drivers/net/ethernet/ibm/ibmveth.c

   1 /*
   2  * IBM Power Virtual Ethernet Device Driver
   3  *
   4  * This program is free software; you can redistribute it and/or modify
   5  * it under the terms of the GNU General Public License as published by
   6  * the Free Software Foundation; either version 2 of the License, or
   7  * (at your option) any later version.
   8  *
   9  * This program is distributed in the hope that it will be useful,
  10  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  11  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
  12  * GNU General Public License for more details.
  13  *
  14  * You should have received a copy of the GNU General Public License
  15  * along with this program; if not, write to the Free Software
  16  * Foundation, Inc., 59 Temple Place - Suite 330, Boston, MA 02111-1307, USA.
  17  *
  18  * Copyright (C) IBM Corporation, 2003, 2010
  19  *
  20  * Authors: Dave Larson <larson1@us.ibm.com>
  21  *          Santiago Leon <santil@linux.vnet.ibm.com>
  22  *          Brian King <brking@linux.vnet.ibm.com>
  23  *          Robert Jennings <rcj@linux.vnet.ibm.com>
  24  *          Anton Blanchard <anton@au.ibm.com>
  25  */
  26
  27 #include <linux/module.h>
  28 #include <linux/moduleparam.h>
  29 #include <linux/types.h>
  30 #include <linux/errno.h>
  31 #include <linux/dma-mapping.h>
  32 #include <linux/kernel.h>
  33 #include <linux/netdevice.h>
  34 #include <linux/etherdevice.h>
  35 #include <linux/skbuff.h>
  36 #include <linux/init.h>
  37 #include <linux/interrupt.h>
  38 #include <linux/mm.h>
  39 #include <linux/pm.h>
  40 #include <linux/ethtool.h>
  41 #include <linux/in.h>
  42 #include <linux/ip.h>
  43 #include <linux/ipv6.h>
  44 #include <linux/slab.h>
  45 #include <asm/hvcall.h>
  46 #include <linux/atomic.h>
  47 #include <asm/vio.h>
  48 #include <asm/iommu.h>
  49 #include <asm/firmware.h>
  50
  51 #include "ibmveth.h"
  52
  53 static irqreturn_t ibmveth_interrupt(int irq, void *dev_instance);
  54 static void ibmveth_rxq_harvest_buffer(struct ibmveth_adapter *adapter);
  55 static unsigned long ibmveth_get_desired_dma(struct vio_dev *vdev);
  56
  57 static struct kobj_type ktype_veth_pool;
  58
  59
  60 static const char ibmveth_driver_name[] = "ibmveth";
  61 static const char ibmveth_driver_string[] = "IBM Power Virtual Ethernet Driver";
  62 #define ibmveth_driver_version "1.04"
  63
  64 MODULE_AUTHOR("Santiago Leon <santil@linux.vnet.ibm.com>");
  65 MODULE_DESCRIPTION("IBM Power Virtual Ethernet Driver");
  66 MODULE_LICENSE("GPL");
  67 MODULE_VERSION(ibmveth_driver_version);
  68
  69 static unsigned int tx_copybreak __read_mostly = 128;
  70 module_param(tx_copybreak, uint, 0644);
  71 MODULE_PARM_DESC(tx_copybreak,
  72         "Maximum size of packet that is copied to a new buffer on transmit");
  73
  74 static unsigned int rx_copybreak __read_mostly = 128;
  75 module_param(rx_copybreak, uint, 0644);
  76 MODULE_PARM_DESC(rx_copybreak,
  77         "Maximum size of packet that is copied to a new buffer on receive");
  78
  79 static unsigned int rx_flush __read_mostly = 0;
  80 module_param(rx_flush, uint, 0644);
  81 MODULE_PARM_DESC(rx_flush, "Flush receive buffers before use");
  82
  83 struct ibmveth_stat {
  84         char name[ETH_GSTRING_LEN];
  85         int offset;
  86 };
  87
  88 #define IBMVETH_STAT_OFF(stat) offsetof(struct ibmveth_adapter, stat)
  89 #define IBMVETH_GET_STAT(a, off) *((u64 *)(((unsigned long)(a)) + off))
  90
  91 struct ibmveth_stat ibmveth_stats[] = {
  92         { "replenish_task_cycles", IBMVETH_STAT_OFF(replenish_task_cycles) },
  93         { "replenish_no_mem", IBMVETH_STAT_OFF(replenish_no_mem) },
  94         { "replenish_add_buff_failure",
  95                         IBMVETH_STAT_OFF(replenish_add_buff_failure) },
  96         { "replenish_add_buff_success",
  97                         IBMVETH_STAT_OFF(replenish_add_buff_success) },
  98         { "rx_invalid_buffer", IBMVETH_STAT_OFF(rx_invalid_buffer) },
  99         { "rx_no_buffer", IBMVETH_STAT_OFF(rx_no_buffer) },
 100         { "tx_map_failed", IBMVETH_STAT_OFF(tx_map_failed) },
 101         { "tx_send_failed", IBMVETH_STAT_OFF(tx_send_failed) },
 102         { "fw_enabled_ipv4_csum", IBMVETH_STAT_OFF(fw_ipv4_csum_support) },
 103         { "fw_enabled_ipv6_csum", IBMVETH_STAT_OFF(fw_ipv6_csum_support) },
 104 };
 105
 106 /* simple methods of getting data from the current rxq entry */
 107 static inline u32 ibmveth_rxq_flags(struct ibmveth_adapter *adapter)
 108 {
 109         return adapter->rx_queue.queue_addr[adapter->rx_queue.index].flags_off;
 110 }
 111
 112 static inline int ibmveth_rxq_toggle(struct ibmveth_adapter *adapter)
 113 {
 114         return (ibmveth_rxq_flags(adapter) & IBMVETH_RXQ_TOGGLE) >>
 115                         IBMVETH_RXQ_TOGGLE_SHIFT;
 116 }
 117
 118 static inline int ibmveth_rxq_pending_buffer(struct ibmveth_adapter *adapter)
 119 {
 120         return ibmveth_rxq_toggle(adapter) == adapter->rx_queue.toggle;
 121 }
 122
 123 static inline int ibmveth_rxq_buffer_valid(struct ibmveth_adapter *adapter)
 124 {
 125         return ibmveth_rxq_flags(adapter) & IBMVETH_RXQ_VALID;
 126 }
 127
 128 static inline int ibmveth_rxq_frame_offset(struct ibmveth_adapter *adapter)
 129 {
 130         return ibmveth_rxq_flags(adapter) & IBMVETH_RXQ_OFF_MASK;
 131 }
 132
 133 static inline int ibmveth_rxq_frame_length(struct ibmveth_adapter *adapter)
 134 {
 135         return adapter->rx_queue.queue_addr[adapter->rx_queue.index].length;
 136 }
 137
 138 static inline int ibmveth_rxq_csum_good(struct ibmveth_adapter *adapter)
 139 {
 140         return ibmveth_rxq_flags(adapter) & IBMVETH_RXQ_CSUM_GOOD;
 141 }
 142
 143 /* setup the initial settings for a buffer pool */
 144 static void ibmveth_init_buffer_pool(struct ibmveth_buff_pool *pool,
 145                                      u32 pool_index, u32 pool_size,
 146                                      u32 buff_size, u32 pool_active)
 147 {
 148         pool->size = pool_size;
 149         pool->index = pool_index;
 150         pool->buff_size = buff_size;
 151         pool->threshold = pool_size * 7 / 8;
 152         pool->active = pool_active;
 153 }
 154
 155 /* allocate and setup an buffer pool - called during open */
 156 static int ibmveth_alloc_buffer_pool(struct ibmveth_buff_pool *pool)
 157 {
 158         int i;
 159
 160         pool->free_map = kmalloc(sizeof(u16) * pool->size, GFP_KERNEL);
 161
 162         if (!pool->free_map)
 163                 return -1;
 164
 165         pool->dma_addr = kmalloc(sizeof(dma_addr_t) * pool->size, GFP_KERNEL);
 166         if (!pool->dma_addr) {
 167                 kfree(pool->free_map);
 168                 pool->free_map = NULL;
 169                 return -1;
 170         }
 171
 172         pool->skbuff = kcalloc(pool->size, sizeof(void *), GFP_KERNEL);
 173
 174         if (!pool->skbuff) {
 175                 kfree(pool->dma_addr);
 176                 pool->dma_addr = NULL;
 177
 178                 kfree(pool->free_map);
 179                 pool->free_map = NULL;
 180                 return -1;
 181         }
 182
 183         memset(pool->dma_addr, 0, sizeof(dma_addr_t) * pool->size);
 184
 185         for (i = 0; i < pool->size; ++i)
 186                 pool->free_map[i] = i;
 187
 188         atomic_set(&pool->available, 0);
 189         pool->producer_index = 0;
 190         pool->consumer_index = 0;
 191
 192         return 0;
 193 }
 194
 195 static inline void ibmveth_flush_buffer(void *addr, unsigned long length)
 196 {
 197         unsigned long offset;
 198
 199         for (offset = 0; offset < length; offset += SMP_CACHE_BYTES)
 200                 asm("dcbfl %0,%1" :: "b" (addr), "r" (offset));
 201 }
 202
 203 /* replenish the buffers for a pool.  note that we don't need to
 204  * skb_reserve these since they are used for incoming...
