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/types.h>
  28 #include <linux/errno.h>
  29 #include <linux/dma-mapping.h>
  30 #include <linux/kernel.h>
  31 #include <linux/netdevice.h>
  32 #include <linux/etherdevice.h>
  33 #include <linux/skbuff.h>
  34 #include <linux/init.h>
  35 #include <linux/interrupt.h>
  36 #include <linux/mm.h>
  37 #include <linux/pm.h>
  38 #include <linux/ethtool.h>
  39 #include <linux/in.h>
  40 #include <linux/ip.h>
  41 #include <linux/ipv6.h>
  42 #include <linux/slab.h>
  43 #include <asm/hvcall.h>
  44 #include <linux/atomic.h>
  45 #include <asm/vio.h>
  46 #include <asm/iommu.h>
  47 #include <asm/firmware.h>
  48 #include <net/tcp.h>
  49 #include <net/ip6_checksum.h>
  50
  51 #include "ibmveth.h"
  52
  53 static irqreturn_t ibmveth_interrupt(int irq, void *dev_instance);
  54 static void ibmveth_rxq_harvest_buffer(struct ibmveth_adapter *adapter);
  55 static unsigned long ibmveth_get_desired_dma(struct vio_dev *vdev);
  56
  57 static struct kobj_type ktype_veth_pool;
  58
  59
  60 static const char ibmveth_driver_name[] = "ibmveth";
  61 static const char ibmveth_driver_string[] = "IBM Power Virtual Ethernet Driver";
  62 #define ibmveth_driver_version "1.06"
  63
  64 MODULE_AUTHOR("Santiago Leon <santil@linux.vnet.ibm.com>");
  65 MODULE_DESCRIPTION("IBM Power Virtual Ethernet Driver");
  66 MODULE_LICENSE("GPL");
  67 MODULE_VERSION(ibmveth_driver_version);
  68
  69 static unsigned int tx_copybreak __read_mostly = 128;
  70 module_param(tx_copybreak, uint, 0644);
  71 MODULE_PARM_DESC(tx_copybreak,
  72         "Maximum size of packet that is copied to a new buffer on transmit");
  73
  74 static unsigned int rx_copybreak __read_mostly = 128;
  75 module_param(rx_copybreak, uint, 0644);
  76 MODULE_PARM_DESC(rx_copybreak,
  77         "Maximum size of packet that is copied to a new buffer on receive");
  78
  79 static unsigned int rx_flush __read_mostly = 0;
  80 module_param(rx_flush, uint, 0644);
  81 MODULE_PARM_DESC(rx_flush, "Flush receive buffers before use");
  82
  83 static bool old_large_send __read_mostly;
  84 module_param(old_large_send, bool, 0444);
  85 MODULE_PARM_DESC(old_large_send,
  86         "Use old large send method on firmware that supports the new method");
  87
  88 struct ibmveth_stat {
  89         char name[ETH_GSTRING_LEN];
  90         int offset;
  91 };
  92
  93 #define IBMVETH_STAT_OFF(stat) offsetof(struct ibmveth_adapter, stat)
  94 #define IBMVETH_GET_STAT(a, off) *((u64 *)(((unsigned long)(a)) + off))
  95
  96 struct ibmveth_stat ibmveth_stats[] = {
  97         { "replenish_task_cycles", IBMVETH_STAT_OFF(replenish_task_cycles) },
  98         { "replenish_no_mem", IBMVETH_STAT_OFF(replenish_no_mem) },
  99         { "replenish_add_buff_failure",
 100                         IBMVETH_STAT_OFF(replenish_add_buff_failure) },
 101         { "replenish_add_buff_success",
 102                         IBMVETH_STAT_OFF(replenish_add_buff_success) },
 103         { "rx_invalid_buffer", IBMVETH_STAT_OFF(rx_invalid_buffer) },
 104         { "rx_no_buffer", IBMVETH_STAT_OFF(rx_no_buffer) },
 105         { "tx_map_failed", IBMVETH_STAT_OFF(tx_map_failed) },
 106         { "tx_send_failed", IBMVETH_STAT_OFF(tx_send_failed) },
 107         { "fw_enabled_ipv4_csum", IBMVETH_STAT_OFF(fw_ipv4_csum_support) },
 108         { "fw_enabled_ipv6_csum", IBMVETH_STAT_OFF(fw_ipv6_csum_support) },
 109         { "tx_large_packets", IBMVETH_STAT_OFF(tx_large_packets) },
 110         { "rx_large_packets", IBMVETH_STAT_OFF(rx_large_packets) },
 111         { "fw_enabled_large_send", IBMVETH_STAT_OFF(fw_large_send_support) }
 112 };
 113
 114 /* simple methods of getting data from the current rxq entry */
 115 static inline u32 ibmveth_rxq_flags(struct ibmveth_adapter *adapter)
 116 {
 117         return be32_to_cpu(adapter->rx_queue.queue_addr[adapter->rx_queue.index].flags_off);
 118 }
 119
 120 static inline int ibmveth_rxq_toggle(struct ibmveth_adapter *adapter)
 121 {
 122         return (ibmveth_rxq_flags(adapter) & IBMVETH_RXQ_TOGGLE) >>
 123                         IBMVETH_RXQ_TOGGLE_SHIFT;
 124 }
 125
 126 static inline int ibmveth_rxq_pending_buffer(struct ibmveth_adapter *adapter)
 127 {
 128         return ibmveth_rxq_toggle(adapter) == adapter->rx_queue.toggle;
 129 }
 130
 131 static inline int ibmveth_rxq_buffer_valid(struct ibmveth_adapter *adapter)
 132 {
 133         return ibmveth_rxq_flags(adapter) & IBMVETH_RXQ_VALID;
 134 }
 135
 136 static inline int ibmveth_rxq_frame_offset(struct ibmveth_adapter *adapter)
 137 {
 138         return ibmveth_rxq_flags(adapter) & IBMVETH_RXQ_OFF_MASK;
 139 }
 140
 141 static inline int ibmveth_rxq_large_packet(struct ibmveth_adapter *adapter)
 142 {
 143         return ibmveth_rxq_flags(adapter) & IBMVETH_RXQ_LRG_PKT;
 144 }
 145
 146 static inline int ibmveth_rxq_frame_length(struct ibmveth_adapter *adapter)
 147 {
 148         return be32_to_cpu(adapter->rx_queue.queue_addr[adapter->rx_queue.index].length);
 149 }
 150
 151 static inline int ibmveth_rxq_csum_good(struct ibmveth_adapter *adapter)
 152 {
 153         return ibmveth_rxq_flags(adapter) & IBMVETH_RXQ_CSUM_GOOD;
 154 }
 155
 156 /* setup the initial settings for a buffer pool */
 157 static void ibmveth_init_buffer_pool(struct ibmveth_buff_pool *pool,
 158                                      u32 pool_index, u32 pool_size,
 159                                      u32 buff_size, u32 pool_active)
 160 {
 161         pool->size = pool_size;
 162         pool->index = pool_index;
 163         pool->buff_size = buff_size;
 164         pool->threshold = pool_size * 7 / 8;
 165         pool->active = pool_active;
 166 }
 167
 168 /* allocate and setup an buffer pool - called during open */
 169 static int ibmveth_alloc_buffer_pool(struct ibmveth_buff_pool *pool)
 170 {
 171         int i;
 172
 173         pool->free_map = kmalloc_array(pool->size, sizeof(u16), GFP_KERNEL);
 174
 175         if (!pool->free_map)
 176                 return -1;
 177
 178         pool->dma_addr = kcalloc(pool->size, sizeof(dma_addr_t), GFP_KERNEL);
 179         if (!pool->dma_addr) {
 180                 kfree(pool->free_map);
 181                 pool->free_map = NULL;
 182                 return -1;
 183         }
 184
 185         pool->skbuff = kcalloc(pool->size, sizeof(void *), GFP_KERNEL);
 186
 187         if (!pool->skbuff) {
 188                 kfree(pool->dma_addr);
 189                 pool->dma_addr = NULL;
 190
 191                 kfree(pool->free_map);
 192                 pool->free_map = NULL;
 193                 return -1;
 194         }
 195
 196         for (i = 0; i < pool->size; ++i)
 197                 pool->free_map[i] = i;
 198
 199         atomic_set(&pool->available, 0);
 200         pool->producer_index = 0;
 201         pool->consumer_index = 0;
 202
 203         return 0;
 204 }
 205
 206 static inline void ibmveth_flush_buffer(void *addr, unsigned long length)
 207 {
 208         unsigned long offset;
 209
 210         for (offset = 0; offset < length; offset += SMP_CACHE_BYTES)
 211                 asm("dcbfl %0,%1" :: "b" (addr), "r" (offset));
 212 }
 213
 214 /* replenish the buffers for a pool.  note that we don't need to
 215  * skb_reserve these since they are used for incoming...
