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, see <http://www.gnu.org/licenses/>.
  16  *
  17  * Copyright (C) IBM Corporation, 2003, 2010
  18  *
  19  * Authors: Dave Larson <larson1@us.ibm.com>
  20  *          Santiago Leon <santil@linux.vnet.ibm.com>
  21  *          Brian King <brking@linux.vnet.ibm.com>
  22  *          Robert Jennings <rcj@linux.vnet.ibm.com>
  23  *          Anton Blanchard <anton@au.ibm.com>
  24  */
  25
  26 #include <linux/module.h>
  27 #include <linux/moduleparam.h>
  28 #include <linux/types.h>
  29 #include <linux/errno.h>
  30 #include <linux/dma-mapping.h>
  31 #include <linux/kernel.h>
  32 #include <linux/netdevice.h>
  33 #include <linux/etherdevice.h>
  34 #include <linux/skbuff.h>
  35 #include <linux/init.h>
  36 #include <linux/interrupt.h>
  37 #include <linux/mm.h>
  38 #include <linux/pm.h>
  39 #include <linux/ethtool.h>
  40 #include <linux/in.h>
  41 #include <linux/ip.h>
  42 #include <linux/ipv6.h>
  43 #include <linux/slab.h>
  44 #include <asm/hvcall.h>
  45 #include <linux/atomic.h>
  46 #include <asm/vio.h>
  47 #include <asm/iommu.h>
  48 #include <asm/firmware.h>
  49
  50 #include "ibmveth.h"
  51
  52 static irqreturn_t ibmveth_interrupt(int irq, void *dev_instance);
  53 static void ibmveth_rxq_harvest_buffer(struct ibmveth_adapter *adapter);
  54 static unsigned long ibmveth_get_desired_dma(struct vio_dev *vdev);
  55
  56 static struct kobj_type ktype_veth_pool;
  57
  58
  59 static const char ibmveth_driver_name[] = "ibmveth";
  60 static const char ibmveth_driver_string[] = "IBM Power Virtual Ethernet Driver";
  61 #define ibmveth_driver_version "1.05"
  62
  63 MODULE_AUTHOR("Santiago Leon <santil@linux.vnet.ibm.com>");
  64 MODULE_DESCRIPTION("IBM Power Virtual Ethernet Driver");
  65 MODULE_LICENSE("GPL");
  66 MODULE_VERSION(ibmveth_driver_version);
  67
  68 static unsigned int tx_copybreak __read_mostly = 128;
  69 module_param(tx_copybreak, uint, 0644);
  70 MODULE_PARM_DESC(tx_copybreak,
  71         "Maximum size of packet that is copied to a new buffer on transmit");
  72
  73 static unsigned int rx_copybreak __read_mostly = 128;
  74 module_param(rx_copybreak, uint, 0644);
  75 MODULE_PARM_DESC(rx_copybreak,
  76         "Maximum size of packet that is copied to a new buffer on receive");
  77
  78 static unsigned int rx_flush __read_mostly = 0;
  79 module_param(rx_flush, uint, 0644);
  80 MODULE_PARM_DESC(rx_flush, "Flush receive buffers before use");
  81
  82 struct ibmveth_stat {
  83         char name[ETH_GSTRING_LEN];
  84         int offset;
  85 };
  86
  87 #define IBMVETH_STAT_OFF(stat) offsetof(struct ibmveth_adapter, stat)
  88 #define IBMVETH_GET_STAT(a, off) *((u64 *)(((unsigned long)(a)) + off))
  89
  90 struct ibmveth_stat ibmveth_stats[] = {
  91         { "replenish_task_cycles", IBMVETH_STAT_OFF(replenish_task_cycles) },
  92         { "replenish_no_mem", IBMVETH_STAT_OFF(replenish_no_mem) },
  93         { "replenish_add_buff_failure",
  94                         IBMVETH_STAT_OFF(replenish_add_buff_failure) },
  95         { "replenish_add_buff_success",
  96                         IBMVETH_STAT_OFF(replenish_add_buff_success) },
  97         { "rx_invalid_buffer", IBMVETH_STAT_OFF(rx_invalid_buffer) },
  98         { "rx_no_buffer", IBMVETH_STAT_OFF(rx_no_buffer) },
  99         { "tx_map_failed", IBMVETH_STAT_OFF(tx_map_failed) },
 100         { "tx_send_failed", IBMVETH_STAT_OFF(tx_send_failed) },
 101         { "fw_enabled_ipv4_csum", IBMVETH_STAT_OFF(fw_ipv4_csum_support) },
 102         { "fw_enabled_ipv6_csum", IBMVETH_STAT_OFF(fw_ipv6_csum_support) },
 103         { "tx_large_packets", IBMVETH_STAT_OFF(tx_large_packets) },
 104         { "rx_large_packets", IBMVETH_STAT_OFF(rx_large_packets) }
 105 };
 106
 107 /* simple methods of getting data from the current rxq entry */
 108 static inline u32 ibmveth_rxq_flags(struct ibmveth_adapter *adapter)
 109 {
 110         return be32_to_cpu(adapter->rx_queue.queue_addr[adapter->rx_queue.index].flags_off);
 111 }
 112
 113 static inline int ibmveth_rxq_toggle(struct ibmveth_adapter *adapter)
 114 {
 115         return (ibmveth_rxq_flags(adapter) & IBMVETH_RXQ_TOGGLE) >>
 116                         IBMVETH_RXQ_TOGGLE_SHIFT;
 117 }
 118
 119 static inline int ibmveth_rxq_pending_buffer(struct ibmveth_adapter *adapter)
 120 {
 121         return ibmveth_rxq_toggle(adapter) == adapter->rx_queue.toggle;
 122 }
 123
 124 static inline int ibmveth_rxq_buffer_valid(struct ibmveth_adapter *adapter)
 125 {
 126         return ibmveth_rxq_flags(adapter) & IBMVETH_RXQ_VALID;
 127 }
 128
 129 static inline int ibmveth_rxq_frame_offset(struct ibmveth_adapter *adapter)
 130 {
 131         return ibmveth_rxq_flags(adapter) & IBMVETH_RXQ_OFF_MASK;
 132 }
 133
 134 static inline int ibmveth_rxq_frame_length(struct ibmveth_adapter *adapter)
 135 {
 136         return be32_to_cpu(adapter->rx_queue.queue_addr[adapter->rx_queue.index].length);
 137 }
 138
 139 static inline int ibmveth_rxq_csum_good(struct ibmveth_adapter *adapter)
 140 {
 141         return ibmveth_rxq_flags(adapter) & IBMVETH_RXQ_CSUM_GOOD;
 142 }
 143
 144 /* setup the initial settings for a buffer pool */
 145 static void ibmveth_init_buffer_pool(struct ibmveth_buff_pool *pool,
 146                                      u32 pool_index, u32 pool_size,
 147                                      u32 buff_size, u32 pool_active)
 148 {
 149         pool->size = pool_size;
 150         pool->index = pool_index;
 151         pool->buff_size = buff_size;
 152         pool->threshold = pool_size * 7 / 8;
 153         pool->active = pool_active;
 154 }
 155
 156 /* allocate and setup an buffer pool - called during open */
 157 static int ibmveth_alloc_buffer_pool(struct ibmveth_buff_pool *pool)
 158 {
 159         int i;
 160
 161         pool->free_map = kmalloc(sizeof(u16) * pool->size, GFP_KERNEL);
 162
 163         if (!pool->free_map)
 164                 return -1;
 165
 166         pool->dma_addr = kmalloc(sizeof(dma_addr_t) * pool->size, GFP_KERNEL);
 167         if (!pool->dma_addr) {
 168                 kfree(pool->free_map);
 169                 pool->free_map = NULL;
 170                 return -1;
 171         }
 172
 173         pool->skbuff = kcalloc(pool->size, sizeof(void *), GFP_KERNEL);
 174
 175         if (!pool->skbuff) {
 176                 kfree(pool->dma_addr);
 177                 pool->dma_addr = NULL;
 178
 179                 kfree(pool->free_map);
 180                 pool->free_map = NULL;
 181                 return -1;
 182         }
 183
 184         memset(pool->dma_addr, 0, sizeof(dma_addr_t) * pool->size);
 185
 186         for (i = 0; i < pool->size; ++i)
 187                 pool->free_map[i] = i;
 188
 189         atomic_set(&pool->available, 0);
 190         pool->producer_index = 0;
 191         pool->consumer_index = 0;
 192
 193         return 0;
 194 }
 195
 196 static inline void ibmveth_flush_buffer(void *addr, unsigned long length)
 197 {
 198         unsigned long offset;
 199
 200         for (offset = 0; offset < length; offset += SMP_CACHE_BYTES)
 201                 asm("dcbfl %0,%1" :: "b" (addr), "r" (offset));
 202 }
 203
 204 /* replenish the buffers for a pool.  note that we don't need to
 205  * skb_reserve these since they are used for incoming...
