drivers/net/ethernet/ibm/ibmveth.c

   1 // SPDX-License-Identifier: GPL-2.0-or-later
   2 /*
   3  * IBM Power Virtual Ethernet Device Driver
   4  *
   5  * Copyright (C) IBM Corporation, 2003, 2010
   6  *
   7  * Authors: Dave Larson <[email protected]>
   8  *          Santiago Leon <[email protected]>
   9  *          Brian King <[email protected]>
  10  *          Robert Jennings <[email protected]>
  11  *          Anton Blanchard <[email protected]>
  12  */
  13
  14 #include <linux/module.h>
  15 #include <linux/types.h>
  16 #include <linux/errno.h>
  17 #include <linux/dma-mapping.h>
  18 #include <linux/kernel.h>
  19 #include <linux/netdevice.h>
  20 #include <linux/etherdevice.h>
  21 #include <linux/skbuff.h>
  22 #include <linux/init.h>
  23 #include <linux/interrupt.h>
  24 #include <linux/mm.h>
  25 #include <linux/pm.h>
  26 #include <linux/ethtool.h>
  27 #include <linux/in.h>
  28 #include <linux/ip.h>
  29 #include <linux/ipv6.h>
  30 #include <linux/slab.h>
  31 #include <asm/hvcall.h>
  32 #include <linux/atomic.h>
  33 #include <asm/vio.h>
  34 #include <asm/iommu.h>
  35 #include <asm/firmware.h>
  36 #include <net/tcp.h>
  37 #include <net/ip6_checksum.h>
  38
  39 #include "ibmveth.h"
  40
  41 static irqreturn_t ibmveth_interrupt(int irq, void *dev_instance);
  42 static void ibmveth_rxq_harvest_buffer(struct ibmveth_adapter *adapter);
  43 static unsigned long ibmveth_get_desired_dma(struct vio_dev *vdev);
  44
  45 static struct kobj_type ktype_veth_pool;
  46
  47
  48 static const char ibmveth_driver_name[] = "ibmveth";
  49 static const char ibmveth_driver_string[] = "IBM Power Virtual Ethernet Driver";
  50 #define ibmveth_driver_version "1.06"
  51
  52 MODULE_AUTHOR("Santiago Leon <[email protected]>");
  53 MODULE_DESCRIPTION("IBM Power Virtual Ethernet Driver");
  54 MODULE_LICENSE("GPL");
  55 MODULE_VERSION(ibmveth_driver_version);
  56
  57 static unsigned int tx_copybreak __read_mostly = 128;
  58 module_param(tx_copybreak, uint, 0644);
  59 MODULE_PARM_DESC(tx_copybreak,
  60         "Maximum size of packet that is copied to a new buffer on transmit");
  61
  62 static unsigned int rx_copybreak __read_mostly = 128;
  63 module_param(rx_copybreak, uint, 0644);
  64 MODULE_PARM_DESC(rx_copybreak,
  65         "Maximum size of packet that is copied to a new buffer on receive");
  66
  67 static unsigned int rx_flush __read_mostly = 0;
  68 module_param(rx_flush, uint, 0644);
  69 MODULE_PARM_DESC(rx_flush, "Flush receive buffers before use");
  70
  71 static bool old_large_send __read_mostly;
  72 module_param(old_large_send, bool, 0444);
  73 MODULE_PARM_DESC(old_large_send,
  74         "Use old large send method on firmware that supports the new method");
  75
  76 struct ibmveth_stat {
  77         char name[ETH_GSTRING_LEN];
  78         int offset;
  79 };
  80
  81 #define IBMVETH_STAT_OFF(stat) offsetof(struct ibmveth_adapter, stat)
  82 #define IBMVETH_GET_STAT(a, off) *((u64 *)(((unsigned long)(a)) + off))
  83
  84 static struct ibmveth_stat ibmveth_stats[] = {
  85         { "replenish_task_cycles", IBMVETH_STAT_OFF(replenish_task_cycles) },
  86         { "replenish_no_mem", IBMVETH_STAT_OFF(replenish_no_mem) },
  87         { "replenish_add_buff_failure",
  88                         IBMVETH_STAT_OFF(replenish_add_buff_failure) },
  89         { "replenish_add_buff_success",
  90                         IBMVETH_STAT_OFF(replenish_add_buff_success) },
  91         { "rx_invalid_buffer", IBMVETH_STAT_OFF(rx_invalid_buffer) },
  92         { "rx_no_buffer", IBMVETH_STAT_OFF(rx_no_buffer) },
  93         { "tx_map_failed", IBMVETH_STAT_OFF(tx_map_failed) },
  94         { "tx_send_failed", IBMVETH_STAT_OFF(tx_send_failed) },
  95         { "fw_enabled_ipv4_csum", IBMVETH_STAT_OFF(fw_ipv4_csum_support) },
  96         { "fw_enabled_ipv6_csum", IBMVETH_STAT_OFF(fw_ipv6_csum_support) },
  97         { "tx_large_packets", IBMVETH_STAT_OFF(tx_large_packets) },
  98         { "rx_large_packets", IBMVETH_STAT_OFF(rx_large_packets) },
  99         { "fw_enabled_large_send", IBMVETH_STAT_OFF(fw_large_send_support) }
 100 };
 101
 102 /* simple methods of getting data from the current rxq entry */
 103 static inline u32 ibmveth_rxq_flags(struct ibmveth_adapter *adapter)
 104 {
 105         return be32_to_cpu(adapter->rx_queue.queue_addr[adapter->rx_queue.index].flags_off);
 106 }
 107
 108 static inline int ibmveth_rxq_toggle(struct ibmveth_adapter *adapter)
 109 {
 110         return (ibmveth_rxq_flags(adapter) & IBMVETH_RXQ_TOGGLE) >>
 111                         IBMVETH_RXQ_TOGGLE_SHIFT;
 112 }
 113
 114 static inline int ibmveth_rxq_pending_buffer(struct ibmveth_adapter *adapter)
 115 {
 116         return ibmveth_rxq_toggle(adapter) == adapter->rx_queue.toggle;
 117 }
 118
 119 static inline int ibmveth_rxq_buffer_valid(struct ibmveth_adapter *adapter)
 120 {
 121         return ibmveth_rxq_flags(adapter) & IBMVETH_RXQ_VALID;
 122 }
 123
 124 static inline int ibmveth_rxq_frame_offset(struct ibmveth_adapter *adapter)
 125 {
 126         return ibmveth_rxq_flags(adapter) & IBMVETH_RXQ_OFF_MASK;
 127 }
 128
 129 static inline int ibmveth_rxq_large_packet(struct ibmveth_adapter *adapter)
 130 {
 131         return ibmveth_rxq_flags(adapter) & IBMVETH_RXQ_LRG_PKT;
 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_array(pool->size, sizeof(u16), GFP_KERNEL);
 162
 163         if (!pool->free_map)
 164                 return -1;
 165
 166         pool->dma_addr = kcalloc(pool->size, sizeof(dma_addr_t), 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         for (i = 0; i < pool->size; ++i)
 185                 pool->free_map[i] = i;
 186
 187         atomic_set(&pool->available, 0);
 188         pool->producer_index = 0;
 189         pool->consumer_index = 0;
 190
 191         return 0;
 192 }
 193
 194 static inline void ibmveth_flush_buffer(void *addr, unsigned long length)
 195 {
 196         unsigned long offset;
 197
 198         for (offset = 0; offset < length; offset += SMP_CACHE_BYTES)
 199                 asm("dcbfl %0,%1" :: "b" (addr), "r" (offset));
 200 }
 201
 202 /* replenish the buffers for a pool.  note that we don't need to
 203  * skb_reserve these since they are used for incoming...
 204  */
 205 static void ibmveth_replenish_buffer_pool(struct ibmveth_adapter *adapter,
 206                                           struct ibmveth_buff_pool *pool)
 207 {
 208         u32 i;
 209         u32 count = pool->size - atomic_read(&pool->available);
 210         u32 buffers_added = 0;
 211         struct sk_buff *skb;
 212         unsigned int free_index, index;
 213         u64 correlator;
 214         unsigned long lpar_rc;
 215         dma_addr_t dma_addr;
 216
 217         mb();
 218
 219         for (i = 0; i < count; ++i) {
 220                 union ibmveth_buf_desc desc;
 221
 222                 skb = netdev_alloc_skb(adapter->netdev, pool->buff_size);
 223
 224                 if (!skb) {
 225                         netdev_dbg(adapter->netdev,
 226                                    "replenish: unable to allocate skb\n");
 227                         adapter->replenish_no_mem++;
 228                         break;
 229                 }
 230
 231                 free_index = pool->consumer_index;
 232                 pool->consumer_index++;
 233                 if (pool->consumer_index >= pool->size)
 234                         pool->consumer_index = 0;
 235                 index = pool->free_map[free_index];
 236
 237                 BUG_ON(index == IBM_VETH_INVALID_MAP);
 238                 BUG_ON(pool->skbuff[index] != NULL);
 239
 240                 dma_addr = dma_map_single(&adapter->vdev->dev, skb->data,
 241                                 pool->buff_size, DMA_FROM_DEVICE);
 242
 243                 if (dma_mapping_error(&adapter->vdev->dev, dma_addr))
 244                         goto failure;
 245
 246                 pool->free_map[free_index] = IBM_VETH_INVALID_MAP;
 247                 pool->dma_addr[index] = dma_addr;
 248                 pool->skbuff[index] = skb;
 249
 250                 correlator = ((u64)pool->index << 32) | index;
 251                 *(u64 *)skb->data = correlator;
 252
 253                 desc.fields.flags_len = IBMVETH_BUF_VALID | pool->buff_size;
 254                 desc.fields.address = dma_addr;
 255
 256                 if (rx_flush) {
 257                         unsigned int len = min(pool->buff_size,
 258                                                 adapter->netdev->mtu +
 259                                                 IBMVETH_BUFF_OH);
 260                         ibmveth_flush_buffer(skb->data, len);
 261                 }
 262                 lpar_rc = h_add_logical_lan_buffer(adapter->vdev->unit_address,
 263                                                    desc.desc);
 264
 265                 if (lpar_rc != H_SUCCESS) {
 266                         goto failure;
 267                 } else {
 268                         buffers_added++;
 269                         adapter->replenish_add_buff_success++;
 270                 }
 271         }
 272
 273         mb();
 274         atomic_add(buffers_added, &(pool->available));
 275         return;
 276
 277 failure:
 278         pool->free_map[free_index] = index;
 279         pool->skbuff[index] = NULL;
 280         if (pool->consumer_index == 0)
 281                 pool->consumer_index = pool->size - 1;
 282         else
 283                 pool->consumer_index--;
 284         if (!dma_mapping_error(&adapter->vdev->dev, dma_addr))
 285                 dma_unmap_single(&adapter->vdev->dev,
 286                                  pool->dma_addr[index], pool->buff_size,
 287                                  DMA_FROM_DEVICE);
 288         dev_kfree_skb_any(skb);
 289         adapter->replenish_add_buff_failure++;
 290
 291         mb();
 292         atomic_add(buffers_added, &(pool->available));
 293 }
 294
 295 /*
 296  * The final 8 bytes of the buffer list is a counter of frames dropped
 297  * because there was not a buffer in the buffer list capable of holding
 298  * the frame.