 216  */
 217 static void ibmveth_replenish_buffer_pool(struct ibmveth_adapter *adapter,
 218                                           struct ibmveth_buff_pool *pool)
 219 {
 220         u32 i;
 221         u32 count = pool->size - atomic_read(&pool->available);
 222         u32 buffers_added = 0;
 223         struct sk_buff *skb;
 224         unsigned int free_index, index;
 225         u64 correlator;
 226         unsigned long lpar_rc;
 227         dma_addr_t dma_addr;
 228
 229         mb();
 230
 231         for (i = 0; i < count; ++i) {
 232                 union ibmveth_buf_desc desc;
 233
 234                 skb = netdev_alloc_skb(adapter->netdev, pool->buff_size);
 235
 236                 if (!skb) {
 237                         netdev_dbg(adapter->netdev,
 238                                    "replenish: unable to allocate skb\n");
 239                         adapter->replenish_no_mem++;
 240                         break;
 241                 }
 242
 243                 free_index = pool->consumer_index;
 244                 pool->consumer_index++;
 245                 if (pool->consumer_index >= pool->size)
 246                         pool->consumer_index = 0;
 247                 index = pool->free_map[free_index];
 248
 249                 BUG_ON(index == IBM_VETH_INVALID_MAP);
 250                 BUG_ON(pool->skbuff[index] != NULL);
 251
 252                 dma_addr = dma_map_single(&adapter->vdev->dev, skb->data,
 253                                 pool->buff_size, DMA_FROM_DEVICE);
 254
 255                 if (dma_mapping_error(&adapter->vdev->dev, dma_addr))
 256                         goto failure;
 257
 258                 pool->free_map[free_index] = IBM_VETH_INVALID_MAP;
 259                 pool->dma_addr[index] = dma_addr;
 260                 pool->skbuff[index] = skb;
 261
 262                 correlator = ((u64)pool->index << 32) | index;
 263                 *(u64 *)skb->data = correlator;
 264
 265                 desc.fields.flags_len = IBMVETH_BUF_VALID | pool->buff_size;
 266                 desc.fields.address = dma_addr;
 267
 268                 if (rx_flush) {
 269                         unsigned int len = min(pool->buff_size,
 270                                                 adapter->netdev->mtu +
 271                                                 IBMVETH_BUFF_OH);
 272                         ibmveth_flush_buffer(skb->data, len);
 273                 }
 274                 lpar_rc = h_add_logical_lan_buffer(adapter->vdev->unit_address,
 275                                                    desc.desc);
 276
 277                 if (lpar_rc != H_SUCCESS) {
 278                         goto failure;
 279                 } else {
 280                         buffers_added++;
 281                         adapter->replenish_add_buff_success++;
 282                 }
 283         }
 284
 285         mb();
 286         atomic_add(buffers_added, &(pool->available));
 287         return;
 288
 289 failure:
 290         pool->free_map[free_index] = index;
 291         pool->skbuff[index] = NULL;
 292         if (pool->consumer_index == 0)
 293                 pool->consumer_index = pool->size - 1;
 294         else
 295                 pool->consumer_index--;
 296         if (!dma_mapping_error(&adapter->vdev->dev, dma_addr))
 297                 dma_unmap_single(&adapter->vdev->dev,
 298                                  pool->dma_addr[index], pool->buff_size,
 299                                  DMA_FROM_DEVICE);
 300         dev_kfree_skb_any(skb);
 301         adapter->replenish_add_buff_failure++;
 302
 303         mb();
 304         atomic_add(buffers_added, &(pool->available));
 305 }
 306
 307 /*
 308  * The final 8 bytes of the buffer list is a counter of frames dropped
 309  * because there was not a buffer in the buffer list capable of holding
 310  * the frame.