 206  */
 207 static void ibmveth_replenish_buffer_pool(struct ibmveth_adapter *adapter,
 208                                           struct ibmveth_buff_pool *pool)
 209 {
 210         u32 i;
 211         u32 count = pool->size - atomic_read(&pool->available);
 212         u32 buffers_added = 0;
 213         struct sk_buff *skb;
 214         unsigned int free_index, index;
 215         u64 correlator;
 216         unsigned long lpar_rc;
 217         dma_addr_t dma_addr;
 218
 219         mb();
 220
 221         for (i = 0; i < count; ++i) {
 222                 union ibmveth_buf_desc desc;
 223
 224                 skb = netdev_alloc_skb(adapter->netdev, pool->buff_size);
 225
 226                 if (!skb) {
 227                         netdev_dbg(adapter->netdev,
 228                                    "replenish: unable to allocate skb\n");
 229                         adapter->replenish_no_mem++;
 230                         break;
 231                 }
 232
 233                 free_index = pool->consumer_index;
 234                 pool->consumer_index++;
 235                 if (pool->consumer_index >= pool->size)
 236                         pool->consumer_index = 0;
 237                 index = pool->free_map[free_index];
 238
 239                 BUG_ON(index == IBM_VETH_INVALID_MAP);
 240                 BUG_ON(pool->skbuff[index] != NULL);
 241
 242                 dma_addr = dma_map_single(&adapter->vdev->dev, skb->data,
 243                                 pool->buff_size, DMA_FROM_DEVICE);
 244
 245                 if (dma_mapping_error(&adapter->vdev->dev, dma_addr))
 246                         goto failure;
 247
 248                 pool->free_map[free_index] = IBM_VETH_INVALID_MAP;
 249                 pool->dma_addr[index] = dma_addr;
 250                 pool->skbuff[index] = skb;
 251
 252                 correlator = ((u64)pool->index << 32) | index;
 253                 *(u64 *)skb->data = correlator;
 254
 255                 desc.fields.flags_len = IBMVETH_BUF_VALID | pool->buff_size;
 256                 desc.fields.address = dma_addr;
 257
 258                 if (rx_flush) {
 259                         unsigned int len = min(pool->buff_size,
 260                                                 adapter->netdev->mtu +
 261                                                 IBMVETH_BUFF_OH);
 262                         ibmveth_flush_buffer(skb->data, len);
 263                 }
 264                 lpar_rc = h_add_logical_lan_buffer(adapter->vdev->unit_address,
 265                                                    desc.desc);
 266
 267                 if (lpar_rc != H_SUCCESS) {
 268                         goto failure;
 269                 } else {
 270                         buffers_added++;
 271                         adapter->replenish_add_buff_success++;
 272                 }
 273         }
 274
 275         mb();
 276         atomic_add(buffers_added, &(pool->available));
 277         return;
 278
 279 failure:
 280         pool->free_map[free_index] = index;
 281         pool->skbuff[index] = NULL;
 282         if (pool->consumer_index == 0)
 283                 pool->consumer_index = pool->size - 1;
 284         else
 285                 pool->consumer_index--;
 286         if (!dma_mapping_error(&adapter->vdev->dev, dma_addr))
 287                 dma_unmap_single(&adapter->vdev->dev,
 288                                  pool->dma_addr[index], pool->buff_size,
 289                                  DMA_FROM_DEVICE);
 290         dev_kfree_skb_any(skb);
 291         adapter->replenish_add_buff_failure++;
 292
 293         mb();
 294         atomic_add(buffers_added, &(pool->available));
 295 }
 296
 297 /*
 298  * The final 8 bytes of the buffer list is a counter of frames dropped
 299  * because there was not a buffer in the buffer list capable of holding
 300  * the frame.
 301  */
 302 static void ibmveth_update_rx_no_buffer(struct ibmveth_adapter *adapter)
 303 {
 304         __be64 *p = adapter->buffer_list_addr + 4096 - 8;
 305
 306         adapter->rx_no_buffer = be64_to_cpup(p);
 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 = (IBMVETH_NUM_BUFF_POOLS - 1); i >= 0; i--) {
 317                 struct ibmveth_buff_pool *pool = &adapter->rx_buff_pool[i];
 318
 319                 if (pool->active &&
 320                     (atomic_read(&pool->available) < pool->threshold))
 321                         ibmveth_replenish_buffer_pool(adapter, pool);
 322         }
 323
 324         ibmveth_update_rx_no_buffer(adapter);
 325 }
 326
 327 /* empty and free ana buffer pool - also used to do cleanup in error paths */
 328 static void ibmveth_free_buffer_pool(struct ibmveth_adapter *adapter,
 329                                      struct ibmveth_buff_pool *pool)
 330 {
 331         int i;
 332
 333         kfree(pool->free_map);
 334         pool->free_map = NULL;
 335
 336         if (pool->skbuff && pool->dma_addr) {
 337                 for (i = 0; i < pool->size; ++i) {
 338                         struct sk_buff *skb = pool->skbuff[i];
 339                         if (skb) {
 340                                 dma_unmap_single(&adapter->vdev->dev,
 341                                                  pool->dma_addr[i],
 342                                                  pool->buff_size,
 343                                                  DMA_FROM_DEVICE);
 344                                 dev_kfree_skb_any(skb);
 345                                 pool->skbuff[i] = NULL;
 346                         }
 347                 }
 348         }
 349
 350         if (pool->dma_addr) {
 351                 kfree(pool->dma_addr);
 352                 pool->dma_addr = NULL;
 353         }
 354
 355         if (pool->skbuff) {
 356                 kfree(pool->skbuff);
 357                 pool->skbuff = NULL;
 358         }
 359 }
 360
 361 /* remove a buffer from a pool */
 362 static void ibmveth_remove_buffer_from_pool(struct ibmveth_adapter *adapter,
 363                                             u64 correlator)
 364 {
 365         unsigned int pool  = correlator >> 32;
 366         unsigned int index = correlator & 0xffffffffUL;
 367         unsigned int free_index;
 368         struct sk_buff *skb;
 369
 370         BUG_ON(pool >= IBMVETH_NUM_BUFF_POOLS);
 371         BUG_ON(index >= adapter->rx_buff_pool[pool].size);
 372
 373         skb = adapter->rx_buff_pool[pool].skbuff[index];
 374
 375         BUG_ON(skb == NULL);
 376
 377         adapter->rx_buff_pool[pool].skbuff[index] = NULL;
 378
 379         dma_unmap_single(&adapter->vdev->dev,
 380                          adapter->rx_buff_pool[pool].dma_addr[index],
 381                          adapter->rx_buff_pool[pool].buff_size,
 382                          DMA_FROM_DEVICE);
 383
 384         free_index = adapter->rx_buff_pool[pool].producer_index;
 385         adapter->rx_buff_pool[pool].producer_index++;
 386         if (adapter->rx_buff_pool[pool].producer_index >=
 387             adapter->rx_buff_pool[pool].size)
 388                 adapter->rx_buff_pool[pool].producer_index = 0;
 389         adapter->rx_buff_pool[pool].free_map[free_index] = index;
 390
 391         mb();
 392
 393         atomic_dec(&(adapter->rx_buff_pool[pool].available));
 394 }
 395
 396 /* get the current buffer on the rx queue */
 397 static inline struct sk_buff *ibmveth_rxq_get_buffer(struct ibmveth_adapter *adapter)
 398 {
 399         u64 correlator = adapter->rx_queue.queue_addr[adapter->rx_queue.index].correlator;
 400         unsigned int pool = correlator >> 32;
 401         unsigned int index = correlator & 0xffffffffUL;
 402
 403         BUG_ON(pool >= IBMVETH_NUM_BUFF_POOLS);
 404         BUG_ON(index >= adapter->rx_buff_pool[pool].size);
 405
 406         return adapter->rx_buff_pool[pool].skbuff[index];
 407 }
 408
 409 /* recycle the current buffer on the rx queue */
 410 static int ibmveth_rxq_recycle_buffer(struct ibmveth_adapter *adapter)
 411 {
 412         u32 q_index = adapter->rx_queue.index;
 413         u64 correlator = adapter->rx_queue.queue_addr[q_index].correlator;
 414         unsigned int pool = correlator >> 32;
 415         unsigned int index = correlator & 0xffffffffUL;
 416         union ibmveth_buf_desc desc;
 417         unsigned long lpar_rc;
 418         int ret = 1;
 419
 420         BUG_ON(pool >= IBMVETH_NUM_BUFF_POOLS);
 421         BUG_ON(index >= adapter->rx_buff_pool[pool].size);
 422
 423         if (!adapter->rx_buff_pool[pool].active) {
 424                 ibmveth_rxq_harvest_buffer(adapter);