 299  */
 300 static void ibmveth_update_rx_no_buffer(struct ibmveth_adapter *adapter)
 301 {
 302         __be64 *p = adapter->buffer_list_addr + 4096 - 8;
 303
 304         adapter->rx_no_buffer = be64_to_cpup(p);
 305 }
 306
 307 /* replenish routine */
 308 static void ibmveth_replenish_task(struct ibmveth_adapter *adapter)
 309 {
 310         int i;
 311
 312         adapter->replenish_task_cycles++;
 313
 314         for (i = (IBMVETH_NUM_BUFF_POOLS - 1); i >= 0; i--) {
 315                 struct ibmveth_buff_pool *pool = &adapter->rx_buff_pool[i];
 316
 317                 if (pool->active &&
 318                     (atomic_read(&pool->available) < pool->threshold))
 319                         ibmveth_replenish_buffer_pool(adapter, pool);
 320         }
 321
 322         ibmveth_update_rx_no_buffer(adapter);
 323 }
 324
 325 /* empty and free ana buffer pool - also used to do cleanup in error paths */
 326 static void ibmveth_free_buffer_pool(struct ibmveth_adapter *adapter,
 327                                      struct ibmveth_buff_pool *pool)
 328 {
 329         int i;
 330
 331         kfree(pool->free_map);
 332         pool->free_map = NULL;
 333
 334         if (pool->skbuff && pool->dma_addr) {
 335                 for (i = 0; i < pool->size; ++i) {
 336                         struct sk_buff *skb = pool->skbuff[i];
 337                         if (skb) {
 338                                 dma_unmap_single(&adapter->vdev->dev,
 339                                                  pool->dma_addr[i],
 340                                                  pool->buff_size,
 341                                                  DMA_FROM_DEVICE);
 342                                 dev_kfree_skb_any(skb);
 343                                 pool->skbuff[i] = NULL;
 344                         }
 345                 }
 346         }
 347
 348         if (pool->dma_addr) {
 349                 kfree(pool->dma_addr);
 350                 pool->dma_addr = NULL;
 351         }
 352
 353         if (pool->skbuff) {
 354                 kfree(pool->skbuff);
 355                 pool->skbuff = NULL;
 356         }
 357 }
 358
 359 /* remove a buffer from a pool */
 360 static void ibmveth_remove_buffer_from_pool(struct ibmveth_adapter *adapter,
 361                                             u64 correlator)
 362 {
 363         unsigned int pool  = correlator >> 32;
 364         unsigned int index = correlator & 0xffffffffUL;
 365         unsigned int free_index;
 366         struct sk_buff *skb;
 367
 368         BUG_ON(pool >= IBMVETH_NUM_BUFF_POOLS);
 369         BUG_ON(index >= adapter->rx_buff_pool[pool].size);
 370
 371         skb = adapter->rx_buff_pool[pool].skbuff[index];
 372
 373         BUG_ON(skb == NULL);
 374
 375         adapter->rx_buff_pool[pool].skbuff[index] = NULL;
 376
 377         dma_unmap_single(&adapter->vdev->dev,
 378                          adapter->rx_buff_pool[pool].dma_addr[index],
 379                          adapter->rx_buff_pool[pool].buff_size,
 380                          DMA_FROM_DEVICE);
 381
 382         free_index = adapter->rx_buff_pool[pool].producer_index;
 383         adapter->rx_buff_pool[pool].producer_index++;
 384         if (adapter->rx_buff_pool[pool].producer_index >=
 385             adapter->rx_buff_pool[pool].size)
 386                 adapter->rx_buff_pool[pool].producer_index = 0;
 387         adapter->rx_buff_pool[pool].free_map[free_index] = index;
 388
 389         mb();
 390
 391         atomic_dec(&(adapter->rx_buff_pool[pool].available));
 392 }
 393
 394 /* get the current buffer on the rx queue */
 395 static inline struct sk_buff *ibmveth_rxq_get_buffer(struct ibmveth_adapter *adapter)
 396 {
 397         u64 correlator = adapter->rx_queue.queue_addr[adapter->rx_queue.index].correlator;
 398         unsigned int pool = correlator >> 32;
 399         unsigned int index = correlator & 0xffffffffUL;
 400
 401         BUG_ON(pool >= IBMVETH_NUM_BUFF_POOLS);
 402         BUG_ON(index >= adapter->rx_buff_pool[pool].size);
 403
 404         return adapter->rx_buff_pool[pool].skbuff[index];
 405 }
 406
 407 /* recycle the current buffer on the rx queue */
 408 static int ibmveth_rxq_recycle_buffer(struct ibmveth_adapter *adapter)
 409 {
 410         u32 q_index = adapter->rx_queue.index;
 411         u64 correlator = adapter->rx_queue.queue_addr[q_index].correlator;
 412         unsigned int pool = correlator >> 32;
 413         unsigned int index = correlator & 0xffffffffUL;
 414         union ibmveth_buf_desc desc;
 415         unsigned long lpar_rc;
 416         int ret = 1;
 417
 418         BUG_ON(pool >= IBMVETH_NUM_BUFF_POOLS);
 419         BUG_ON(index >= adapter->rx_buff_pool[pool].size);
 420
 421         if (!adapter->rx_buff_pool[pool].active) {
 422                 ibmveth_rxq_harvest_buffer(adapter);
 423                 ibmveth_free_buffer_pool(adapter, &adapter->rx_buff_pool[pool]);
 424                 goto out;
 425         }
 426
 427         desc.fields.flags_len = IBMVETH_BUF_VALID |
 428                 adapter->rx_buff_pool[pool].buff_size;
 429         desc.fields.address = adapter->rx_buff_pool[pool].dma_addr[index];
 430
 431         lpar_rc = h_add_logical_lan_buffer(adapter->vdev->unit_address, desc.desc);
 432
 433         if (lpar_rc != H_SUCCESS) {
 434                 netdev_dbg(adapter->netdev, "h_add_logical_lan_buffer failed "
 435                            "during recycle rc=%ld", lpar_rc);
 436                 ibmveth_remove_buffer_from_pool(adapter, adapter->rx_queue.queue_addr[adapter->rx_queue.index].correlator);
 437                 ret = 0;
 438         }
 439
 440         if (++adapter->rx_queue.index == adapter->rx_queue.num_slots) {
 441                 adapter->rx_queue.index = 0;
 442                 adapter->rx_queue.toggle = !adapter->rx_queue.toggle;
 443         }
 444
 445 out:
 446         return ret;
 447 }
 448
 449 static void ibmveth_rxq_harvest_buffer(struct ibmveth_adapter *adapter)
 450 {
 451         ibmveth_remove_buffer_from_pool(adapter, adapter->rx_queue.queue_addr[adapter->rx_queue.index].correlator);
 452
 453         if (++adapter->rx_queue.index == adapter->rx_queue.num_slots) {
 454                 adapter->rx_queue.index = 0;
 455                 adapter->rx_queue.toggle = !adapter->rx_queue.toggle;
 456         }
 457 }
 458
 459 static int ibmveth_register_logical_lan(struct ibmveth_adapter *adapter,
 460         union ibmveth_buf_desc rxq_desc, u64 mac_address)
 461 {
 462         int rc, try_again = 1;
 463
 464         /*
 465          * After a kexec the adapter will still be open, so our attempt to
 466          * open it will fail. So if we get a failure we free the adapter and
 467          * try again, but only once.
