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