Pileus Git - ~andy/linux/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/pm.h>
  49 #include <linux/ethtool.h>
  50 #include <linux/proc_fs.h>
  51 #include <linux/in.h>
  52 #include <linux/ip.h>
  53 #include <linux/slab.h>
  54 #include <net/net_namespace.h>
  55 #include <asm/hvcall.h>
  56 #include <asm/atomic.h>
  57 #include <asm/vio.h>
  58 #include <asm/iommu.h>
  59 #include <asm/uaccess.h>
  60 #include <asm/firmware.h>
  61 #include <linux/seq_file.h>
  62
  63 #include "ibmveth.h"
  64
  65 #undef DEBUG
  66
  67 #define ibmveth_printk(fmt, args...) \
  68   printk(KERN_DEBUG "%s: " fmt, __FILE__, ## args)
  69
  70 #define ibmveth_error_printk(fmt, args...) \
  71   printk(KERN_ERR "(%s:%3.3d ua:%x) ERROR: " fmt, __FILE__, __LINE__ , adapter->vdev->unit_address, ## args)
  72
  73 #ifdef DEBUG
  74 #define ibmveth_debug_printk_no_adapter(fmt, args...) \
  75   printk(KERN_DEBUG "(%s:%3.3d): " fmt, __FILE__, __LINE__ , ## args)
  76 #define ibmveth_debug_printk(fmt, args...) \
  77   printk(KERN_DEBUG "(%s:%3.3d ua:%x): " fmt, __FILE__, __LINE__ , adapter->vdev->unit_address, ## args)
  78 #define ibmveth_assert(expr) \
  79   if(!(expr)) {                                   \
  80     printk(KERN_DEBUG "assertion failed (%s:%3.3d ua:%x): %s\n", __FILE__, __LINE__, adapter->vdev->unit_address, #expr); \
  81     BUG(); \
  82   }
  83 #else
  84 #define ibmveth_debug_printk_no_adapter(fmt, args...)
  85 #define ibmveth_debug_printk(fmt, args...)
  86 #define ibmveth_assert(expr)
  87 #endif
  88
  89 static int ibmveth_open(struct net_device *dev);
  90 static int ibmveth_close(struct net_device *dev);
  91 static int ibmveth_ioctl(struct net_device *dev, struct ifreq *ifr, int cmd);
  92 static int ibmveth_poll(struct napi_struct *napi, int budget);
  93 static int ibmveth_start_xmit(struct sk_buff *skb, struct net_device *dev);
  94 static void ibmveth_set_multicast_list(struct net_device *dev);
  95 static int ibmveth_change_mtu(struct net_device *dev, int new_mtu);
  96 static void ibmveth_proc_register_driver(void);
  97 static void ibmveth_proc_unregister_driver(void);
  98 static void ibmveth_proc_register_adapter(struct ibmveth_adapter *adapter);
  99 static void ibmveth_proc_unregister_adapter(struct ibmveth_adapter *adapter);
 100 static irqreturn_t ibmveth_interrupt(int irq, void *dev_instance);
 101 static void ibmveth_rxq_harvest_buffer(struct ibmveth_adapter *adapter);
 102 static unsigned long ibmveth_get_desired_dma(struct vio_dev *vdev);
 103 static struct kobj_type ktype_veth_pool;
 104
 105
 106 #ifdef CONFIG_PROC_FS
 107 #define IBMVETH_PROC_DIR "ibmveth"
 108 static struct proc_dir_entry *ibmveth_proc_dir;
 109 #endif
 110
 111 static const char ibmveth_driver_name[] = "ibmveth";
 112 static const char ibmveth_driver_string[] = "IBM i/pSeries Virtual Ethernet Driver";
 113 #define ibmveth_driver_version "1.03"
 114
 115 MODULE_AUTHOR("Santiago Leon <santil@us.ibm.com>");
 116 MODULE_DESCRIPTION("IBM i/pSeries Virtual Ethernet Driver");
 117 MODULE_LICENSE("GPL");
 118 MODULE_VERSION(ibmveth_driver_version);
 119
 120 static unsigned int tx_copybreak __read_mostly = 128;
 121 module_param(tx_copybreak, uint, 0644);
 122 MODULE_PARM_DESC(tx_copybreak,
 123         "Maximum size of packet that is copied to a new buffer on transmit");
 124
 125 struct ibmveth_stat {
 126         char name[ETH_GSTRING_LEN];
 127         int offset;
 128 };
 129
 130 #define IBMVETH_STAT_OFF(stat) offsetof(struct ibmveth_adapter, stat)
 131 #define IBMVETH_GET_STAT(a, off) *((u64 *)(((unsigned long)(a)) + off))
 132
 133 struct ibmveth_stat ibmveth_stats[] = {
 134         { "replenish_task_cycles", IBMVETH_STAT_OFF(replenish_task_cycles) },
 135         { "replenish_no_mem", IBMVETH_STAT_OFF(replenish_no_mem) },
 136         { "replenish_add_buff_failure", IBMVETH_STAT_OFF(replenish_add_buff_failure) },
 137         { "replenish_add_buff_success", IBMVETH_STAT_OFF(replenish_add_buff_success) },
 138         { "rx_invalid_buffer", IBMVETH_STAT_OFF(rx_invalid_buffer) },
 139         { "rx_no_buffer", IBMVETH_STAT_OFF(rx_no_buffer) },
 140         { "tx_map_failed", IBMVETH_STAT_OFF(tx_map_failed) },
 141         { "tx_send_failed", IBMVETH_STAT_OFF(tx_send_failed) },
 142 };
 143
 144 /* simple methods of getting data from the current rxq entry */
 145 static inline u32 ibmveth_rxq_flags(struct ibmveth_adapter *adapter)
 146 {
 147         return adapter->rx_queue.queue_addr[adapter->rx_queue.index].flags_off;
 148 }
 149
 150 static inline int ibmveth_rxq_toggle(struct ibmveth_adapter *adapter)
 151 {
 152         return (ibmveth_rxq_flags(adapter) & IBMVETH_RXQ_TOGGLE) >> IBMVETH_RXQ_TOGGLE_SHIFT;
 153 }
 154
 155 static inline int ibmveth_rxq_pending_buffer(struct ibmveth_adapter *adapter)
 156 {
 157         return (ibmveth_rxq_toggle(adapter) == adapter->rx_queue.toggle);
 158 }
 159
 160 static inline int ibmveth_rxq_buffer_valid(struct ibmveth_adapter *adapter)
 161 {
 162         return (ibmveth_rxq_flags(adapter) & IBMVETH_RXQ_VALID);
 163 }
 164
 165 static inline int ibmveth_rxq_frame_offset(struct ibmveth_adapter *adapter)
 166 {
 167         return (ibmveth_rxq_flags(adapter) & IBMVETH_RXQ_OFF_MASK);
 168 }
 169
 170 static inline int ibmveth_rxq_frame_length(struct ibmveth_adapter *adapter)
 171 {
 172         return (adapter->rx_queue.queue_addr[adapter->rx_queue.index].length);
 173 }
 174
 175 static inline int ibmveth_rxq_csum_good(struct ibmveth_adapter *adapter)
 176 {
 177         return (ibmveth_rxq_flags(adapter) & IBMVETH_RXQ_CSUM_GOOD);
 178 }
 179
 180 /* setup the initial settings for a buffer pool */
 181 static void ibmveth_init_buffer_pool(struct ibmveth_buff_pool *pool, u32 pool_index, u32 pool_size, u32 buff_size, u32 pool_active)
 182 {
 183         pool->size = pool_size;
 184         pool->index = pool_index;
 185         pool->buff_size = buff_size;
 186         pool->threshold = pool_size * 7 / 8;
 187         pool->active = pool_active;
 188 }
 189
 190 /* allocate and setup an buffer pool - called during open */
 191 static int ibmveth_alloc_buffer_pool(struct ibmveth_buff_pool *pool)
 192 {
 193         int i;
 194
 195         pool->free_map = kmalloc(sizeof(u16) * pool->size, GFP_KERNEL);
 196
 197         if(!pool->free_map) {
 198                 return -1;
 199         }
 200
 201         pool->dma_addr = kmalloc(sizeof(dma_addr_t) * pool->size, GFP_KERNEL);
 202         if(!pool->dma_addr) {
 203                 kfree(pool->free_map);
 204                 pool->free_map = NULL;
 205                 return -1;
 206         }
 207
 208         pool->skbuff = kcalloc(pool->size, sizeof(void *), GFP_KERNEL);
 209
 210         if(!pool->skbuff) {
 211                 kfree(pool->dma_addr);
 212                 pool->dma_addr = NULL;
 213
 214                 kfree(pool->free_map);
 215                 pool->free_map = NULL;
 216                 return -1;
 217         }
 218
 219         memset(pool->dma_addr, 0, sizeof(dma_addr_t) * pool->size);
 220
 221         for(i = 0; i < pool->size; ++i) {
 222                 pool->free_map[i] = i;
 223         }
 224
 225         atomic_set(&pool->available, 0);
 226         pool->producer_index = 0;
 227         pool->consumer_index = 0;
 228
 229         return 0;
 230 }
 231
 232 /* replenish the buffers for a pool.  note that we don't need to
 233  * skb_reserve these since they are used for incoming...