 311  */
 312 static void ibmveth_update_rx_no_buffer(struct ibmveth_adapter *adapter)
 313 {
 314         __be64 *p = adapter->buffer_list_addr + 4096 - 8;
 315
 316         adapter->rx_no_buffer = be64_to_cpup(p);
 317 }
 318
 319 /* replenish routine */
 320 static void ibmveth_replenish_task(struct ibmveth_adapter *adapter)
 321 {
 322         int i;
 323
 324         adapter->replenish_task_cycles++;
 325
 326         for (i = (IBMVETH_NUM_BUFF_POOLS - 1); i >= 0; i--) {
 327                 struct ibmveth_buff_pool *pool = &adapter->rx_buff_pool[i];
 328
 329                 if (pool->active &&
 330                     (atomic_read(&pool->available) < pool->threshold))
 331                         ibmveth_replenish_buffer_pool(adapter, pool);
 332         }
 333
 334         ibmveth_update_rx_no_buffer(adapter);
 335 }
 336
 337 /* empty and free ana buffer pool - also used to do cleanup in error paths */
 338 static void ibmveth_free_buffer_pool(struct ibmveth_adapter *adapter,
 339                                      struct ibmveth_buff_pool *pool)
 340 {
 341         int i;
 342
 343         kfree(pool->free_map);
 344         pool->free_map = NULL;
 345
 346         if (pool->skbuff && pool->dma_addr) {
 347                 for (i = 0; i < pool->size; ++i) {
 348                         struct sk_buff *skb = pool->skbuff[i];
 349                         if (skb) {
 350                                 dma_unmap_single(&adapter->vdev->dev,
 351                                                  pool->dma_addr[i],
 352                                                  pool->buff_size,
 353                                                  DMA_FROM_DEVICE);
 354                                 dev_kfree_skb_any(skb);
 355                                 pool->skbuff[i] = NULL;
 356                         }
 357                 }
 358         }
 359
 360         if (pool->dma_addr) {
 361                 kfree(pool->dma_addr);
 362                 pool->dma_addr = NULL;
 363         }
 364
 365         if (pool->skbuff) {
 366                 kfree(pool->skbuff);
 367                 pool->skbuff = NULL;
 368         }
 369 }
 370
 371 /* remove a buffer from a pool */
 372 static void ibmveth_remove_buffer_from_pool(struct ibmveth_adapter *adapter,
 373                                             u64 correlator)
 374 {
 375         unsigned int pool  = correlator >> 32;
 376         unsigned int index = correlator & 0xffffffffUL;
 377         unsigned int free_index;
 378         struct sk_buff *skb;
 379
 380         BUG_ON(pool >= IBMVETH_NUM_BUFF_POOLS);
 381         BUG_ON(index >= adapter->rx_buff_pool[pool].size);
 382
 383         skb = adapter->rx_buff_pool[pool].skbuff[index];
 384
 385         BUG_ON(skb == NULL);
 386
 387         adapter->rx_buff_pool[pool].skbuff[index] = NULL;
 388
 389         dma_unmap_single(&adapter->vdev->dev,
 390                          adapter->rx_buff_pool[pool].dma_addr[index],
 391                          adapter->rx_buff_pool[pool].buff_size,
 392                          DMA_FROM_DEVICE);
 393
 394         free_index = adapter->rx_buff_pool[pool].producer_index;
 395         adapter->rx_buff_pool[pool].producer_index++;
 396         if (adapter->rx_buff_pool[pool].producer_index >=
 397             adapter->rx_buff_pool[pool].size)
 398                 adapter->rx_buff_pool[pool].producer_index = 0;
 399         adapter->rx_buff_pool[pool].free_map[free_index] = index;
 400
 401         mb();
 402
 403         atomic_dec(&(adapter->rx_buff_pool[pool].available));
 404 }
 405
 406 /* get the current buffer on the rx queue */
 407 static inline struct sk_buff *ibmveth_rxq_get_buffer(struct ibmveth_adapter *adapter)
 408 {
 409         u64 correlator = adapter->rx_queue.queue_addr[adapter->rx_queue.index].correlator;
 410         unsigned int pool = correlator >> 32;
 411         unsigned int index = correlator & 0xffffffffUL;
 412
 413         BUG_ON(pool >= IBMVETH_NUM_BUFF_POOLS);
 414         BUG_ON(index >= adapter->rx_buff_pool[pool].size);
 415
 416         return adapter->rx_buff_pool[pool].skbuff[index];
 417 }
 418
 419 /* recycle the current buffer on the rx queue */
 420 static int ibmveth_rxq_recycle_buffer(struct ibmveth_adapter *adapter)
 421 {
 422         u32 q_index = adapter->rx_queue.index;
 423         u64 correlator = adapter->rx_queue.queue_addr[q_index].correlator;
 424         unsigned int pool = correlator >> 32;
 425         unsigned int index = correlator & 0xffffffffUL;
 426         union ibmveth_buf_desc desc;
 427         unsigned long lpar_rc;
 428         int ret = 1;
 429
 430         BUG_ON(pool >= IBMVETH_NUM_BUFF_POOLS);
 431         BUG_ON(index >= adapter->rx_buff_pool[pool].size);
 432
 433         if (!adapter->rx_buff_pool[pool].active) {
 434                 ibmveth_rxq_harvest_buffer(adapter);
 435                 ibmveth_free_buffer_pool(adapter, &adapter->rx_buff_pool[pool]);
 436                 goto out;
 437         }
 438
 439         desc.fields.flags_len = IBMVETH_BUF_VALID |
 440                 adapter->rx_buff_pool[pool].buff_size;
 441         desc.fields.address = adapter->rx_buff_pool[pool].dma_addr[index];
 442
 443         lpar_rc = h_add_logical_lan_buffer(adapter->vdev->unit_address, desc.desc);
 444
 445         if (lpar_rc != H_SUCCESS) {
 446                 netdev_dbg(adapter->netdev, "h_add_logical_lan_buffer failed "
 447                            "during recycle rc=%ld", lpar_rc);
 448                 ibmveth_remove_buffer_from_pool(adapter, adapter->rx_queue.queue_addr[adapter->rx_queue.index].correlator);
 449                 ret = 0;
 450         }
 451
 452         if (++adapter->rx_queue.index == adapter->rx_queue.num_slots) {
 453                 adapter->rx_queue.index = 0;
 454                 adapter->rx_queue.toggle = !adapter->rx_queue.toggle;
 455         }
 456
 457 out:
 458         return ret;
 459 }
 460
 461 static void ibmveth_rxq_harvest_buffer(struct ibmveth_adapter *adapter)
 462 {
 463         ibmveth_remove_buffer_from_pool(adapter, adapter->rx_queue.queue_addr[adapter->rx_queue.index].correlator);
 464
 465         if (++adapter->rx_queue.index == adapter->rx_queue.num_slots) {
 466                 adapter->rx_queue.index = 0;
 467                 adapter->rx_queue.toggle = !adapter->rx_queue.toggle;
 468         }
 469 }
 470
 471 static int ibmveth_register_logical_lan(struct ibmveth_adapter *adapter,
 472         union ibmveth_buf_desc rxq_desc, u64 mac_address)
 473 {
 474         int rc, try_again = 1;
 475
 476         /*
 477          * After a kexec the adapter will still be open, so our attempt to
 478          * open it will fail. So if we get a failure we free the adapter and
 479          * try again, but only once.