 425                 ibmveth_free_buffer_pool(adapter, &adapter->rx_buff_pool[pool]);
 426                 goto out;
 427         }
 428
 429         desc.fields.flags_len = IBMVETH_BUF_VALID |
 430                 adapter->rx_buff_pool[pool].buff_size;
 431         desc.fields.address = adapter->rx_buff_pool[pool].dma_addr[index];
 432
 433         lpar_rc = h_add_logical_lan_buffer(adapter->vdev->unit_address, desc.desc);
 434
 435         if (lpar_rc != H_SUCCESS) {
 436                 netdev_dbg(adapter->netdev, "h_add_logical_lan_buffer failed "
 437                            "during recycle rc=%ld", lpar_rc);
 438                 ibmveth_remove_buffer_from_pool(adapter, adapter->rx_queue.queue_addr[adapter->rx_queue.index].correlator);
 439                 ret = 0;
 440         }
 441
 442         if (++adapter->rx_queue.index == adapter->rx_queue.num_slots) {
 443                 adapter->rx_queue.index = 0;
 444                 adapter->rx_queue.toggle = !adapter->rx_queue.toggle;
 445         }
 446
 447 out:
 448         return ret;
 449 }
 450
 451 static void ibmveth_rxq_harvest_buffer(struct ibmveth_adapter *adapter)
 452 {
 453         ibmveth_remove_buffer_from_pool(adapter, adapter->rx_queue.queue_addr[adapter->rx_queue.index].correlator);
 454
 455         if (++adapter->rx_queue.index == adapter->rx_queue.num_slots) {
 456                 adapter->rx_queue.index = 0;
 457                 adapter->rx_queue.toggle = !adapter->rx_queue.toggle;
 458         }
 459 }
 460
 461 static void ibmveth_cleanup(struct ibmveth_adapter *adapter)
 462 {
 463         int i;
 464         struct device *dev = &adapter->vdev->dev;
 465
 466         if (adapter->buffer_list_addr != NULL) {
 467                 if (!dma_mapping_error(dev, adapter->buffer_list_dma)) {
 468                         dma_unmap_single(dev, adapter->buffer_list_dma, 4096,
 469                                         DMA_BIDIRECTIONAL);
 470                         adapter->buffer_list_dma = DMA_ERROR_CODE;
 471                 }
 472                 free_page((unsigned long)adapter->buffer_list_addr);
 473                 adapter->buffer_list_addr = NULL;
 474         }
 475
 476         if (adapter->filter_list_addr != NULL) {
 477                 if (!dma_mapping_error(dev, adapter->filter_list_dma)) {
 478                         dma_unmap_single(dev, adapter->filter_list_dma, 4096,
 479                                         DMA_BIDIRECTIONAL);
 480                         adapter->filter_list_dma = DMA_ERROR_CODE;
 481                 }
 482                 free_page((unsigned long)adapter->filter_list_addr);
 483                 adapter->filter_list_addr = NULL;
 484         }
 485
 486         if (adapter->rx_queue.queue_addr != NULL) {
 487                 dma_free_coherent(dev, adapter->rx_queue.queue_len,
 488                                   adapter->rx_queue.queue_addr,
 489                                   adapter->rx_queue.queue_dma);
 490                 adapter->rx_queue.queue_addr = NULL;
 491         }
 492
 493         for (i = 0; i < IBMVETH_NUM_BUFF_POOLS; i++)
 494                 if (adapter->rx_buff_pool[i].active)
 495                         ibmveth_free_buffer_pool(adapter,
 496                                                  &adapter->rx_buff_pool[i]);
 497
 498         if (adapter->bounce_buffer != NULL) {
 499                 if (!dma_mapping_error(dev, adapter->bounce_buffer_dma)) {
 500                         dma_unmap_single(&adapter->vdev->dev,
 501                                         adapter->bounce_buffer_dma,
 502                                         adapter->netdev->mtu + IBMVETH_BUFF_OH,
 503                                         DMA_BIDIRECTIONAL);
 504                         adapter->bounce_buffer_dma = DMA_ERROR_CODE;
 505                 }
 506                 kfree(adapter->bounce_buffer);
 507                 adapter->bounce_buffer = NULL;
 508         }
 509 }
 510
 511 static int ibmveth_register_logical_lan(struct ibmveth_adapter *adapter,
 512         union ibmveth_buf_desc rxq_desc, u64 mac_address)
 513 {
 514         int rc, try_again = 1;
 515
 516         /*
 517          * After a kexec the adapter will still be open, so our attempt to
 518          * open it will fail. So if we get a failure we free the adapter and
 519          * try again, but only once.
 520          */
 521 retry:
 522         rc = h_register_logical_lan(adapter->vdev->unit_address,
 523                                     adapter->buffer_list_dma, rxq_desc.desc,
 524                                     adapter->filter_list_dma, mac_address);
 525
 526         if (rc != H_SUCCESS && try_again) {
 527                 do {
 528                         rc = h_free_logical_lan(adapter->vdev->unit_address);
 529                 } while (H_IS_LONG_BUSY(rc) || (rc == H_BUSY));
 530
 531                 try_again = 0;
 532                 goto retry;
 533         }
 534
 535         return rc;
 536 }
 537
 538 static u64 ibmveth_encode_mac_addr(u8 *mac)
 539 {
 540         int i;
 541         u64 encoded = 0;
 542
 543         for (i = 0; i < ETH_ALEN; i++)
 544                 encoded = (encoded << 8) | mac[i];
 545
 546         return encoded;
 547 }
 548
 549 static int ibmveth_open(struct net_device *netdev)
 550 {
 551         struct ibmveth_adapter *adapter = netdev_priv(netdev);
 552         u64 mac_address;
 553         int rxq_entries = 1;
 554         unsigned long lpar_rc;
 555         int rc;
 556         union ibmveth_buf_desc rxq_desc;
 557         int i;
 558         struct device *dev;
 559
 560         netdev_dbg(netdev, "open starting\n");
 561
 562         napi_enable(&adapter->napi);
 563
 564         for(i = 0; i < IBMVETH_NUM_BUFF_POOLS; i++)
 565                 rxq_entries += adapter->rx_buff_pool[i].size;
 566
 567         adapter->buffer_list_addr = (void*) get_zeroed_page(GFP_KERNEL);
 568         adapter->filter_list_addr = (void*) get_zeroed_page(GFP_KERNEL);
 569
 570         if (!adapter->buffer_list_addr || !adapter->filter_list_addr) {
 571                 netdev_err(netdev, "unable to allocate filter or buffer list "
 572                            "pages\n");
 573                 rc = -ENOMEM;
 574                 goto err_out;
 575         }
 576
 577         dev = &adapter->vdev->dev;
 578
 579         adapter->rx_queue.queue_len = sizeof(struct ibmveth_rx_q_entry) *
 580                                                 rxq_entries;
 581         adapter->rx_queue.queue_addr =
 582                 dma_alloc_coherent(dev, adapter->rx_queue.queue_len,
 583                                    &adapter->rx_queue.queue_dma, GFP_KERNEL);
 584         if (!adapter->rx_queue.queue_addr) {
 585                 rc = -ENOMEM;
 586                 goto err_out;
 587         }
 588
 589         adapter->buffer_list_dma = dma_map_single(dev,
 590                         adapter->buffer_list_addr, 4096, DMA_BIDIRECTIONAL);
 591         adapter->filter_list_dma = dma_map_single(dev,
 592                         adapter->filter_list_addr, 4096, DMA_BIDIRECTIONAL);
 593
 594         if ((dma_mapping_error(dev, adapter->buffer_list_dma)) ||
 595             (dma_mapping_error(dev, adapter->filter_list_dma))) {
 596                 netdev_err(netdev, "unable to map filter or buffer list "
 597                            "pages\n");
 598                 rc = -ENOMEM;
 599                 goto err_out;
 600         }
 601
 602         adapter->rx_queue.index = 0;
 603         adapter->rx_queue.num_slots = rxq_entries;
 604         adapter->rx_queue.toggle = 1;
 605
 606         mac_address = ibmveth_encode_mac_addr(netdev->dev_addr);
 607
 608         rxq_desc.fields.flags_len = IBMVETH_BUF_VALID |
 609                                         adapter->rx_queue.queue_len;
 610         rxq_desc.fields.address = adapter->rx_queue.queue_dma;
 611
 612         netdev_dbg(netdev, "buffer list @ 0x%p\n", adapter->buffer_list_addr);
 613         netdev_dbg(netdev, "filter list @ 0x%p\n", adapter->filter_list_addr);
 614         netdev_dbg(netdev, "receive q   @ 0x%p\n", adapter->rx_queue.queue_addr);
 615
 616         h_vio_signal(adapter->vdev->unit_address, VIO_IRQ_DISABLE);
 617
 618         lpar_rc = ibmveth_register_logical_lan(adapter, rxq_desc, mac_address);
 619
 620         if (lpar_rc != H_SUCCESS) {
 621                 netdev_err(netdev, "h_register_logical_lan failed with %ld\n",
 622                            lpar_rc);
 623                 netdev_err(netdev, "buffer TCE:0x%llx filter TCE:0x%llx rxq "
 624                            "desc:0x%llx MAC:0x%llx\n",