 234  */
 235 static void ibmveth_replenish_buffer_pool(struct ibmveth_adapter *adapter, struct ibmveth_buff_pool *pool)
 236 {
 237         u32 i;
 238         u32 count = pool->size - atomic_read(&pool->available);
 239         u32 buffers_added = 0;
 240         struct sk_buff *skb;
 241         unsigned int free_index, index;
 242         u64 correlator;
 243         unsigned long lpar_rc;
 244         dma_addr_t dma_addr;
 245
 246         mb();
 247
 248         for(i = 0; i < count; ++i) {
 249                 union ibmveth_buf_desc desc;
 250
 251                 skb = alloc_skb(pool->buff_size, GFP_ATOMIC);
 252
 253                 if(!skb) {
 254                         ibmveth_debug_printk("replenish: unable to allocate skb\n");
 255                         adapter->replenish_no_mem++;
 256                         break;
 257                 }
 258
 259                 free_index = pool->consumer_index;
 260                 pool->consumer_index++;
 261                 if (pool->consumer_index >= pool->size)
 262                         pool->consumer_index = 0;
 263                 index = pool->free_map[free_index];
 264
 265                 ibmveth_assert(index != IBM_VETH_INVALID_MAP);
 266                 ibmveth_assert(pool->skbuff[index] == NULL);
 267
 268                 dma_addr = dma_map_single(&adapter->vdev->dev, skb->data,
 269                                 pool->buff_size, DMA_FROM_DEVICE);
 270
 271                 if (dma_mapping_error(&adapter->vdev->dev, dma_addr))
 272                         goto failure;
 273
 274                 pool->free_map[free_index] = IBM_VETH_INVALID_MAP;
 275                 pool->dma_addr[index] = dma_addr;
 276                 pool->skbuff[index] = skb;
 277
 278                 correlator = ((u64)pool->index << 32) | index;
 279                 *(u64*)skb->data = correlator;
 280
 281                 desc.fields.flags_len = IBMVETH_BUF_VALID | pool->buff_size;
 282                 desc.fields.address = dma_addr;
 283
 284                 lpar_rc = h_add_logical_lan_buffer(adapter->vdev->unit_address, desc.desc);
 285
 286                 if (lpar_rc != H_SUCCESS)
 287                         goto failure;
 288                 else {
 289                         buffers_added++;
 290                         adapter->replenish_add_buff_success++;
 291                 }
 292         }
 293
 294         mb();
 295         atomic_add(buffers_added, &(pool->available));
 296         return;
 297
 298 failure:
 299         pool->free_map[free_index] = index;
 300         pool->skbuff[index] = NULL;
 301         if (pool->consumer_index == 0)
 302                 pool->consumer_index = pool->size - 1;
 303         else
 304                 pool->consumer_index--;
 305         if (!dma_mapping_error(&adapter->vdev->dev, dma_addr))
 306                 dma_unmap_single(&adapter->vdev->dev,
 307                                  pool->dma_addr[index], pool->buff_size,
 308                                  DMA_FROM_DEVICE);
 309         dev_kfree_skb_any(skb);
 310         adapter->replenish_add_buff_failure++;
 311
 312         mb();
 313         atomic_add(buffers_added, &(pool->available));
 314 }
 315
 316 /* replenish routine */
 317 static void ibmveth_replenish_task(struct ibmveth_adapter *adapter)
 318 {
 319         int i;
 320
 321         adapter->replenish_task_cycles++;
 322
 323         for (i = (IbmVethNumBufferPools - 1); i >= 0; i--) {
 324                 struct ibmveth_buff_pool *pool = &adapter->rx_buff_pool[i];
 325
 326                 if (pool->active &&
 327                     (atomic_read(&pool->available) < pool->threshold))
 328                         ibmveth_replenish_buffer_pool(adapter, pool);
 329         }
 330
 331         adapter->rx_no_buffer = *(u64*)(((char*)adapter->buffer_list_addr) + 4096 - 8);
 332 }
 333
 334 /* empty and free ana buffer pool - also used to do cleanup in error paths */
 335 static void ibmveth_free_buffer_pool(struct ibmveth_adapter *adapter, struct ibmveth_buff_pool *pool)
 336 {
 337         int i;
 338
 339         kfree(pool->free_map);
 340         pool->free_map = NULL;
 341
 342         if(pool->skbuff && pool->dma_addr) {
 343                 for(i = 0; i < pool->size; ++i) {
 344                         struct sk_buff *skb = pool->skbuff[i];
 345                         if(skb) {
 346                                 dma_unmap_single(&adapter->vdev->dev,
 347                                                  pool->dma_addr[i],
 348                                                  pool->buff_size,
 349                                                  DMA_FROM_DEVICE);
 350                                 dev_kfree_skb_any(skb);
 351                                 pool->skbuff[i] = NULL;
 352                         }
 353                 }
 354         }
 355
 356         if(pool->dma_addr) {
 357                 kfree(pool->dma_addr);
 358                 pool->dma_addr = NULL;
 359         }
 360
 361         if(pool->skbuff) {
 362                 kfree(pool->skbuff);
 363                 pool->skbuff = NULL;
 364         }
 365 }
 366
 367 /* remove a buffer from a pool */
 368 static void ibmveth_remove_buffer_from_pool(struct ibmveth_adapter *adapter, u64 correlator)
 369 {
 370         unsigned int pool  = correlator >> 32;
 371         unsigned int index = correlator & 0xffffffffUL;
 372         unsigned int free_index;
 373         struct sk_buff *skb;
 374
 375         ibmveth_assert(pool < IbmVethNumBufferPools);
 376         ibmveth_assert(index < adapter->rx_buff_pool[pool].size);
 377
 378         skb = adapter->rx_buff_pool[pool].skbuff[index];
 379
 380         ibmveth_assert(skb != NULL);
 381
 382         adapter->rx_buff_pool[pool].skbuff[index] = NULL;
 383
 384         dma_unmap_single(&adapter->vdev->dev,
 385                          adapter->rx_buff_pool[pool].dma_addr[index],
 386                          adapter->rx_buff_pool[pool].buff_size,
 387                          DMA_FROM_DEVICE);
 388
 389         free_index = adapter->rx_buff_pool[pool].producer_index;
 390         adapter->rx_buff_pool[pool].producer_index++;
 391         if (adapter->rx_buff_pool[pool].producer_index >=
 392             adapter->rx_buff_pool[pool].size)
 393                 adapter->rx_buff_pool[pool].producer_index = 0;
 394         adapter->rx_buff_pool[pool].free_map[free_index] = index;
 395
 396         mb();
 397
 398         atomic_dec(&(adapter->rx_buff_pool[pool].available));
 399 }
 400
 401 /* get the current buffer on the rx queue */
 402 static inline struct sk_buff *ibmveth_rxq_get_buffer(struct ibmveth_adapter *adapter)
 403 {
 404         u64 correlator = adapter->rx_queue.queue_addr[adapter->rx_queue.index].correlator;
 405         unsigned int pool = correlator >> 32;
 406         unsigned int index = correlator & 0xffffffffUL;
 407
 408         ibmveth_assert(pool < IbmVethNumBufferPools);
 409         ibmveth_assert(index < adapter->rx_buff_pool[pool].size);
 410
 411         return adapter->rx_buff_pool[pool].skbuff[index];
 412 }
 413
 414 /* recycle the current buffer on the rx queue */
 415 static void ibmveth_rxq_recycle_buffer(struct ibmveth_adapter *adapter)
 416 {
 417         u32 q_index = adapter->rx_queue.index;
 418         u64 correlator = adapter->rx_queue.queue_addr[q_index].correlator;
 419         unsigned int pool = correlator >> 32;
 420         unsigned int index = correlator & 0xffffffffUL;
 421         union ibmveth_buf_desc desc;
 422         unsigned long lpar_rc;
 423
 424         ibmveth_assert(pool < IbmVethNumBufferPools);
 425         ibmveth_assert(index < adapter->rx_buff_pool[pool].size);
 426
 427         if(!adapter->rx_buff_pool[pool].active) {
 428                 ibmveth_rxq_harvest_buffer(adapter);
 429                 ibmveth_free_buffer_pool(adapter, &adapter->rx_buff_pool[pool]);
 430                 return;
 431         }
 432
 433         desc.fields.flags_len = IBMVETH_BUF_VALID |
 434                 adapter->rx_buff_pool[pool].buff_size;
 435         desc.fields.address = adapter->rx_buff_pool[pool].dma_addr[index];
 436
 437         lpar_rc = h_add_logical_lan_buffer(adapter->vdev->unit_address, desc.desc);