 480          */
 481 retry:
 482         rc = h_register_logical_lan(adapter->vdev->unit_address,
 483                                     adapter->buffer_list_dma, rxq_desc.desc,
 484                                     adapter->filter_list_dma, mac_address);
 485
 486         if (rc != H_SUCCESS && try_again) {
 487                 do {
 488                         rc = h_free_logical_lan(adapter->vdev->unit_address);
 489                 } while (H_IS_LONG_BUSY(rc) || (rc == H_BUSY));
 490
 491                 try_again = 0;
 492                 goto retry;
 493         }
 494
 495         return rc;
 496 }
 497
 498 static u64 ibmveth_encode_mac_addr(u8 *mac)
 499 {
 500         int i;
 501         u64 encoded = 0;
 502
 503         for (i = 0; i < ETH_ALEN; i++)
 504                 encoded = (encoded << 8) | mac[i];
 505
 506         return encoded;
 507 }
 508
 509 static int ibmveth_open(struct net_device *netdev)
 510 {
 511         struct ibmveth_adapter *adapter = netdev_priv(netdev);
 512         u64 mac_address;
 513         int rxq_entries = 1;
 514         unsigned long lpar_rc;
 515         int rc;
 516         union ibmveth_buf_desc rxq_desc;
 517         int i;
 518         struct device *dev;
 519
 520         netdev_dbg(netdev, "open starting\n");
 521
 522         napi_enable(&adapter->napi);
 523
 524         for(i = 0; i < IBMVETH_NUM_BUFF_POOLS; i++)
 525                 rxq_entries += adapter->rx_buff_pool[i].size;
 526
 527         rc = -ENOMEM;
 528         adapter->buffer_list_addr = (void*) get_zeroed_page(GFP_KERNEL);
 529         if (!adapter->buffer_list_addr) {
 530                 netdev_err(netdev, "unable to allocate list pages\n");
 531                 goto out;
 532         }
 533
 534         adapter->filter_list_addr = (void*) get_zeroed_page(GFP_KERNEL);
 535         if (!adapter->filter_list_addr) {
 536                 netdev_err(netdev, "unable to allocate filter pages\n");
 537                 goto out_free_buffer_list;
 538         }
 539
 540         dev = &adapter->vdev->dev;
 541
 542         adapter->rx_queue.queue_len = sizeof(struct ibmveth_rx_q_entry) *
 543                                                 rxq_entries;
 544         adapter->rx_queue.queue_addr =
 545                 dma_alloc_coherent(dev, adapter->rx_queue.queue_len,
 546                                    &adapter->rx_queue.queue_dma, GFP_KERNEL);
 547         if (!adapter->rx_queue.queue_addr)
 548                 goto out_free_filter_list;
 549
 550         adapter->buffer_list_dma = dma_map_single(dev,
 551                         adapter->buffer_list_addr, 4096, DMA_BIDIRECTIONAL);
 552         if (dma_mapping_error(dev, adapter->buffer_list_dma)) {
 553                 netdev_err(netdev, "unable to map buffer list pages\n");
 554                 goto out_free_queue_mem;
 555         }
 556
 557         adapter->filter_list_dma = dma_map_single(dev,
 558                         adapter->filter_list_addr, 4096, DMA_BIDIRECTIONAL);
 559         if (dma_mapping_error(dev, adapter->filter_list_dma)) {
 560                 netdev_err(netdev, "unable to map filter list pages\n");
 561                 goto out_unmap_buffer_list;
 562         }
 563
 564         adapter->rx_queue.index = 0;
 565         adapter->rx_queue.num_slots = rxq_entries;
 566         adapter->rx_queue.toggle = 1;
 567
 568         mac_address = ibmveth_encode_mac_addr(netdev->dev_addr);
 569
 570         rxq_desc.fields.flags_len = IBMVETH_BUF_VALID |
 571                                         adapter->rx_queue.queue_len;
 572         rxq_desc.fields.address = adapter->rx_queue.queue_dma;
 573
 574         netdev_dbg(netdev, "buffer list @ 0x%p\n", adapter->buffer_list_addr);
 575         netdev_dbg(netdev, "filter list @ 0x%p\n", adapter->filter_list_addr);
 576         netdev_dbg(netdev, "receive q   @ 0x%p\n", adapter->rx_queue.queue_addr);
 577
 578         h_vio_signal(adapter->vdev->unit_address, VIO_IRQ_DISABLE);
 579
 580         lpar_rc = ibmveth_register_logical_lan(adapter, rxq_desc, mac_address);
 581
 582         if (lpar_rc != H_SUCCESS) {
 583                 netdev_err(netdev, "h_register_logical_lan failed with %ld\n",
 584                            lpar_rc);
 585                 netdev_err(netdev, "buffer TCE:0x%llx filter TCE:0x%llx rxq "
 586                            "desc:0x%llx MAC:0x%llx\n",
 587                                      adapter->buffer_list_dma,
 588                                      adapter->filter_list_dma,
 589                                      rxq_desc.desc,
 590                                      mac_address);
 591                 rc = -ENONET;
 592                 goto out_unmap_filter_list;
 593         }
 594
 595         for (i = 0; i < IBMVETH_NUM_BUFF_POOLS; i++) {
 596                 if (!adapter->rx_buff_pool[i].active)
 597                         continue;
 598                 if (ibmveth_alloc_buffer_pool(&adapter->rx_buff_pool[i])) {
 599                         netdev_err(netdev, "unable to alloc pool\n");
 600                         adapter->rx_buff_pool[i].active = 0;
 601                         rc = -ENOMEM;
 602                         goto out_free_buffer_pools;
 603                 }
 604         }
 605
 606         netdev_dbg(netdev, "registering irq 0x%x\n", netdev->irq);
 607         rc = request_irq(netdev->irq, ibmveth_interrupt, 0, netdev->name,
 608                          netdev);
 609         if (rc != 0) {
 610                 netdev_err(netdev, "unable to request irq 0x%x, rc %d\n",
 611                            netdev->irq, rc);
 612                 do {
 613                         lpar_rc = h_free_logical_lan(adapter->vdev->unit_address);
 614                 } while (H_IS_LONG_BUSY(lpar_rc) || (lpar_rc == H_BUSY));
 615
 616                 goto out_free_buffer_pools;
 617         }
 618
 619         rc = -ENOMEM;
 620         adapter->bounce_buffer =
 621             kmalloc(netdev->mtu + IBMVETH_BUFF_OH, GFP_KERNEL);
 622         if (!adapter->bounce_buffer)
 623                 goto out_free_irq;
 624
 625         adapter->bounce_buffer_dma =
 626             dma_map_single(&adapter->vdev->dev, adapter->bounce_buffer,
 627                            netdev->mtu + IBMVETH_BUFF_OH, DMA_BIDIRECTIONAL);
 628         if (dma_mapping_error(dev, adapter->bounce_buffer_dma)) {
 629                 netdev_err(netdev, "unable to map bounce buffer\n");
 630                 goto out_free_bounce_buffer;
 631         }
 632
 633         netdev_dbg(netdev, "initial replenish cycle\n");
 634         ibmveth_interrupt(netdev->irq, netdev);
 635
 636         netif_start_queue(netdev);
 637
 638         netdev_dbg(netdev, "open complete\n");
 639
 640         return 0;
 641
 642 out_free_bounce_buffer:
 643         kfree(adapter->bounce_buffer);
 644 out_free_irq:
 645         free_irq(netdev->irq, netdev);
 646 out_free_buffer_pools:
 647         while (--i >= 0) {
 648                 if (adapter->rx_buff_pool[i].active)
 649                         ibmveth_free_buffer_pool(adapter,
 650                                                  &adapter->rx_buff_pool[i]);
 651         }
 652 out_unmap_filter_list:
 653         dma_unmap_single(dev, adapter->filter_list_dma, 4096,
 654                          DMA_BIDIRECTIONAL);
 655 out_unmap_buffer_list:
 656         dma_unmap_single(dev, adapter->buffer_list_dma, 4096,
 657                          DMA_BIDIRECTIONAL);
 658 out_free_queue_mem:
 659         dma_free_coherent(dev, adapter->rx_queue.queue_len,
 660                           adapter->rx_queue.queue_addr,
 661                           adapter->rx_queue.queue_dma);