 625                                      adapter->buffer_list_dma,
 626                                      adapter->filter_list_dma,
 627                                      rxq_desc.desc,
 628                                      mac_address);
 629                 rc = -ENONET;
 630                 goto err_out;
 631         }
 632
 633         for (i = 0; i < IBMVETH_NUM_BUFF_POOLS; i++) {
 634                 if (!adapter->rx_buff_pool[i].active)
 635                         continue;
 636                 if (ibmveth_alloc_buffer_pool(&adapter->rx_buff_pool[i])) {
 637                         netdev_err(netdev, "unable to alloc pool\n");
 638                         adapter->rx_buff_pool[i].active = 0;
 639                         rc = -ENOMEM;
 640                         goto err_out;
 641                 }
 642         }
 643
 644         netdev_dbg(netdev, "registering irq 0x%x\n", netdev->irq);
 645         rc = request_irq(netdev->irq, ibmveth_interrupt, 0, netdev->name,
 646                          netdev);
 647         if (rc != 0) {
 648                 netdev_err(netdev, "unable to request irq 0x%x, rc %d\n",
 649                            netdev->irq, rc);
 650                 do {
 651                         lpar_rc = h_free_logical_lan(adapter->vdev->unit_address);
 652                 } while (H_IS_LONG_BUSY(lpar_rc) || (lpar_rc == H_BUSY));
 653
 654                 goto err_out;
 655         }
 656
 657         adapter->bounce_buffer =
 658             kmalloc(netdev->mtu + IBMVETH_BUFF_OH, GFP_KERNEL);
 659         if (!adapter->bounce_buffer) {
 660                 rc = -ENOMEM;
 661                 goto err_out_free_irq;
 662         }
 663         adapter->bounce_buffer_dma =
 664             dma_map_single(&adapter->vdev->dev, adapter->bounce_buffer,
 665                            netdev->mtu + IBMVETH_BUFF_OH, DMA_BIDIRECTIONAL);
 666         if (dma_mapping_error(dev, adapter->bounce_buffer_dma)) {
 667                 netdev_err(netdev, "unable to map bounce buffer\n");
 668                 rc = -ENOMEM;
 669                 goto err_out_free_irq;
 670         }
 671
 672         netdev_dbg(netdev, "initial replenish cycle\n");
 673         ibmveth_interrupt(netdev->irq, netdev);
 674
 675         netif_start_queue(netdev);
 676
 677         netdev_dbg(netdev, "open complete\n");
 678
 679         return 0;
 680
 681 err_out_free_irq:
 682         free_irq(netdev->irq, netdev);
 683 err_out:
 684         ibmveth_cleanup(adapter);
 685         napi_disable(&adapter->napi);
 686         return rc;
 687 }
 688
 689 static int ibmveth_close(struct net_device *netdev)
 690 {
 691         struct ibmveth_adapter *adapter = netdev_priv(netdev);
 692         long lpar_rc;
 693
 694         netdev_dbg(netdev, "close starting\n");
 695
 696         napi_disable(&adapter->napi);
 697
 698         if (!adapter->pool_config)
 699                 netif_stop_queue(netdev);
 700
 701         h_vio_signal(adapter->vdev->unit_address, VIO_IRQ_DISABLE);
 702
 703         do {
 704                 lpar_rc = h_free_logical_lan(adapter->vdev->unit_address);
 705         } while (H_IS_LONG_BUSY(lpar_rc) || (lpar_rc == H_BUSY));
 706
 707         if (lpar_rc != H_SUCCESS) {
 708                 netdev_err(netdev, "h_free_logical_lan failed with %lx, "
 709                            "continuing with close\n", lpar_rc);
 710         }
 711
 712         free_irq(netdev->irq, netdev);
 713
 714         ibmveth_update_rx_no_buffer(adapter);
 715
 716         ibmveth_cleanup(adapter);
 717
 718         netdev_dbg(netdev, "close complete\n");
 719
 720         return 0;
 721 }
 722
 723 static int netdev_get_settings(struct net_device *dev, struct ethtool_cmd *cmd)
 724 {
 725         cmd->supported = (SUPPORTED_1000baseT_Full | SUPPORTED_Autoneg |
 726                                 SUPPORTED_FIBRE);
 727         cmd->advertising = (ADVERTISED_1000baseT_Full | ADVERTISED_Autoneg |
 728                                 ADVERTISED_FIBRE);
 729         ethtool_cmd_speed_set(cmd, SPEED_1000);
 730         cmd->duplex = DUPLEX_FULL;
 731         cmd->port = PORT_FIBRE;
 732         cmd->phy_address = 0;
 733         cmd->transceiver = XCVR_INTERNAL;
 734         cmd->autoneg = AUTONEG_ENABLE;
 735         cmd->maxtxpkt = 0;
 736         cmd->maxrxpkt = 1;
 737         return 0;
 738 }
 739
 740 static void netdev_get_drvinfo(struct net_device *dev,
 741                                struct ethtool_drvinfo *info)
 742 {
 743         strlcpy(info->driver, ibmveth_driver_name, sizeof(info->driver));
 744         strlcpy(info->version, ibmveth_driver_version, sizeof(info->version));
 745 }
 746
 747 static netdev_features_t ibmveth_fix_features(struct net_device *dev,
 748         netdev_features_t features)
 749 {
 750         /*
 751          * Since the ibmveth firmware interface does not have the
 752          * concept of separate tx/rx checksum offload enable, if rx
 753          * checksum is disabled we also have to disable tx checksum
 754          * offload. Once we disable rx checksum offload, we are no
 755          * longer allowed to send tx buffers that are not properly
 756          * checksummed.
 757          */
 758
 759         if (!(features & NETIF_F_RXCSUM))
 760                 features &= ~NETIF_F_ALL_CSUM;
 761
 762         return features;
 763 }
 764
 765 static int ibmveth_set_csum_offload(struct net_device *dev, u32 data)
 766 {
 767         struct ibmveth_adapter *adapter = netdev_priv(dev);
 768         unsigned long set_attr, clr_attr, ret_attr;
 769         unsigned long set_attr6, clr_attr6;
 770         long ret, ret4, ret6;
 771         int rc1 = 0, rc2 = 0;
 772         int restart = 0;
 773
 774         if (netif_running(dev)) {
 775                 restart = 1;
 776                 adapter->pool_config = 1;
 777                 ibmveth_close(dev);
 778                 adapter->pool_config = 0;
 779         }
 780
 781         set_attr = 0;
 782         clr_attr = 0;
 783         set_attr6 = 0;
 784         clr_attr6 = 0;
 785
 786         if (data) {
 787                 set_attr = IBMVETH_ILLAN_IPV4_TCP_CSUM;
 788                 set_attr6 = IBMVETH_ILLAN_IPV6_TCP_CSUM;
 789         } else {
 790                 clr_attr = IBMVETH_ILLAN_IPV4_TCP_CSUM;
 791                 clr_attr6 = IBMVETH_ILLAN_IPV6_TCP_CSUM;
 792         }
 793
 794         ret = h_illan_attributes(adapter->vdev->unit_address, 0, 0, &ret_attr);
 795
 796         if (ret == H_SUCCESS && !(ret_attr & IBMVETH_ILLAN_ACTIVE_TRUNK) &&
 797             !(ret_attr & IBMVETH_ILLAN_TRUNK_PRI_MASK) &&
 798             (ret_attr & IBMVETH_ILLAN_PADDED_PKT_CSUM)) {
 799                 ret4 = h_illan_attributes(adapter->vdev->unit_address, clr_attr,
 800                                          set_attr, &ret_attr);
 801
 802                 if (ret4 != H_SUCCESS) {
 803                         netdev_err(dev, "unable to change IPv4 checksum "
 804                                         "offload settings. %d rc=%ld\n",
 805                                         data, ret4);
 806
 807                         h_illan_attributes(adapter->vdev->unit_address,
 808                                            set_attr, clr_attr, &ret_attr);
 809
 810                         if (data == 1)
 811                                 dev->features &= ~NETIF_F_IP_CSUM;
 812
 813                 } else {
 814                         adapter->fw_ipv4_csum_support = data;
 815                 }
 816
 817                 ret6 = h_illan_attributes(adapter->vdev->unit_address,
 818                                          clr_attr6, set_attr6, &ret_attr);
 819
 820                 if (ret6 != H_SUCCESS) {
 821                         netdev_err(dev, "unable to change IPv6 checksum "
 822                                         "offload settings. %d rc=%ld\n",
 823                                         data, ret6);
 824
 825                         h_illan_attributes(adapter->vdev->unit_address,
 826                                            set_attr6, clr_attr6, &ret_attr);
 827
 828                         if (data == 1)
 829                                 dev->features &= ~NETIF_F_IPV6_CSUM;
 830
 831                 } else
 832                         adapter->fw_ipv6_csum_support = data;
 833
 834                 if (ret4 == H_SUCCESS || ret6 == H_SUCCESS)
 835                         adapter->rx_csum = data;
 836                 else
 837                         rc1 = -EIO;
 838         } else {
 839                 rc1 = -EIO;
 840                 netdev_err(dev, "unable to change checksum offload settings."