 205  */
 206 static void ibmveth_replenish_buffer_pool(struct ibmveth_adapter *adapter,
 207                                           struct ibmveth_buff_pool *pool)
 208 {
 209         u32 i;
 210         u32 count = pool->size - atomic_read(&pool->available);
 211         u32 buffers_added = 0;
 212         struct sk_buff *skb;
 213         unsigned int free_index, index;
 214         u64 correlator;
 215         unsigned long lpar_rc;
 216         dma_addr_t dma_addr;
 217
 218         mb();
 219
 220         for (i = 0; i < count; ++i) {
 221                 union ibmveth_buf_desc desc;
 222
 223                 skb = netdev_alloc_skb(adapter->netdev, pool->buff_size);
 224
 225                 if (!skb) {
 226                         netdev_dbg(adapter->netdev,
 227                                    "replenish: unable to allocate skb\n");
 228                         adapter->replenish_no_mem++;
 229                         break;
 230                 }
 231
 232                 free_index = pool->consumer_index;
 233                 pool->consumer_index++;
 234                 if (pool->consumer_index >= pool->size)
 235                         pool->consumer_index = 0;
 236                 index = pool->free_map[free_index];
 237
 238                 BUG_ON(index == IBM_VETH_INVALID_MAP);
 239                 BUG_ON(pool->skbuff[index] != NULL);
 240
 241                 dma_addr = dma_map_single(&adapter->vdev->dev, skb->data,
 242                                 pool->buff_size, DMA_FROM_DEVICE);
 243
 244                 if (dma_mapping_error(&adapter->vdev->dev, dma_addr))
 245                         goto failure;
 246
 247                 pool->free_map[free_index] = IBM_VETH_INVALID_MAP;
 248                 pool->dma_addr[index] = dma_addr;
 249                 pool->skbuff[index] = skb;
 250
 251                 correlator = ((u64)pool->index << 32) | index;
 252                 *(u64 *)skb->data = correlator;
 253
 254                 desc.fields.flags_len = IBMVETH_BUF_VALID | pool->buff_size;
 255                 desc.fields.address = dma_addr;
 256
 257                 if (rx_flush) {
 258                         unsigned int len = min(pool->buff_size,
 259                                                 adapter->netdev->mtu +
 260                                                 IBMVETH_BUFF_OH);
 261                         ibmveth_flush_buffer(skb->data, len);
 262                 }
 263                 lpar_rc = h_add_logical_lan_buffer(adapter->vdev->unit_address,
 264                                                    desc.desc);
 265
 266                 if (lpar_rc != H_SUCCESS) {
 267                         goto failure;
 268                 } else {
 269                         buffers_added++;
 270                         adapter->replenish_add_buff_success++;
 271                 }
 272         }
 273
 274         mb();
 275         atomic_add(buffers_added, &(pool->available));
 276         return;
 277
 278 failure:
 279         pool->free_map[free_index] = index;
 280         pool->skbuff[index] = NULL;
 281         if (pool->consumer_index == 0)
 282                 pool->consumer_index = pool->size - 1;
 283         else
 284                 pool->consumer_index--;
 285         if (!dma_mapping_error(&adapter->vdev->dev, dma_addr))
 286                 dma_unmap_single(&adapter->vdev->dev,
 287                                  pool->dma_addr[index], pool->buff_size,
 288                                  DMA_FROM_DEVICE);
 289         dev_kfree_skb_any(skb);
 290         adapter->replenish_add_buff_failure++;
 291
 292         mb();
 293         atomic_add(buffers_added, &(pool->available));
 294 }
 295
 296 /* replenish routine */
 297 static void ibmveth_replenish_task(struct ibmveth_adapter *adapter)
 298 {
 299         int i;
 300
 301         adapter->replenish_task_cycles++;
 302
 303         for (i = (IBMVETH_NUM_BUFF_POOLS - 1); i >= 0; i--) {
 304                 struct ibmveth_buff_pool *pool = &adapter->rx_buff_pool[i];
 305
 306                 if (pool->active &&
 307                     (atomic_read(&pool->available) < pool->threshold))
 308                         ibmveth_replenish_buffer_pool(adapter, pool);
 309         }
 310
 311         adapter->rx_no_buffer = *(u64 *)(((char*)adapter->buffer_list_addr) +
 312                                                 4096 - 8);
 313 }
 314
 315 /* empty and free ana buffer pool - also used to do cleanup in error paths */
 316 static void ibmveth_free_buffer_pool(struct ibmveth_adapter *adapter,
 317                                      struct ibmveth_buff_pool *pool)
 318 {
 319         int i;
 320
 321         kfree(pool->free_map);
 322         pool->free_map = NULL;
 323
 324         if (pool->skbuff && pool->dma_addr) {
 325                 for (i = 0; i < pool->size; ++i) {
 326                         struct sk_buff *skb = pool->skbuff[i];
 327                         if (skb) {
 328                                 dma_unmap_single(&adapter->vdev->dev,
 329                                                  pool->dma_addr[i],
 330                                                  pool->buff_size,
 331                                                  DMA_FROM_DEVICE);
 332                                 dev_kfree_skb_any(skb);
 333                                 pool->skbuff[i] = NULL;
 334                         }
 335                 }
 336         }
 337
 338         if (pool->dma_addr) {
 339                 kfree(pool->dma_addr);
 340                 pool->dma_addr = NULL;
 341         }
 342
 343         if (pool->skbuff) {
 344                 kfree(pool->skbuff);
 345                 pool->skbuff = NULL;
 346         }
 347 }
 348
 349 /* remove a buffer from a pool */
 350 static void ibmveth_remove_buffer_from_pool(struct ibmveth_adapter *adapter,
 351                                             u64 correlator)
 352 {
 353         unsigned int pool  = correlator >> 32;
 354         unsigned int index = correlator & 0xffffffffUL;
 355         unsigned int free_index;
 356         struct sk_buff *skb;
 357
 358         BUG_ON(pool >= IBMVETH_NUM_BUFF_POOLS);
 359         BUG_ON(index >= adapter->rx_buff_pool[pool].size);
 360
 361         skb = adapter->rx_buff_pool[pool].skbuff[index];
 362
 363         BUG_ON(skb == NULL);
 364
 365         adapter->rx_buff_pool[pool].skbuff[index] = NULL;
 366
 367         dma_unmap_single(&adapter->vdev->dev,
 368                          adapter->rx_buff_pool[pool].dma_addr[index],
 369                          adapter->rx_buff_pool[pool].buff_size,
 370                          DMA_FROM_DEVICE);
 371
 372         free_index = adapter->rx_buff_pool[pool].producer_index;
 373         adapter->rx_buff_pool[pool].producer_index++;
 374         if (adapter->rx_buff_pool[pool].producer_index >=
 375             adapter->rx_buff_pool[pool].size)
 376                 adapter->rx_buff_pool[pool].producer_index = 0;
 377         adapter->rx_buff_pool[pool].free_map[free_index] = index;
 378
 379         mb();
 380
 381         atomic_dec(&(adapter->rx_buff_pool[pool].available));
 382 }
 383
 384 /* get the current buffer on the rx queue */
 385 static inline struct sk_buff *ibmveth_rxq_get_buffer(struct ibmveth_adapter *adapter)
 386 {
 387         u64 correlator = adapter->rx_queue.queue_addr[adapter->rx_queue.index].correlator;
 388         unsigned int pool = correlator >> 32;
 389         unsigned int index = correlator & 0xffffffffUL;
 390
 391         BUG_ON(pool >= IBMVETH_NUM_BUFF_POOLS);
 392         BUG_ON(index >= adapter->rx_buff_pool[pool].size);
 393
 394         return adapter->rx_buff_pool[pool].skbuff[index];