 438
 439         if(lpar_rc != H_SUCCESS) {
 440                 ibmveth_debug_printk("h_add_logical_lan_buffer failed during recycle rc=%ld", lpar_rc);
 441                 ibmveth_remove_buffer_from_pool(adapter, adapter->rx_queue.queue_addr[adapter->rx_queue.index].correlator);
 442         }
 443
 444         if(++adapter->rx_queue.index == adapter->rx_queue.num_slots) {
 445                 adapter->rx_queue.index = 0;
 446                 adapter->rx_queue.toggle = !adapter->rx_queue.toggle;
 447         }
 448 }
 449
 450 static void ibmveth_rxq_harvest_buffer(struct ibmveth_adapter *adapter)
 451 {
 452         ibmveth_remove_buffer_from_pool(adapter, adapter->rx_queue.queue_addr[adapter->rx_queue.index].correlator);
 453
 454         if(++adapter->rx_queue.index == adapter->rx_queue.num_slots) {
 455                 adapter->rx_queue.index = 0;
 456                 adapter->rx_queue.toggle = !adapter->rx_queue.toggle;
 457         }
 458 }
 459
 460 static void ibmveth_cleanup(struct ibmveth_adapter *adapter)
 461 {
 462         int i;
 463         struct device *dev = &adapter->vdev->dev;
 464
 465         if(adapter->buffer_list_addr != NULL) {
 466                 if (!dma_mapping_error(dev, adapter->buffer_list_dma)) {
 467                         dma_unmap_single(dev, adapter->buffer_list_dma, 4096,
 468                                         DMA_BIDIRECTIONAL);
 469                         adapter->buffer_list_dma = DMA_ERROR_CODE;
 470                 }
 471                 free_page((unsigned long)adapter->buffer_list_addr);
 472                 adapter->buffer_list_addr = NULL;
 473         }
 474
 475         if(adapter->filter_list_addr != NULL) {
 476                 if (!dma_mapping_error(dev, adapter->filter_list_dma)) {
 477                         dma_unmap_single(dev, adapter->filter_list_dma, 4096,
 478                                         DMA_BIDIRECTIONAL);
 479                         adapter->filter_list_dma = DMA_ERROR_CODE;
 480                 }
 481                 free_page((unsigned long)adapter->filter_list_addr);
 482                 adapter->filter_list_addr = NULL;
 483         }
 484
 485         if(adapter->rx_queue.queue_addr != NULL) {
 486                 if (!dma_mapping_error(dev, adapter->rx_queue.queue_dma)) {
 487                         dma_unmap_single(dev,
 488                                         adapter->rx_queue.queue_dma,
 489                                         adapter->rx_queue.queue_len,
 490                                         DMA_BIDIRECTIONAL);
 491                         adapter->rx_queue.queue_dma = DMA_ERROR_CODE;
 492                 }
 493                 kfree(adapter->rx_queue.queue_addr);
 494                 adapter->rx_queue.queue_addr = NULL;
 495         }
 496
 497         for(i = 0; i<IbmVethNumBufferPools; i++)
 498                 if (adapter->rx_buff_pool[i].active)
 499                         ibmveth_free_buffer_pool(adapter,
 500                                                  &adapter->rx_buff_pool[i]);
 501
 502         if (adapter->bounce_buffer != NULL) {
 503                 if (!dma_mapping_error(dev, adapter->bounce_buffer_dma)) {
 504                         dma_unmap_single(&adapter->vdev->dev,
 505                                         adapter->bounce_buffer_dma,
 506                                         adapter->netdev->mtu + IBMVETH_BUFF_OH,
 507                                         DMA_BIDIRECTIONAL);
 508                         adapter->bounce_buffer_dma = DMA_ERROR_CODE;
 509                 }
 510                 kfree(adapter->bounce_buffer);
 511                 adapter->bounce_buffer = NULL;
 512         }
 513 }
 514
 515 static int ibmveth_register_logical_lan(struct ibmveth_adapter *adapter,
 516         union ibmveth_buf_desc rxq_desc, u64 mac_address)
 517 {
 518         int rc, try_again = 1;
 519
 520         /* After a kexec the adapter will still be open, so our attempt to
 521         * open it will fail. So if we get a failure we free the adapter and
 522         * try again, but only once. */
 523 retry:
 524         rc = h_register_logical_lan(adapter->vdev->unit_address,
 525                                     adapter->buffer_list_dma, rxq_desc.desc,
 526                                     adapter->filter_list_dma, mac_address);
 527
 528         if (rc != H_SUCCESS && try_again) {
 529                 do {
 530                         rc = h_free_logical_lan(adapter->vdev->unit_address);
 531                 } while (H_IS_LONG_BUSY(rc) || (rc == H_BUSY));
 532
 533                 try_again = 0;
 534                 goto retry;
 535         }
 536
 537         return rc;
 538 }
 539
 540 static int ibmveth_open(struct net_device *netdev)
 541 {
 542         struct ibmveth_adapter *adapter = netdev_priv(netdev);
 543         u64 mac_address = 0;
 544         int rxq_entries = 1;
 545         unsigned long lpar_rc;
 546         int rc;
 547         union ibmveth_buf_desc rxq_desc;
 548         int i;
 549         struct device *dev;
 550
 551         ibmveth_debug_printk("open starting\n");
 552
 553         napi_enable(&adapter->napi);
 554
 555         for(i = 0; i<IbmVethNumBufferPools; i++)
 556                 rxq_entries += adapter->rx_buff_pool[i].size;
 557
 558         adapter->buffer_list_addr = (void*) get_zeroed_page(GFP_KERNEL);
 559         adapter->filter_list_addr = (void*) get_zeroed_page(GFP_KERNEL);
 560
 561         if(!adapter->buffer_list_addr || !adapter->filter_list_addr) {
 562                 ibmveth_error_printk("unable to allocate filter or buffer list pages\n");
 563                 ibmveth_cleanup(adapter);
 564                 napi_disable(&adapter->napi);
 565                 return -ENOMEM;
 566         }
 567
 568         adapter->rx_queue.queue_len = sizeof(struct ibmveth_rx_q_entry) * rxq_entries;
 569         adapter->rx_queue.queue_addr = kmalloc(adapter->rx_queue.queue_len, GFP_KERNEL);
 570
 571         if(!adapter->rx_queue.queue_addr) {
 572                 ibmveth_error_printk("unable to allocate rx queue pages\n");
 573                 ibmveth_cleanup(adapter);
 574                 napi_disable(&adapter->napi);
 575                 return -ENOMEM;
 576         }
 577
 578         dev = &adapter->vdev->dev;
 579
 580         adapter->buffer_list_dma = dma_map_single(dev,
 581                         adapter->buffer_list_addr, 4096, DMA_BIDIRECTIONAL);
 582         adapter->filter_list_dma = dma_map_single(dev,
 583                         adapter->filter_list_addr, 4096, DMA_BIDIRECTIONAL);
 584         adapter->rx_queue.queue_dma = dma_map_single(dev,
 585                         adapter->rx_queue.queue_addr,
 586                         adapter->rx_queue.queue_len, DMA_BIDIRECTIONAL);
 587
 588         if ((dma_mapping_error(dev, adapter->buffer_list_dma)) ||
 589             (dma_mapping_error(dev, adapter->filter_list_dma)) ||
 590             (dma_mapping_error(dev, adapter->rx_queue.queue_dma))) {
 591                 ibmveth_error_printk("unable to map filter or buffer list pages\n");
 592                 ibmveth_cleanup(adapter);
 593                 napi_disable(&adapter->napi);
 594                 return -ENOMEM;
 595         }
 596
 597         adapter->rx_queue.index = 0;
 598         adapter->rx_queue.num_slots = rxq_entries;
 599         adapter->rx_queue.toggle = 1;
 600
 601         memcpy(&mac_address, netdev->dev_addr, netdev->addr_len);
 602         mac_address = mac_address >> 16;
 603
 604         rxq_desc.fields.flags_len = IBMVETH_BUF_VALID | adapter->rx_queue.queue_len;
 605         rxq_desc.fields.address = adapter->rx_queue.queue_dma;
 606
 607         ibmveth_debug_printk("buffer list @ 0x%p\n", adapter->buffer_list_addr);
 608         ibmveth_debug_printk("filter list @ 0x%p\n", adapter->filter_list_addr);
 609         ibmveth_debug_printk("receive q   @ 0x%p\n", adapter->rx_queue.queue_addr);
 610
 611         h_vio_signal(adapter->vdev->unit_address, VIO_IRQ_DISABLE);
 612
 613         lpar_rc = ibmveth_register_logical_lan(adapter, rxq_desc, mac_address);
 614
 615         if(lpar_rc != H_SUCCESS) {