 662 out_free_filter_list:
 663         free_page((unsigned long)adapter->filter_list_addr);
 664 out_free_buffer_list:
 665         free_page((unsigned long)adapter->buffer_list_addr);
 666 out:
 667         napi_disable(&adapter->napi);
 668         return rc;
 669 }
 670
 671 static int ibmveth_close(struct net_device *netdev)
 672 {
 673         struct ibmveth_adapter *adapter = netdev_priv(netdev);
 674         struct device *dev = &adapter->vdev->dev;
 675         long lpar_rc;
 676         int i;
 677
 678         netdev_dbg(netdev, "close starting\n");
 679
 680         napi_disable(&adapter->napi);
 681
 682         if (!adapter->pool_config)
 683                 netif_stop_queue(netdev);
 684
 685         h_vio_signal(adapter->vdev->unit_address, VIO_IRQ_DISABLE);
 686
 687         do {
 688                 lpar_rc = h_free_logical_lan(adapter->vdev->unit_address);
 689         } while (H_IS_LONG_BUSY(lpar_rc) || (lpar_rc == H_BUSY));
 690
 691         if (lpar_rc != H_SUCCESS) {
 692                 netdev_err(netdev, "h_free_logical_lan failed with %lx, "
 693                            "continuing with close\n", lpar_rc);
 694         }
 695
 696         free_irq(netdev->irq, netdev);
 697
 698         ibmveth_update_rx_no_buffer(adapter);
 699
 700         dma_unmap_single(dev, adapter->buffer_list_dma, 4096,
 701                          DMA_BIDIRECTIONAL);
 702         free_page((unsigned long)adapter->buffer_list_addr);
 703
 704         dma_unmap_single(dev, adapter->filter_list_dma, 4096,
 705                          DMA_BIDIRECTIONAL);
 706         free_page((unsigned long)adapter->filter_list_addr);
 707
 708         dma_free_coherent(dev, adapter->rx_queue.queue_len,
 709                           adapter->rx_queue.queue_addr,
 710                           adapter->rx_queue.queue_dma);
 711
 712         for (i = 0; i < IBMVETH_NUM_BUFF_POOLS; i++)
 713                 if (adapter->rx_buff_pool[i].active)
 714                         ibmveth_free_buffer_pool(adapter,
 715                                                  &adapter->rx_buff_pool[i]);
 716
 717         dma_unmap_single(&adapter->vdev->dev, adapter->bounce_buffer_dma,
 718                          adapter->netdev->mtu + IBMVETH_BUFF_OH,
 719                          DMA_BIDIRECTIONAL);
 720         kfree(adapter->bounce_buffer);
 721
 722         netdev_dbg(netdev, "close complete\n");
 723
 724         return 0;
 725 }
 726
 727 static int netdev_get_link_ksettings(struct net_device *dev,
 728                                      struct ethtool_link_ksettings *cmd)
 729 {
 730         u32 supported, advertising;
 731
 732         supported = (SUPPORTED_1000baseT_Full | SUPPORTED_Autoneg |
 733                                 SUPPORTED_FIBRE);
 734         advertising = (ADVERTISED_1000baseT_Full | ADVERTISED_Autoneg |
 735                                 ADVERTISED_FIBRE);
 736         cmd->base.speed = SPEED_1000;
 737         cmd->base.duplex = DUPLEX_FULL;
 738         cmd->base.port = PORT_FIBRE;
 739         cmd->base.phy_address = 0;
 740         cmd->base.autoneg = AUTONEG_ENABLE;
 741
 742         ethtool_convert_legacy_u32_to_link_mode(cmd->link_modes.supported,
 743                                                 supported);
 744         ethtool_convert_legacy_u32_to_link_mode(cmd->link_modes.advertising,
 745                                                 advertising);
 746
 747         return 0;
 748 }
 749
 750 static void netdev_get_drvinfo(struct net_device *dev,
 751                                struct ethtool_drvinfo *info)
 752 {
 753         strlcpy(info->driver, ibmveth_driver_name, sizeof(info->driver));
 754         strlcpy(info->version, ibmveth_driver_version, sizeof(info->version));
 755 }
 756
 757 static netdev_features_t ibmveth_fix_features(struct net_device *dev,
 758         netdev_features_t features)
 759 {
 760         /*
 761          * Since the ibmveth firmware interface does not have the
 762          * concept of separate tx/rx checksum offload enable, if rx
 763          * checksum is disabled we also have to disable tx checksum
 764          * offload. Once we disable rx checksum offload, we are no
 765          * longer allowed to send tx buffers that are not properly
 766          * checksummed.
 767          */
 768
 769         if (!(features & NETIF_F_RXCSUM))
 770                 features &= ~NETIF_F_CSUM_MASK;
 771
 772         return features;
 773 }
 774
 775 static int ibmveth_set_csum_offload(struct net_device *dev, u32 data)
 776 {
 777         struct ibmveth_adapter *adapter = netdev_priv(dev);
 778         unsigned long set_attr, clr_attr, ret_attr;
 779         unsigned long set_attr6, clr_attr6;
 780         long ret, ret4, ret6;
 781         int rc1 = 0, rc2 = 0;
 782         int restart = 0;
 783
 784         if (netif_running(dev)) {
 785                 restart = 1;
 786                 adapter->pool_config = 1;
 787                 ibmveth_close(dev);
 788                 adapter->pool_config = 0;
 789         }
 790
 791         set_attr = 0;
 792         clr_attr = 0;
 793         set_attr6 = 0;
 794         clr_attr6 = 0;
 795
 796         if (data) {
 797                 set_attr = IBMVETH_ILLAN_IPV4_TCP_CSUM;
 798                 set_attr6 = IBMVETH_ILLAN_IPV6_TCP_CSUM;
 799         } else {
 800                 clr_attr = IBMVETH_ILLAN_IPV4_TCP_CSUM;
 801                 clr_attr6 = IBMVETH_ILLAN_IPV6_TCP_CSUM;
 802         }
 803
 804         ret = h_illan_attributes(adapter->vdev->unit_address, 0, 0, &ret_attr);
 805
 806         if (ret == H_SUCCESS &&
 807             (ret_attr & IBMVETH_ILLAN_PADDED_PKT_CSUM)) {
 808                 ret4 = h_illan_attributes(adapter->vdev->unit_address, clr_attr,
 809                                          set_attr, &ret_attr);
 810
 811                 if (ret4 != H_SUCCESS) {
 812                         netdev_err(dev, "unable to change IPv4 checksum "
 813                                         "offload settings. %d rc=%ld\n",
 814                                         data, ret4);
 815
 816                         h_illan_attributes(adapter->vdev->unit_address,
 817                                            set_attr, clr_attr, &ret_attr);
 818
 819                         if (data == 1)
 820                                 dev->features &= ~NETIF_F_IP_CSUM;
 821
 822                 } else {
 823                         adapter->fw_ipv4_csum_support = data;
 824                 }
 825
 826                 ret6 = h_illan_attributes(adapter->vdev->unit_address,
 827                                          clr_attr6, set_attr6, &ret_attr);
 828
 829                 if (ret6 != H_SUCCESS) {
 830                         netdev_err(dev, "unable to change IPv6 checksum "
 831                                         "offload settings. %d rc=%ld\n",
 832                                         data, ret6);
 833
 834                         h_illan_attributes(adapter->vdev->unit_address,
 835                                            set_attr6, clr_attr6, &ret_attr);
 836
 837                         if (data == 1)
 838                                 dev->features &= ~NETIF_F_IPV6_CSUM;
 839
 840                 } else
 841                         adapter->fw_ipv6_csum_support = data;
 842
 843                 if (ret4 == H_SUCCESS || ret6 == H_SUCCESS)
 844                         adapter->rx_csum = data;
 845                 else
 846                         rc1 = -EIO;
 847         } else {
 848                 rc1 = -EIO;
 849                 netdev_err(dev, "unable to change checksum offload settings."