 395 }
 396
 397 /* recycle the current buffer on the rx queue */
 398 static int ibmveth_rxq_recycle_buffer(struct ibmveth_adapter *adapter)
 399 {
 400         u32 q_index = adapter->rx_queue.index;
 401         u64 correlator = adapter->rx_queue.queue_addr[q_index].correlator;
 402         unsigned int pool = correlator >> 32;
 403         unsigned int index = correlator & 0xffffffffUL;
 404         union ibmveth_buf_desc desc;
 405         unsigned long lpar_rc;
 406         int ret = 1;
 407
 408         BUG_ON(pool >= IBMVETH_NUM_BUFF_POOLS);
 409         BUG_ON(index >= adapter->rx_buff_pool[pool].size);
 410
 411         if (!adapter->rx_buff_pool[pool].active) {
 412                 ibmveth_rxq_harvest_buffer(adapter);
 413                 ibmveth_free_buffer_pool(adapter, &adapter->rx_buff_pool[pool]);
 414                 goto out;
 415         }
 416
 417         desc.fields.flags_len = IBMVETH_BUF_VALID |
 418                 adapter->rx_buff_pool[pool].buff_size;
 419         desc.fields.address = adapter->rx_buff_pool[pool].dma_addr[index];
 420
 421         lpar_rc = h_add_logical_lan_buffer(adapter->vdev->unit_address, desc.desc);
 422
 423         if (lpar_rc != H_SUCCESS) {
 424                 netdev_dbg(adapter->netdev, "h_add_logical_lan_buffer failed "
 425                            "during recycle rc=%ld", lpar_rc);
 426                 ibmveth_remove_buffer_from_pool(adapter, adapter->rx_queue.queue_addr[adapter->rx_queue.index].correlator);
 427                 ret = 0;
 428         }
 429
 430         if (++adapter->rx_queue.index == adapter->rx_queue.num_slots) {
 431                 adapter->rx_queue.index = 0;
 432                 adapter->rx_queue.toggle = !adapter->rx_queue.toggle;
 433         }
 434
 435 out:
 436         return ret;
 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 < IBMVETH_NUM_BUFF_POOLS; 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         /*
 510          * After a kexec the adapter will still be open, so our attempt to
 511          * open it will fail. So if we get a failure we free the adapter and
 512          * try again, but only once.
 513          */
 514 retry:
 515         rc = h_register_logical_lan(adapter->vdev->unit_address,
 516                                     adapter->buffer_list_dma, rxq_desc.desc,
 517                                     adapter->filter_list_dma, mac_address);
 518
 519         if (rc != H_SUCCESS && try_again) {
 520                 do {
 521                         rc = h_free_logical_lan(adapter->vdev->unit_address);
 522                 } while (H_IS_LONG_BUSY(rc) || (rc == H_BUSY));
 523
 524                 try_again = 0;
 525                 goto retry;
 526         }
 527
 528         return rc;
 529 }
 530
 531 static int ibmveth_open(struct net_device *netdev)
 532 {
 533         struct ibmveth_adapter *adapter = netdev_priv(netdev);
 534         u64 mac_address = 0;
 535         int rxq_entries = 1;
 536         unsigned long lpar_rc;
 537         int rc;
 538         union ibmveth_buf_desc rxq_desc;
 539         int i;
 540         struct device *dev;
 541
 542         netdev_dbg(netdev, "open starting\n");
 543
 544         napi_enable(&adapter->napi);
 545
 546         for(i = 0; i < IBMVETH_NUM_BUFF_POOLS; i++)
 547                 rxq_entries += adapter->rx_buff_pool[i].size;
 548
 549         adapter->buffer_list_addr = (void*) get_zeroed_page(GFP_KERNEL);
 550         adapter->filter_list_addr = (void*) get_zeroed_page(GFP_KERNEL);
 551
 552         if (!adapter->buffer_list_addr || !adapter->filter_list_addr) {
 553                 netdev_err(netdev, "unable to allocate filter or buffer list "
 554                            "pages\n");
 555                 rc = -ENOMEM;
 556                 goto err_out;
 557         }
 558
 559         adapter->rx_queue.queue_len = sizeof(struct ibmveth_rx_q_entry) *
 560                                                 rxq_entries;
 561         adapter->rx_queue.queue_addr = kmalloc(adapter->rx_queue.queue_len,
 562                                                 GFP_KERNEL);
 563
 564         if (!adapter->rx_queue.queue_addr) {
 565                 netdev_err(netdev, "unable to allocate rx queue pages\n");
 566                 rc = -ENOMEM;
 567                 goto err_out;
 568         }
 569
 570         dev = &adapter->vdev->dev;
 571
 572         adapter->buffer_list_dma = dma_map_single(dev,
 573                         adapter->buffer_list_addr, 4096, DMA_BIDIRECTIONAL);
 574         adapter->filter_list_dma = dma_map_single(dev,
 575                         adapter->filter_list_addr, 4096, DMA_BIDIRECTIONAL);
 576         adapter->rx_queue.queue_dma = dma_map_single(dev,
 577                         adapter->rx_queue.queue_addr,
 578                         adapter->rx_queue.queue_len, DMA_BIDIRECTIONAL);
 579
 580         if ((dma_mapping_error(dev, adapter->buffer_list_dma)) ||
 581             (dma_mapping_error(dev, adapter->filter_list_dma)) ||
 582             (dma_mapping_error(dev, adapter->rx_queue.queue_dma))) {
 583                 netdev_err(netdev, "unable to map filter or buffer list "
 584                            "pages\n");
 585                 rc = -ENOMEM;
 586                 goto err_out;
 587         }
 588
 589         adapter->rx_queue.index = 0;
 590         adapter->rx_queue.num_slots = rxq_entries;
 591         adapter->rx_queue.toggle = 1;
 592
 593         memcpy(&mac_address, netdev->dev_addr, netdev->addr_len);
 594         mac_address = mac_address >> 16;
 595
 596         rxq_desc.fields.flags_len = IBMVETH_BUF_VALID |
 597                                         adapter->rx_queue.queue_len;
 598         rxq_desc.fields.address = adapter->rx_queue.queue_dma;
 599
 600         netdev_dbg(netdev, "buffer list @ 0x%p\n", adapter->buffer_list_addr);
 601         netdev_dbg(netdev, "filter list @ 0x%p\n", adapter->filter_list_addr);
 602         netdev_dbg(netdev, "receive q   @ 0x%p\n", adapter->rx_queue.queue_addr);
 603
 604         h_vio_signal(adapter->vdev->unit_address, VIO_IRQ_DISABLE);
 605
 606         lpar_rc = ibmveth_register_logical_lan(adapter, rxq_desc, mac_address);
 607
 608         if (lpar_rc != H_SUCCESS) {
 609                 netdev_err(netdev, "h_register_logical_lan failed with %ld\n",
 610                            lpar_rc);
 611                 netdev_err(netdev, "buffer TCE:0x%llx filter TCE:0x%llx rxq "
 612                            "desc:0x%llx MAC:0x%llx\n",
 613                                      adapter->buffer_list_dma,
 614                                      adapter->filter_list_dma,
 615                                      rxq_desc.desc,
 616                                      mac_address);
 617                 rc = -ENONET;
 618                 goto err_out;
 619         }
 620
 621         for (i = 0; i < IBMVETH_NUM_BUFF_POOLS; i++) {
 622                 if (!adapter->rx_buff_pool[i].active)
 623                         continue;
 624                 if (ibmveth_alloc_buffer_pool(&adapter->rx_buff_pool[i])) {
 625                         netdev_err(netdev, "unable to alloc pool\n");
 626                         adapter->rx_buff_pool[i].active = 0;
 627                         rc = -ENOMEM;
 628                         goto err_out;
 629                 }
 630         }
 631
 632         netdev_dbg(netdev, "registering irq 0x%x\n", netdev->irq);
 633         rc = request_irq(netdev->irq, ibmveth_interrupt, 0, netdev->name,
 634                          netdev);
 635         if (rc != 0) {
 636                 netdev_err(netdev, "unable to request irq 0x%x, rc %d\n",