 616                 ibmveth_error_printk("h_register_logical_lan failed with %ld\n", lpar_rc);
 617                 ibmveth_error_printk("buffer TCE:0x%llx filter TCE:0x%llx rxq desc:0x%llx MAC:0x%llx\n",
 618                                      adapter->buffer_list_dma,
 619                                      adapter->filter_list_dma,
 620                                      rxq_desc.desc,
 621                                      mac_address);
 622                 ibmveth_cleanup(adapter);
 623                 napi_disable(&adapter->napi);
 624                 return -ENONET;
 625         }
 626
 627         for(i = 0; i<IbmVethNumBufferPools; i++) {
 628                 if(!adapter->rx_buff_pool[i].active)
 629                         continue;
 630                 if (ibmveth_alloc_buffer_pool(&adapter->rx_buff_pool[i])) {
 631                         ibmveth_error_printk("unable to alloc pool\n");
 632                         adapter->rx_buff_pool[i].active = 0;
 633                         ibmveth_cleanup(adapter);
 634                         napi_disable(&adapter->napi);
 635                         return -ENOMEM ;
 636                 }
 637         }
 638
 639         ibmveth_debug_printk("registering irq 0x%x\n", netdev->irq);
 640         if((rc = request_irq(netdev->irq, ibmveth_interrupt, 0, netdev->name, netdev)) != 0) {
 641                 ibmveth_error_printk("unable to request irq 0x%x, rc %d\n", netdev->irq, rc);
 642                 do {
 643                         rc = h_free_logical_lan(adapter->vdev->unit_address);
 644                 } while (H_IS_LONG_BUSY(rc) || (rc == H_BUSY));
 645
 646                 ibmveth_cleanup(adapter);
 647                 napi_disable(&adapter->napi);
 648                 return rc;
 649         }
 650
 651         adapter->bounce_buffer =
 652             kmalloc(netdev->mtu + IBMVETH_BUFF_OH, GFP_KERNEL);
 653         if (!adapter->bounce_buffer) {
 654                 ibmveth_error_printk("unable to allocate bounce buffer\n");
 655                 ibmveth_cleanup(adapter);
 656                 napi_disable(&adapter->napi);
 657                 return -ENOMEM;
 658         }
 659         adapter->bounce_buffer_dma =
 660             dma_map_single(&adapter->vdev->dev, adapter->bounce_buffer,
 661                            netdev->mtu + IBMVETH_BUFF_OH, DMA_BIDIRECTIONAL);
 662         if (dma_mapping_error(dev, adapter->bounce_buffer_dma)) {
 663                 ibmveth_error_printk("unable to map bounce buffer\n");
 664                 ibmveth_cleanup(adapter);
 665                 napi_disable(&adapter->napi);
 666                 return -ENOMEM;
 667         }
 668
 669         ibmveth_debug_printk("initial replenish cycle\n");
 670         ibmveth_interrupt(netdev->irq, netdev);
 671
 672         netif_start_queue(netdev);
 673
 674         ibmveth_debug_printk("open complete\n");
 675
 676         return 0;
 677 }
 678
 679 static int ibmveth_close(struct net_device *netdev)
 680 {
 681         struct ibmveth_adapter *adapter = netdev_priv(netdev);
 682         long lpar_rc;
 683
 684         ibmveth_debug_printk("close starting\n");
 685
 686         napi_disable(&adapter->napi);
 687
 688         if (!adapter->pool_config)
 689                 netif_stop_queue(netdev);
 690
 691         h_vio_signal(adapter->vdev->unit_address, VIO_IRQ_DISABLE);
 692
 693         do {
 694                 lpar_rc = h_free_logical_lan(adapter->vdev->unit_address);
 695         } while (H_IS_LONG_BUSY(lpar_rc) || (lpar_rc == H_BUSY));
 696
 697         if(lpar_rc != H_SUCCESS)
 698         {
 699                 ibmveth_error_printk("h_free_logical_lan failed with %lx, continuing with close\n",
 700                                      lpar_rc);
 701         }
 702
 703         free_irq(netdev->irq, netdev);
 704
 705         adapter->rx_no_buffer = *(u64*)(((char*)adapter->buffer_list_addr) + 4096 - 8);
 706
 707         ibmveth_cleanup(adapter);
 708
 709         ibmveth_debug_printk("close complete\n");
 710
 711         return 0;
 712 }
 713
 714 static int netdev_get_settings(struct net_device *dev, struct ethtool_cmd *cmd) {
 715         cmd->supported = (SUPPORTED_1000baseT_Full | SUPPORTED_Autoneg | SUPPORTED_FIBRE);
 716         cmd->advertising = (ADVERTISED_1000baseT_Full | ADVERTISED_Autoneg | ADVERTISED_FIBRE);
 717         cmd->speed = SPEED_1000;
 718         cmd->duplex = DUPLEX_FULL;
 719         cmd->port = PORT_FIBRE;
 720         cmd->phy_address = 0;
 721         cmd->transceiver = XCVR_INTERNAL;
 722         cmd->autoneg = AUTONEG_ENABLE;
 723         cmd->maxtxpkt = 0;
 724         cmd->maxrxpkt = 1;
 725         return 0;
 726 }
 727
 728 static void netdev_get_drvinfo (struct net_device *dev, struct ethtool_drvinfo *info) {
 729         strncpy(info->driver, ibmveth_driver_name, sizeof(info->driver) - 1);
 730         strncpy(info->version, ibmveth_driver_version, sizeof(info->version) - 1);
 731 }
 732
 733 static u32 netdev_get_link(struct net_device *dev) {
 734         return 1;
 735 }
 736
 737 static void ibmveth_set_rx_csum_flags(struct net_device *dev, u32 data)
 738 {
 739         struct ibmveth_adapter *adapter = netdev_priv(dev);
 740
 741         if (data)
 742                 adapter->rx_csum = 1;
 743         else {
 744                 /*
 745                  * Since the ibmveth firmware interface does not have the concept of
 746                  * separate tx/rx checksum offload enable, if rx checksum is disabled
 747                  * we also have to disable tx checksum offload. Once we disable rx
 748                  * checksum offload, we are no longer allowed to send tx buffers that
 749                  * are not properly checksummed.
 750                  */
 751                 adapter->rx_csum = 0;
 752                 dev->features &= ~NETIF_F_IP_CSUM;
 753         }
 754 }
 755
 756 static void ibmveth_set_tx_csum_flags(struct net_device *dev, u32 data)
 757 {
 758         struct ibmveth_adapter *adapter = netdev_priv(dev);
 759
 760         if (data) {
 761                 dev->features |= NETIF_F_IP_CSUM;
 762                 adapter->rx_csum = 1;
 763         } else
 764                 dev->features &= ~NETIF_F_IP_CSUM;
 765 }
 766
 767 static int ibmveth_set_csum_offload(struct net_device *dev, u32 data,
 768                                     void (*done) (struct net_device *, u32))
 769 {
 770         struct ibmveth_adapter *adapter = netdev_priv(dev);
 771         unsigned long set_attr, clr_attr, ret_attr;
 772         long ret;
 773         int rc1 = 0, rc2 = 0;
 774         int restart = 0;
 775
 776         if (netif_running(dev)) {
 777                 restart = 1;
 778                 adapter->pool_config = 1;
 779                 ibmveth_close(dev);
 780                 adapter->pool_config = 0;
 781         }
 782
 783         set_attr = 0;
 784         clr_attr = 0;
 785
 786         if (data)
 787                 set_attr = IBMVETH_ILLAN_IPV4_TCP_CSUM;
 788         else
 789                 clr_attr = IBMVETH_ILLAN_IPV4_TCP_CSUM;
 790
 791         ret = h_illan_attributes(adapter->vdev->unit_address, 0, 0, &ret_attr);
 792
 793         if (ret == H_SUCCESS && !(ret_attr & IBMVETH_ILLAN_ACTIVE_TRUNK) &&
 794             !(ret_attr & IBMVETH_ILLAN_TRUNK_PRI_MASK) &&
 795             (ret_attr & IBMVETH_ILLAN_PADDED_PKT_CSUM)) {
 796                 ret = h_illan_attributes(adapter->vdev->unit_address, clr_attr,
 797                                          set_attr, &ret_attr);
 798
 799                 if (ret != H_SUCCESS) {
 800                         rc1 = -EIO;
 801                         ibmveth_error_printk("unable to change checksum offload settings."
 802                                              " %d rc=%ld\n", data, ret);
 803
 804                         ret = h_illan_attributes(adapter->vdev->unit_address,
 805                                                  set_attr, clr_attr, &ret_attr);
 806                 } else
 807                         done(dev, data);
 808         } else {
 809                 rc1 = -EIO;
 810                 ibmveth_error_printk("unable to change checksum offload settings."