 468          */
 469 retry:
 470         rc = h_register_logical_lan(adapter->vdev->unit_address,
 471                                     adapter->buffer_list_dma, rxq_desc.desc,
 472                                     adapter->filter_list_dma, mac_address);
 473
 474         if (rc != H_SUCCESS && try_again) {
 475                 do {
 476                         rc = h_free_logical_lan(adapter->vdev->unit_address);
 477                 } while (H_IS_LONG_BUSY(rc) || (rc == H_BUSY));
 478
 479                 try_again = 0;
 480                 goto retry;
 481         }
 482
 483         return rc;
 484 }
 485
 486 static int ibmveth_open(struct net_device *netdev)
 487 {
 488         struct ibmveth_adapter *adapter = netdev_priv(netdev);
 489         u64 mac_address;
 490         int rxq_entries = 1;
 491         unsigned long lpar_rc;
 492         int rc;
 493         union ibmveth_buf_desc rxq_desc;
 494         int i;
 495         struct device *dev;
 496
 497         netdev_dbg(netdev, "open starting\n");
 498
 499         napi_enable(&adapter->napi);
 500
 501         for(i = 0; i < IBMVETH_NUM_BUFF_POOLS; i++)
 502                 rxq_entries += adapter->rx_buff_pool[i].size;
 503
 504         rc = -ENOMEM;
 505         adapter->buffer_list_addr = (void*) get_zeroed_page(GFP_KERNEL);
 506         if (!adapter->buffer_list_addr) {
 507                 netdev_err(netdev, "unable to allocate list pages\n");
 508                 goto out;
 509         }
 510
 511         adapter->filter_list_addr = (void*) get_zeroed_page(GFP_KERNEL);
 512         if (!adapter->filter_list_addr) {
 513                 netdev_err(netdev, "unable to allocate filter pages\n");
 514                 goto out_free_buffer_list;
 515         }
 516
 517         dev = &adapter->vdev->dev;
 518
 519         adapter->rx_queue.queue_len = sizeof(struct ibmveth_rx_q_entry) *
 520                                                 rxq_entries;
 521         adapter->rx_queue.queue_addr =
 522                 dma_alloc_coherent(dev, adapter->rx_queue.queue_len,
 523                                    &adapter->rx_queue.queue_dma, GFP_KERNEL);
 524         if (!adapter->rx_queue.queue_addr)
 525                 goto out_free_filter_list;
 526
 527         adapter->buffer_list_dma = dma_map_single(dev,
 528                         adapter->buffer_list_addr, 4096, DMA_BIDIRECTIONAL);
 529         if (dma_mapping_error(dev, adapter->buffer_list_dma)) {
 530                 netdev_err(netdev, "unable to map buffer list pages\n");
 531                 goto out_free_queue_mem;
 532         }
 533
 534         adapter->filter_list_dma = dma_map_single(dev,
 535                         adapter->filter_list_addr, 4096, DMA_BIDIRECTIONAL);
 536         if (dma_mapping_error(dev, adapter->filter_list_dma)) {
 537                 netdev_err(netdev, "unable to map filter list pages\n");
 538                 goto out_unmap_buffer_list;
 539         }
 540
 541         adapter->rx_queue.index = 0;
 542         adapter->rx_queue.num_slots = rxq_entries;
 543         adapter->rx_queue.toggle = 1;
 544
 545         mac_address = ether_addr_to_u64(netdev->dev_addr);
 546
 547         rxq_desc.fields.flags_len = IBMVETH_BUF_VALID |
 548                                         adapter->rx_queue.queue_len;
 549         rxq_desc.fields.address = adapter->rx_queue.queue_dma;
 550
 551         netdev_dbg(netdev, "buffer list @ 0x%p\n", adapter->buffer_list_addr);
 552         netdev_dbg(netdev, "filter list @ 0x%p\n", adapter->filter_list_addr);
 553         netdev_dbg(netdev, "receive q   @ 0x%p\n", adapter->rx_queue.queue_addr);
 554
 555         h_vio_signal(adapter->vdev->unit_address, VIO_IRQ_DISABLE);
 556
 557         lpar_rc = ibmveth_register_logical_lan(adapter, rxq_desc, mac_address);
 558
 559         if (lpar_rc != H_SUCCESS) {
 560                 netdev_err(netdev, "h_register_logical_lan failed with %ld\n",
 561                            lpar_rc);
 562                 netdev_err(netdev, "buffer TCE:0x%llx filter TCE:0x%llx rxq "
 563                            "desc:0x%llx MAC:0x%llx\n",
 564                                      adapter->buffer_list_dma,
 565                                      adapter->filter_list_dma,
 566                                      rxq_desc.desc,
 567                                      mac_address);
 568                 rc = -ENONET;
 569                 goto out_unmap_filter_list;
 570         }
 571
 572         for (i = 0; i < IBMVETH_NUM_BUFF_POOLS; i++) {
 573                 if (!adapter->rx_buff_pool[i].active)
 574                         continue;
 575                 if (ibmveth_alloc_buffer_pool(&adapter->rx_buff_pool[i])) {
 576                         netdev_err(netdev, "unable to alloc pool\n");
 577                         adapter->rx_buff_pool[i].active = 0;
 578                         rc = -ENOMEM;
 579                         goto out_free_buffer_pools;
 580                 }
 581         }
 582
 583         netdev_dbg(netdev, "registering irq 0x%x\n", netdev->irq);
 584         rc = request_irq(netdev->irq, ibmveth_interrupt, 0, netdev->name,
 585                          netdev);
 586         if (rc != 0) {
 587                 netdev_err(netdev, "unable to request irq 0x%x, rc %d\n",
 588                            netdev->irq, rc);
 589                 do {
 590                         lpar_rc = h_free_logical_lan(adapter->vdev->unit_address);
 591                 } while (H_IS_LONG_BUSY(lpar_rc) || (lpar_rc == H_BUSY));
 592
 593                 goto out_free_buffer_pools;
 594         }
 595
 596         rc = -ENOMEM;
 597
 598         adapter->bounce_buffer = dma_alloc_coherent(&adapter->vdev->dev,
 599                                                     netdev->mtu + IBMVETH_BUFF_OH,
 600                                                     &adapter->bounce_buffer_dma, GFP_KERNEL);
 601         if (!adapter->bounce_buffer) {
 602                 netdev_err(netdev, "unable to alloc bounce buffer\n");
 603                 goto out_free_irq;
 604         }
 605
 606         netdev_dbg(netdev, "initial replenish cycle\n");
 607         ibmveth_interrupt(netdev->irq, netdev);
 608
 609         netif_start_queue(netdev);
 610
 611         netdev_dbg(netdev, "open complete\n");
 612
 613         return 0;
 614
 615 out_free_irq:
 616         free_irq(netdev->irq, netdev);
 617 out_free_buffer_pools:
 618         while (--i >= 0) {
 619                 if (adapter->rx_buff_pool[i].active)
 620                         ibmveth_free_buffer_pool(adapter,
 621                                                  &adapter->rx_buff_pool[i]);
 622         }
 623 out_unmap_filter_list:
 624         dma_unmap_single(dev, adapter->filter_list_dma, 4096,
 625                          DMA_BIDIRECTIONAL);
 626 out_unmap_buffer_list:
 627         dma_unmap_single(dev, adapter->buffer_list_dma, 4096,
 628                          DMA_BIDIRECTIONAL);
 629 out_free_queue_mem:
 630         dma_free_coherent(dev, adapter->rx_queue.queue_len,
 631                           adapter->rx_queue.queue_addr,
 632                           adapter->rx_queue.queue_dma);
 633 out_free_filter_list:
 634         free_page((unsigned long)adapter->filter_list_addr);
 635 out_free_buffer_list:
 636         free_page((unsigned long)adapter->buffer_list_addr);
 637 out:
 638         napi_disable(&adapter->napi);
 639         return rc;
 640 }
 641
 642 static int ibmveth_close(struct net_device *netdev)
 643 {
 644         struct ibmveth_adapter *adapter = netdev_priv(netdev);
 645         struct device *dev = &adapter->vdev->dev;
 646         long lpar_rc;
 647         int i;
 648
 649         netdev_dbg(netdev, "close starting\n");
 650
 651         napi_disable(&adapter->napi);
 652
 653         if (!adapter->pool_config)
 654                 netif_stop_queue(netdev);
 655
 656         h_vio_signal(adapter->vdev->unit_address, VIO_IRQ_DISABLE);
 657
 658         do {
 659                 lpar_rc = h_free_logical_lan(adapter->vdev->unit_address);
 660         } while (H_IS_LONG_BUSY(lpar_rc) || (lpar_rc == H_BUSY));
 661
 662         if (lpar_rc != H_SUCCESS) {
 663                 netdev_err(netdev, "h_free_logical_lan failed with %lx, "
 664                            "continuing with close\n", lpar_rc);
 665         }
 666
 667         free_irq(netdev->irq, netdev);
 668
 669         ibmveth_update_rx_no_buffer(adapter);
 670
 671         dma_unmap_single(dev, adapter->buffer_list_dma, 4096,
 672                          DMA_BIDIRECTIONAL);
 673         free_page((unsigned long)adapter->buffer_list_addr);
 674
 675         dma_unmap_single(dev, adapter->filter_list_dma, 4096,
 676                          DMA_BIDIRECTIONAL);
 677         free_page((unsigned long)adapter->filter_list_addr);
 678
 679         dma_free_coherent(dev, adapter->rx_queue.queue_len,
 680                           adapter->rx_queue.queue_addr,
 681                           adapter->rx_queue.queue_dma);
 682
 683         for (i = 0; i < IBMVETH_NUM_BUFF_POOLS; i++)
 684                 if (adapter->rx_buff_pool[i].active)
 685                         ibmveth_free_buffer_pool(adapter,
 686                                                  &adapter->rx_buff_pool[i]);
 687
 688         dma_free_coherent(&adapter->vdev->dev,
 689                           adapter->netdev->mtu + IBMVETH_BUFF_OH,
 690                           adapter->bounce_buffer, adapter->bounce_buffer_dma);
 691
 692         netdev_dbg(netdev, "close complete\n");
 693
 694         return 0;
 695 }
 696
 697 static int ibmveth_set_link_ksettings(struct net_device *dev,
 698                                       const struct ethtool_link_ksettings *cmd)
 699 {
 700         struct ibmveth_adapter *adapter = netdev_priv(dev);
 701
 702         return ethtool_virtdev_set_link_ksettings(dev, cmd,
 703                                                   &adapter->speed,
 704                                                   &adapter->duplex);
 705 }
 706
 707 static int ibmveth_get_link_ksettings(struct net_device *dev,
 708                                       struct ethtool_link_ksettings *cmd)
 709 {
 710         struct ibmveth_adapter *adapter = netdev_priv(dev);
 711
 712         cmd->base.speed = adapter->speed;
 713         cmd->base.duplex = adapter->duplex;
 714         cmd->base.port = PORT_OTHER;
 715
 716         return 0;
 717 }
 718
 719 static void ibmveth_init_link_settings(struct net_device *dev)
 720 {
 721         struct ibmveth_adapter *adapter = netdev_priv(dev);
 722
 723         adapter->speed = SPEED_1000;
 724         adapter->duplex = DUPLEX_FULL;
 725 }
 726
 727 static void netdev_get_drvinfo(struct net_device *dev,
 728                                struct ethtool_drvinfo *info)
 729 {
 730         strlcpy(info->driver, ibmveth_driver_name, sizeof(info->driver));
 731         strlcpy(info->version, ibmveth_driver_version, sizeof(info->version));
 732 }
 733
 734 static netdev_features_t ibmveth_fix_features(struct net_device *dev,
 735         netdev_features_t features)
 736 {
 737         /*
 738          * Since the ibmveth firmware interface does not have the
 739          * concept of separate tx/rx checksum offload enable, if rx
 740          * checksum is disabled we also have to disable tx checksum
 741          * offload. Once we disable rx checksum offload, we are no
 742          * longer allowed to send tx buffers that are not properly
 743          * checksummed.