 841                                      " %d rc=%ld ret_attr=%lx\n", data, ret,
 842                                      ret_attr);
 843         }
 844
 845         if (restart)
 846                 rc2 = ibmveth_open(dev);
 847
 848         return rc1 ? rc1 : rc2;
 849 }
 850
 851 static int ibmveth_set_features(struct net_device *dev,
 852         netdev_features_t features)
 853 {
 854         struct ibmveth_adapter *adapter = netdev_priv(dev);
 855         int rx_csum = !!(features & NETIF_F_RXCSUM);
 856         int rc;
 857         netdev_features_t changed = features ^ dev->features;
 858
 859         if (features & NETIF_F_TSO & changed)
 860                 netdev_info(dev, "TSO feature requires all partitions to have updated driver");
 861
 862         if (rx_csum == adapter->rx_csum)
 863                 return 0;
 864
 865         rc = ibmveth_set_csum_offload(dev, rx_csum);
 866         if (rc && !adapter->rx_csum)
 867                 dev->features = features & ~(NETIF_F_ALL_CSUM | NETIF_F_RXCSUM);
 868
 869         return rc;
 870 }
 871
 872 static void ibmveth_get_strings(struct net_device *dev, u32 stringset, u8 *data)
 873 {
 874         int i;
 875
 876         if (stringset != ETH_SS_STATS)
 877                 return;
 878
 879         for (i = 0; i < ARRAY_SIZE(ibmveth_stats); i++, data += ETH_GSTRING_LEN)
 880                 memcpy(data, ibmveth_stats[i].name, ETH_GSTRING_LEN);
 881 }
 882
 883 static int ibmveth_get_sset_count(struct net_device *dev, int sset)
 884 {
 885         switch (sset) {
 886         case ETH_SS_STATS:
 887                 return ARRAY_SIZE(ibmveth_stats);
 888         default:
 889                 return -EOPNOTSUPP;
 890         }
 891 }
 892
 893 static void ibmveth_get_ethtool_stats(struct net_device *dev,
 894                                       struct ethtool_stats *stats, u64 *data)
 895 {
 896         int i;
 897         struct ibmveth_adapter *adapter = netdev_priv(dev);
 898
 899         for (i = 0; i < ARRAY_SIZE(ibmveth_stats); i++)
 900                 data[i] = IBMVETH_GET_STAT(adapter, ibmveth_stats[i].offset);
 901 }
 902
 903 static const struct ethtool_ops netdev_ethtool_ops = {
 904         .get_drvinfo            = netdev_get_drvinfo,
 905         .get_settings           = netdev_get_settings,
 906         .get_link               = ethtool_op_get_link,
 907         .get_strings            = ibmveth_get_strings,
 908         .get_sset_count         = ibmveth_get_sset_count,
 909         .get_ethtool_stats      = ibmveth_get_ethtool_stats,
 910 };
 911
 912 static int ibmveth_ioctl(struct net_device *dev, struct ifreq *ifr, int cmd)
 913 {
 914         return -EOPNOTSUPP;
 915 }
 916
 917 #define page_offset(v) ((unsigned long)(v) & ((1 << 12) - 1))
 918
 919 static int ibmveth_send(struct ibmveth_adapter *adapter,
 920                         union ibmveth_buf_desc *descs)
 921 {
 922         unsigned long correlator;
 923         unsigned int retry_count;
 924         unsigned long ret;
 925
 926         /*
 927          * The retry count sets a maximum for the number of broadcast and
 928          * multicast destinations within the system.
 929          */
 930         retry_count = 1024;
 931         correlator = 0;
 932         do {
 933                 ret = h_send_logical_lan(adapter->vdev->unit_address,
 934                                              descs[0].desc, descs[1].desc,
 935                                              descs[2].desc, descs[3].desc,
 936                                              descs[4].desc, descs[5].desc,
 937                                              correlator, &correlator);
 938         } while ((ret == H_BUSY) && (retry_count--));
 939
 940         if (ret != H_SUCCESS && ret != H_DROPPED) {
 941                 netdev_err(adapter->netdev, "tx: h_send_logical_lan failed "
 942                            "with rc=%ld\n", ret);
 943                 return 1;
 944         }
 945
 946         return 0;
 947 }
 948
 949 static netdev_tx_t ibmveth_start_xmit(struct sk_buff *skb,
 950                                       struct net_device *netdev)
 951 {
 952         struct ibmveth_adapter *adapter = netdev_priv(netdev);
 953         unsigned int desc_flags;
 954         union ibmveth_buf_desc descs[6];
 955         int last, i;
 956         int force_bounce = 0;
 957         dma_addr_t dma_addr;
 958
 959         /*
 960          * veth handles a maximum of 6 segments including the header, so
 961          * we have to linearize the skb if there are more than this.
 962          */
 963         if (skb_shinfo(skb)->nr_frags > 5 && __skb_linearize(skb)) {
 964                 netdev->stats.tx_dropped++;
 965                 goto out;
 966         }
 967
 968         /* veth can't checksum offload UDP */
 969         if (skb->ip_summed == CHECKSUM_PARTIAL &&
 970             ((skb->protocol == htons(ETH_P_IP) &&
 971               ip_hdr(skb)->protocol != IPPROTO_TCP) ||
 972              (skb->protocol == htons(ETH_P_IPV6) &&
 973               ipv6_hdr(skb)->nexthdr != IPPROTO_TCP)) &&
 974             skb_checksum_help(skb)) {
 975
 976                 netdev_err(netdev, "tx: failed to checksum packet\n");
 977                 netdev->stats.tx_dropped++;
 978                 goto out;
 979         }
 980
 981         desc_flags = IBMVETH_BUF_VALID;
 982
 983         if (skb->ip_summed == CHECKSUM_PARTIAL) {
 984                 unsigned char *buf = skb_transport_header(skb) +
 985                                                 skb->csum_offset;
 986
 987                 desc_flags |= (IBMVETH_BUF_NO_CSUM | IBMVETH_BUF_CSUM_GOOD);
 988
 989                 /* Need to zero out the checksum */
 990                 buf[0] = 0;
 991                 buf[1] = 0;
 992         }
 993
 994 retry_bounce:
 995         memset(descs, 0, sizeof(descs));
 996
 997         /*
 998          * If a linear packet is below the rx threshold then
 999          * copy it into the static bounce buffer. This avoids the
1000          * cost of a TCE insert and remove.
1001          */
1002         if (force_bounce || (!skb_is_nonlinear(skb) &&
1003                                 (skb->len < tx_copybreak))) {
1004                 skb_copy_from_linear_data(skb, adapter->bounce_buffer,
1005                                           skb->len);
1006
1007                 descs[0].fields.flags_len = desc_flags | skb->len;
1008                 descs[0].fields.address = adapter->bounce_buffer_dma;
1009
1010                 if (ibmveth_send(adapter, descs)) {
1011                         adapter->tx_send_failed++;
1012                         netdev->stats.tx_dropped++;
1013                 } else {
1014                         netdev->stats.tx_packets++;
1015                         netdev->stats.tx_bytes += skb->len;
1016                 }
1017
1018                 goto out;
1019         }
1020
1021         /* Map the header */
1022         dma_addr = dma_map_single(&adapter->vdev->dev, skb->data,
1023                                   skb_headlen(skb), DMA_TO_DEVICE);
1024         if (dma_mapping_error(&adapter->vdev->dev, dma_addr))
1025                 goto map_failed;
1026
1027         descs[0].fields.flags_len = desc_flags | skb_headlen(skb);
1028         descs[0].fields.address = dma_addr;
1029
1030         /* Map the frags */
1031         for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
1032                 const skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
1033
1034                 dma_addr = skb_frag_dma_map(&adapter->vdev->dev, frag, 0,
1035                                             skb_frag_size(frag), DMA_TO_DEVICE);
1036
1037                 if (dma_mapping_error(&adapter->vdev->dev, dma_addr))
1038                         goto map_failed_frags;
1039
1040                 descs[i+1].fields.flags_len = desc_flags | skb_frag_size(frag);
1041                 descs[i+1].fields.address = dma_addr;
1042         }
1043
1044         if (skb_is_gso(skb) && !skb_is_gso_v6(skb)) {
1045                 /* Put -1 in the IP checksum to tell phyp it
1046                  *  is a largesend packet and put the mss in the TCP checksum.