 811                                      " %d rc=%ld ret_attr=%lx\n", data, ret, ret_attr);
 812         }
 813
 814         if (restart)
 815                 rc2 = ibmveth_open(dev);
 816
 817         return rc1 ? rc1 : rc2;
 818 }
 819
 820 static int ibmveth_set_rx_csum(struct net_device *dev, u32 data)
 821 {
 822         struct ibmveth_adapter *adapter = netdev_priv(dev);
 823
 824         if ((data && adapter->rx_csum) || (!data && !adapter->rx_csum))
 825                 return 0;
 826
 827         return ibmveth_set_csum_offload(dev, data, ibmveth_set_rx_csum_flags);
 828 }
 829
 830 static int ibmveth_set_tx_csum(struct net_device *dev, u32 data)
 831 {
 832         struct ibmveth_adapter *adapter = netdev_priv(dev);
 833         int rc = 0;
 834
 835         if (data && (dev->features & NETIF_F_IP_CSUM))
 836                 return 0;
 837         if (!data && !(dev->features & NETIF_F_IP_CSUM))
 838                 return 0;
 839
 840         if (data && !adapter->rx_csum)
 841                 rc = ibmveth_set_csum_offload(dev, data, ibmveth_set_tx_csum_flags);
 842         else
 843                 ibmveth_set_tx_csum_flags(dev, data);
 844
 845         return rc;
 846 }
 847
 848 static u32 ibmveth_get_rx_csum(struct net_device *dev)
 849 {
 850         struct ibmveth_adapter *adapter = netdev_priv(dev);
 851         return adapter->rx_csum;
 852 }
 853
 854 static void ibmveth_get_strings(struct net_device *dev, u32 stringset, u8 *data)
 855 {
 856         int i;
 857
 858         if (stringset != ETH_SS_STATS)
 859                 return;
 860
 861         for (i = 0; i < ARRAY_SIZE(ibmveth_stats); i++, data += ETH_GSTRING_LEN)
 862                 memcpy(data, ibmveth_stats[i].name, ETH_GSTRING_LEN);
 863 }
 864
 865 static int ibmveth_get_sset_count(struct net_device *dev, int sset)
 866 {
 867         switch (sset) {
 868         case ETH_SS_STATS:
 869                 return ARRAY_SIZE(ibmveth_stats);
 870         default:
 871                 return -EOPNOTSUPP;
 872         }
 873 }
 874
 875 static void ibmveth_get_ethtool_stats(struct net_device *dev,
 876                                       struct ethtool_stats *stats, u64 *data)
 877 {
 878         int i;
 879         struct ibmveth_adapter *adapter = netdev_priv(dev);
 880
 881         for (i = 0; i < ARRAY_SIZE(ibmveth_stats); i++)
 882                 data[i] = IBMVETH_GET_STAT(adapter, ibmveth_stats[i].offset);
 883 }
 884
 885 static const struct ethtool_ops netdev_ethtool_ops = {
 886         .get_drvinfo            = netdev_get_drvinfo,
 887         .get_settings           = netdev_get_settings,
 888         .get_link               = netdev_get_link,
 889         .set_tx_csum            = ibmveth_set_tx_csum,
 890         .get_rx_csum            = ibmveth_get_rx_csum,
 891         .set_rx_csum            = ibmveth_set_rx_csum,
 892         .get_strings            = ibmveth_get_strings,
 893         .get_sset_count         = ibmveth_get_sset_count,
 894         .get_ethtool_stats      = ibmveth_get_ethtool_stats,
 895 };
 896
 897 static int ibmveth_ioctl(struct net_device *dev, struct ifreq *ifr, int cmd)
 898 {
 899         return -EOPNOTSUPP;
 900 }
 901
 902 #define page_offset(v) ((unsigned long)(v) & ((1 << 12) - 1))
 903
 904 static netdev_tx_t ibmveth_start_xmit(struct sk_buff *skb,
 905                                       struct net_device *netdev)
 906 {
 907         struct ibmveth_adapter *adapter = netdev_priv(netdev);
 908         union ibmveth_buf_desc desc;
 909         unsigned long lpar_rc;
 910         unsigned long correlator;
 911         unsigned int retry_count;
 912         unsigned int tx_dropped = 0;
 913         unsigned int tx_bytes = 0;
 914         unsigned int tx_packets = 0;
 915         unsigned int tx_send_failed = 0;
 916         unsigned int tx_map_failed = 0;
 917         int used_bounce = 0;
 918         unsigned long data_dma_addr;
 919
 920         desc.fields.flags_len = IBMVETH_BUF_VALID | skb->len;
 921
 922         if (skb->ip_summed == CHECKSUM_PARTIAL &&
 923             ip_hdr(skb)->protocol != IPPROTO_TCP && skb_checksum_help(skb)) {
 924                 ibmveth_error_printk("tx: failed to checksum packet\n");
 925                 tx_dropped++;
 926                 goto out;
 927         }
 928
 929         if (skb->ip_summed == CHECKSUM_PARTIAL) {
 930                 unsigned char *buf = skb_transport_header(skb) + skb->csum_offset;
 931
 932                 desc.fields.flags_len |= (IBMVETH_BUF_NO_CSUM | IBMVETH_BUF_CSUM_GOOD);
 933
 934                 /* Need to zero out the checksum */
 935                 buf[0] = 0;
 936                 buf[1] = 0;
 937         }
 938
 939         if (skb->len < tx_copybreak) {
 940                 used_bounce = 1;
 941         } else {
 942                 data_dma_addr = dma_map_single(&adapter->vdev->dev, skb->data,
 943                                                skb->len, DMA_TO_DEVICE);
 944                 if (dma_mapping_error(&adapter->vdev->dev, data_dma_addr)) {
 945                         if (!firmware_has_feature(FW_FEATURE_CMO))
 946                                 ibmveth_error_printk("tx: unable to map "
 947                                                      "xmit buffer\n");
 948                         tx_map_failed++;
 949                         used_bounce = 1;
 950                 }
 951         }
 952
 953         if (used_bounce) {
 954                 skb_copy_from_linear_data(skb, adapter->bounce_buffer,
 955                                           skb->len);
 956                 desc.fields.address = adapter->bounce_buffer_dma;
 957         } else
 958                 desc.fields.address = data_dma_addr;
 959
 960         /* send the frame. Arbitrarily set retrycount to 1024 */
 961         correlator = 0;
 962         retry_count = 1024;
 963         do {
 964                 lpar_rc = h_send_logical_lan(adapter->vdev->unit_address,
 965                                              desc.desc, 0, 0, 0, 0, 0,
 966                                              correlator, &correlator);
 967         } while ((lpar_rc == H_BUSY) && (retry_count--));
 968
 969         if(lpar_rc != H_SUCCESS && lpar_rc != H_DROPPED) {
 970                 ibmveth_error_printk("tx: h_send_logical_lan failed with rc=%ld\n", lpar_rc);
 971                 ibmveth_error_printk("tx: valid=%d, len=%d, address=0x%08x\n",
 972                                      (desc.fields.flags_len & IBMVETH_BUF_VALID) ? 1 : 0,
 973                                      skb->len, desc.fields.address);
 974                 tx_send_failed++;
 975                 tx_dropped++;
 976         } else {
 977                 tx_packets++;
 978                 tx_bytes += skb->len;
 979         }
 980
 981         if (!used_bounce)
 982                 dma_unmap_single(&adapter->vdev->dev, data_dma_addr,
 983                                  skb->len, DMA_TO_DEVICE);
 984
 985 out:
 986         netdev->stats.tx_dropped += tx_dropped;
 987         netdev->stats.tx_bytes += tx_bytes;
 988         netdev->stats.tx_packets += tx_packets;
 989         adapter->tx_send_failed += tx_send_failed;
 990         adapter->tx_map_failed += tx_map_failed;
 991
 992         dev_kfree_skb(skb);
 993         return NETDEV_TX_OK;
 994 }
 995
 996 static int ibmveth_poll(struct napi_struct *napi, int budget)
 997 {
 998         struct ibmveth_adapter *adapter = container_of(napi, struct ibmveth_adapter, napi);
 999         struct net_device *netdev = adapter->netdev;
1000         int frames_processed = 0;
1001         unsigned long lpar_rc;
1002
1003  restart_poll:
1004         do {
1005                 struct sk_buff *skb;
1006
1007                 if (!ibmveth_rxq_pending_buffer(adapter))
1008                         break;
1009
1010                 rmb();
1011                 if (!ibmveth_rxq_buffer_valid(adapter)) {
1012                         wmb(); /* suggested by larson1 */
1013                         adapter->rx_invalid_buffer++;
1014                         ibmveth_debug_printk("recycling invalid buffer\n");
1015                         ibmveth_rxq_recycle_buffer(adapter);
1016                 } else {
1017                         int length = ibmveth_rxq_frame_length(adapter);
1018                         int offset = ibmveth_rxq_frame_offset(adapter);
1019                         int csum_good = ibmveth_rxq_csum_good(adapter);
1020
1021                         skb = ibmveth_rxq_get_buffer(adapter);
1022                         if (csum_good)
1023                                 skb->ip_summed = CHECKSUM_UNNECESSARY;
1024
1025                         ibmveth_rxq_harvest_buffer(adapter);
1026
1027                         skb_reserve(skb, offset);
1028                         skb_put(skb, length);
1029                         skb->protocol = eth_type_trans(skb, netdev);
1030
1031                         netif_receive_skb(skb); /* send it up */
1032
1033                         netdev->stats.rx_packets++;
1034                         netdev->stats.rx_bytes += length;
1035                         frames_processed++;
1036                 }
1037         } while (frames_processed < budget);
1038
1039         ibmveth_replenish_task(adapter);
1040
1041         if (frames_processed < budget) {
1042                 /* We think we are done - reenable interrupts,
1043                  * then check once more to make sure we are done.