 637                            netdev->irq, rc);
 638                 do {
 639                         lpar_rc = h_free_logical_lan(adapter->vdev->unit_address);
 640                 } while (H_IS_LONG_BUSY(lpar_rc) || (lpar_rc == H_BUSY));
 641
 642                 goto err_out;
 643         }
 644
 645         adapter->bounce_buffer =
 646             kmalloc(netdev->mtu + IBMVETH_BUFF_OH, GFP_KERNEL);
 647         if (!adapter->bounce_buffer) {
 648                 netdev_err(netdev, "unable to allocate bounce buffer\n");
 649                 rc = -ENOMEM;
 650                 goto err_out_free_irq;
 651         }
 652         adapter->bounce_buffer_dma =
 653             dma_map_single(&adapter->vdev->dev, adapter->bounce_buffer,
 654                            netdev->mtu + IBMVETH_BUFF_OH, DMA_BIDIRECTIONAL);
 655         if (dma_mapping_error(dev, adapter->bounce_buffer_dma)) {
 656                 netdev_err(netdev, "unable to map bounce buffer\n");
 657                 rc = -ENOMEM;
 658                 goto err_out_free_irq;
 659         }
 660
 661         netdev_dbg(netdev, "initial replenish cycle\n");
 662         ibmveth_interrupt(netdev->irq, netdev);
 663
 664         netif_start_queue(netdev);
 665
 666         netdev_dbg(netdev, "open complete\n");
 667
 668         return 0;
 669
 670 err_out_free_irq:
 671         free_irq(netdev->irq, netdev);
 672 err_out:
 673         ibmveth_cleanup(adapter);
 674         napi_disable(&adapter->napi);
 675         return rc;
 676 }
 677
 678 static int ibmveth_close(struct net_device *netdev)
 679 {
 680         struct ibmveth_adapter *adapter = netdev_priv(netdev);
 681         long lpar_rc;
 682
 683         netdev_dbg(netdev, "close starting\n");
 684
 685         napi_disable(&adapter->napi);
 686
 687         if (!adapter->pool_config)
 688                 netif_stop_queue(netdev);
 689
 690         h_vio_signal(adapter->vdev->unit_address, VIO_IRQ_DISABLE);
 691
 692         do {
 693                 lpar_rc = h_free_logical_lan(adapter->vdev->unit_address);
 694         } while (H_IS_LONG_BUSY(lpar_rc) || (lpar_rc == H_BUSY));
 695
 696         if (lpar_rc != H_SUCCESS) {
 697                 netdev_err(netdev, "h_free_logical_lan failed with %lx, "
 698                            "continuing with close\n", lpar_rc);
 699         }
 700
 701         free_irq(netdev->irq, netdev);
 702
 703         adapter->rx_no_buffer = *(u64 *)(((char *)adapter->buffer_list_addr) +
 704                                                 4096 - 8);
 705
 706         ibmveth_cleanup(adapter);
 707
 708         netdev_dbg(netdev, "close complete\n");
 709
 710         return 0;
 711 }
 712
 713 static int netdev_get_settings(struct net_device *dev, struct ethtool_cmd *cmd)
 714 {
 715         cmd->supported = (SUPPORTED_1000baseT_Full | SUPPORTED_Autoneg |
 716                                 SUPPORTED_FIBRE);
 717         cmd->advertising = (ADVERTISED_1000baseT_Full | ADVERTISED_Autoneg |
 718                                 ADVERTISED_FIBRE);
 719         ethtool_cmd_speed_set(cmd, SPEED_1000);
 720         cmd->duplex = DUPLEX_FULL;
 721         cmd->port = PORT_FIBRE;
 722         cmd->phy_address = 0;
 723         cmd->transceiver = XCVR_INTERNAL;
 724         cmd->autoneg = AUTONEG_ENABLE;
 725         cmd->maxtxpkt = 0;
 726         cmd->maxrxpkt = 1;
 727         return 0;
 728 }
 729
 730 static void netdev_get_drvinfo(struct net_device *dev,
 731                                struct ethtool_drvinfo *info)
 732 {
 733         strncpy(info->driver, ibmveth_driver_name, sizeof(info->driver) - 1);
 734         strncpy(info->version, ibmveth_driver_version,
 735                 sizeof(info->version) - 1);
 736 }
 737
 738 static netdev_features_t ibmveth_fix_features(struct net_device *dev,
 739         netdev_features_t features)
 740 {
 741         /*
 742          * Since the ibmveth firmware interface does not have the
 743          * concept of separate tx/rx checksum offload enable, if rx
 744          * checksum is disabled we also have to disable tx checksum
 745          * offload. Once we disable rx checksum offload, we are no
 746          * longer allowed to send tx buffers that are not properly
 747          * checksummed.
 748          */
 749
 750         if (!(features & NETIF_F_RXCSUM))
 751                 features &= ~NETIF_F_ALL_CSUM;
 752
 753         return features;
 754 }
 755
 756 static int ibmveth_set_csum_offload(struct net_device *dev, u32 data)
 757 {
 758         struct ibmveth_adapter *adapter = netdev_priv(dev);
 759         unsigned long set_attr, clr_attr, ret_attr;
 760         unsigned long set_attr6, clr_attr6;
 761         long ret, ret4, ret6;
 762         int rc1 = 0, rc2 = 0;
 763         int restart = 0;
 764
 765         if (netif_running(dev)) {
 766                 restart = 1;
 767                 adapter->pool_config = 1;
 768                 ibmveth_close(dev);
 769                 adapter->pool_config = 0;
 770         }
 771
 772         set_attr = 0;
 773         clr_attr = 0;
 774         set_attr6 = 0;
 775         clr_attr6 = 0;
 776
 777         if (data) {
 778                 set_attr = IBMVETH_ILLAN_IPV4_TCP_CSUM;
 779                 set_attr6 = IBMVETH_ILLAN_IPV6_TCP_CSUM;
 780         } else {
 781                 clr_attr = IBMVETH_ILLAN_IPV4_TCP_CSUM;
 782                 clr_attr6 = IBMVETH_ILLAN_IPV6_TCP_CSUM;
 783         }
 784
 785         ret = h_illan_attributes(adapter->vdev->unit_address, 0, 0, &ret_attr);
 786
 787         if (ret == H_SUCCESS && !(ret_attr & IBMVETH_ILLAN_ACTIVE_TRUNK) &&
 788             !(ret_attr & IBMVETH_ILLAN_TRUNK_PRI_MASK) &&
 789             (ret_attr & IBMVETH_ILLAN_PADDED_PKT_CSUM)) {
 790                 ret4 = h_illan_attributes(adapter->vdev->unit_address, clr_attr,
 791                                          set_attr, &ret_attr);
 792
 793                 if (ret4 != H_SUCCESS) {
 794                         netdev_err(dev, "unable to change IPv4 checksum "
 795                                         "offload settings. %d rc=%ld\n",
 796                                         data, ret4);
 797
 798                         h_illan_attributes(adapter->vdev->unit_address,
 799                                            set_attr, clr_attr, &ret_attr);
 800
 801                         if (data == 1)
 802                                 dev->features &= ~NETIF_F_IP_CSUM;
 803
 804                 } else {
 805                         adapter->fw_ipv4_csum_support = data;
 806                 }
 807
 808                 ret6 = h_illan_attributes(adapter->vdev->unit_address,
 809                                          clr_attr6, set_attr6, &ret_attr);
 810
 811                 if (ret6 != H_SUCCESS) {
 812                         netdev_err(dev, "unable to change IPv6 checksum "
 813                                         "offload settings. %d rc=%ld\n",
 814                                         data, ret6);
 815
 816                         h_illan_attributes(adapter->vdev->unit_address,
 817                                            set_attr6, clr_attr6, &ret_attr);
 818
 819                         if (data == 1)
 820                                 dev->features &= ~NETIF_F_IPV6_CSUM;
 821
 822                 } else
 823                         adapter->fw_ipv6_csum_support = data;
 824
 825                 if (ret4 == H_SUCCESS || ret6 == H_SUCCESS)
 826                         adapter->rx_csum = data;
 827                 else
 828                         rc1 = -EIO;
 829         } else {
 830                 rc1 = -EIO;
 831                 netdev_err(dev, "unable to change checksum offload settings."