 744          */
 745
 746         if (!(features & NETIF_F_RXCSUM))
 747                 features &= ~NETIF_F_CSUM_MASK;
 748
 749         return features;
 750 }
 751
 752 static int ibmveth_set_csum_offload(struct net_device *dev, u32 data)
 753 {
 754         struct ibmveth_adapter *adapter = netdev_priv(dev);
 755         unsigned long set_attr, clr_attr, ret_attr;
 756         unsigned long set_attr6, clr_attr6;
 757         long ret, ret4, ret6;
 758         int rc1 = 0, rc2 = 0;
 759         int restart = 0;
 760
 761         if (netif_running(dev)) {
 762                 restart = 1;
 763                 adapter->pool_config = 1;
 764                 ibmveth_close(dev);
 765                 adapter->pool_config = 0;
 766         }
 767
 768         set_attr = 0;
 769         clr_attr = 0;
 770         set_attr6 = 0;
 771         clr_attr6 = 0;
 772
 773         if (data) {
 774                 set_attr = IBMVETH_ILLAN_IPV4_TCP_CSUM;
 775                 set_attr6 = IBMVETH_ILLAN_IPV6_TCP_CSUM;
 776         } else {
 777                 clr_attr = IBMVETH_ILLAN_IPV4_TCP_CSUM;
 778                 clr_attr6 = IBMVETH_ILLAN_IPV6_TCP_CSUM;
 779         }
 780
 781         ret = h_illan_attributes(adapter->vdev->unit_address, 0, 0, &ret_attr);
 782
 783         if (ret == H_SUCCESS &&
 784             (ret_attr & IBMVETH_ILLAN_PADDED_PKT_CSUM)) {
 785                 ret4 = h_illan_attributes(adapter->vdev->unit_address, clr_attr,
 786                                          set_attr, &ret_attr);
 787
 788                 if (ret4 != H_SUCCESS) {
 789                         netdev_err(dev, "unable to change IPv4 checksum "
 790                                         "offload settings. %d rc=%ld\n",
 791                                         data, ret4);
 792
 793                         h_illan_attributes(adapter->vdev->unit_address,
 794                                            set_attr, clr_attr, &ret_attr);
 795
 796                         if (data == 1)
 797                                 dev->features &= ~NETIF_F_IP_CSUM;
 798
 799                 } else {
 800                         adapter->fw_ipv4_csum_support = data;
 801                 }
 802
 803                 ret6 = h_illan_attributes(adapter->vdev->unit_address,
 804                                          clr_attr6, set_attr6, &ret_attr);
 805
 806                 if (ret6 != H_SUCCESS) {
 807                         netdev_err(dev, "unable to change IPv6 checksum "
 808                                         "offload settings. %d rc=%ld\n",
 809                                         data, ret6);
 810
 811                         h_illan_attributes(adapter->vdev->unit_address,
 812                                            set_attr6, clr_attr6, &ret_attr);
 813
 814                         if (data == 1)
 815                                 dev->features &= ~NETIF_F_IPV6_CSUM;
 816
 817                 } else
 818                         adapter->fw_ipv6_csum_support = data;
 819
 820                 if (ret4 == H_SUCCESS || ret6 == H_SUCCESS)
 821                         adapter->rx_csum = data;
 822                 else
 823                         rc1 = -EIO;
 824         } else {
 825                 rc1 = -EIO;
 826                 netdev_err(dev, "unable to change checksum offload settings."
 827                                      " %d rc=%ld ret_attr=%lx\n", data, ret,
 828                                      ret_attr);
 829         }
 830
 831         if (restart)
 832                 rc2 = ibmveth_open(dev);
 833
 834         return rc1 ? rc1 : rc2;
 835 }
 836
 837 static int ibmveth_set_tso(struct net_device *dev, u32 data)
 838 {
 839         struct ibmveth_adapter *adapter = netdev_priv(dev);
 840         unsigned long set_attr, clr_attr, ret_attr;
 841         long ret1, ret2;
 842         int rc1 = 0, rc2 = 0;
 843         int restart = 0;
 844
 845         if (netif_running(dev)) {
 846                 restart = 1;
 847                 adapter->pool_config = 1;
 848                 ibmveth_close(dev);
 849                 adapter->pool_config = 0;
 850         }
 851
 852         set_attr = 0;
 853         clr_attr = 0;
 854
 855         if (data)
 856                 set_attr = IBMVETH_ILLAN_LRG_SR_ENABLED;
 857         else
 858                 clr_attr = IBMVETH_ILLAN_LRG_SR_ENABLED;
 859
 860         ret1 = h_illan_attributes(adapter->vdev->unit_address, 0, 0, &ret_attr);
 861
 862         if (ret1 == H_SUCCESS && (ret_attr & IBMVETH_ILLAN_LRG_SND_SUPPORT) &&
 863             !old_large_send) {
 864                 ret2 = h_illan_attributes(adapter->vdev->unit_address, clr_attr,
 865                                           set_attr, &ret_attr);
 866
 867                 if (ret2 != H_SUCCESS) {
 868                         netdev_err(dev, "unable to change tso settings. %d rc=%ld\n",
 869                                    data, ret2);
 870
 871                         h_illan_attributes(adapter->vdev->unit_address,
 872                                            set_attr, clr_attr, &ret_attr);
 873
 874                         if (data == 1)
 875                                 dev->features &= ~(NETIF_F_TSO | NETIF_F_TSO6);
 876                         rc1 = -EIO;
 877
 878                 } else {
 879                         adapter->fw_large_send_support = data;
 880                         adapter->large_send = data;
 881                 }
 882         } else {
 883                 /* Older firmware version of large send offload does not
 884                  * support tcp6/ipv6
 885                  */
 886                 if (data == 1) {
 887                         dev->features &= ~NETIF_F_TSO6;
 888                         netdev_info(dev, "TSO feature requires all partitions to have updated driver");
 889                 }
 890                 adapter->large_send = data;
 891         }
 892
 893         if (restart)
 894                 rc2 = ibmveth_open(dev);
 895
 896         return rc1 ? rc1 : rc2;
 897 }
 898
 899 static int ibmveth_set_features(struct net_device *dev,
 900         netdev_features_t features)
 901 {
 902         struct ibmveth_adapter *adapter = netdev_priv(dev);
 903         int rx_csum = !!(features & NETIF_F_RXCSUM);
 904         int large_send = !!(features & (NETIF_F_TSO | NETIF_F_TSO6));
 905         int rc1 = 0, rc2 = 0;
 906
 907         if (rx_csum != adapter->rx_csum) {
 908                 rc1 = ibmveth_set_csum_offload(dev, rx_csum);
 909                 if (rc1 && !adapter->rx_csum)
 910                         dev->features =
 911                                 features & ~(NETIF_F_CSUM_MASK |
 912                                              NETIF_F_RXCSUM);
 913         }
 914
 915         if (large_send != adapter->large_send) {
 916                 rc2 = ibmveth_set_tso(dev, large_send);
 917                 if (rc2 && !adapter->large_send)
 918                         dev->features =
 919                                 features & ~(NETIF_F_TSO | NETIF_F_TSO6);
 920         }
 921
 922         return rc1 ? rc1 : rc2;
 923 }
 924
 925 static void ibmveth_get_strings(struct net_device *dev, u32 stringset, u8 *data)
 926 {
 927         int i;
 928
 929         if (stringset != ETH_SS_STATS)
 930                 return;
 931
 932         for (i = 0; i < ARRAY_SIZE(ibmveth_stats); i++, data += ETH_GSTRING_LEN)
 933                 memcpy(data, ibmveth_stats[i].name, ETH_GSTRING_LEN);
 934 }
 935
 936 static int ibmveth_get_sset_count(struct net_device *dev, int sset)
 937 {
 938         switch (sset) {
 939         case ETH_SS_STATS:
 940                 return ARRAY_SIZE(ibmveth_stats);
 941         default:
 942                 return -EOPNOTSUPP;
 943         }
 944 }
 945
 946 static void ibmveth_get_ethtool_stats(struct net_device *dev,
 947                                       struct ethtool_stats *stats, u64 *data)
 948 {
 949         int i;
 950         struct ibmveth_adapter *adapter = netdev_priv(dev);
 951
 952         for (i = 0; i < ARRAY_SIZE(ibmveth_stats); i++)
 953                 data[i] = IBMVETH_GET_STAT(adapter, ibmveth_stats[i].offset);
 954 }
 955
 956 static const struct ethtool_ops netdev_ethtool_ops = {
 957         .get_drvinfo                     = netdev_get_drvinfo,
 958         .get_link                        = ethtool_op_get_link,
 959         .get_strings                     = ibmveth_get_strings,
 960         .get_sset_count                  = ibmveth_get_sset_count,
 961         .get_ethtool_stats               = ibmveth_get_ethtool_stats,
 962         .get_link_ksettings              = ibmveth_get_link_ksettings,
 963         .set_link_ksettings              = ibmveth_set_link_ksettings,
 964 };
 965
 966 static int ibmveth_ioctl(struct net_device *dev, struct ifreq *ifr, int cmd)
 967 {
 968         return -EOPNOTSUPP;
 969 }
 970
 971 static int ibmveth_send(struct ibmveth_adapter *adapter,
 972                         union ibmveth_buf_desc *descs, unsigned long mss)
 973 {
 974         unsigned long correlator;
 975         unsigned int retry_count;
 976         unsigned long ret;
 977
 978         /*
 979          * The retry count sets a maximum for the number of broadcast and
 980          * multicast destinations within the system.