1044                  */
1045                 lpar_rc = h_vio_signal(adapter->vdev->unit_address,
1046                                        VIO_IRQ_ENABLE);
1047
1048                 ibmveth_assert(lpar_rc == H_SUCCESS);
1049
1050                 napi_complete(napi);
1051
1052                 if (ibmveth_rxq_pending_buffer(adapter) &&
1053                     napi_reschedule(napi)) {
1054                         lpar_rc = h_vio_signal(adapter->vdev->unit_address,
1055                                                VIO_IRQ_DISABLE);
1056                         goto restart_poll;
1057                 }
1058         }
1059
1060         return frames_processed;
1061 }
1062
1063 static irqreturn_t ibmveth_interrupt(int irq, void *dev_instance)
1064 {
1065         struct net_device *netdev = dev_instance;
1066         struct ibmveth_adapter *adapter = netdev_priv(netdev);
1067         unsigned long lpar_rc;
1068
1069         if (napi_schedule_prep(&adapter->napi)) {
1070                 lpar_rc = h_vio_signal(adapter->vdev->unit_address,
1071                                        VIO_IRQ_DISABLE);
1072                 ibmveth_assert(lpar_rc == H_SUCCESS);
1073                 __napi_schedule(&adapter->napi);
1074         }
1075         return IRQ_HANDLED;
1076 }
1077
1078 static void ibmveth_set_multicast_list(struct net_device *netdev)
1079 {
1080         struct ibmveth_adapter *adapter = netdev_priv(netdev);
1081         unsigned long lpar_rc;
1082
1083         if ((netdev->flags & IFF_PROMISC) ||
1084             (netdev_mc_count(netdev) > adapter->mcastFilterSize)) {
1085                 lpar_rc = h_multicast_ctrl(adapter->vdev->unit_address,
1086                                            IbmVethMcastEnableRecv |
1087                                            IbmVethMcastDisableFiltering,
1088                                            0);
1089                 if(lpar_rc != H_SUCCESS) {
1090                         ibmveth_error_printk("h_multicast_ctrl rc=%ld when entering promisc mode\n", lpar_rc);
1091                 }
1092         } else {
1093                 struct netdev_hw_addr *ha;
1094                 /* clear the filter table & disable filtering */
1095                 lpar_rc = h_multicast_ctrl(adapter->vdev->unit_address,
1096                                            IbmVethMcastEnableRecv |
1097                                            IbmVethMcastDisableFiltering |
1098                                            IbmVethMcastClearFilterTable,
1099                                            0);
1100                 if(lpar_rc != H_SUCCESS) {
1101                         ibmveth_error_printk("h_multicast_ctrl rc=%ld when attempting to clear filter table\n", lpar_rc);
1102                 }
1103                 /* add the addresses to the filter table */
1104                 netdev_for_each_mc_addr(ha, netdev) {
1105                         // add the multicast address to the filter table
1106                         unsigned long mcast_addr = 0;
1107                         memcpy(((char *)&mcast_addr)+2, ha->addr, 6);
1108                         lpar_rc = h_multicast_ctrl(adapter->vdev->unit_address,
1109                                                    IbmVethMcastAddFilter,
1110                                                    mcast_addr);
1111                         if(lpar_rc != H_SUCCESS) {
1112                                 ibmveth_error_printk("h_multicast_ctrl rc=%ld when adding an entry to the filter table\n", lpar_rc);
1113                         }
1114                 }
1115
1116                 /* re-enable filtering */
1117                 lpar_rc = h_multicast_ctrl(adapter->vdev->unit_address,
1118                                            IbmVethMcastEnableFiltering,
1119                                            0);
1120                 if(lpar_rc != H_SUCCESS) {
1121                         ibmveth_error_printk("h_multicast_ctrl rc=%ld when enabling filtering\n", lpar_rc);
1122                 }
1123         }
1124 }
1125
1126 static int ibmveth_change_mtu(struct net_device *dev, int new_mtu)
1127 {
1128         struct ibmveth_adapter *adapter = netdev_priv(dev);
1129         struct vio_dev *viodev = adapter->vdev;
1130         int new_mtu_oh = new_mtu + IBMVETH_BUFF_OH;
1131         int i, rc;
1132         int need_restart = 0;
1133
1134         if (new_mtu < IBMVETH_MAX_MTU)
1135                 return -EINVAL;
1136
1137         for (i = 0; i < IbmVethNumBufferPools; i++)
1138                 if (new_mtu_oh < adapter->rx_buff_pool[i].buff_size)
1139                         break;
1140
1141         if (i == IbmVethNumBufferPools)
1142                 return -EINVAL;
1143
1144         /* Deactivate all the buffer pools so that the next loop can activate
1145            only the buffer pools necessary to hold the new MTU */
1146         if (netif_running(adapter->netdev)) {
1147                 need_restart = 1;
1148                 adapter->pool_config = 1;
1149                 ibmveth_close(adapter->netdev);
1150                 adapter->pool_config = 0;
1151         }
1152
1153         /* Look for an active buffer pool that can hold the new MTU */
1154         for(i = 0; i<IbmVethNumBufferPools; i++) {
1155                 adapter->rx_buff_pool[i].active = 1;
1156
1157                 if (new_mtu_oh < adapter->rx_buff_pool[i].buff_size) {
1158                         dev->mtu = new_mtu;
1159                         vio_cmo_set_dev_desired(viodev,
1160                                                 ibmveth_get_desired_dma
1161                                                 (viodev));
1162                         if (need_restart) {
1163                                 return ibmveth_open(adapter->netdev);
1164                         }
1165                         return 0;
1166                 }
1167         }
1168
1169         if (need_restart && (rc = ibmveth_open(adapter->netdev)))
1170                 return rc;
1171
1172         return -EINVAL;
1173 }
1174
1175 #ifdef CONFIG_NET_POLL_CONTROLLER
1176 static void ibmveth_poll_controller(struct net_device *dev)
1177 {
1178         ibmveth_replenish_task(netdev_priv(dev));
1179         ibmveth_interrupt(dev->irq, dev);
1180 }
1181 #endif
1182
1183 /**
1184  * ibmveth_get_desired_dma - Calculate IO memory desired by the driver
1185  *
1186  * @vdev: struct vio_dev for the device whose desired IO mem is to be returned
1187  *
1188  * Return value:
1189  *      Number of bytes of IO data the driver will need to perform well.
1190  */
1191 static unsigned long ibmveth_get_desired_dma(struct vio_dev *vdev)
1192 {
1193         struct net_device *netdev = dev_get_drvdata(&vdev->dev);
1194         struct ibmveth_adapter *adapter;
1195         unsigned long ret;
1196         int i;
1197         int rxqentries = 1;
1198
1199         /* netdev inits at probe time along with the structures we need below*/
1200         if (netdev == NULL)
1201                 return IOMMU_PAGE_ALIGN(IBMVETH_IO_ENTITLEMENT_DEFAULT);
1202
1203         adapter = netdev_priv(netdev);
1204
1205         ret = IBMVETH_BUFF_LIST_SIZE + IBMVETH_FILT_LIST_SIZE;
1206         ret += IOMMU_PAGE_ALIGN(netdev->mtu);
1207
1208         for (i = 0; i < IbmVethNumBufferPools; i++) {
1209                 /* add the size of the active receive buffers */
1210                 if (adapter->rx_buff_pool[i].active)