1047                  */
1048                 ip_hdr(skb)->check = 0xffff;
1049                 tcp_hdr(skb)->check = cpu_to_be16(skb_shinfo(skb)->gso_size);
1050                 adapter->tx_large_packets++;
1051         }
1052
1053         if (ibmveth_send(adapter, descs)) {
1054                 adapter->tx_send_failed++;
1055                 netdev->stats.tx_dropped++;
1056         } else {
1057                 netdev->stats.tx_packets++;
1058                 netdev->stats.tx_bytes += skb->len;
1059         }
1060
1061         dma_unmap_single(&adapter->vdev->dev,
1062                          descs[0].fields.address,
1063                          descs[0].fields.flags_len & IBMVETH_BUF_LEN_MASK,
1064                          DMA_TO_DEVICE);
1065
1066         for (i = 1; i < skb_shinfo(skb)->nr_frags + 1; i++)
1067                 dma_unmap_page(&adapter->vdev->dev, descs[i].fields.address,
1068                                descs[i].fields.flags_len & IBMVETH_BUF_LEN_MASK,
1069                                DMA_TO_DEVICE);
1070
1071 out:
1072         dev_consume_skb_any(skb);
1073         return NETDEV_TX_OK;
1074
1075 map_failed_frags:
1076         last = i+1;
1077         for (i = 0; i < last; i++)
1078                 dma_unmap_page(&adapter->vdev->dev, descs[i].fields.address,
1079                                descs[i].fields.flags_len & IBMVETH_BUF_LEN_MASK,
1080                                DMA_TO_DEVICE);
1081
1082 map_failed:
1083         if (!firmware_has_feature(FW_FEATURE_CMO))
1084                 netdev_err(netdev, "tx: unable to map xmit buffer\n");
1085         adapter->tx_map_failed++;
1086         skb_linearize(skb);
1087         force_bounce = 1;
1088         goto retry_bounce;
1089 }
1090
1091 static int ibmveth_poll(struct napi_struct *napi, int budget)
1092 {
1093         struct ibmveth_adapter *adapter =
1094                         container_of(napi, struct ibmveth_adapter, napi);
1095         struct net_device *netdev = adapter->netdev;
1096         int frames_processed = 0;
1097         unsigned long lpar_rc;
1098         struct iphdr *iph;
1099
1100 restart_poll:
1101         while (frames_processed < budget) {
1102                 if (!ibmveth_rxq_pending_buffer(adapter))
1103                         break;
1104
1105                 smp_rmb();
1106                 if (!ibmveth_rxq_buffer_valid(adapter)) {
1107                         wmb(); /* suggested by larson1 */
1108                         adapter->rx_invalid_buffer++;
1109                         netdev_dbg(netdev, "recycling invalid buffer\n");
1110                         ibmveth_rxq_recycle_buffer(adapter);
1111                 } else {
1112                         struct sk_buff *skb, *new_skb;
1113                         int length = ibmveth_rxq_frame_length(adapter);
1114                         int offset = ibmveth_rxq_frame_offset(adapter);
1115                         int csum_good = ibmveth_rxq_csum_good(adapter);
1116
1117                         skb = ibmveth_rxq_get_buffer(adapter);
1118
1119                         new_skb = NULL;
1120                         if (length < rx_copybreak)
1121                                 new_skb = netdev_alloc_skb(netdev, length);
1122
1123                         if (new_skb) {
1124                                 skb_copy_to_linear_data(new_skb,
1125                                                         skb->data + offset,
1126                                                         length);
1127                                 if (rx_flush)
1128                                         ibmveth_flush_buffer(skb->data,
1129                                                 length + offset);
1130                                 if (!ibmveth_rxq_recycle_buffer(adapter))
1131                                         kfree_skb(skb);
1132                                 skb = new_skb;
1133                         } else {
1134                                 ibmveth_rxq_harvest_buffer(adapter);
1135                                 skb_reserve(skb, offset);
1136                         }
1137
1138                         skb_put(skb, length);
1139                         skb->protocol = eth_type_trans(skb, netdev);
1140
1141                         if (csum_good) {
1142                                 skb->ip_summed = CHECKSUM_UNNECESSARY;
1143                                 if (be16_to_cpu(skb->protocol) == ETH_P_IP) {
1144                                         iph = (struct iphdr *)skb->data;
1145
1146                                         /* If the IP checksum is not offloaded and if the packet
1147                                          *  is large send, the checksum must be rebuilt.
1148                                          */
1149                                         if (iph->check == 0xffff) {
1150                                                 iph->check = 0;
1151                                                 iph->check = ip_fast_csum((unsigned char *)iph, iph->ihl);
1152                                                 adapter->rx_large_packets++;
1153                                         }
1154                                 }
1155                         }
1156
1157                         napi_gro_receive(napi, skb);    /* send it up */
1158
1159                         netdev->stats.rx_packets++;
1160                         netdev->stats.rx_bytes += length;
1161                         frames_processed++;
1162                 }
1163         }
1164
1165         ibmveth_replenish_task(adapter);
1166
1167         if (frames_processed < budget) {
1168                 napi_complete(napi);
1169
1170                 /* We think we are done - reenable interrupts,
1171                  * then check once more to make sure we are done.
1172                  */
1173                 lpar_rc = h_vio_signal(adapter->vdev->unit_address,
1174                                        VIO_IRQ_ENABLE);
1175
1176                 BUG_ON(lpar_rc != H_SUCCESS);
1177
1178                 if (ibmveth_rxq_pending_buffer(adapter) &&
1179                     napi_reschedule(napi)) {
1180                         lpar_rc = h_vio_signal(adapter->vdev->unit_address,
1181                                                VIO_IRQ_DISABLE);
1182                         goto restart_poll;
1183                 }
1184         }
1185
1186         return frames_processed;
1187 }
1188
1189 static irqreturn_t ibmveth_interrupt(int irq, void *dev_instance)
1190 {
1191         struct net_device *netdev = dev_instance;
1192         struct ibmveth_adapter *adapter = netdev_priv(netdev);
1193         unsigned long lpar_rc;
1194
1195         if (napi_schedule_prep(&adapter->napi)) {
1196                 lpar_rc = h_vio_signal(adapter->vdev->unit_address,
1197                                        VIO_IRQ_DISABLE);
1198                 BUG_ON(lpar_rc != H_SUCCESS);
1199                 __napi_schedule(&adapter->napi);
1200         }
1201         return IRQ_HANDLED;
1202 }
1203
1204 static void ibmveth_set_multicast_list(struct net_device *netdev)
1205 {
1206         struct ibmveth_adapter *adapter = netdev_priv(netdev);
1207         unsigned long lpar_rc;
1208
1209         if ((netdev->flags & IFF_PROMISC) ||
1210             (netdev_mc_count(netdev) > adapter->mcastFilterSize)) {
1211                 lpar_rc = h_multicast_ctrl(adapter->vdev->unit_address,
1212                                            IbmVethMcastEnableRecv |
1213                                            IbmVethMcastDisableFiltering,
1214                                            0);
1215                 if (lpar_rc != H_SUCCESS) {
1216                         netdev_err(netdev, "h_multicast_ctrl rc=%ld when "
1217                                    "entering promisc mode\n", lpar_rc);
1218                 }
1219         } else {
1220                 struct netdev_hw_addr *ha;
1221                 /* clear the filter table & disable filtering */
1222                 lpar_rc = h_multicast_ctrl(adapter->vdev->unit_address,
1223                                            IbmVethMcastEnableRecv |
1224                                            IbmVethMcastDisableFiltering |
1225                                            IbmVethMcastClearFilterTable,
1226                                            0);
1227                 if (lpar_rc != H_SUCCESS) {
1228                         netdev_err(netdev, "h_multicast_ctrl rc=%ld when "
1229                                    "attempting to clear filter table\n",
1230                                    lpar_rc);
1231                 }
1232                 /* add the addresses to the filter table */
1233                 netdev_for_each_mc_addr(ha, netdev) {
1234                         /* add the multicast address to the filter table */
1235                         u64 mcast_addr;
1236                         mcast_addr = ibmveth_encode_mac_addr(ha->addr);
1237                         lpar_rc = h_multicast_ctrl(adapter->vdev->unit_address,
1238                                                    IbmVethMcastAddFilter,
1239                                                    mcast_addr);
1240                         if (lpar_rc != H_SUCCESS) {
1241                                 netdev_err(netdev, "h_multicast_ctrl rc=%ld "
1242                                            "when adding an entry to the filter "
1243                                            "table\n", lpar_rc);
1244                         }
1245                 }
1246
1247                 /* re-enable filtering */
1248                 lpar_rc = h_multicast_ctrl(adapter->vdev->unit_address,
1249                                            IbmVethMcastEnableFiltering,
1250                                            0);
1251                 if (lpar_rc != H_SUCCESS) {
1252                         netdev_err(netdev, "h_multicast_ctrl rc=%ld when "
1253                                    "enabling filtering\n", lpar_rc);
1254                 }
1255         }
1256 }
1257
1258 static int ibmveth_change_mtu(struct net_device *dev, int new_mtu)
1259 {
1260         struct ibmveth_adapter *adapter = netdev_priv(dev);
1261         struct vio_dev *viodev = adapter->vdev;
1262         int new_mtu_oh = new_mtu + IBMVETH_BUFF_OH;
1263         int i, rc;
1264         int need_restart = 0;
1265
1266         if (new_mtu < IBMVETH_MIN_MTU)
1267                 return -EINVAL;
1268
1269         for (i = 0; i < IBMVETH_NUM_BUFF_POOLS; i++)
1270                 if (new_mtu_oh <= adapter->rx_buff_pool[i].buff_size)
1271                         break;
1272
1273         if (i == IBMVETH_NUM_BUFF_POOLS)
1274                 return -EINVAL;
1275
1276         /* Deactivate all the buffer pools so that the next loop can activate
1277            only the buffer pools necessary to hold the new MTU */
1278         if (netif_running(adapter->netdev)) {
1279                 need_restart = 1;
1280                 adapter->pool_config = 1;
1281                 ibmveth_close(adapter->netdev);
1282                 adapter->pool_config = 0;
1283         }
1284
1285         /* Look for an active buffer pool that can hold the new MTU */
1286         for (i = 0; i < IBMVETH_NUM_BUFF_POOLS; i++) {
1287                 adapter->rx_buff_pool[i].active = 1;
1288
1289                 if (new_mtu_oh <= adapter->rx_buff_pool[i].buff_size) {
1290                         dev->mtu = new_mtu;
1291                         vio_cmo_set_dev_desired(viodev,
1292                                                 ibmveth_get_desired_dma
1293                                                 (viodev));
1294                         if (need_restart) {
1295                                 return ibmveth_open(adapter->netdev);
1296                         }
1297                         return 0;
1298                 }
1299         }
1300
1301         if (need_restart && (rc = ibmveth_open(adapter->netdev)))
1302                 return rc;
1303
1304         return -EINVAL;
1305 }
1306
1307 #ifdef CONFIG_NET_POLL_CONTROLLER
1308 static void ibmveth_poll_controller(struct net_device *dev)
1309 {
1310         ibmveth_replenish_task(netdev_priv(dev));
1311         ibmveth_interrupt(dev->irq, dev);
1312 }
1313 #endif
1314
1315 /**
1316  * ibmveth_get_desired_dma - Calculate IO memory desired by the driver
1317  *
1318  * @vdev: struct vio_dev for the device whose desired IO mem is to be returned
1319  *
1320  * Return value:
1321  *      Number of bytes of IO data the driver will need to perform well.