 981          */
 982         retry_count = 1024;
 983         correlator = 0;
 984         do {
 985                 ret = h_send_logical_lan(adapter->vdev->unit_address,
 986                                              descs[0].desc, descs[1].desc,
 987                                              descs[2].desc, descs[3].desc,
 988                                              descs[4].desc, descs[5].desc,
 989                                              correlator, &correlator, mss,
 990                                              adapter->fw_large_send_support);
 991         } while ((ret == H_BUSY) && (retry_count--));
 992
 993         if (ret != H_SUCCESS && ret != H_DROPPED) {
 994                 netdev_err(adapter->netdev, "tx: h_send_logical_lan failed "
 995                            "with rc=%ld\n", ret);
 996                 return 1;
 997         }
 998
 999         return 0;
1000 }
1001
1002 static int ibmveth_is_packet_unsupported(struct sk_buff *skb,
1003                                          struct net_device *netdev)
1004 {
1005         struct ethhdr *ether_header;
1006         int ret = 0;
1007
1008         ether_header = eth_hdr(skb);
1009
1010         if (ether_addr_equal(ether_header->h_dest, netdev->dev_addr)) {
1011                 netdev_dbg(netdev, "veth doesn't support loopback packets, dropping packet.\n");
1012                 netdev->stats.tx_dropped++;
1013                 ret = -EOPNOTSUPP;
1014         }
1015
1016         return ret;
1017 }
1018
1019 static netdev_tx_t ibmveth_start_xmit(struct sk_buff *skb,
1020                                       struct net_device *netdev)
1021 {
1022         struct ibmveth_adapter *adapter = netdev_priv(netdev);
1023         unsigned int desc_flags;
1024         union ibmveth_buf_desc descs[6];
1025         int last, i;
1026         int force_bounce = 0;
1027         dma_addr_t dma_addr;
1028         unsigned long mss = 0;
1029
1030         if (ibmveth_is_packet_unsupported(skb, netdev))
1031                 goto out;
1032
1033         /* veth doesn't handle frag_list, so linearize the skb.
1034          * When GRO is enabled SKB's can have frag_list.
1035          */
1036         if (adapter->is_active_trunk &&
1037             skb_has_frag_list(skb) && __skb_linearize(skb)) {
1038                 netdev->stats.tx_dropped++;
1039                 goto out;
1040         }
1041
1042         /*
1043          * veth handles a maximum of 6 segments including the header, so
1044          * we have to linearize the skb if there are more than this.
1045          */
1046         if (skb_shinfo(skb)->nr_frags > 5 && __skb_linearize(skb)) {
1047                 netdev->stats.tx_dropped++;
1048                 goto out;
1049         }
1050
1051         /* veth can't checksum offload UDP */
1052         if (skb->ip_summed == CHECKSUM_PARTIAL &&
1053             ((skb->protocol == htons(ETH_P_IP) &&
1054               ip_hdr(skb)->protocol != IPPROTO_TCP) ||
1055              (skb->protocol == htons(ETH_P_IPV6) &&
1056               ipv6_hdr(skb)->nexthdr != IPPROTO_TCP)) &&
1057             skb_checksum_help(skb)) {
1058
1059                 netdev_err(netdev, "tx: failed to checksum packet\n");
1060                 netdev->stats.tx_dropped++;
1061                 goto out;
1062         }
1063
1064         desc_flags = IBMVETH_BUF_VALID;
1065
1066         if (skb->ip_summed == CHECKSUM_PARTIAL) {
1067                 unsigned char *buf = skb_transport_header(skb) +
1068                                                 skb->csum_offset;
1069
1070                 desc_flags |= (IBMVETH_BUF_NO_CSUM | IBMVETH_BUF_CSUM_GOOD);
1071
1072                 /* Need to zero out the checksum */
1073                 buf[0] = 0;
1074                 buf[1] = 0;
1075
1076                 if (skb_is_gso(skb) && adapter->fw_large_send_support)
1077                         desc_flags |= IBMVETH_BUF_LRG_SND;
1078         }
1079
1080 retry_bounce:
1081         memset(descs, 0, sizeof(descs));
1082
1083         /*
1084          * If a linear packet is below the rx threshold then
1085          * copy it into the static bounce buffer. This avoids the
1086          * cost of a TCE insert and remove.
1087          */
1088         if (force_bounce || (!skb_is_nonlinear(skb) &&
1089                                 (skb->len < tx_copybreak))) {
1090                 skb_copy_from_linear_data(skb, adapter->bounce_buffer,
1091                                           skb->len);
1092
1093                 descs[0].fields.flags_len = desc_flags | skb->len;
1094                 descs[0].fields.address = adapter->bounce_buffer_dma;
1095
1096                 if (ibmveth_send(adapter, descs, 0)) {
1097                         adapter->tx_send_failed++;
1098                         netdev->stats.tx_dropped++;
1099                 } else {
1100                         netdev->stats.tx_packets++;
1101                         netdev->stats.tx_bytes += skb->len;
1102                 }
1103
1104                 goto out;
1105         }
1106
1107         /* Map the header */
1108         dma_addr = dma_map_single(&adapter->vdev->dev, skb->data,
1109                                   skb_headlen(skb), DMA_TO_DEVICE);
1110         if (dma_mapping_error(&adapter->vdev->dev, dma_addr))
1111                 goto map_failed;
1112
1113         descs[0].fields.flags_len = desc_flags | skb_headlen(skb);
1114         descs[0].fields.address = dma_addr;
1115
1116         /* Map the frags */
1117         for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
1118                 const skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
1119
1120                 dma_addr = skb_frag_dma_map(&adapter->vdev->dev, frag, 0,
1121                                             skb_frag_size(frag), DMA_TO_DEVICE);
1122
1123                 if (dma_mapping_error(&adapter->vdev->dev, dma_addr))
1124                         goto map_failed_frags;
1125
1126                 descs[i+1].fields.flags_len = desc_flags | skb_frag_size(frag);
1127                 descs[i+1].fields.address = dma_addr;
1128         }
1129
1130         if (skb->ip_summed == CHECKSUM_PARTIAL && skb_is_gso(skb)) {
1131                 if (adapter->fw_large_send_support) {
1132                         mss = (unsigned long)skb_shinfo(skb)->gso_size;
1133                         adapter->tx_large_packets++;
1134                 } else if (!skb_is_gso_v6(skb)) {
1135                         /* Put -1 in the IP checksum to tell phyp it
1136                          * is a largesend packet. Put the mss in
1137                          * the TCP checksum.
1138                          */
1139                         ip_hdr(skb)->check = 0xffff;
1140                         tcp_hdr(skb)->check =
1141                                 cpu_to_be16(skb_shinfo(skb)->gso_size);
1142                         adapter->tx_large_packets++;
1143                 }
1144         }
1145
1146         if (ibmveth_send(adapter, descs, mss)) {
1147                 adapter->tx_send_failed++;
1148                 netdev->stats.tx_dropped++;
1149         } else {
1150                 netdev->stats.tx_packets++;
1151                 netdev->stats.tx_bytes += skb->len;
1152         }
1153
1154         dma_unmap_single(&adapter->vdev->dev,
1155                          descs[0].fields.address,
1156                          descs[0].fields.flags_len & IBMVETH_BUF_LEN_MASK,
1157                          DMA_TO_DEVICE);
1158
1159         for (i = 1; i < skb_shinfo(skb)->nr_frags + 1; i++)
1160                 dma_unmap_page(&adapter->vdev->dev, descs[i].fields.address,
1161                                descs[i].fields.flags_len & IBMVETH_BUF_LEN_MASK,
1162                                DMA_TO_DEVICE);
1163
1164 out:
1165         dev_consume_skb_any(skb);
1166         return NETDEV_TX_OK;
1167
1168 map_failed_frags:
1169         last = i+1;
1170         for (i = 1; i < last; i++)
1171                 dma_unmap_page(&adapter->vdev->dev, descs[i].fields.address,
1172                                descs[i].fields.flags_len & IBMVETH_BUF_LEN_MASK,
1173                                DMA_TO_DEVICE);
1174
1175         dma_unmap_single(&adapter->vdev->dev,
1176                          descs[0].fields.address,
1177                          descs[0].fields.flags_len & IBMVETH_BUF_LEN_MASK,
1178                          DMA_TO_DEVICE);
1179 map_failed:
1180         if (!firmware_has_feature(FW_FEATURE_CMO))
1181                 netdev_err(netdev, "tx: unable to map xmit buffer\n");
1182         adapter->tx_map_failed++;
1183         if (skb_linearize(skb)) {
1184                 netdev->stats.tx_dropped++;
1185                 goto out;
1186         }
1187         force_bounce = 1;
1188         goto retry_bounce;
1189 }
1190
1191 static void ibmveth_rx_mss_helper(struct sk_buff *skb, u16 mss, int lrg_pkt)
1192 {
1193         struct tcphdr *tcph;
1194         int offset = 0;
1195         int hdr_len;
1196
1197         /* only TCP packets will be aggregated */
1198         if (skb->protocol == htons(ETH_P_IP)) {
1199                 struct iphdr *iph = (struct iphdr *)skb->data;
1200
1201                 if (iph->protocol == IPPROTO_TCP) {
1202                         offset = iph->ihl * 4;
1203                         skb_shinfo(skb)->gso_type = SKB_GSO_TCPV4;
1204                 } else {
1205                         return;
1206                 }
1207         } else if (skb->protocol == htons(ETH_P_IPV6)) {
1208                 struct ipv6hdr *iph6 = (struct ipv6hdr *)skb->data;
1209
1210                 if (iph6->nexthdr == IPPROTO_TCP) {
1211                         offset = sizeof(struct ipv6hdr);
1212                         skb_shinfo(skb)->gso_type = SKB_GSO_TCPV6;
1213                 } else {
1214                         return;
1215                 }
1216         } else {
1217                 return;
1218         }
1219         /* if mss is not set through Large Packet bit/mss in rx buffer,
1220          * expect that the mss will be written to the tcp header checksum.