 850                                      " %d rc=%ld ret_attr=%lx\n", data, ret,
 851                                      ret_attr);
 852         }
 853
 854         if (restart)
 855                 rc2 = ibmveth_open(dev);
 856
 857         return rc1 ? rc1 : rc2;
 858 }
 859
 860 static int ibmveth_set_tso(struct net_device *dev, u32 data)
 861 {
 862         struct ibmveth_adapter *adapter = netdev_priv(dev);
 863         unsigned long set_attr, clr_attr, ret_attr;
 864         long ret1, ret2;
 865         int rc1 = 0, rc2 = 0;
 866         int restart = 0;
 867
 868         if (netif_running(dev)) {
 869                 restart = 1;
 870                 adapter->pool_config = 1;
 871                 ibmveth_close(dev);
 872                 adapter->pool_config = 0;
 873         }
 874
 875         set_attr = 0;
 876         clr_attr = 0;
 877
 878         if (data)
 879                 set_attr = IBMVETH_ILLAN_LRG_SR_ENABLED;
 880         else
 881                 clr_attr = IBMVETH_ILLAN_LRG_SR_ENABLED;
 882
 883         ret1 = h_illan_attributes(adapter->vdev->unit_address, 0, 0, &ret_attr);
 884
 885         if (ret1 == H_SUCCESS && (ret_attr & IBMVETH_ILLAN_LRG_SND_SUPPORT) &&
 886             !old_large_send) {
 887                 ret2 = h_illan_attributes(adapter->vdev->unit_address, clr_attr,
 888                                           set_attr, &ret_attr);
 889
 890                 if (ret2 != H_SUCCESS) {
 891                         netdev_err(dev, "unable to change tso settings. %d rc=%ld\n",
 892                                    data, ret2);
 893
 894                         h_illan_attributes(adapter->vdev->unit_address,
 895                                            set_attr, clr_attr, &ret_attr);
 896
 897                         if (data == 1)
 898                                 dev->features &= ~(NETIF_F_TSO | NETIF_F_TSO6);
 899                         rc1 = -EIO;
 900
 901                 } else {
 902                         adapter->fw_large_send_support = data;
 903                         adapter->large_send = data;
 904                 }
 905         } else {
 906                 /* Older firmware version of large send offload does not
 907                  * support tcp6/ipv6
 908                  */
 909                 if (data == 1) {
 910                         dev->features &= ~NETIF_F_TSO6;
 911                         netdev_info(dev, "TSO feature requires all partitions to have updated driver");
 912                 }
 913                 adapter->large_send = data;
 914         }
 915
 916         if (restart)
 917                 rc2 = ibmveth_open(dev);
 918
 919         return rc1 ? rc1 : rc2;
 920 }
 921
 922 static int ibmveth_set_features(struct net_device *dev,
 923         netdev_features_t features)
 924 {
 925         struct ibmveth_adapter *adapter = netdev_priv(dev);
 926         int rx_csum = !!(features & NETIF_F_RXCSUM);
 927         int large_send = !!(features & (NETIF_F_TSO | NETIF_F_TSO6));
 928         int rc1 = 0, rc2 = 0;
 929
 930         if (rx_csum != adapter->rx_csum) {
 931                 rc1 = ibmveth_set_csum_offload(dev, rx_csum);
 932                 if (rc1 && !adapter->rx_csum)
 933                         dev->features =
 934                                 features & ~(NETIF_F_CSUM_MASK |
 935                                              NETIF_F_RXCSUM);
 936         }
 937
 938         if (large_send != adapter->large_send) {
 939                 rc2 = ibmveth_set_tso(dev, large_send);
 940                 if (rc2 && !adapter->large_send)
 941                         dev->features =
 942                                 features & ~(NETIF_F_TSO | NETIF_F_TSO6);
 943         }
 944
 945         return rc1 ? rc1 : rc2;
 946 }
 947
 948 static void ibmveth_get_strings(struct net_device *dev, u32 stringset, u8 *data)
 949 {
 950         int i;
 951
 952         if (stringset != ETH_SS_STATS)
 953                 return;
 954
 955         for (i = 0; i < ARRAY_SIZE(ibmveth_stats); i++, data += ETH_GSTRING_LEN)
 956                 memcpy(data, ibmveth_stats[i].name, ETH_GSTRING_LEN);
 957 }
 958
 959 static int ibmveth_get_sset_count(struct net_device *dev, int sset)
 960 {
 961         switch (sset) {
 962         case ETH_SS_STATS:
 963                 return ARRAY_SIZE(ibmveth_stats);
 964         default:
 965                 return -EOPNOTSUPP;
 966         }
 967 }
 968
 969 static void ibmveth_get_ethtool_stats(struct net_device *dev,
 970                                       struct ethtool_stats *stats, u64 *data)
 971 {
 972         int i;
 973         struct ibmveth_adapter *adapter = netdev_priv(dev);
 974
 975         for (i = 0; i < ARRAY_SIZE(ibmveth_stats); i++)
 976                 data[i] = IBMVETH_GET_STAT(adapter, ibmveth_stats[i].offset);
 977 }
 978
 979 static const struct ethtool_ops netdev_ethtool_ops = {
 980         .get_drvinfo            = netdev_get_drvinfo,
 981         .get_link               = ethtool_op_get_link,
 982         .get_strings            = ibmveth_get_strings,
 983         .get_sset_count         = ibmveth_get_sset_count,
 984         .get_ethtool_stats      = ibmveth_get_ethtool_stats,
 985         .get_link_ksettings     = netdev_get_link_ksettings,
 986 };
 987
 988 static int ibmveth_ioctl(struct net_device *dev, struct ifreq *ifr, int cmd)
 989 {
 990         return -EOPNOTSUPP;
 991 }
 992
 993 #define page_offset(v) ((unsigned long)(v) & ((1 << 12) - 1))
 994
 995 static int ibmveth_send(struct ibmveth_adapter *adapter,
 996                         union ibmveth_buf_desc *descs, unsigned long mss)
 997 {
 998         unsigned long correlator;
 999         unsigned int retry_count;
1000         unsigned long ret;
1001
1002         /*
1003          * The retry count sets a maximum for the number of broadcast and
1004          * multicast destinations within the system.