1322  */
1323 static unsigned long ibmveth_get_desired_dma(struct vio_dev *vdev)
1324 {
1325         struct net_device *netdev = dev_get_drvdata(&vdev->dev);
1326         struct ibmveth_adapter *adapter;
1327         struct iommu_table *tbl;
1328         unsigned long ret;
1329         int i;
1330         int rxqentries = 1;
1331
1332         tbl = get_iommu_table_base(&vdev->dev);
1333
1334         /* netdev inits at probe time along with the structures we need below*/
1335         if (netdev == NULL)
1336                 return IOMMU_PAGE_ALIGN(IBMVETH_IO_ENTITLEMENT_DEFAULT, tbl);
1337
1338         adapter = netdev_priv(netdev);
1339
1340         ret = IBMVETH_BUFF_LIST_SIZE + IBMVETH_FILT_LIST_SIZE;
1341         ret += IOMMU_PAGE_ALIGN(netdev->mtu, tbl);
1342
1343         for (i = 0; i < IBMVETH_NUM_BUFF_POOLS; i++) {
1344                 /* add the size of the active receive buffers */
1345                 if (adapter->rx_buff_pool[i].active)
1346                         ret +=
1347                             adapter->rx_buff_pool[i].size *
1348                             IOMMU_PAGE_ALIGN(adapter->rx_buff_pool[i].
1349                                              buff_size, tbl);
1350                 rxqentries += adapter->rx_buff_pool[i].size;
1351         }
1352         /* add the size of the receive queue entries */
1353         ret += IOMMU_PAGE_ALIGN(
1354                 rxqentries * sizeof(struct ibmveth_rx_q_entry), tbl);
1355
1356         return ret;
1357 }
1358
1359 static int ibmveth_set_mac_addr(struct net_device *dev, void *p)
1360 {
1361         struct ibmveth_adapter *adapter = netdev_priv(dev);
1362         struct sockaddr *addr = p;
1363         u64 mac_address;
1364         int rc;
1365
1366         if (!is_valid_ether_addr(addr->sa_data))
1367                 return -EADDRNOTAVAIL;
1368
1369         mac_address = ibmveth_encode_mac_addr(addr->sa_data);
1370         rc = h_change_logical_lan_mac(adapter->vdev->unit_address, mac_address);
1371         if (rc) {
1372                 netdev_err(adapter->netdev, "h_change_logical_lan_mac failed with rc=%d\n", rc);
1373                 return rc;
1374         }
1375
1376         ether_addr_copy(dev->dev_addr, addr->sa_data);
1377
1378         return 0;
1379 }
1380
1381 static const struct net_device_ops ibmveth_netdev_ops = {
1382         .ndo_open               = ibmveth_open,
1383         .ndo_stop               = ibmveth_close,
1384         .ndo_start_xmit         = ibmveth_start_xmit,
1385         .ndo_set_rx_mode        = ibmveth_set_multicast_list,
1386         .ndo_do_ioctl           = ibmveth_ioctl,
1387         .ndo_change_mtu         = ibmveth_change_mtu,
1388         .ndo_fix_features       = ibmveth_fix_features,
1389         .ndo_set_features       = ibmveth_set_features,
1390         .ndo_validate_addr      = eth_validate_addr,
1391         .ndo_set_mac_address    = ibmveth_set_mac_addr,
1392 #ifdef CONFIG_NET_POLL_CONTROLLER
1393         .ndo_poll_controller    = ibmveth_poll_controller,
1394 #endif
1395 };
1396
1397 static int ibmveth_probe(struct vio_dev *dev, const struct vio_device_id *id)
1398 {
1399         int rc, i, mac_len;
1400         struct net_device *netdev;
1401         struct ibmveth_adapter *adapter;
1402         unsigned char *mac_addr_p;
1403         unsigned int *mcastFilterSize_p;
1404
1405         dev_dbg(&dev->dev, "entering ibmveth_probe for UA 0x%x\n",
1406                 dev->unit_address);
1407
1408         mac_addr_p = (unsigned char *)vio_get_attribute(dev, VETH_MAC_ADDR,
1409                                                         &mac_len);
1410         if (!mac_addr_p) {
1411                 dev_err(&dev->dev, "Can't find VETH_MAC_ADDR attribute\n");
1412                 return -EINVAL;
1413         }
1414         /* Workaround for old/broken pHyp */
1415         if (mac_len == 8)
1416                 mac_addr_p += 2;
1417         else if (mac_len != 6) {
1418                 dev_err(&dev->dev, "VETH_MAC_ADDR attribute wrong len %d\n",
1419                         mac_len);
1420                 return -EINVAL;
1421         }
1422
1423         mcastFilterSize_p = (unsigned int *)vio_get_attribute(dev,
1424                                                 VETH_MCAST_FILTER_SIZE, NULL);
1425         if (!mcastFilterSize_p) {
1426                 dev_err(&dev->dev, "Can't find VETH_MCAST_FILTER_SIZE "
1427                         "attribute\n");
1428                 return -EINVAL;
1429         }
1430
1431         netdev = alloc_etherdev(sizeof(struct ibmveth_adapter));
1432
1433         if (!netdev)
1434                 return -ENOMEM;
1435
1436         adapter = netdev_priv(netdev);
1437         dev_set_drvdata(&dev->dev, netdev);
1438
1439         adapter->vdev = dev;
1440         adapter->netdev = netdev;
1441         adapter->mcastFilterSize = *mcastFilterSize_p;
1442         adapter->pool_config = 0;
1443
1444         netif_napi_add(netdev, &adapter->napi, ibmveth_poll, 16);
1445
1446         netdev->irq = dev->irq;
1447         netdev->netdev_ops = &ibmveth_netdev_ops;
1448         netdev->ethtool_ops = &netdev_ethtool_ops;
1449         SET_NETDEV_DEV(netdev, &dev->dev);
1450         netdev->hw_features = NETIF_F_SG | NETIF_F_RXCSUM |
1451                 NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM;
1452         netdev->features |= netdev->hw_features;
1453
1454         /* TSO is disabled by default */
1455         netdev->hw_features |= NETIF_F_TSO;
1456
1457         memcpy(netdev->dev_addr, mac_addr_p, ETH_ALEN);
1458
1459         if (firmware_has_feature(FW_FEATURE_CMO))
1460                 memcpy(pool_count, pool_count_cmo, sizeof(pool_count));
1461
1462         for (i = 0; i < IBMVETH_NUM_BUFF_POOLS; i++) {
1463                 struct kobject *kobj = &adapter->rx_buff_pool[i].kobj;
1464                 int error;
1465
1466                 ibmveth_init_buffer_pool(&adapter->rx_buff_pool[i], i,
1467                                          pool_count[i], pool_size[i],
1468                                          pool_active[i]);
1469                 error = kobject_init_and_add(kobj, &ktype_veth_pool,
1470                                              &dev->dev.kobj, "pool%d", i);
1471                 if (!error)
1472                         kobject_uevent(kobj, KOBJ_ADD);
1473         }
1474
1475         netdev_dbg(netdev, "adapter @ 0x%p\n", adapter);
1476
1477         adapter->buffer_list_dma = DMA_ERROR_CODE;
1478         adapter->filter_list_dma = DMA_ERROR_CODE;
1479         adapter->rx_queue.queue_dma = DMA_ERROR_CODE;
1480
1481         netdev_dbg(netdev, "registering netdev...\n");
1482
1483         ibmveth_set_features(netdev, netdev->features);
1484
1485         rc = register_netdev(netdev);
1486
1487         if (rc) {
1488                 netdev_dbg(netdev, "failed to register netdev rc=%d\n", rc);
1489                 free_netdev(netdev);
1490                 return rc;
1491         }
1492
1493         netdev_dbg(netdev, "registered\n");
1494
1495         return 0;
1496 }
1497
1498 static int ibmveth_remove(struct vio_dev *dev)
1499 {
1500         struct net_device *netdev = dev_get_drvdata(&dev->dev);
1501         struct ibmveth_adapter *adapter = netdev_priv(netdev);
1502         int i;
1503
1504         for (i = 0; i < IBMVETH_NUM_BUFF_POOLS; i++)
1505                 kobject_put(&adapter->rx_buff_pool[i].kobj);
1506
1507         unregister_netdev(netdev);
1508
1509         free_netdev(netdev);
1510         dev_set_drvdata(&dev->dev, NULL);
1511
1512         return 0;
1513 }
1514
1515 static struct attribute veth_active_attr;
1516 static struct attribute veth_num_attr;
1517 static struct attribute veth_size_attr;
1518
1519 static ssize_t veth_pool_show(struct kobject *kobj,