1221          */
1222         tcph = (struct tcphdr *)(skb->data + offset);
1223         if (lrg_pkt) {
1224                 skb_shinfo(skb)->gso_size = mss;
1225         } else if (offset) {
1226                 skb_shinfo(skb)->gso_size = ntohs(tcph->check);
1227                 tcph->check = 0;
1228         }
1229
1230         if (skb_shinfo(skb)->gso_size) {
1231                 hdr_len = offset + tcph->doff * 4;
1232                 skb_shinfo(skb)->gso_segs =
1233                                 DIV_ROUND_UP(skb->len - hdr_len,
1234                                              skb_shinfo(skb)->gso_size);
1235         }
1236 }
1237
1238 static void ibmveth_rx_csum_helper(struct sk_buff *skb,
1239                                    struct ibmveth_adapter *adapter)
1240 {
1241         struct iphdr *iph = NULL;
1242         struct ipv6hdr *iph6 = NULL;
1243         __be16 skb_proto = 0;
1244         u16 iphlen = 0;
1245         u16 iph_proto = 0;
1246         u16 tcphdrlen = 0;
1247
1248         skb_proto = be16_to_cpu(skb->protocol);
1249
1250         if (skb_proto == ETH_P_IP) {
1251                 iph = (struct iphdr *)skb->data;
1252
1253                 /* If the IP checksum is not offloaded and if the packet
1254                  *  is large send, the checksum must be rebuilt.
1255                  */
1256                 if (iph->check == 0xffff) {
1257                         iph->check = 0;
1258                         iph->check = ip_fast_csum((unsigned char *)iph,
1259                                                   iph->ihl);
1260                 }
1261
1262                 iphlen = iph->ihl * 4;
1263                 iph_proto = iph->protocol;
1264         } else if (skb_proto == ETH_P_IPV6) {
1265                 iph6 = (struct ipv6hdr *)skb->data;
1266                 iphlen = sizeof(struct ipv6hdr);
1267                 iph_proto = iph6->nexthdr;
1268         }
1269
1270         /* When CSO is enabled the TCP checksum may have be set to NULL by
1271          * the sender given that we zeroed out TCP checksum field in
1272          * transmit path (refer ibmveth_start_xmit routine). In this case set
1273          * up CHECKSUM_PARTIAL. If the packet is forwarded, the checksum will
1274          * then be recalculated by the destination NIC (CSO must be enabled
1275          * on the destination NIC).
1276          *
1277          * In an OVS environment, when a flow is not cached, specifically for a
1278          * new TCP connection, the first packet information is passed up to
1279          * the user space for finding a flow. During this process, OVS computes
1280          * checksum on the first packet when CHECKSUM_PARTIAL flag is set.
1281          *
1282          * So, re-compute TCP pseudo header checksum when configured for
1283          * trunk mode.
1284          */
1285         if (iph_proto == IPPROTO_TCP) {
1286                 struct tcphdr *tcph = (struct tcphdr *)(skb->data + iphlen);
1287                 if (tcph->check == 0x0000) {
1288                         /* Recompute TCP pseudo header checksum  */
1289                         if (adapter->is_active_trunk) {
1290                                 tcphdrlen = skb->len - iphlen;
1291                                 if (skb_proto == ETH_P_IP)
1292                                         tcph->check =
1293                                          ~csum_tcpudp_magic(iph->saddr,
1294                                         iph->daddr, tcphdrlen, iph_proto, 0);
1295                                 else if (skb_proto == ETH_P_IPV6)
1296                                         tcph->check =
1297                                          ~csum_ipv6_magic(&iph6->saddr,
1298                                         &iph6->daddr, tcphdrlen, iph_proto, 0);
1299                         }
1300                         /* Setup SKB fields for checksum offload */
1301                         skb_partial_csum_set(skb, iphlen,
1302                                              offsetof(struct tcphdr, check));
1303                         skb_reset_network_header(skb);
1304                 }
1305         }
1306 }
1307
1308 static int ibmveth_poll(struct napi_struct *napi, int budget)
1309 {
1310         struct ibmveth_adapter *adapter =
1311                         container_of(napi, struct ibmveth_adapter, napi);
1312         struct net_device *netdev = adapter->netdev;
1313         int frames_processed = 0;
1314         unsigned long lpar_rc;
1315         u16 mss = 0;
1316
1317         while (frames_processed < budget) {
1318                 if (!ibmveth_rxq_pending_buffer(adapter))
1319                         break;
1320
1321                 smp_rmb();
1322                 if (!ibmveth_rxq_buffer_valid(adapter)) {
1323                         wmb(); /* suggested by larson1 */
1324                         adapter->rx_invalid_buffer++;
1325                         netdev_dbg(netdev, "recycling invalid buffer\n");
1326                         ibmveth_rxq_recycle_buffer(adapter);
1327                 } else {
1328                         struct sk_buff *skb, *new_skb;
1329                         int length = ibmveth_rxq_frame_length(adapter);
1330                         int offset = ibmveth_rxq_frame_offset(adapter);
1331                         int csum_good = ibmveth_rxq_csum_good(adapter);
1332                         int lrg_pkt = ibmveth_rxq_large_packet(adapter);
1333                         __sum16 iph_check = 0;
1334
1335                         skb = ibmveth_rxq_get_buffer(adapter);
1336
1337                         /* if the large packet bit is set in the rx queue
1338                          * descriptor, the mss will be written by PHYP eight
1339                          * bytes from the start of the rx buffer, which is
1340                          * skb->data at this stage
1341                          */
1342                         if (lrg_pkt) {
1343                                 __be64 *rxmss = (__be64 *)(skb->data + 8);
1344
1345                                 mss = (u16)be64_to_cpu(*rxmss);
1346                         }
1347
1348                         new_skb = NULL;
1349                         if (length < rx_copybreak)
1350                                 new_skb = netdev_alloc_skb(netdev, length);
1351
1352                         if (new_skb) {
1353                                 skb_copy_to_linear_data(new_skb,
1354                                                         skb->data + offset,
1355                                                         length);
1356                                 if (rx_flush)
1357                                         ibmveth_flush_buffer(skb->data,
1358                                                 length + offset);
1359                                 if (!ibmveth_rxq_recycle_buffer(adapter))
1360                                         kfree_skb(skb);
1361                                 skb = new_skb;
1362                         } else {
1363                                 ibmveth_rxq_harvest_buffer(adapter);
1364                                 skb_reserve(skb, offset);
1365                         }
1366
1367                         skb_put(skb, length);
1368                         skb->protocol = eth_type_trans(skb, netdev);
1369
1370                         /* PHYP without PLSO support places a -1 in the ip
1371                          * checksum for large send frames.
1372                          */
1373                         if (skb->protocol == cpu_to_be16(ETH_P_IP)) {
1374                                 struct iphdr *iph = (struct iphdr *)skb->data;
1375
1376                                 iph_check = iph->check;
1377                         }
1378
1379                         if ((length > netdev->mtu + ETH_HLEN) ||
1380                             lrg_pkt || iph_check == 0xffff) {
1381                                 ibmveth_rx_mss_helper(skb, mss, lrg_pkt);
1382                                 adapter->rx_large_packets++;
1383                         }
1384
1385                         if (csum_good) {
1386                                 skb->ip_summed = CHECKSUM_UNNECESSARY;
1387                                 ibmveth_rx_csum_helper(skb, adapter);
1388                         }
1389
1390                         napi_gro_receive(napi, skb);    /* send it up */
1391
1392                         netdev->stats.rx_packets++;
1393                         netdev->stats.rx_bytes += length;
1394                         frames_processed++;
1395                 }
1396         }
1397
1398         ibmveth_replenish_task(adapter);
1399
1400         if (frames_processed < budget) {
1401                 napi_complete_done(napi, frames_processed);
1402
1403                 /* We think we are done - reenable interrupts,
1404                  * then check once more to make sure we are done.