1005          */
1006         retry_count = 1024;
1007         correlator = 0;
1008         do {
1009                 ret = h_send_logical_lan(adapter->vdev->unit_address,
1010                                              descs[0].desc, descs[1].desc,
1011                                              descs[2].desc, descs[3].desc,
1012                                              descs[4].desc, descs[5].desc,
1013                                              correlator, &correlator, mss,
1014                                              adapter->fw_large_send_support);
1015         } while ((ret == H_BUSY) && (retry_count--));
1016
1017         if (ret != H_SUCCESS && ret != H_DROPPED) {
1018                 netdev_err(adapter->netdev, "tx: h_send_logical_lan failed "
1019                            "with rc=%ld\n", ret);
1020                 return 1;
1021         }
1022
1023         return 0;
1024 }
1025
1026 static netdev_tx_t ibmveth_start_xmit(struct sk_buff *skb,
1027                                       struct net_device *netdev)
1028 {
1029         struct ibmveth_adapter *adapter = netdev_priv(netdev);
1030         unsigned int desc_flags;
1031         union ibmveth_buf_desc descs[6];
1032         int last, i;
1033         int force_bounce = 0;
1034         dma_addr_t dma_addr;
1035         unsigned long mss = 0;
1036
1037         /* veth doesn't handle frag_list, so linearize the skb.
1038          * When GRO is enabled SKB's can have frag_list.
1039          */
1040         if (adapter->is_active_trunk &&
1041             skb_has_frag_list(skb) && __skb_linearize(skb)) {
1042                 netdev->stats.tx_dropped++;
1043                 goto out;
1044         }
1045
1046         /*
1047          * veth handles a maximum of 6 segments including the header, so
1048          * we have to linearize the skb if there are more than this.
1049          */
1050         if (skb_shinfo(skb)->nr_frags > 5 && __skb_linearize(skb)) {
1051                 netdev->stats.tx_dropped++;
1052                 goto out;
1053         }
1054
1055         /* veth can't checksum offload UDP */
1056         if (skb->ip_summed == CHECKSUM_PARTIAL &&
1057             ((skb->protocol == htons(ETH_P_IP) &&
1058               ip_hdr(skb)->protocol != IPPROTO_TCP) ||
1059              (skb->protocol == htons(ETH_P_IPV6) &&
1060               ipv6_hdr(skb)->nexthdr != IPPROTO_TCP)) &&
1061             skb_checksum_help(skb)) {
1062
1063                 netdev_err(netdev, "tx: failed to checksum packet\n");
1064                 netdev->stats.tx_dropped++;
1065                 goto out;
1066         }
1067
1068         desc_flags = IBMVETH_BUF_VALID;
1069
1070         if (skb->ip_summed == CHECKSUM_PARTIAL) {
1071                 unsigned char *buf = skb_transport_header(skb) +
1072                                                 skb->csum_offset;
1073
1074                 desc_flags |= (IBMVETH_BUF_NO_CSUM | IBMVETH_BUF_CSUM_GOOD);
1075
1076                 /* Need to zero out the checksum */
1077                 buf[0] = 0;
1078                 buf[1] = 0;
1079
1080                 if (skb_is_gso(skb) && adapter->fw_large_send_support)
1081                         desc_flags |= IBMVETH_BUF_LRG_SND;
1082         }
1083
1084 retry_bounce:
1085         memset(descs, 0, sizeof(descs));
1086
1087         /*
1088          * If a linear packet is below the rx threshold then
1089          * copy it into the static bounce buffer. This avoids the
1090          * cost of a TCE insert and remove.
1091          */
1092         if (force_bounce || (!skb_is_nonlinear(skb) &&
1093                                 (skb->len < tx_copybreak))) {
1094                 skb_copy_from_linear_data(skb, adapter->bounce_buffer,
1095                                           skb->len);
1096
1097                 descs[0].fields.flags_len = desc_flags | skb->len;
1098                 descs[0].fields.address = adapter->bounce_buffer_dma;
1099
1100                 if (ibmveth_send(adapter, descs, 0)) {
1101                         adapter->tx_send_failed++;
1102                         netdev->stats.tx_dropped++;
1103                 } else {
1104                         netdev->stats.tx_packets++;
1105                         netdev->stats.tx_bytes += skb->len;
1106                 }
1107
1108                 goto out;
1109         }
1110
1111         /* Map the header */
1112         dma_addr = dma_map_single(&adapter->vdev->dev, skb->data,
1113                                   skb_headlen(skb), DMA_TO_DEVICE);
1114         if (dma_mapping_error(&adapter->vdev->dev, dma_addr))
1115                 goto map_failed;
1116
1117         descs[0].fields.flags_len = desc_flags | skb_headlen(skb);
1118         descs[0].fields.address = dma_addr;
1119
1120         /* Map the frags */
1121         for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
1122                 const skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
1123
1124                 dma_addr = skb_frag_dma_map(&adapter->vdev->dev, frag, 0,
1125                                             skb_frag_size(frag), DMA_TO_DEVICE);
1126
1127                 if (dma_mapping_error(&adapter->vdev->dev, dma_addr))
1128                         goto map_failed_frags;
1129
1130                 descs[i+1].fields.flags_len = desc_flags | skb_frag_size(frag);
1131                 descs[i+1].fields.address = dma_addr;
1132         }
1133
1134         if (skb->ip_summed == CHECKSUM_PARTIAL && skb_is_gso(skb)) {
1135                 if (adapter->fw_large_send_support) {
1136                         mss = (unsigned long)skb_shinfo(skb)->gso_size;
1137                         adapter->tx_large_packets++;
1138                 } else if (!skb_is_gso_v6(skb)) {
1139                         /* Put -1 in the IP checksum to tell phyp it
1140                          * is a largesend packet. Put the mss in
1141                          * the TCP checksum.