 832                                      " %d rc=%ld ret_attr=%lx\n", data, ret,
 833                                      ret_attr);
 834         }
 835
 836         if (restart)
 837                 rc2 = ibmveth_open(dev);
 838
 839         return rc1 ? rc1 : rc2;
 840 }
 841
 842 static int ibmveth_set_features(struct net_device *dev,
 843         netdev_features_t features)
 844 {
 845         struct ibmveth_adapter *adapter = netdev_priv(dev);
 846         int rx_csum = !!(features & NETIF_F_RXCSUM);
 847         int rc;
 848
 849         if (rx_csum == adapter->rx_csum)
 850                 return 0;
 851
 852         rc = ibmveth_set_csum_offload(dev, rx_csum);
 853         if (rc && !adapter->rx_csum)
 854                 dev->features = features & ~(NETIF_F_ALL_CSUM | NETIF_F_RXCSUM);
 855
 856         return rc;
 857 }
 858
 859 static void ibmveth_get_strings(struct net_device *dev, u32 stringset, u8 *data)
 860 {
 861         int i;
 862
 863         if (stringset != ETH_SS_STATS)
 864                 return;
 865
 866         for (i = 0; i < ARRAY_SIZE(ibmveth_stats); i++, data += ETH_GSTRING_LEN)
 867                 memcpy(data, ibmveth_stats[i].name, ETH_GSTRING_LEN);
 868 }
 869
 870 static int ibmveth_get_sset_count(struct net_device *dev, int sset)
 871 {
 872         switch (sset) {
 873         case ETH_SS_STATS:
 874                 return ARRAY_SIZE(ibmveth_stats);
 875         default:
 876                 return -EOPNOTSUPP;
 877         }
 878 }
 879
 880 static void ibmveth_get_ethtool_stats(struct net_device *dev,
 881                                       struct ethtool_stats *stats, u64 *data)
 882 {
 883         int i;
 884         struct ibmveth_adapter *adapter = netdev_priv(dev);
 885
 886         for (i = 0; i < ARRAY_SIZE(ibmveth_stats); i++)
 887                 data[i] = IBMVETH_GET_STAT(adapter, ibmveth_stats[i].offset);
 888 }
 889
 890 static const struct ethtool_ops netdev_ethtool_ops = {
 891         .get_drvinfo            = netdev_get_drvinfo,
 892         .get_settings           = netdev_get_settings,
 893         .get_link               = ethtool_op_get_link,
 894         .get_strings            = ibmveth_get_strings,
 895         .get_sset_count         = ibmveth_get_sset_count,
 896         .get_ethtool_stats      = ibmveth_get_ethtool_stats,
 897 };
 898
 899 static int ibmveth_ioctl(struct net_device *dev, struct ifreq *ifr, int cmd)
 900 {
 901         return -EOPNOTSUPP;
 902 }
 903
 904 #define page_offset(v) ((unsigned long)(v) & ((1 << 12) - 1))
 905
 906 static int ibmveth_send(struct ibmveth_adapter *adapter,
 907                         union ibmveth_buf_desc *descs)
 908 {
 909         unsigned long correlator;
 910         unsigned int retry_count;
 911         unsigned long ret;
 912
 913         /*
 914          * The retry count sets a maximum for the number of broadcast and
 915          * multicast destinations within the system.
 916          */
 917         retry_count = 1024;
 918         correlator = 0;
 919         do {
 920                 ret = h_send_logical_lan(adapter->vdev->unit_address,
 921                                              descs[0].desc, descs[1].desc,
 922                                              descs[2].desc, descs[3].desc,
 923                                              descs[4].desc, descs[5].desc,
 924                                              correlator, &correlator);
 925         } while ((ret == H_BUSY) && (retry_count--));
 926
 927         if (ret != H_SUCCESS && ret != H_DROPPED) {
 928                 netdev_err(adapter->netdev, "tx: h_send_logical_lan failed "
 929                            "with rc=%ld\n", ret);
 930                 return 1;
 931         }
 932
 933         return 0;
 934 }
 935
 936 static netdev_tx_t ibmveth_start_xmit(struct sk_buff *skb,
 937                                       struct net_device *netdev)
 938 {
 939         struct ibmveth_adapter *adapter = netdev_priv(netdev);
 940         unsigned int desc_flags;
 941         union ibmveth_buf_desc descs[6];
 942         int last, i;
 943         int force_bounce = 0;
 944         dma_addr_t dma_addr;
 945
 946         /*
 947          * veth handles a maximum of 6 segments including the header, so
 948          * we have to linearize the skb if there are more than this.
 949          */
 950         if (skb_shinfo(skb)->nr_frags > 5 && __skb_linearize(skb)) {
 951                 netdev->stats.tx_dropped++;
 952                 goto out;
 953         }
 954
 955         /* veth can't checksum offload UDP */
 956         if (skb->ip_summed == CHECKSUM_PARTIAL &&
 957             ((skb->protocol == htons(ETH_P_IP) &&
 958               ip_hdr(skb)->protocol != IPPROTO_TCP) ||
 959              (skb->protocol == htons(ETH_P_IPV6) &&
 960               ipv6_hdr(skb)->nexthdr != IPPROTO_TCP)) &&
 961             skb_checksum_help(skb)) {
 962
 963                 netdev_err(netdev, "tx: failed to checksum packet\n");
 964                 netdev->stats.tx_dropped++;
 965                 goto out;
 966         }
 967
 968         desc_flags = IBMVETH_BUF_VALID;
 969
 970         if (skb->ip_summed == CHECKSUM_PARTIAL) {
 971                 unsigned char *buf = skb_transport_header(skb) +
 972                                                 skb->csum_offset;
 973
 974                 desc_flags |= (IBMVETH_BUF_NO_CSUM | IBMVETH_BUF_CSUM_GOOD);
 975
 976                 /* Need to zero out the checksum */
 977                 buf[0] = 0;
 978                 buf[1] = 0;
 979         }
 980
 981 retry_bounce:
 982         memset(descs, 0, sizeof(descs));
 983
 984         /*
 985          * If a linear packet is below the rx threshold then
 986          * copy it into the static bounce buffer. This avoids the
 987          * cost of a TCE insert and remove.