1405                  */
1406                 lpar_rc = h_vio_signal(adapter->vdev->unit_address,
1407                                        VIO_IRQ_ENABLE);
1408
1409                 BUG_ON(lpar_rc != H_SUCCESS);
1410
1411                 if (ibmveth_rxq_pending_buffer(adapter) &&
1412                     napi_reschedule(napi)) {
1413                         lpar_rc = h_vio_signal(adapter->vdev->unit_address,
1414                                                VIO_IRQ_DISABLE);
1415                 }
1416         }
1417
1418         return frames_processed;
1419 }
1420
1421 static irqreturn_t ibmveth_interrupt(int irq, void *dev_instance)
1422 {
1423         struct net_device *netdev = dev_instance;
1424         struct ibmveth_adapter *adapter = netdev_priv(netdev);
1425         unsigned long lpar_rc;
1426
1427         if (napi_schedule_prep(&adapter->napi)) {
1428                 lpar_rc = h_vio_signal(adapter->vdev->unit_address,
1429                                        VIO_IRQ_DISABLE);
1430                 BUG_ON(lpar_rc != H_SUCCESS);
1431                 __napi_schedule(&adapter->napi);
1432         }
1433         return IRQ_HANDLED;
1434 }
1435
1436 static void ibmveth_set_multicast_list(struct net_device *netdev)
1437 {
1438         struct ibmveth_adapter *adapter = netdev_priv(netdev);
1439         unsigned long lpar_rc;
1440
1441         if ((netdev->flags & IFF_PROMISC) ||
1442             (netdev_mc_count(netdev) > adapter->mcastFilterSize)) {
1443                 lpar_rc = h_multicast_ctrl(adapter->vdev->unit_address,
1444                                            IbmVethMcastEnableRecv |
1445                                            IbmVethMcastDisableFiltering,
1446                                            0);
1447                 if (lpar_rc != H_SUCCESS) {
1448                         netdev_err(netdev, "h_multicast_ctrl rc=%ld when "
1449                                    "entering promisc mode\n", lpar_rc);
1450                 }
1451         } else {
1452                 struct netdev_hw_addr *ha;
1453                 /* clear the filter table & disable filtering */
1454                 lpar_rc = h_multicast_ctrl(adapter->vdev->unit_address,
1455                                            IbmVethMcastEnableRecv |
1456                                            IbmVethMcastDisableFiltering |
1457                                            IbmVethMcastClearFilterTable,
1458                                            0);
1459                 if (lpar_rc != H_SUCCESS) {
1460                         netdev_err(netdev, "h_multicast_ctrl rc=%ld when "
1461                                    "attempting to clear filter table\n",
1462                                    lpar_rc);
1463                 }
1464                 /* add the addresses to the filter table */
1465                 netdev_for_each_mc_addr(ha, netdev) {
1466                         /* add the multicast address to the filter table */
1467                         u64 mcast_addr;
1468                         mcast_addr = ether_addr_to_u64(ha->addr);
1469                         lpar_rc = h_multicast_ctrl(adapter->vdev->unit_address,
1470                                                    IbmVethMcastAddFilter,
1471                                                    mcast_addr);
1472                         if (lpar_rc != H_SUCCESS) {
1473                                 netdev_err(netdev, "h_multicast_ctrl rc=%ld "
1474                                            "when adding an entry to the filter "
1475                                            "table\n", lpar_rc);
1476                         }
1477                 }
1478
1479                 /* re-enable filtering */
1480                 lpar_rc = h_multicast_ctrl(adapter->vdev->unit_address,
1481                                            IbmVethMcastEnableFiltering,
1482                                            0);
1483                 if (lpar_rc != H_SUCCESS) {
1484                         netdev_err(netdev, "h_multicast_ctrl rc=%ld when "
1485                                    "enabling filtering\n", lpar_rc);
1486                 }
1487         }
1488 }
1489
1490 static int ibmveth_change_mtu(struct net_device *dev, int new_mtu)
1491 {
1492         struct ibmveth_adapter *adapter = netdev_priv(dev);
1493         struct vio_dev *viodev = adapter->vdev;
1494         int new_mtu_oh = new_mtu + IBMVETH_BUFF_OH;
1495         int i, rc;
1496         int need_restart = 0;
1497
1498         for (i = 0; i < IBMVETH_NUM_BUFF_POOLS; i++)
1499                 if (new_mtu_oh <= adapter->rx_buff_pool[i].buff_size)
1500                         break;
1501
1502         if (i == IBMVETH_NUM_BUFF_POOLS)
1503                 return -EINVAL;
1504
1505         /* Deactivate all the buffer pools so that the next loop can activate
1506            only the buffer pools necessary to hold the new MTU */
1507         if (netif_running(adapter->netdev)) {
1508                 need_restart = 1;
1509                 adapter->pool_config = 1;
1510                 ibmveth_close(adapter->netdev);
1511                 adapter->pool_config = 0;
1512         }
1513
1514         /* Look for an active buffer pool that can hold the new MTU */
1515         for (i = 0; i < IBMVETH_NUM_BUFF_POOLS; i++) {
1516                 adapter->rx_buff_pool[i].active = 1;
1517
1518                 if (new_mtu_oh <= adapter->rx_buff_pool[i].buff_size) {
1519                         dev->mtu = new_mtu;
1520                         vio_cmo_set_dev_desired(viodev,
1521                                                 ibmveth_get_desired_dma
1522                                                 (viodev));
1523                         if (need_restart) {
1524                                 return ibmveth_open(adapter->netdev);
1525                         }
1526                         return 0;
1527                 }
1528         }
1529
1530         if (need_restart && (rc = ibmveth_open(adapter->netdev)))
1531                 return rc;
1532
1533         return -EINVAL;
1534 }
1535
1536 #ifdef CONFIG_NET_POLL_CONTROLLER
1537 static void ibmveth_poll_controller(struct net_device *dev)
1538 {
1539         ibmveth_replenish_task(netdev_priv(dev));
1540         ibmveth_interrupt(dev->irq, dev);
1541 }
1542 #endif
1543
1544 /**
1545  * ibmveth_get_desired_dma - Calculate IO memory desired by the driver
1546  *
1547  * @vdev: struct vio_dev for the device whose desired IO mem is to be returned
1548  *
1549  * Return value:
1550  *      Number of bytes of IO data the driver will need to perform well.
1551  */
1552 static unsigned long ibmveth_get_desired_dma(struct vio_dev *vdev)
1553 {
1554         struct net_device *netdev = dev_get_drvdata(&vdev->dev);
1555         struct ibmveth_adapter *adapter;
1556         struct iommu_table *tbl;
1557         unsigned long ret;
1558         int i;
1559         int rxqentries = 1;
1560
1561         tbl = get_iommu_table_base(&vdev->dev);
1562
1563         /* netdev inits at probe time along with the structures we need below*/
1564         if (netdev == NULL)
1565                 return IOMMU_PAGE_ALIGN(IBMVETH_IO_ENTITLEMENT_DEFAULT, tbl);
1566
1567         adapter = netdev_priv(netdev);
1568
1569         ret = IBMVETH_BUFF_LIST_SIZE + IBMVETH_FILT_LIST_SIZE;
1570         ret += IOMMU_PAGE_ALIGN(netdev->mtu, tbl);
1571
1572         for (i = 0; i < IBMVETH_NUM_BUFF_POOLS; i++) {
1573                 /* add the size of the active receive buffers */
1574                 if (adapter->rx_buff_pool[i].active)
1575                         ret +=
1576                             adapter->rx_buff_pool[i].size *
1577                             IOMMU_PAGE_ALIGN(adapter->rx_buff_pool[i].
1578                                              buff_size, tbl);
1579                 rxqentries += adapter->rx_buff_pool[i].size;
1580         }
1581         /* add the size of the receive queue entries */
1582         ret += IOMMU_PAGE_ALIGN(
1583                 rxqentries * sizeof(struct ibmveth_rx_q_entry), tbl);
1584
1585         return ret;
1586 }
1587
1588 static int ibmveth_set_mac_addr(struct net_device *dev, void *p)
1589 {
1590         struct ibmveth_adapter *adapter = netdev_priv(dev);
1591         struct sockaddr *addr = p;
1592         u64 mac_address;
1593         int rc;
1594
1595         if (!is_valid_ether_addr(addr->sa_data))
1596                 return -EADDRNOTAVAIL;
1597
1598         mac_address = ether_addr_to_u64(addr->sa_data);
1599         rc = h_change_logical_lan_mac(adapter->vdev->unit_address, mac_address);
1600         if (rc) {
1601                 netdev_err(adapter->netdev, "h_change_logical_lan_mac failed with rc=%d\n", rc);
1602                 return rc;
1603         }
1604
1605         eth_hw_addr_set(dev, addr->sa_data);
1606
1607         return 0;
1608 }
1609
1610 static const struct net_device_ops ibmveth_netdev_ops = {
1611         .ndo_open               = ibmveth_open,
1612         .ndo_stop               = ibmveth_close,
1613         .ndo_start_xmit         = ibmveth_start_xmit,
1614         .ndo_set_rx_mode        = ibmveth_set_multicast_list,
1615         .ndo_eth_ioctl          = ibmveth_ioctl,
1616         .ndo_change_mtu         = ibmveth_change_mtu,
1617         .ndo_fix_features       = ibmveth_fix_features,
1618         .ndo_set_features       = ibmveth_set_features,
1619         .ndo_validate_addr      = eth_validate_addr,
1620         .ndo_set_mac_address    = ibmveth_set_mac_addr,
1621 #ifdef CONFIG_NET_POLL_CONTROLLER
1622         .ndo_poll_controller    = ibmveth_poll_controller,
1623 #endif
1624 };
1625
1626 static int ibmveth_probe(struct vio_dev *dev, const struct vio_device_id *id)
1627 {
1628         int rc, i, mac_len;
1629         struct net_device *netdev;
1630         struct ibmveth_adapter *adapter;
1631         unsigned char *mac_addr_p;
1632         __be32 *mcastFilterSize_p;
1633         long ret;
1634         unsigned long ret_attr;
1635
1636         dev_dbg(&dev->dev, "entering ibmveth_probe for UA 0x%x\n",
1637                 dev->unit_address);
1638
1639         mac_addr_p = (unsigned char *)vio_get_attribute(dev, VETH_MAC_ADDR,
1640                                                         &mac_len);
1641         if (!mac_addr_p) {
1642                 dev_err(&dev->dev, "Can't find VETH_MAC_ADDR attribute\n");
1643                 return -EINVAL;
1644         }
1645         /* Workaround for old/broken pHyp */
1646         if (mac_len == 8)
1647                 mac_addr_p += 2;
1648         else if (mac_len != 6) {
1649                 dev_err(&dev->dev, "VETH_MAC_ADDR attribute wrong len %d\n",
1650                         mac_len);
1651                 return -EINVAL;
1652         }
1653
1654         mcastFilterSize_p = (__be32 *)vio_get_attribute(dev,
1655                                                         VETH_MCAST_FILTER_SIZE,
1656                                                         NULL);
1657         if (!mcastFilterSize_p) {
1658                 dev_err(&dev->dev, "Can't find VETH_MCAST_FILTER_SIZE "
1659                         "attribute\n");
1660                 return -EINVAL;
1661         }
1662
1663         netdev = alloc_etherdev(sizeof(struct ibmveth_adapter));
1664
1665         if (!netdev)
1666                 return -ENOMEM;
1667
1668         adapter = netdev_priv(netdev);
1669         dev_set_drvdata(&dev->dev, netdev);
1670
1671         adapter->vdev = dev;
1672         adapter->netdev = netdev;