1211                         ret +=
1212                             adapter->rx_buff_pool[i].size *
1213                             IOMMU_PAGE_ALIGN(adapter->rx_buff_pool[i].
1214                                     buff_size);
1215                 rxqentries += adapter->rx_buff_pool[i].size;
1216         }
1217         /* add the size of the receive queue entries */
1218         ret += IOMMU_PAGE_ALIGN(rxqentries * sizeof(struct ibmveth_rx_q_entry));
1219
1220         return ret;
1221 }
1222
1223 static const struct net_device_ops ibmveth_netdev_ops = {
1224         .ndo_open               = ibmveth_open,
1225         .ndo_stop               = ibmveth_close,
1226         .ndo_start_xmit         = ibmveth_start_xmit,
1227         .ndo_set_multicast_list = ibmveth_set_multicast_list,
1228         .ndo_do_ioctl           = ibmveth_ioctl,
1229         .ndo_change_mtu         = ibmveth_change_mtu,
1230         .ndo_validate_addr      = eth_validate_addr,
1231         .ndo_set_mac_address    = eth_mac_addr,
1232 #ifdef CONFIG_NET_POLL_CONTROLLER
1233         .ndo_poll_controller    = ibmveth_poll_controller,
1234 #endif
1235 };
1236
1237 static int __devinit ibmveth_probe(struct vio_dev *dev, const struct vio_device_id *id)
1238 {
1239         int rc, i;
1240         long ret;
1241         struct net_device *netdev;
1242         struct ibmveth_adapter *adapter;
1243         unsigned long set_attr, ret_attr;
1244
1245         unsigned char *mac_addr_p;
1246         unsigned int *mcastFilterSize_p;
1247
1248
1249         ibmveth_debug_printk_no_adapter("entering ibmveth_probe for UA 0x%x\n",
1250                                         dev->unit_address);
1251
1252         mac_addr_p = (unsigned char *) vio_get_attribute(dev,
1253                                                 VETH_MAC_ADDR, NULL);
1254         if(!mac_addr_p) {
1255                 printk(KERN_ERR "(%s:%3.3d) ERROR: Can't find VETH_MAC_ADDR "
1256                                 "attribute\n", __FILE__, __LINE__);
1257                 return 0;
1258         }
1259
1260         mcastFilterSize_p = (unsigned int *) vio_get_attribute(dev,
1261                                                 VETH_MCAST_FILTER_SIZE, NULL);
1262         if(!mcastFilterSize_p) {
1263                 printk(KERN_ERR "(%s:%3.3d) ERROR: Can't find "
1264                                 "VETH_MCAST_FILTER_SIZE attribute\n",
1265                                 __FILE__, __LINE__);
1266                 return 0;
1267         }
1268
1269         netdev = alloc_etherdev(sizeof(struct ibmveth_adapter));
1270
1271         if(!netdev)
1272                 return -ENOMEM;
1273
1274         adapter = netdev_priv(netdev);
1275         dev_set_drvdata(&dev->dev, netdev);
1276
1277         adapter->vdev = dev;
1278         adapter->netdev = netdev;
1279         adapter->mcastFilterSize= *mcastFilterSize_p;
1280         adapter->pool_config = 0;
1281
1282         netif_napi_add(netdev, &adapter->napi, ibmveth_poll, 16);
1283
1284         /*      Some older boxes running PHYP non-natively have an OF that
1285                 returns a 8-byte local-mac-address field (and the first
1286                 2 bytes have to be ignored) while newer boxes' OF return
1287                 a 6-byte field. Note that IEEE 1275 specifies that
1288                 local-mac-address must be a 6-byte field.
1289                 The RPA doc specifies that the first byte must be 10b, so
1290                 we'll just look for it to solve this 8 vs. 6 byte field issue */
1291
1292         if ((*mac_addr_p & 0x3) != 0x02)
1293                 mac_addr_p += 2;
1294
1295         adapter->mac_addr = 0;
1296         memcpy(&adapter->mac_addr, mac_addr_p, 6);
1297
1298         netdev->irq = dev->irq;
1299         netdev->netdev_ops = &ibmveth_netdev_ops;
1300         netdev->ethtool_ops = &netdev_ethtool_ops;
1301         SET_NETDEV_DEV(netdev, &dev->dev);
1302
1303         memcpy(netdev->dev_addr, &adapter->mac_addr, netdev->addr_len);
1304
1305         for(i = 0; i<IbmVethNumBufferPools; i++) {
1306                 struct kobject *kobj = &adapter->rx_buff_pool[i].kobj;
1307                 int error;
1308
1309                 ibmveth_init_buffer_pool(&adapter->rx_buff_pool[i], i,
1310                                          pool_count[i], pool_size[i],
1311                                          pool_active[i]);
1312                 error = kobject_init_and_add(kobj, &ktype_veth_pool,
1313                                              &dev->dev.kobj, "pool%d", i);
1314                 if (!error)
1315                         kobject_uevent(kobj, KOBJ_ADD);
1316         }
1317
1318         ibmveth_debug_printk("adapter @ 0x%p\n", adapter);
1319
1320         adapter->buffer_list_dma = DMA_ERROR_CODE;
1321         adapter->filter_list_dma = DMA_ERROR_CODE;
1322         adapter->rx_queue.queue_dma = DMA_ERROR_CODE;
1323
1324         ibmveth_debug_printk("registering netdev...\n");
1325
1326         ret = h_illan_attributes(dev->unit_address, 0, 0, &ret_attr);
1327
1328         if (ret == H_SUCCESS && !(ret_attr & IBMVETH_ILLAN_ACTIVE_TRUNK) &&
1329             !(ret_attr & IBMVETH_ILLAN_TRUNK_PRI_MASK) &&
1330             (ret_attr & IBMVETH_ILLAN_PADDED_PKT_CSUM)) {
1331                 set_attr = IBMVETH_ILLAN_IPV4_TCP_CSUM;
1332
1333                 ret = h_illan_attributes(dev->unit_address, 0, set_attr, &ret_attr);
1334
1335                 if (ret == H_SUCCESS) {
1336                         adapter->rx_csum = 1;
1337                         netdev->features |= NETIF_F_IP_CSUM;
1338                 } else
1339                         ret = h_illan_attributes(dev->unit_address, set_attr, 0, &ret_attr);
1340         }
1341
1342         rc = register_netdev(netdev);
1343
1344         if(rc) {
1345                 ibmveth_debug_printk("failed to register netdev rc=%d\n", rc);
1346                 free_netdev(netdev);
1347                 return rc;
1348         }
1349
1350         ibmveth_debug_printk("registered\n");
1351
1352         ibmveth_proc_register_adapter(adapter);
1353
1354         return 0;
1355 }
1356
1357 static int __devexit ibmveth_remove(struct vio_dev *dev)
1358 {
1359         struct net_device *netdev = dev_get_drvdata(&dev->dev);
1360         struct ibmveth_adapter *adapter = netdev_priv(netdev);
1361         int i;
1362
1363         for(i = 0; i<IbmVethNumBufferPools; i++)
1364                 kobject_put(&adapter->rx_buff_pool[i].kobj);
1365
1366         unregister_netdev(netdev);
1367
1368         ibmveth_proc_unregister_adapter(adapter);
1369
1370         free_netdev(netdev);
1371         dev_set_drvdata(&dev->dev, NULL);
1372
1373         return 0;
1374 }
1375
1376 #ifdef CONFIG_PROC_FS
1377 static void ibmveth_proc_register_driver(void)
1378 {
1379         ibmveth_proc_dir = proc_mkdir(IBMVETH_PROC_DIR, init_net.proc_net);
1380         if (ibmveth_proc_dir) {
1381         }
1382 }
1383
1384 static void ibmveth_proc_unregister_driver(void)
1385 {
1386         remove_proc_entry(IBMVETH_PROC_DIR, init_net.proc_net);
1387 }
1388
1389 static int ibmveth_show(struct seq_file *seq, void *v)
1390 {
1391         struct ibmveth_adapter *adapter = seq->private;
1392         char *current_mac = (char *) adapter->netdev->dev_addr;
1393         char *firmware_mac = (char *) &adapter->mac_addr;
1394
1395         seq_printf(seq, "%s %s\n\n", ibmveth_driver_string, ibmveth_driver_version);
1396
1397         seq_printf(seq, "Unit Address:    0x%x\n", adapter->vdev->unit_address);
1398         seq_printf(seq, "Current MAC:     %pM\n", current_mac);
1399         seq_printf(seq, "Firmware MAC:    %pM\n", firmware_mac);
1400
1401         seq_printf(seq, "\nAdapter Statistics:\n");
1402         seq_printf(seq, "  TX:  vio_map_single failres:      %lld\n", adapter->tx_map_failed);
1403         seq_printf(seq, "       send failures:               %lld\n", adapter->tx_send_failed);
1404         seq_printf(seq, "  RX:  replenish task cycles:       %lld\n", adapter->replenish_task_cycles);
1405         seq_printf(seq, "       alloc_skb_failures:          %lld\n", adapter->replenish_no_mem);
1406         seq_printf(seq, "       add buffer failures:         %lld\n", adapter->replenish_add_buff_failure);
1407         seq_printf(seq, "       invalid buffers:             %lld\n", adapter->rx_invalid_buffer);
1408         seq_printf(seq, "       no buffers:                  %lld\n", adapter->rx_no_buffer);
1409
1410         return 0;
1411 }
1412
1413 static int ibmveth_proc_open(struct inode *inode, struct file *file)
1414 {
1415         return single_open(file, ibmveth_show, PDE(inode)->data);
1416 }
1417
1418 static const struct file_operations ibmveth_proc_fops = {
1419         .owner   = THIS_MODULE,
1420         .open    = ibmveth_proc_open,
1421         .read    = seq_read,
1422         .llseek  = seq_lseek,
1423         .release = single_release,
1424 };
1425
1426 static void ibmveth_proc_register_adapter(struct ibmveth_adapter *adapter)
1427 {
1428         struct proc_dir_entry *entry;
1429         if (ibmveth_proc_dir) {
