ret_attr;
1228
1229         unsigned char *mac_addr_p;
1230         unsigned int *mcastFilterSize_p;
1231
1232
1233         ibmveth_debug_printk_no_adapter("entering ibmveth_probe for UA 0x%x\n",
1234                                         dev->unit_address);
1235
1236         mac_addr_p = (unsigned char *) vio_get_attribute(dev,
1237                                                 VETH_MAC_ADDR, NULL);
1238         if(!mac_addr_p) {
1239                 printk(KERN_ERR "(%s:%3.3d) ERROR: Can't find VETH_MAC_ADDR "
1240                                 "attribute\n", __FILE__, __LINE__);
1241                 return 0;
1242         }
1243
1244         mcastFilterSize_p = (unsigned int *) vio_get_attribute(dev,
1245                                                 VETH_MCAST_FILTER_SIZE, NULL);
1246         if(!mcastFilterSize_p) {
1247                 printk(KERN_ERR "(%s:%3.3d) ERROR: Can't find "
1248                                 "VETH_MCAST_FILTER_SIZE attribute\n",
1249                                 __FILE__, __LINE__);
1250                 return 0;
1251         }
1252
1253         netdev = alloc_etherdev(sizeof(struct ibmveth_adapter));
1254
1255         if(!netdev)
1256                 return -ENOMEM;
1257
1258         adapter = netdev_priv(netdev);
1259         dev_set_drvdata(&dev->dev, netdev);
1260
1261         adapter->vdev = dev;
1262         adapter->netdev = netdev;
1263         adapter->mcastFilterSize= *mcastFilterSize_p;
1264         adapter->pool_config = 0;
1265
1266         netif_napi_add(netdev, &adapter->napi, ibmveth_poll, 16);
1267
1268         /*      Some older boxes running PHYP non-natively have an OF that
1269                 returns a 8-byte local-mac-address field (and the first
1270                 2 bytes have to be ignored) while newer boxes' OF return
1271                 a 6-byte field. Note that IEEE 1275 specifies that
1272                 local-mac-address must be a 6-byte field.
1273                 The RPA doc specifies that the first byte must be 10b, so
1274                 we'll just look for it to solve this 8 vs. 6 byte field issue */
1275
1276         if ((*mac_addr_p & 0x3) != 0x02)
1277                 mac_addr_p += 2;
1278
1279         adapter->mac_addr = 0;
1280         memcpy(&adapter->mac_addr, mac_addr_p, 6);
1281
1282         netdev->irq = dev->irq;
1283         netdev->netdev_ops = &ibmveth_netdev_ops;
1284         netdev->ethtool_ops = &netdev_ethtool_ops;
1285         SET_NETDEV_DEV(netdev, &dev->dev);
1286         netdev->features |= NETIF_F_LLTX;
1287         spin_lock_init(&adapter->stats_lock);
1288
1289         memcpy(netdev->dev_addr, &adapter->mac_addr, netdev->addr_len);
1290
1291         for(i = 0; i<IbmVethNumBufferPools; i++) {
1292                 struct kobject *kobj = &adapter->rx_buff_pool[i].kobj;
1293                 int error;
1294
1295                 ibmveth_init_buffer_pool(&adapter->rx_buff_pool[i], i,
1296                                          pool_count[i], pool_size[i],
1297                                          pool_active[i]);
1298                 error = kobject_init_and_add(kobj, &ktype_veth_pool,
1299                                              &dev->dev.kobj, "pool%d", i);
1300                 if (!error)
1301                         kobject_uevent(kobj, KOBJ_ADD);
1302         }
1303
1304         ibmveth_debug_printk("adapter @ 0x%p\n", adapter);
1305
1306         adapter->buffer_list_dma = DMA_ERROR_CODE;
1307         adapter->filter_list_dma = DMA_ERROR_CODE;
1308         adapter->rx_queue.queue_dma = DMA_ERROR_CODE;
1309
1310         ibmveth_debug_printk("registering netdev...\n");
1311
1312         ret = h_illan_attributes(dev->unit_address, 0, 0, &ret_attr);
1313
1314         if (ret == H_SUCCESS && !(ret_attr & IBMVETH_ILLAN_ACTIVE_TRUNK) &&
1315             !(ret_attr & IBMVETH_ILLAN_TRUNK_PRI_MASK) &&
1316             (ret_attr & IBMVETH_ILLAN_PADDED_PKT_CSUM)) {
1317                 set_attr = IBMVETH_ILLAN_IPV4_TCP_CSUM;
1318
1319                 ret = h_illan_attributes(dev->unit_address, 0, set_attr, &ret_attr);
1320
1321                 if (ret == H_SUCCESS) {