1673         adapter->mcastFilterSize = be32_to_cpu(*mcastFilterSize_p);
1674         adapter->pool_config = 0;
1675         ibmveth_init_link_settings(netdev);
1676
1677         netif_napi_add(netdev, &adapter->napi, ibmveth_poll, 16);
1678
1679         netdev->irq = dev->irq;
1680         netdev->netdev_ops = &ibmveth_netdev_ops;
1681         netdev->ethtool_ops = &netdev_ethtool_ops;
1682         SET_NETDEV_DEV(netdev, &dev->dev);
1683         netdev->hw_features = NETIF_F_SG;
1684         if (vio_get_attribute(dev, "ibm,illan-options", NULL) != NULL) {
1685                 netdev->hw_features |= NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM |
1686                                        NETIF_F_RXCSUM;
1687         }
1688
1689         netdev->features |= netdev->hw_features;
1690
1691         ret = h_illan_attributes(adapter->vdev->unit_address, 0, 0, &ret_attr);
1692
1693         /* If running older firmware, TSO should not be enabled by default */
1694         if (ret == H_SUCCESS && (ret_attr & IBMVETH_ILLAN_LRG_SND_SUPPORT) &&
1695             !old_large_send) {
1696                 netdev->hw_features |= NETIF_F_TSO | NETIF_F_TSO6;
1697                 netdev->features |= netdev->hw_features;
1698         } else {
1699                 netdev->hw_features |= NETIF_F_TSO;
1700         }
1701
1702         adapter->is_active_trunk = false;
1703         if (ret == H_SUCCESS && (ret_attr & IBMVETH_ILLAN_ACTIVE_TRUNK)) {
1704                 adapter->is_active_trunk = true;
1705                 netdev->hw_features |= NETIF_F_FRAGLIST;
1706                 netdev->features |= NETIF_F_FRAGLIST;
1707         }
1708
1709         netdev->min_mtu = IBMVETH_MIN_MTU;
1710         netdev->max_mtu = ETH_MAX_MTU - IBMVETH_BUFF_OH;
1711
1712         eth_hw_addr_set(netdev, mac_addr_p);
1713
1714         if (firmware_has_feature(FW_FEATURE_CMO))
1715                 memcpy(pool_count, pool_count_cmo, sizeof(pool_count));
1716
1717         for (i = 0; i < IBMVETH_NUM_BUFF_POOLS; i++) {
1718                 struct kobject *kobj = &adapter->rx_buff_pool[i].kobj;
1719                 int error;
1720
1721                 ibmveth_init_buffer_pool(&adapter->rx_buff_pool[i], i,
1722                                          pool_count[i], pool_size[i],
1723                                          pool_active[i]);
1724                 error = kobject_init_and_add(kobj, &ktype_veth_pool,
1725                                              &dev->dev.kobj, "pool%d", i);
1726                 if (!error)
1727                         kobject_uevent(kobj, KOBJ_ADD);
1728         }
1729
1730         netdev_dbg(netdev, "adapter @ 0x%p\n", adapter);
1731         netdev_dbg(netdev, "registering netdev...\n");
1732
1733         ibmveth_set_features(netdev, netdev->features);
1734
1735         rc = register_netdev(netdev);
1736
1737         if (rc) {
1738                 netdev_dbg(netdev, "failed to register netdev rc=%d\n", rc);
1739                 free_netdev(netdev);
1740                 return rc;
1741         }
1742
1743         netdev_dbg(netdev, "registered\n");
1744
1745         return 0;
1746 }
1747
1748 static void ibmveth_remove(struct vio_dev *dev)
1749 {
1750         struct net_device *netdev = dev_get_drvdata(&dev->dev);
1751         struct ibmveth_adapter *adapter = netdev_priv(netdev);
1752         int i;
1753
1754         for (i = 0; i < IBMVETH_NUM_BUFF_POOLS; i++)
1755                 kobject_put(&adapter->rx_buff_pool[i].kobj);
1756
1757         unregister_netdev(netdev);
1758
1759         free_netdev(netdev);
1760         dev_set_drvdata(&dev->dev, NULL);
1761 }
1762
1763 static struct attribute veth_active_attr;
1764 static struct attribute veth_num_attr;
1765 static struct attribute veth_size_attr;
1766
1767 static ssize_t veth_pool_show(struct kobject *kobj,
1768                               struct attribute *attr, char *buf)
1769 {
1770         struct ibmveth_buff_pool *pool = container_of(kobj,
1771                                                       struct ibmveth_buff_pool,
1772                                                       kobj);
1773
1774         if (attr == &veth_active_attr)
1775                 return sprintf(buf, "%d\n", pool->active);
1776         else if (attr == &veth_num_attr)
1777                 return sprintf(buf, "%d\n", pool->size);
1778         else if (attr == &veth_size_attr)
1779                 return sprintf(buf, "%d\n", pool->buff_size);
1780         return 0;
1781 }
1782
1783 static ssize_t veth_pool_store(struct kobject *kobj, struct attribute *attr,
1784                                const char *buf, size_t count)
1785 {
1786         struct ibmveth_buff_pool *pool = container_of(kobj,
1787                                                       struct ibmveth_buff_pool,
1788                                                       kobj);
1789         struct net_device *netdev = dev_get_drvdata(kobj_to_dev(kobj->parent));
1790         struct ibmveth_adapter *adapter = netdev_priv(netdev);
1791         long value = simple_strtol(buf, NULL, 10);
1792         long rc;
1793
1794         if (attr == &veth_active_attr) {
1795                 if (value && !pool->active) {
1796                         if (netif_running(netdev)) {
1797                                 if (ibmveth_alloc_buffer_pool(pool)) {
1798                                         netdev_err(netdev,
1799                                                    "unable to alloc pool\n");
1800                                         return -ENOMEM;
1801                                 }
1802                                 pool->active = 1;
1803                                 adapter->pool_config = 1;
1804                                 ibmveth_close(netdev);
1805                                 adapter->pool_config = 0;
1806                                 if ((rc = ibmveth_open(netdev)))
1807                                         return rc;
1808                         } else {
1809                                 pool->active = 1;
1810                         }
1811                 } else if (!value && pool->active) {
1812                         int mtu = netdev->mtu + IBMVETH_BUFF_OH;
1813                         int i;
1814                         /* Make sure there is a buffer pool with buffers that
1815                            can hold a packet of the size of the MTU */
1816                         for (i = 0; i < IBMVETH_NUM_BUFF_POOLS; i++) {
1817                                 if (pool == &adapter->rx_buff_pool[i])
1818                                         continue;
1819                                 if (!adapter->rx_buff_pool[i].active)
1820                                         continue;
1821                                 if (mtu <= adapter->rx_buff_pool[i].buff_size)
1822                                         break;
1823                         }
1824
1825                         if (i == IBMVETH_NUM_BUFF_POOLS) {
1826                                 netdev_err(netdev, "no active pool >= MTU\n");
1827                                 return -EPERM;
1828                         }
1829
1830                         if (netif_running(netdev)) {
1831                                 adapter->pool_config = 1;
1832                                 ibmveth_close(netdev);
1833                                 pool->active = 0;
1834                                 adapter->pool_config = 0;
1835                                 if ((rc = ibmveth_open(netdev)))
1836                                         return rc;
1837                         }
1838                         pool->active = 0;
1839                 }
1840         } else if (attr == &veth_num_attr) {
1841                 if (value <= 0 || value > IBMVETH_MAX_POOL_COUNT) {
1842                         return -EINVAL;
1843                 } else {
1844                         if (netif_running(netdev)) {
1845                                 adapter->pool_config = 1;
1846                                 ibmveth_close(netdev);
1847                                 adapter->pool_config = 0;
1848                                 pool->size = value;
1849                                 if ((rc = ibmveth_open(netdev)))
1850                                         return rc;
1851                         } else {
1852                                 pool->size = value;
1853                         }
1854                 }
1855         } else if (attr == &veth_size_attr) {
1856                 if (value <= IBMVETH_BUFF_OH || value > IBMVETH_MAX_BUF_SIZE) {
1857                         return -EINVAL;
1858                 } else {
1859                         if (netif_running(netdev)) {
1860                                 adapter->pool_config = 1;
1861                                 ibmveth_close(netdev);
1862                                 adapter->pool_config = 0;
1863                                 pool->buff_size = value;
1864                                 if ((rc = ibmveth_open(netdev)))
1865                                         return rc;
1866                         } else {
1867                                 pool->buff_size = value;
1868                         }
1869                 }
1870         }
1871
1872         /* kick the interrupt handler to allocate/deallocate pools */
1873         ibmveth_interrupt(netdev->irq, netdev);
1874         return count;
1875 }
1876
1877
1878 #define ATTR(_name, _mode)                              \
1879         struct attribute veth_##_name##_attr = {        \
1880         .name = __stringify(_name), .mode = _mode,      \
1881         };
1882
1883 static ATTR(active, 0644);
1884 static ATTR(num, 0644);
1885 static ATTR(size, 0644);
1886
1887 static struct attribute *veth_pool_attrs[] = {
1888         &veth_active_attr,
1889         &veth_num_attr,
1890         &veth_size_attr,
1891         NULL,
1892 };
1893
1894 static const struct sysfs_ops veth_pool_ops = {
1895         .show   = veth_pool_show,
1896         .store  = veth_pool_store,
1897 };
1898
1899 static struct kobj_type ktype_veth_pool = {
1900         .release        = NULL,
1901         .sysfs_ops      = &veth_pool_ops,
1902         .default_attrs  = veth_pool_attrs,
1903 };
1904
1905 static int ibmveth_resume(struct device *dev)
1906 {
1907         struct net_device *netdev = dev_get_drvdata(dev);
1908         ibmveth_interrupt(netdev->irq, netdev);
1909         return 0;
1910 }
1911
1912 static const struct vio_device_id ibmveth_device_table[] = {
1913         { "network", "IBM,l-lan"},
1914         { "", "" }
1915 };
1916 MODULE_DEVICE_TABLE(vio, ibmveth_device_table);
1917
1918 static const struct dev_pm_ops ibmveth_pm_ops = {
1919         .resume = ibmveth_resume
1920 };
1921
1922 static struct vio_driver ibmveth_driver = {
1923         .id_table       = ibmveth_device_table,
1924         .probe          = ibmveth_probe,
1925         .remove         = ibmveth_remove,
1926         .get_desired_dma = ibmveth_get_desired_dma,
1927         .name           = ibmveth_driver_name,
1928         .pm             = &ibmveth_pm_ops,
1929 };
1930
1931 static int __init ibmveth_module_init(void)
1932 {
1933         printk(KERN_DEBUG "%s: %s %s\n", ibmveth_driver_name,
1934                ibmveth_driver_string, ibmveth_driver_version);
1935
1936         return vio_register_driver(&ibmveth_driver);
1937 }
1938
1939 static void __exit ibmveth_module_exit(void)
1940 {
1941         vio_unregister_driver(&ibmveth_driver);
1942 }
1943
1944 module_init(ibmveth_module_init);
1945 module_exit(ibmveth_module_exit);