 988          */
 989         if (force_bounce || (!skb_is_nonlinear(skb) &&
 990                                 (skb->len < tx_copybreak))) {
 991                 skb_copy_from_linear_data(skb, adapter->bounce_buffer,
 992                                           skb->len);
 993
 994                 descs[0].fields.flags_len = desc_flags | skb->len;
 995                 descs[0].fields.address = adapter->bounce_buffer_dma;
 996
 997                 if (ibmveth_send(adapter, descs)) {
 998                         adapter->tx_send_failed++;
 999                         netdev->stats.tx_dropped++;
1000                 } else {
1001                         netdev->stats.tx_packets++;
1002                         netdev->stats.tx_bytes += skb->len;
1003                 }
1004
1005                 goto out;
1006         }
1007
1008         /* Map the header */
1009         dma_addr = dma_map_single(&adapter->vdev->dev, skb->data,
1010                                   skb_headlen(skb), DMA_TO_DEVICE);
1011         if (dma_mapping_error(&adapter->vdev->dev, dma_addr))
1012                 goto map_failed;
1013
1014         descs[0].fields.flags_len = desc_flags | skb_headlen(skb);
1015         descs[0].fields.address = dma_addr;
1016
1017         /* Map the frags */
1018         for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
1019                 const skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
1020
1021                 dma_addr = skb_frag_dma_map(&adapter->vdev->dev, frag, 0,
1022                                             skb_frag_size(frag), DMA_TO_DEVICE);
1023
1024                 if (dma_mapping_error(&adapter->vdev->dev, dma_addr))
1025                         goto map_failed_frags;
1026
1027                 descs[i+1].fields.flags_len = desc_flags | skb_frag_size(frag);
1028                 descs[i+1].fields.address = dma_addr;
1029         }
1030
1031         if (ibmveth_send(adapter, descs)) {
1032                 adapter->tx_send_failed++;
1033                 netdev->stats.tx_dropped++;
1034         } else {
1035                 netdev->stats.tx_packets++;
1036                 netdev->stats.tx_bytes += skb->len;
1037         }
1038
1039         dma_unmap_single(&adapter->vdev->dev,
1040                          descs[0].fields.address,
1041                          descs[0].fields.flags_len & IBMVETH_BUF_LEN_MASK,
1042                          DMA_TO_DEVICE);
1043
1044         for (i = 1; i < skb_shinfo(skb)->nr_frags + 1; i++)
1045                 dma_unmap_page(&adapter->vdev->dev, descs[i].fields.address,
1046                                descs[i].fields.flags_len & IBMVETH_BUF_LEN_MASK,
1047                                DMA_TO_DEVICE);
1048
1049 out:
1050         dev_kfree_skb(skb);
1051         return NETDEV_TX_OK;
1052
1053 map_failed_frags:
1054         last = i+1;
1055         for (i = 0; i < last; i++)
1056                 dma_unmap_page(&adapter->vdev->dev, descs[i].fields.address,
1057                                descs[i].fields.flags_len & IBMVETH_BUF_LEN_MASK,
1058                                DMA_TO_DEVICE);
1059
1060 map_failed:
1061         if (!firmware_has_feature(FW_FEATURE_CMO))
1062                 netdev_err(netdev, "tx: unable to map xmit buffer\n");
1063         adapter->tx_map_failed++;
1064         skb_linearize(skb);
1065         force_bounce = 1;
1066         goto retry_bounce;
1067 }
1068
1069 static int ibmveth_poll(struct napi_struct *napi, int budget)
1070 {
1071         struct ibmveth_adapter *adapter =
1072                         container_of(napi, struct ibmveth_adapter, napi);
1073         struct net_device *netdev = adapter->netdev;
1074         int frames_processed = 0;
1075         unsigned long lpar_rc;
1076
1077 restart_poll:
1078         do {
1079                 if (!ibmveth_rxq_pending_buffer(adapter))
1080                         break;
1081
1082                 smp_rmb();
1083                 if (!ibmveth_rxq_buffer_valid(adapter)) {
1084                         wmb(); /* suggested by larson1 */
1085                         adapter->rx_invalid_buffer++;
1086                         netdev_dbg(netdev, "recycling invalid buffer\n");
1087                         ibmveth_rxq_recycle_buffer(adapter);
1088                 } else {
1089                         struct sk_buff *skb, *new_skb;
1090                         int length = ibmveth_rxq_frame_length(adapter);
1091                         int offset = ibmveth_rxq_frame_offset(adapter);
1092                         int csum_good = ibmveth_rxq_csum_good(adapter);
1093
1094                         skb = ibmveth_rxq_get_buffer(adapter);
1095
1096                         new_skb = NULL;
1097                         if (length < rx_copybreak)
1098                                 new_skb = netdev_alloc_skb(netdev, length);
1099
1100                         if (new_skb) {
1101                                 skb_copy_to_linear_data(new_skb,
1102                                                         skb->data + offset,
1103                                                         length);
1104                                 if (rx_flush)
1105                                         ibmveth_flush_buffer(skb->data,
1106                                                 length + offset);
1107                                 if (!ibmveth_rxq_recycle_buffer(adapter))
1108                                         kfree_skb(skb);
1109                                 skb = new_skb;
1110                         } else {
1111                                 ibmveth_rxq_harvest_buffer(adapter);
1112                                 skb_reserve(skb, offset);
1113                         }
1114
1115                         skb_put(skb, length);
1116                         skb->protocol = eth_type_trans(skb, netdev);
1117
1118                         if (csum_good)
1119                                 skb->ip_summed = CHECKSUM_UNNECESSARY;
1120
1121                         netif_receive_skb(skb); /* send it up */
1122
1123                         netdev->stats.rx_packets++;
1124                         netdev->stats.rx_bytes += length;
1125                         frames_processed++;
1126                 }
1127         } while (frames_processed < budget);
1128
1129         ibmveth_replenish_task(adapter);
1130
1131         if (frames_processed < budget) {
1132                 /* We think we are done - reenable interrupts,
1133                  * then check once more to make sure we are done.
1134                  */
1135                 lpar_rc = h_vio_signal(adapter->vdev->unit_address,
1136                                        VIO_IRQ_ENABLE);
1137
1138                 BUG_ON(lpar_rc != H_SUCCESS);
1139
1140                 napi_complete(napi);
1141
1142                 if (ibmveth_rxq_pending_buffer(adapter) &&
1143                     napi_reschedule(napi)) {
1144                         lpar_rc = h_vio_signal(adapter->vdev->unit_address,
1145                                                VIO_IRQ_DISABLE);
1146                         goto restart_poll;
1147                 }
1148         }
1149
1150         return frames_processed;
1151 }
1152
1153 static irqreturn_t ibmveth_interrupt(int irq, void *dev_instance)
1154 {
1155         struct net_device *netdev = dev_instance;
1156         struct ibmveth_adapter *adapter = netdev_priv(netdev);
1157         unsigned long lpar_rc;
1158
1159         if (napi_schedule_prep(&adapter->napi)) {
1160                 lpar_rc = h_vio_signal(adapter->vdev->unit_address,
1161                                        VIO_IRQ_DISABLE);
1162                 BUG_ON(lpar_rc != H_SUCCESS);
1163                 __napi_schedule(&adapter->napi);
1164         }
1165         return IRQ_HANDLED;
1166 }
1167
1168 static void ibmveth_set_multicast_list(struct net_device *netdev)
1169 {
1170         struct ibmveth_adapter *adapter = netdev_priv(netdev);
1171         unsigned long lpar_rc;
1172
1173         if ((netdev->flags & IFF_PROMISC) ||
1174             (netdev_mc_count(netdev) > adapter->mcastFilterSize)) {
1175                 lpar_rc = h_multicast_ctrl(adapter->vdev->unit_address,
1176                                            IbmVethMcastEnableRecv |
1177                                            IbmVethMcastDisableFiltering,
1178                                            0);
1179                 if (lpar_rc != H_SUCCESS) {
1180                         netdev_err(netdev, "h_multicast_ctrl rc=%ld when "
1181                                    "entering promisc mode\n", lpar_rc);
1182                 }
1183         } else {
1184                 struct netdev_hw_addr *ha;
1185                 /* clear the filter table & disable filtering */
1186                 lpar_rc = h_multicast_ctrl(adapter->vdev->unit_address,
1187                                            IbmVethMcastEnableRecv |
1188                                            IbmVethMcastDisableFiltering |
1189                                            IbmVethMcastClearFilterTable,
1190                                            0);
1191                 if (lpar_rc != H_SUCCESS) {
1192                         netdev_err(netdev, "h_multicast_ctrl rc=%ld when "
1193                                    "attempting to clear filter table\n",
1194                                    lpar_rc);
1195                 }
1196                 /* add the addresses to the filter table */
1197                 netdev_for_each_mc_addr(ha, netdev) {
1198                         /* add the multicast address to the filter table */
1199                         unsigned long mcast_addr = 0;
1200                         memcpy(((char *)&mcast_addr)+2, ha->addr, 6);
1201                         lpar_rc = h_multicast_ctrl(adapter->vdev->unit_address,
1202                                                    IbmVethMcastAddFilter,
1203                                                    mcast_addr);
1204                         if (lpar_rc != H_SUCCESS) {
1205                                 netdev_err(netdev, "h_multicast_ctrl rc=%ld "
1206                                            "when adding an entry to the filter "
1207                                            "table\n", lpar_rc);
1208                         }
1209                 }
1210
1211                 /* re-enable filtering */
1212                 lpar_rc = h_multicast_ctrl(adapter->vdev->unit_address,
1213                                            IbmVethMcastEnableFiltering,
1214                                            0);
1215                 if (lpar_rc != H_SUCCESS) {
1216                         netdev_err(netdev, "h_multicast_ctrl rc=%ld when "
1217                                    "enabling filtering\n", lpar_rc);
1218                 }
1219         }
1220 }
1221
1222 static int ibmveth_change_mtu(struct net_device *dev, int new_mtu)
1223 {
1224         struct ibmveth_adapter *adapter = netdev_priv(dev);
1225         struct vio_dev *viodev = adapter->vdev;
1226         int new_mtu_oh = new_mtu + IBMVETH_BUFF_OH;
1227         int i, rc;
1228         int need_restart = 0;
1229
1230         if (new_mtu < IBMVETH_MIN_MTU)
1231                 return -EINVAL;
1232
1233         for (i = 0; i < IBMVETH_NUM_BUFF_POOLS; i++)
1234                 if (new_mtu_oh < adapter->rx_buff_pool[i].buff_size)
1235                         break;
1236
1237         if (i == IBMVETH_NUM_BUFF_POOLS)
1238                 return -EINVAL;
1239
1240         /* Deactivate all the buffer pools so that the next loop can activate
1241            only the buffer pools necessary to hold the new MTU */
1242         if (netif_running(adapter->netdev)) {
1243                 need_restart = 1;
1244                 adapter->pool_config = 1;
1245                 ibmveth_close(adapter->netdev);
1246                 adapter->pool_config = 0;
1247         }
1248
1249         /* Look for an active buffer pool that can hold the new MTU */
1250         for (i = 0; i < IBMVETH_NUM_BUFF_POOLS; i++) {
1251                 adapter->rx_buff_pool[i].active = 1;
1252
1253                 if (new_mtu_oh < adapter->rx_buff_pool[i].buff_size) {
1254                         dev->mtu = new_mtu;
1255                         vio_cmo_set_dev_desired(viodev,
1256                                                 ibmveth_get_desired_dma
1257                                                 (viodev));
1258                         if (need_restart) {
1259                                 return ibmveth_open(adapter->netdev);
1260                         }
1261                         return 0;
1262                 }
1263         }
1264
1265         if (need_restart && (rc = ibmveth_open(adapter->netdev)))
1266                 return rc;
1267
1268         return -EINVAL;
1269 }
1270
1271 #ifdef CONFIG_NET_POLL_CONTROLLER
1272 static void ibmveth_poll_controller(struct net_device *dev)
1273 {
1274         ibmveth_replenish_task(netdev_priv(dev));
1275         ibmveth_interrupt(dev->irq, dev);
1276 }
1277 #endif
1278
1279 /**
1280  * ibmveth_get_desired_dma - Calculate IO memory desired by the driver
1281  *
1282  * @vdev: struct vio_dev for the device whose desired IO mem is to be returned
1283  *
1284  * Return value:
1285  *      Number of bytes of IO data the driver will need to perform well.