1322                         adapter->rx_csum = 1;
1323                         netdev->features |= NETIF_F_IP_CSUM;
1324                 } else
1325                         ret = h_illan_attributes(dev->unit_address, set_attr, 0, &ret_attr);
1326         }
1327
1328         rc = register_netdev(netdev);
1329
1330         if(rc) {
1331                 ibmveth_debug_printk("failed to register netdev rc=%d\n", rc);
1332                 free_netdev(netdev);
1333                 return rc;
1334         }
1335
1336         ibmveth_debug_printk("registered\n");
1337
1338         ibmveth_proc_register_adapter(adapter);
1339
1340         return 0;
1341 }
1342
1343 static int __devexit ibmveth_remove(struct vio_dev *dev)
1344 {
1345         struct net_device *netdev = dev_get_drvdata(&dev->dev);
1346         struct ibmveth_adapter *adapter = netdev_priv(netdev);
1347         int i;
1348
1349         for(i = 0; i<IbmVethNumBufferPools; i++)
1350                 kobject_put(&adapter->rx_buff_pool[i].kobj);
1351
1352         unregister_netdev(netdev);
1353
1354         ibmveth_proc_unregister_adapter(adapter);
1355
1356         free_netdev(netdev);
1357         dev_set_drvdata(&dev->dev, NULL);
1358
1359         return 0;
1360 }
1361
1362 #ifdef CONFIG_PROC_FS
1363 static void ibmveth_proc_register_driver(void)
1364 {
1365         ibmveth_proc_dir = proc_mkdir(IBMVETH_PROC_DIR, init_net.proc_net);
1366         if (ibmveth_proc_dir) {
1367         }
1368 }
1369
1370 static void ibmveth_proc_unregister_driver(void)
1371 {
1372         remove_proc_entry(IBMVETH_PROC_DIR, init_net.proc_net);
1373 }
1374
1375 static int ibmveth_show(struct seq_file *seq, void *v)
1376 {
1377         struct ibmveth_adapter *adapter = seq->private;
1378         char *current_mac = (char *) adapter->netdev->dev_addr;
1379         char *firmware_mac = (char *) &adapter->mac_addr;
1380
1381         seq_printf(seq, "%s %s\n\n", ibmveth_driver_string, ibmveth_driver_version);
1382
1383         seq_printf(seq, "Unit Address:    0x%x\n", adapter->vdev->unit_address);
1384         seq_printf(seq, "Current MAC:     %pM\n", current_mac);
1385         seq_printf(seq, "Firmware MAC:    %pM\n", firmware_mac);
1386
1387         seq_printf(seq, "\nAdapter Statistics:\n");
1388         seq_printf(seq, "  TX:  vio_map_single failres:      %lld\n", adapter->tx_map_failed);
1389         seq_printf(seq, "       send failures:               %lld\n", adapter->tx_send_failed);
1390         seq_printf(seq, "  RX:  replenish task cycles:       %lld\n", adapter->replenish_task_cycles);
1391         seq_printf(seq, "       alloc_skb_failures:          %lld\n", adapter->replenish_no_mem);
1392         seq_printf(seq, "       add buffer failures:         %lld\n", adapter->replenish_add_buff_failure);
1393         seq_printf(seq, "       invalid buffers:             %lld\n", adapter->rx_invalid_buffer);
1394         seq_printf(seq, "       no buffers:                  %lld\n", adapter->rx_no_buffer);
1395
1396         return 0;
1397 }
1398
1399 static int ibmveth_proc_open(struct inode *inode, struct file *file)
1400 {
1401         return single_open(file, ibmveth_show, PDE(inode)->data);
1402 }
1403
1404 static const struct file_operations ibmveth_proc_fops = {
1405         .owner   = THIS_MODULE,
1406         .open    = ibmveth_proc_open,
1407         .read    = seq_read,
1408         .llseek  = seq_lseek,
1409         .release = single_release,
1410 };
1411
1412 static void ibmveth_proc_register_adapter(struct ibmveth_adapter *adapter)
1413 {
1414         struct proc_dir_entry *entry;
1415         if (ibmveth_proc_dir) {
1416                 char u_addr[10];
1417                 sprintf(u_addr, "%x", adapter->vdev->unit_address);
1418                 entry = proc_create_data(u_addr, S_IFREG, ibmveth_proc_dir,
1419                                          &ibmveth_proc_fops, adapter);
1420                 if (!entry)
1421                         ibmveth_error_printk("Cannot create adapter proc entry");
1422         }
1423         return;
1424 }
1425
1426 static void ibmveth_proc_unregister_adapter(struct ibmveth_adapter *adapter)