1430                 char u_addr[10];
1431                 sprintf(u_addr, "%x", adapter->vdev->unit_address);
1432                 entry = proc_create_data(u_addr, S_IFREG, ibmveth_proc_dir,
1433                                          &ibmveth_proc_fops, adapter);
1434                 if (!entry)
1435                         ibmveth_error_printk("Cannot create adapter proc entry");
1436         }
1437 }
1438
1439 static void ibmveth_proc_unregister_adapter(struct ibmveth_adapter *adapter)
1440 {
1441         if (ibmveth_proc_dir) {
1442                 char u_addr[10];
1443                 sprintf(u_addr, "%x", adapter->vdev->unit_address);
1444                 remove_proc_entry(u_addr, ibmveth_proc_dir);
1445         }
1446 }
1447
1448 #else /* CONFIG_PROC_FS */
1449 static void ibmveth_proc_register_adapter(struct ibmveth_adapter *adapter)
1450 {
1451 }
1452
1453 static void ibmveth_proc_unregister_adapter(struct ibmveth_adapter *adapter)
1454 {
1455 }
1456 static void ibmveth_proc_register_driver(void)
1457 {
1458 }
1459
1460 static void ibmveth_proc_unregister_driver(void)
1461 {
1462 }
1463 #endif /* CONFIG_PROC_FS */
1464
1465 static struct attribute veth_active_attr;
1466 static struct attribute veth_num_attr;
1467 static struct attribute veth_size_attr;
1468
1469 static ssize_t veth_pool_show(struct kobject * kobj,
1470                               struct attribute * attr, char * buf)
1471 {
1472         struct ibmveth_buff_pool *pool = container_of(kobj,
1473                                                       struct ibmveth_buff_pool,
1474                                                       kobj);
1475
1476         if (attr == &veth_active_attr)
1477                 return sprintf(buf, "%d\n", pool->active);
1478         else if (attr == &veth_num_attr)
1479                 return sprintf(buf, "%d\n", pool->size);
1480         else if (attr == &veth_size_attr)
1481                 return sprintf(buf, "%d\n", pool->buff_size);
1482         return 0;
1483 }
1484
1485 static ssize_t veth_pool_store(struct kobject * kobj, struct attribute * attr,
1486 const char * buf, size_t count)
1487 {
1488         struct ibmveth_buff_pool *pool = container_of(kobj,
1489                                                       struct ibmveth_buff_pool,
1490                                                       kobj);
1491         struct net_device *netdev = dev_get_drvdata(
1492             container_of(kobj->parent, struct device, kobj));
1493         struct ibmveth_adapter *adapter = netdev_priv(netdev);
1494         long value = simple_strtol(buf, NULL, 10);
1495         long rc;
1496
1497         if (attr == &veth_active_attr) {
1498                 if (value && !pool->active) {
1499                         if (netif_running(netdev)) {
1500                                 if(ibmveth_alloc_buffer_pool(pool)) {
1501                                         ibmveth_error_printk("unable to alloc pool\n");
1502                                         return -ENOMEM;
1503                                 }
1504                                 pool->active = 1;
1505                                 adapter->pool_config = 1;
1506                                 ibmveth_close(netdev);
1507                                 adapter->pool_config = 0;
1508                                 if ((rc = ibmveth_open(netdev)))
1509                                         return rc;
1510                         } else
1511                                 pool->active = 1;
1512                 } else if (!value && pool->active) {
1513                         int mtu = netdev->mtu + IBMVETH_BUFF_OH;
1514                         int i;
1515                         /* Make sure there is a buffer pool with buffers that
1516                            can hold a packet of the size of the MTU */
1517                         for (i = 0; i < IbmVethNumBufferPools; i++) {
1518                                 if (pool == &adapter->rx_buff_pool[i])
1519                                         continue;
1520                                 if (!adapter->rx_buff_pool[i].active)
1521                                         continue;
1522                                 if (mtu <= adapter->rx_buff_pool[i].buff_size)
1523                                         break;
1524                         }
1525
1526                         if (i == IbmVethNumBufferPools) {
1527                                 ibmveth_error_printk("no active pool >= MTU\n");
1528                                 return -EPERM;
1529                         }
1530
1531                         if (netif_running(netdev)) {
1532                                 adapter->pool_config = 1;
1533                                 ibmveth_close(netdev);
1534                                 pool->active = 0;
1535                                 adapter->pool_config = 0;
1536                                 if ((rc = ibmveth_open(netdev)))
1537                                         return rc;
1538                         }
1539                         pool->active = 0;
1540                 }
1541         } else if (attr == &veth_num_attr) {
1542                 if (value <= 0 || value > IBMVETH_MAX_POOL_COUNT)
1543                         return -EINVAL;
1544                 else {
1545                         if (netif_running(netdev)) {
1546                                 adapter->pool_config = 1;
1547                                 ibmveth_close(netdev);
1548                                 adapter->pool_config = 0;
1549                                 pool->size = value;
1550                                 if ((rc = ibmveth_open(netdev)))
1551                                         return rc;
1552                         } else
1553                                 pool->size = value;
1554                 }
1555         } else if (attr == &veth_size_attr) {
1556                 if (value <= IBMVETH_BUFF_OH || value > IBMVETH_MAX_BUF_SIZE)
1557                         return -EINVAL;
1558                 else {
1559                         if (netif_running(netdev)) {
1560                                 adapter->pool_config = 1;
1561                                 ibmveth_close(netdev);
1562                                 adapter->pool_config = 0;
1563                                 pool->buff_size = value;
1564                                 if ((rc = ibmveth_open(netdev)))
1565                                         return rc;
1566                         } else
1567                                 pool->buff_size = value;
1568                 }
1569         }
1570
1571         /* kick the interrupt handler to allocate/deallocate pools */
1572         ibmveth_interrupt(netdev->irq, netdev);
1573         return count;
1574 }
1575
1576
1577 #define ATTR(_name, _mode)      \
1578         struct attribute veth_##_name##_attr = {               \
1579         .name = __stringify(_name), .mode = _mode, \
1580         };
1581
1582 static ATTR(active, 0644);
1583 static ATTR(num, 0644);
1584 static ATTR(size, 0644);
1585
1586 static struct attribute * veth_pool_attrs[] = {
1587         &veth_active_attr,
1588         &veth_num_attr,
1589         &veth_size_attr,
1590         NULL,
1591 };
1592
1593 static const struct sysfs_ops veth_pool_ops = {
1594         .show   = veth_pool_show,
1595         .store  = veth_pool_store,
1596 };
1597
1598 static struct kobj_type ktype_veth_pool = {
1599         .release        = NULL,
1600         .sysfs_ops      = &veth_pool_ops,
1601         .default_attrs  = veth_pool_attrs,
1602 };
1603
1604 static int ibmveth_resume(struct device *dev)
1605 {
1606         struct net_device *netdev = dev_get_drvdata(dev);
1607         ibmveth_interrupt(netdev->irq, netdev);
1608         return 0;
1609 }
1610
1611 static struct vio_device_id ibmveth_device_table[] __devinitdata= {
1612         { "network", "IBM,l-lan"},
1613         { "", "" }
1614 };
1615 MODULE_DEVICE_TABLE(vio, ibmveth_device_table);
1616
1617 static struct dev_pm_ops ibmveth_pm_ops = {
1618         .resume = ibmveth_resume
1619 };
1620
1621 static struct vio_driver ibmveth_driver = {
1622         .id_table       = ibmveth_device_table,
1623         .probe          = ibmveth_probe,
1624         .remove         = ibmveth_remove,
1625         .get_desired_dma = ibmveth_get_desired_dma,
1626         .driver         = {
1627                 .name   = ibmveth_driver_name,
1628                 .owner  = THIS_MODULE,
1629                 .pm = &ibmveth_pm_ops,
1630         }
1631 };
1632
1633 static int __init ibmveth_module_init(void)
1634 {
1635         ibmveth_printk("%s: %s %s\n", ibmveth_driver_name, ibmveth_driver_string, ibmveth_driver_version);
1636
1637         ibmveth_proc_register_driver();
1638
1639         return vio_register_driver(&ibmveth_driver);
1640 }
1641
1642 static void __exit ibmveth_module_exit(void)
1643 {
1644         vio_unregister_driver(&ibmveth_driver);
1645         ibmveth_proc_unregister_driver();
1646 }
1647
1648 module_init(ibmveth_module_init);
1649 module_exit(ibmveth_module_exit);