1286  */
1287 static unsigned long ibmveth_get_desired_dma(struct vio_dev *vdev)
1288 {
1289         struct net_device *netdev = dev_get_drvdata(&vdev->dev);
1290         struct ibmveth_adapter *adapter;
1291         unsigned long ret;
1292         int i;
1293         int rxqentries = 1;
1294
1295         /* netdev inits at probe time along with the structures we need below*/
1296         if (netdev == NULL)
1297                 return IOMMU_PAGE_ALIGN(IBMVETH_IO_ENTITLEMENT_DEFAULT);
1298
1299         adapter = netdev_priv(netdev);
1300
1301         ret = IBMVETH_BUFF_LIST_SIZE + IBMVETH_FILT_LIST_SIZE;
1302         ret += IOMMU_PAGE_ALIGN(netdev->mtu);
1303
1304         for (i = 0; i < IBMVETH_NUM_BUFF_POOLS; i++) {
1305                 /* add the size of the active receive buffers */
1306                 if (adapter->rx_buff_pool[i].active)
1307                         ret +=
1308                             adapter->rx_buff_pool[i].size *
1309                             IOMMU_PAGE_ALIGN(adapter->rx_buff_pool[i].
1310                                     buff_size);
1311                 rxqentries += adapter->rx_buff_pool[i].size;
1312         }
1313         /* add the size of the receive queue entries */
1314         ret += IOMMU_PAGE_ALIGN(rxqentries * sizeof(struct ibmveth_rx_q_entry));
1315
1316         return ret;
1317 }
1318
1319 static const struct net_device_ops ibmveth_netdev_ops = {
1320         .ndo_open               = ibmveth_open,
1321         .ndo_stop               = ibmveth_close,
1322         .ndo_start_xmit         = ibmveth_start_xmit,
1323         .ndo_set_rx_mode        = ibmveth_set_multicast_list,
1324         .ndo_do_ioctl           = ibmveth_ioctl,
1325         .ndo_change_mtu         = ibmveth_change_mtu,
1326         .ndo_fix_features       = ibmveth_fix_features,
1327         .ndo_set_features       = ibmveth_set_features,
1328         .ndo_validate_addr      = eth_validate_addr,
1329         .ndo_set_mac_address    = eth_mac_addr,
1330 #ifdef CONFIG_NET_POLL_CONTROLLER
1331         .ndo_poll_controller    = ibmveth_poll_controller,
1332 #endif
1333 };
1334
1335 static int __devinit ibmveth_probe(struct vio_dev *dev,
1336                                    const struct vio_device_id *id)
1337 {
1338         int rc, i;
1339         struct net_device *netdev;
1340         struct ibmveth_adapter *adapter;
1341         unsigned char *mac_addr_p;
1342         unsigned int *mcastFilterSize_p;
1343
1344         dev_dbg(&dev->dev, "entering ibmveth_probe for UA 0x%x\n",
1345                 dev->unit_address);
1346
1347         mac_addr_p = (unsigned char *)vio_get_attribute(dev, VETH_MAC_ADDR,
1348                                                         NULL);
1349         if (!mac_addr_p) {
1350                 dev_err(&dev->dev, "Can't find VETH_MAC_ADDR attribute\n");
1351                 return -EINVAL;
1352         }
1353
1354         mcastFilterSize_p = (unsigned int *)vio_get_attribute(dev,
1355                                                 VETH_MCAST_FILTER_SIZE, NULL);
1356         if (!mcastFilterSize_p) {
1357                 dev_err(&dev->dev, "Can't find VETH_MCAST_FILTER_SIZE "
1358                         "attribute\n");
1359                 return -EINVAL;
1360         }
1361
1362         netdev = alloc_etherdev(sizeof(struct ibmveth_adapter));
1363
1364         if (!netdev)
1365                 return -ENOMEM;
1366
1367         adapter = netdev_priv(netdev);
1368         dev_set_drvdata(&dev->dev, netdev);
1369
1370         adapter->vdev = dev;
1371         adapter->netdev = netdev;
1372         adapter->mcastFilterSize = *mcastFilterSize_p;
1373         adapter->pool_config = 0;
1374
1375         netif_napi_add(netdev, &adapter->napi, ibmveth_poll, 16);
1376
1377         /*
1378          * Some older boxes running PHYP non-natively have an OF that returns
1379          * a 8-byte local-mac-address field (and the first 2 bytes have to be
1380          * ignored) while newer boxes' OF return a 6-byte field. Note that
1381          * IEEE 1275 specifies that local-mac-address must be a 6-byte field.