1427 {
1428         if (ibmveth_proc_dir) {
1429                 char u_addr[10];
1430                 sprintf(u_addr, "%x", adapter->vdev->unit_address);
1431                 remove_proc_entry(u_addr, ibmveth_proc_dir);
1432         }
1433 }
1434
1435 #else /* CONFIG_PROC_FS */
1436 static void ibmveth_proc_register_adapter(struct ibmveth_adapter *adapter)
1437 {
1438 }
1439
1440 static void ibmveth_proc_unregister_adapter(struct ibmveth_adapter *adapter)
1441 {
1442 }
1443 static void ibmveth_proc_register_driver(void)
1444 {
1445 }
1446
1447 static void ibmveth_proc_unregister_driver(void)
1448 {
1449 }
1450 #endif /* CONFIG_PROC_FS */
1451
1452 static struct attribute veth_active_attr;
1453 static struct attribute veth_num_attr;
1454 static struct attribute veth_size_attr;
1455
1456 static ssize_t veth_pool_show(struct kobject * kobj,
1457                               struct attribute * attr, char * buf)
1458 {
1459         struct ibmveth_buff_pool *pool = container_of(kobj,
1460                                                       struct ibmveth_buff_pool,
1461                                                       kobj);
1462
1463         if (attr == &veth_active_attr)
1464                 return sprintf(buf, "%d\n", pool->active);
1465         else if (attr == &veth_num_attr)
1466                 return sprintf(buf, "%d\n", pool->size);
1467         else if (attr == &veth_size_attr)
1468                 return sprintf(buf, "%d\n", pool->buff_size);
1469         return 0;
1470 }
1471
1472 static ssize_t veth_pool_store(struct kobject * kobj, struct attribute * attr,
1473 const char * buf, size_t count)
1474 {
1475         struct ibmveth_buff_pool *pool = container_of(kobj,
1476                                                       struct ibmveth_buff_pool,
1477                                                       kobj);
1478         struct net_device *netdev = dev_get_drvdata(
1479             container_of(kobj->parent, struct device, kobj));
1480         struct ibmveth_adapter *adapter = netdev_priv(netdev);
1481         long value = simple_strtol(buf, NULL, 10);
1482         long rc;
1483
1484         if (attr == &veth_active_attr) {
1485                 if (value && !pool->active) {
1486                         if (netif_running(netdev)) {
1487                                 if(ibmveth_alloc_buffer_pool(pool)) {
1488                                         ibmveth_error_printk("unable to alloc pool\n");
1489                                         return -ENOMEM;
1490                                 }
1491                                 pool->active = 1;
1492                                 adapter->pool_config = 1;
1493                                 ibmveth_close(netdev);
1494                                 adapter->pool_config = 0;
1495                                 if ((rc = ibmveth_open(netdev)))
1496                                         return rc;
1497                         } else
1498                                 pool->active = 1;
1499                 } else if (!value && pool->active) {
1500                         int mtu = netdev->mtu + IBMVETH_BUFF_OH;
1501                         int i;
1502                         /* Make sure there is a buffer pool with buffers that
1503                            can hold a packet of the size of the MTU */
1504                         for (i = 0; i < IbmVethNumBufferPools; i++) {
1505                                 if (pool == &adapter->rx_buff_pool[i])
1506                                         continue;
1507                                 if (!adapter->rx_buff_pool[i].active)
1508                                         continue;
1509                                 if (mtu <= adapter->rx_buff_pool[i].buff_size)
1510                                         break;
1511                         }
1512
1513                         if (i == IbmVethNumBufferPools) {
1514                                 ibmveth_error_printk("no active pool >= MTU\n");
1515                                 return -EPERM;
1516                         }
1517