1520                               struct attribute *attr, char *buf)
1521 {
1522         struct ibmveth_buff_pool *pool = container_of(kobj,
1523                                                       struct ibmveth_buff_pool,
1524                                                       kobj);
1525
1526         if (attr == &veth_active_attr)
1527                 return sprintf(buf, "%d\n", pool->active);
1528         else if (attr == &veth_num_attr)
1529                 return sprintf(buf, "%d\n", pool->size);
1530         else if (attr == &veth_size_attr)
1531                 return sprintf(buf, "%d\n", pool->buff_size);
1532         return 0;
1533 }
1534
1535 static ssize_t veth_pool_store(struct kobject *kobj, struct attribute *attr,
1536                                const char *buf, size_t count)
1537 {
1538         struct ibmveth_buff_pool *pool = container_of(kobj,
1539                                                       struct ibmveth_buff_pool,
1540                                                       kobj);
1541         struct net_device *netdev = dev_get_drvdata(
1542             container_of(kobj->parent, struct device, kobj));
1543         struct ibmveth_adapter *adapter = netdev_priv(netdev);
1544         long value = simple_strtol(buf, NULL, 10);
1545         long rc;
1546
1547         if (attr == &veth_active_attr) {
1548                 if (value && !pool->active) {
1549                         if (netif_running(netdev)) {
1550                                 if (ibmveth_alloc_buffer_pool(pool)) {
1551                                         netdev_err(netdev,
1552                                                    "unable to alloc pool\n");
1553                                         return -ENOMEM;
1554                                 }
1555                                 pool->active = 1;
1556                                 adapter->pool_config = 1;
1557                                 ibmveth_close(netdev);
1558                                 adapter->pool_config = 0;
1559                                 if ((rc = ibmveth_open(netdev)))
1560                                         return rc;
1561                         } else {
1562                                 pool->active = 1;
1563                         }
1564                 } else if (!value && pool->active) {
1565                         int mtu = netdev->mtu + IBMVETH_BUFF_OH;
1566                         int i;
1567                         /* Make sure there is a buffer pool with buffers that
1568                            can hold a packet of the size of the MTU */
1569                         for (i = 0; i < IBMVETH_NUM_BUFF_POOLS; i++) {
1570                                 if (pool == &adapter->rx_buff_pool[i])
1571                                         continue;
1572                                 if (!adapter->rx_buff_pool[i].active)
1573                                         continue;
1574                                 if (mtu <= adapter->rx_buff_pool[i].buff_size)
1575                                         break;
1576                         }
1577
1578                         if (i == IBMVETH_NUM_BUFF_POOLS) {
1579                                 netdev_err(netdev, "no active pool >= MTU\n");
1580                                 return -EPERM;
1581                         }
1582
1583                         if (netif_running(netdev)) {
1584                                 adapter->pool_config = 1;
1585                                 ibmveth_close(netdev);
1586                                 pool->active = 0;
1587                                 adapter->pool_config = 0;
1588                                 if ((rc = ibmveth_open(netdev)))
1589                                         return rc;
1590                         }
1591                         pool->active = 0;
1592                 }
1593         } else if (attr == &veth_num_attr) {
1594                 if (value <= 0 || value > IBMVETH_MAX_POOL_COUNT) {
1595                         return -EINVAL;
1596                 } else {
1597                         if (netif_running(netdev)) {
1598                                 adapter->pool_config = 1;
1599                                 ibmveth_close(netdev);
1600                                 adapter->pool_config = 0;
1601                                 pool->size = value;
1602                                 if ((rc = ibmveth_open(netdev)))
1603                                         return rc;
1604                         } else {
1605                                 pool->size = value;
1606                         }
1607                 }
1608         } else if (attr == &veth_size_attr) {
1609                 if (value <= IBMVETH_BUFF_OH || value > IBMVETH_MAX_BUF_SIZE) {
1610                         return -EINVAL;
1611                 } else {
1612                         if (netif_running(netdev)) {
1613                                 adapter->pool_config = 1;
1614                                 ibmveth_close(netdev);
1615                                 adapter->pool_config = 0;
1616                                 pool->buff_size = value;
1617                                 if ((rc = ibmveth_open(netdev)))
1618                                         return rc;
1619                         } else {
1620                                 pool->buff_size = value;
1621                         }
1622                 }
1623         }
1624
1625         /* kick the interrupt handler to allocate/deallocate pools */
1626         ibmveth_interrupt(netdev->irq, netdev);
1627         return count;
1628 }
1629
1630
1631 #define ATTR(_name, _mode)                              \
1632         struct attribute veth_##_name##_attr = {        \
1633         .name = __stringify(_name), .mode = _mode,      \
1634         };
1635
1636 static ATTR(active, 0644);
1637 static ATTR(num, 0644);
1638 static ATTR(size, 0644);
1639
1640 static struct attribute *veth_pool_attrs[] = {
1641         &veth_active_attr,
1642         &veth_num_attr,
1643         &veth_size_attr,
1644         NULL,
1645 };
1646
1647 static const struct sysfs_ops veth_pool_ops = {
1648         .show   = veth_pool_show,
1649         .store  = veth_pool_store,
1650 };
1651
1652 static struct kobj_type ktype_veth_pool = {
1653         .release        = NULL,
1654         .sysfs_ops      = &veth_pool_ops,
1655         .default_attrs  = veth_pool_attrs,
1656 };
1657
1658 static int ibmveth_resume(struct device *dev)
1659 {
1660         struct net_device *netdev = dev_get_drvdata(dev);
1661         ibmveth_interrupt(netdev->irq, netdev);
1662         return 0;
1663 }
1664
1665 static struct vio_device_id ibmveth_device_table[] = {
1666         { "network", "IBM,l-lan"},
1667         { "", "" }
1668 };
1669 MODULE_DEVICE_TABLE(vio, ibmveth_device_table);
1670
1671 static struct dev_pm_ops ibmveth_pm_ops = {
1672         .resume = ibmveth_resume
1673 };
1674
1675 static struct vio_driver ibmveth_driver = {
1676         .id_table       = ibmveth_device_table,
1677         .probe          = ibmveth_probe,
1678         .remove         = ibmveth_remove,
1679         .get_desired_dma = ibmveth_get_desired_dma,
1680         .name           = ibmveth_driver_name,
1681         .pm             = &ibmveth_pm_ops,
1682 };
1683
1684 static int __init ibmveth_module_init(void)
1685 {
1686         printk(KERN_DEBUG "%s: %s %s\n", ibmveth_driver_name,
1687                ibmveth_driver_string, ibmveth_driver_version);
1688
1689         return vio_register_driver(&ibmveth_driver);
1690 }
1691
1692 static void __exit ibmveth_module_exit(void)
1693 {
1694         vio_unregister_driver(&ibmveth_driver);
1695 }
1696
1697 module_init(ibmveth_module_init);
1698 module_exit(ibmveth_module_exit);