1382          * The RPA doc specifies that the first byte must be 10b, so we'll
1383          * just look for it to solve this 8 vs. 6 byte field issue
1384          */
1385         if ((*mac_addr_p & 0x3) != 0x02)
1386                 mac_addr_p += 2;
1387
1388         adapter->mac_addr = 0;
1389         memcpy(&adapter->mac_addr, mac_addr_p, 6);
1390
1391         netdev->irq = dev->irq;
1392         netdev->netdev_ops = &ibmveth_netdev_ops;
1393         netdev->ethtool_ops = &netdev_ethtool_ops;
1394         SET_NETDEV_DEV(netdev, &dev->dev);
1395         netdev->hw_features = NETIF_F_SG | NETIF_F_RXCSUM |
1396                 NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM;
1397         netdev->features |= netdev->hw_features;
1398
1399         memcpy(netdev->dev_addr, &adapter->mac_addr, netdev->addr_len);
1400
1401         for (i = 0; i < IBMVETH_NUM_BUFF_POOLS; i++) {
1402                 struct kobject *kobj = &adapter->rx_buff_pool[i].kobj;
1403                 int error;
1404
1405                 ibmveth_init_buffer_pool(&adapter->rx_buff_pool[i], i,
1406                                          pool_count[i], pool_size[i],
1407                                          pool_active[i]);
1408                 error = kobject_init_and_add(kobj, &ktype_veth_pool,
1409                                              &dev->dev.kobj, "pool%d", i);
1410                 if (!error)
1411                         kobject_uevent(kobj, KOBJ_ADD);
1412         }
1413
1414         netdev_dbg(netdev, "adapter @ 0x%p\n", adapter);
1415
1416         adapter->buffer_list_dma = DMA_ERROR_CODE;
1417         adapter->filter_list_dma = DMA_ERROR_CODE;
1418         adapter->rx_queue.queue_dma = DMA_ERROR_CODE;
1419
1420         netdev_dbg(netdev, "registering netdev...\n");
1421
1422         ibmveth_set_features(netdev, netdev->features);
1423
1424         rc = register_netdev(netdev);
1425
1426         if (rc) {
1427                 netdev_dbg(netdev, "failed to register netdev rc=%d\n", rc);
1428                 free_netdev(netdev);
1429                 return rc;
1430         }
1431
1432         netdev_dbg(netdev, "registered\n");
1433
1434         return 0;
1435 }
1436
1437 static int __devexit ibmveth_remove(struct vio_dev *dev)
1438 {
1439         struct net_device *netdev = dev_get_drvdata(&dev->dev);
1440         struct ibmveth_adapter *adapter = netdev_priv(netdev);
1441         int i;
1442
1443         for (i = 0; i < IBMVETH_NUM_BUFF_POOLS; i++)
1444                 kobject_put(&adapter->rx_buff_pool[i].kobj);
1445
1446         unregister_netdev(netdev);
1447
1448         free_netdev(netdev);
1449         dev_set_drvdata(&dev->dev, NULL);
1450
1451         return 0;
1452 }
1453
1454 static struct attribute veth_active_attr;
1455 static struct attribute veth_num_attr;
1456 static struct attribute veth_size_attr;
1457
1458 static ssize_t veth_pool_show(struct kobject *kobj,
1459                               struct attribute *attr, char *buf)
1460 {
1461         struct ibmveth_buff_pool *pool = container_of(kobj,
1462                                                       struct ibmveth_buff_pool,
1463                                                       kobj);
1464
1465         if (attr == &veth_active_attr)
1466                 return sprintf(buf, "%d\n", pool->active);
1467         else if (attr == &veth_num_attr)
1468                 return sprintf(buf, "%d\n", pool->size);
1469         else if (attr == &veth_size_attr)
1470                 return sprintf(buf, "%d\n", pool->buff_size);
1471         return 0;
1472 }
1473
1474 static ssize_t veth_pool_store(struct kobject *kobj, struct attribute *attr,
1475                                const char *buf, size_t count)
1476 {
1477         struct ibmveth_buff_pool *pool = container_of(kobj,
1478                                                       struct ibmveth_buff_pool,
1479                                                       kobj);
1480         struct net_device *netdev = dev_get_drvdata(
1481             container_of(kobj->parent, struct device, kobj));
1482         struct ibmveth_adapter *adapter = netdev_priv(netdev);
1483         long value = simple_strtol(buf, NULL, 10);
1484         long rc;
1485
1486         if (attr == &veth_active_attr) {
1487                 if (value && !pool->active) {
1488                         if (netif_running(netdev)) {
1489                                 if (ibmveth_alloc_buffer_pool(pool)) {
1490                                         netdev_err(netdev,
1491                                                    "unable to alloc pool\n");
1492                                         return -ENOMEM;
1493                                 }
1494                                 pool->active = 1;
1495                                 adapter->pool_config = 1;
1496                                 ibmveth_close(netdev);
1497                                 adapter->pool_config = 0;
1498                                 if ((rc = ibmveth_open(netdev)))
1499                                         return rc;
1500                         } else {
1501                                 pool->active = 1;
1502                         }
1503                 } else if (!value && pool->active) {
1504                         int mtu = netdev->mtu + IBMVETH_BUFF_OH;
1505                         int i;
1506                         /* Make sure there is a buffer pool with buffers that
1507                            can hold a packet of the size of the MTU */
1508                         for (i = 0; i < IBMVETH_NUM_BUFF_POOLS; i++) {
1509                                 if (pool == &adapter->rx_buff_pool[i])
1510                                         continue;
1511                                 if (!adapter->rx_buff_pool[i].active)
1512                                         continue;
1513                                 if (mtu <= adapter->rx_buff_pool[i].buff_size)
1514                                         break;
1515                         }
1516
1517                         if (i == IBMVETH_NUM_BUFF_POOLS) {
1518                                 netdev_err(netdev, "no active pool >= MTU\n");
1519                                 return -EPERM;
1520                         }
1521
1522                         if (netif_running(netdev)) {
1523                                 adapter->pool_config = 1;
1524                                 ibmveth_close(netdev);
1525                                 pool->active = 0;
1526                                 adapter->pool_config = 0;
1527                                 if ((rc = ibmveth_open(netdev)))
1528                                         return rc;
1529                         }
1530                         pool->active = 0;
1531                 }
1532         } else if (attr == &veth_num_attr) {
1533                 if (value <= 0 || value > IBMVETH_MAX_POOL_COUNT) {
1534                         return -EINVAL;
1535                 } else {
1536                         if (netif_running(netdev)) {
1537                                 adapter->pool_config = 1;
1538                                 ibmveth_close(netdev);
1539                                 adapter->pool_config = 0;
1540                                 pool->size = value;
1541                                 if ((rc = ibmveth_open(netdev)))
1542                                         return rc;
1543                         } else {
1544                                 pool->size = value;
1545                         }
1546                 }
1547         } else if (attr == &veth_size_attr) {
1548                 if (value <= IBMVETH_BUFF_OH || value > IBMVETH_MAX_BUF_SIZE) {
1549                         return -EINVAL;
1550                 } else {
1551                         if (netif_running(netdev)) {
1552                                 adapter->pool_config = 1;
1553                                 ibmveth_close(netdev);
1554                                 adapter->pool_config = 0;
1555                                 pool->buff_size = value;
1556                                 if ((rc = ibmveth_open(netdev)))
1557                                         return rc;
1558                         } else {
1559                                 pool->buff_size = value;
1560                         }
1561                 }
1562         }
1563
1564         /* kick the interrupt handler to allocate/deallocate pools */
1565         ibmveth_interrupt(netdev->irq, netdev);
1566         return count;
1567 }
1568
1569
1570 #define ATTR(_name, _mode)                              \
1571         struct attribute veth_##_name##_attr = {        \
1572         .name = __stringify(_name), .mode = _mode,      \
1573         };
1574
1575 static ATTR(active, 0644);
1576 static ATTR(num, 0644);
1577 static ATTR(size, 0644);
1578
1579 static struct attribute *veth_pool_attrs[] = {
1580         &veth_active_attr,
1581         &veth_num_attr,
1582         &veth_size_attr,
1583         NULL,
1584 };
1585
1586 static const struct sysfs_ops veth_pool_ops = {
1587         .show   = veth_pool_show,
1588         .store  = veth_pool_store,
1589 };
1590
1591 static struct kobj_type ktype_veth_pool = {
1592         .release        = NULL,
1593         .sysfs_ops      = &veth_pool_ops,
1594         .default_attrs  = veth_pool_attrs,
1595 };
1596
1597 static int ibmveth_resume(struct device *dev)
1598 {
1599         struct net_device *netdev = dev_get_drvdata(dev);
1600         ibmveth_interrupt(netdev->irq, netdev);
1601         return 0;
1602 }
1603
1604 static struct vio_device_id ibmveth_device_table[] __devinitdata = {
1605         { "network", "IBM,l-lan"},
1606         { "", "" }
1607 };
1608 MODULE_DEVICE_TABLE(vio, ibmveth_device_table);
1609
1610 static struct dev_pm_ops ibmveth_pm_ops = {
1611         .resume = ibmveth_resume
1612 };
1613
1614 static struct vio_driver ibmveth_driver = {
1615         .id_table       = ibmveth_device_table,
1616         .probe          = ibmveth_probe,
1617         .remove         = ibmveth_remove,
1618         .get_desired_dma = ibmveth_get_desired_dma,
1619         .name           = ibmveth_driver_name,
1620         .pm             = &ibmveth_pm_ops,
1621 };
1622
1623 static int __init ibmveth_module_init(void)
1624 {
1625         printk(KERN_DEBUG "%s: %s %s\n", ibmveth_driver_name,
1626                ibmveth_driver_string, ibmveth_driver_version);
1627
1628         return vio_register_driver(&ibmveth_driver);
1629 }
1630
1631 static void __exit ibmveth_module_exit(void)
1632 {
1633         vio_unregister_driver(&ibmveth_driver);
1634 }
1635
1636 module_init(ibmveth_module_init);
1637 module_exit(ibmveth_module_exit);