1518                         if (netif_running(netdev)) {
1519                                 adapter->pool_config = 1;
1520                                 ibmveth_close(netdev);
1521                                 pool->active = 0;
1522                                 adapter->pool_config = 0;
1523                                 if ((rc = ibmveth_open(netdev)))
1524                                         return rc;
1525                         }
1526                         pool->active = 0;
1527                 }
1528         } else if (attr == &veth_num_attr) {
1529                 if (value <= 0 || value > IBMVETH_MAX_POOL_COUNT)
1530                         return -EINVAL;
1531                 else {
1532                         if (netif_running(netdev)) {
1533                                 adapter->pool_config = 1;
1534                                 ibmveth_close(netdev);
1535                                 adapter->pool_config = 0;
1536                                 pool->size = value;
1537                                 if ((rc = ibmveth_open(netdev)))
1538                                         return rc;
1539                         } else
1540                                 pool->size = value;
1541                 }
1542         } else if (attr == &veth_size_attr) {
1543                 if (value <= IBMVETH_BUFF_OH || value > IBMVETH_MAX_BUF_SIZE)
1544                         return -EINVAL;
1545                 else {
1546                         if (netif_running(netdev)) {
1547                                 adapter->pool_config = 1;
1548                                 ibmveth_close(netdev);
1549                                 adapter->pool_config = 0;
1550                                 pool->buff_size = value;
1551                                 if ((rc = ibmveth_open(netdev)))
1552                                         return rc;
1553                         } else
1554                                 pool->buff_size = value;
1555                 }
1556         }
1557
1558         /* kick the interrupt handler to allocate/deallocate pools */
1559         ibmveth_interrupt(netdev->irq, netdev);
1560         return count;
1561 }
1562
1563
1564 #define ATTR(_name, _mode)      \
1565         struct attribute veth_##_name##_attr = {               \
1566         .name = __stringify(_name), .mode = _mode, \
1567         };
1568
1569 static ATTR(active, 0644);
1570 static ATTR(num, 0644);
1571 static ATTR(size, 0644);
1572
1573 static struct attribute * veth_pool_attrs[] = {
1574         &veth_active_attr,
1575         &veth_num_attr,
1576         &veth_size_attr,
1577         NULL,
1578 };
1579
1580 static struct sysfs_ops veth_pool_ops = {
1581         .show   = veth_pool_show,
1582         .store  = veth_pool_store,
1583 };
1584
1585 static struct kobj_type ktype_veth_pool = {
1586         .release        = NULL,
1587         .sysfs_ops      = &veth_pool_ops,
1588         .default_attrs  = veth_pool_attrs,
1589 };
1590
1591
1592 static struct vio_device_id ibmveth_device_table[] __devinitdata= {
1593         { "network", "IBM,l-lan"},
1594         { "", "" }
1595 };
1596 MODULE_DEVICE_TABLE(vio, ibmveth_device_table);
1597
1598 static struct vio_driver ibmveth_driver = {
1599         .id_table       = ibmveth_device_table,
1600         .probe          = ibmveth_probe,
1601         .remove         = ibmveth_remove,
1602         .get_desired_dma = ibmveth_get_desired_dma,
1603         .driver         = {
1604                 .name   = ibmveth_driver_name,
1605                 .owner  = THIS_MODULE,
1606         }
1607 };
1608
1609 static int __init ibmveth_module_init(void)
1610 {
1611         ibmveth_printk("%s: %s %s\n", ibmveth_driver_name, ibmveth_driver_string, ibmveth_driver_version);
1612
1613         ibmveth_proc_register_driver();
1614
1615         return vio_register_driver(&ibmveth_driver);
1616 }
1617
1618 static void __exit ibmveth_module_exit(void)
1619 {
1620         vio_unregister_driver(&ibmveth_driver);
1621         ibmveth_proc_unregister_driver();
1622 }
1623
1624 module_init(ibmveth_module_init);
1625 module_exit(ibmveth_module_exit);