1142                          */
1143                         ip_hdr(skb)->check = 0xffff;
1144                         tcp_hdr(skb)->check =
1145                                 cpu_to_be16(skb_shinfo(skb)->gso_size);
1146                         adapter->tx_large_packets++;
1147                 }
1148         }
1149
1150         if (ibmveth_send(adapter, descs, mss)) {
1151                 adapter->tx_send_failed++;
1152                 netdev->stats.tx_dropped++;
1153         } else {
1154                 netdev->stats.tx_packets++;
1155                 netdev->stats.tx_bytes += skb->len;
1156         }
1157
1158         dma_unmap_single(&adapter->vdev->dev,
1159                          descs[0].fields.address,
1160                          descs[0].fields.flags_len & IBMVETH_BUF_LEN_MASK,
1161                          DMA_TO_DEVICE);
1162
1163         for (i = 1; i < skb_shinfo(skb)->nr_frags + 1; i++)
1164                 dma_unmap_page(&adapter->vdev->dev, descs[i].fields.address,
1165                                descs[i].fields.flags_len & IBMVETH_BUF_LEN_MASK,
1166                                DMA_TO_DEVICE);
1167
1168 out:
1169         dev_consume_skb_any(skb);
1170         return NETDEV_TX_OK;
1171
1172 map_failed_frags:
1173         last = i+1;
1174         for (i = 1; i < last; i++)
1175                 dma_unmap_page(&adapter->vdev->dev, descs[i].fields.address,
1176                                descs[i].fields.flags_len & IBMVETH_BUF_LEN_MASK,
1177                                DMA_TO_DEVICE);
1178
1179         dma_unmap_single(&adapter->vdev->dev,
1180                          descs[0].fields.address,
1181                          descs[0].fields.flags_len & IBMVETH_BUF_LEN_MASK,
1182                          DMA_TO_DEVICE);
1183 map_failed:
1184         if (!firmware_has_feature(FW_FEATURE_CMO))
1185                 netdev_err(netdev, "tx: unable to map xmit buffer\n");
1186         adapter->tx_map_failed++;
1187         if (skb_linearize(skb)) {
1188                 netdev->stats.tx_dropped++;
1189                 goto out;
1190         }
1191         force_bounce = 1;
1192         goto retry_bounce;
1193 }
1194
1195 static void ibmveth_rx_mss_helper(struct sk_buff *skb, u16 mss, int lrg_pkt)
1196 {
1197         struct tcphdr *tcph;
1198         int offset = 0;
1199         int hdr_len;
1200
1201         /* only TCP packets will be aggregated */
1202         if (skb->protocol == htons(ETH_P_IP)) {
1203                 struct iphdr *iph = (struct iphdr *)skb->data;
1204
1205                 if (iph->protocol == IPPROTO_TCP) {
1206                         offset = iph->ihl * 4;
1207                         skb_shinfo(skb)->gso_type = SKB_GSO_TCPV4;
1208                 } else {
1209                         return;
1210                 }
1211         } else if (skb->protocol == htons(ETH_P_IPV6)) {
1212                 struct ipv6hdr *iph6 = (struct ipv6hdr *)skb->data;
1213
1214                 if (iph6->nexthdr == IPPROTO_TCP) {
1215                         offset = sizeof(struct ipv6hdr);
1216                         skb_shinfo(skb)->gso_type = SKB_GSO_TCPV6;
1217                 } else {
1218                         return;
1219                 }
1220         } else {
1221                 return;
1222         }
1223         /* if mss is not set through Large Packet bit/mss in rx buffer,
1224          * expect that the mss will be written to the tcp header checksum.
1225          */
1226         tcph = (struct tcphdr *)(skb->data + offset);
1227         if (lrg_pkt) {
1228                 skb_shinfo(skb)->gso_size = mss;
1229         } else if (offset) {
1230                 skb_shinfo(skb)->gso_size = ntohs(tcph->check);
1231                 tcph->check = 0;
1232         }
1233
1234         if (skb_shinfo(skb)->gso_size) {
1235                 hdr_len = offset + tcph->doff * 4;
1236                 skb_shinfo(skb)->gso_segs =
1237                                 DIV_ROUND_UP(skb->len - hdr_len,
1238                                              skb_shinfo(skb)->gso_size);
1239         }
1240 }
1241
1242 static void ibmveth_rx_csum_helper(struct sk_buff *skb,
1243                                    struct ibmveth_adapter *adapter)
1244 {
1245         struct iphdr *iph = NULL;
1246         struct ipv6hdr *iph6 = NULL;
1247         __be16 skb_proto = 0;
1248         u16 iphlen = 0;
1249         u16 iph_proto = 0;
1250         u16 tcphdrlen = 0;
1251
1252         skb_proto = be16_to_cpu(skb->protocol);
1253
1254         if (skb_proto == ETH_P_IP) {
1255                 iph = (struct iphdr *)skb->data;
1256
1257                 /* If the IP checksum is not offloaded and if the packet
1258                  *  is large send, the checksum must be rebuilt.
1259                  */
1260                 if (iph->check == 0xffff) {
1261                         iph->check = 0;
1262                         iph->check = ip_fast_csum((unsigned char *)iph,
1263                                                   iph->ihl);
1264                 }
1265
1266                 iphlen = iph->ihl * 4;
1267                 iph_proto = iph->protocol;
1268         } else if (skb_proto == ETH_P_IPV6) {
1269                 iph6 = (struct ipv6hdr *)skb->data;
1270                 iphlen = sizeof(struct ipv6hdr);
1271                 iph_proto = iph6->nexthdr;
1272         }
1273
1274         /* In OVS environment, when a flow is not cached, specifically for a
1275          * new TCP connection, the first packet information is passed up
1276          * the user space for finding a flow. During this process, OVS computes
1277          * checksum on the first packet when CHECKSUM_PARTIAL flag is set.
1278          *
1279          * Given that we zeroed out TCP checksum field in transmit path
1280          * (refer ibmveth_start_xmit routine) as we set "no checksum bit",
1281          * OVS computed checksum will be incorrect w/o TCP pseudo checksum
1282          * in the packet. This leads to OVS dropping the packet and hence
1283          * TCP retransmissions are seen.
1284          *
1285          * So, re-compute TCP pseudo header checksum.
1286          */
1287         if (iph_proto == IPPROTO_TCP && adapter->is_active_trunk) {
1288                 struct tcphdr *tcph = (struct tcphdr *)(skb->data + iphlen);
1289
1290                 tcphdrlen = skb->len - iphlen;
1291
1292                 /* Recompute TCP pseudo header checksum */
1293                 if (skb_proto == ETH_P_IP)
1294                         tcph->check = ~csum_tcpudp_magic(iph->saddr,
1295                                         iph->daddr, tcphdrlen, iph_proto, 0);
1296                 else if (skb_proto == ETH_P_IPV6)
1297                         tcph->check = ~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 static int ibmveth_poll(struct napi_struct *napi, int budget)
1308 {
1309         struct ibmveth_adapter *adapter =
1310                         container_of(napi, struct ibmveth_adapter, napi);
1311         struct net_device *netdev = adapter->netdev;
1312         int frames_processed = 0;
1313         unsigned long lpar_rc;
1314         u16 mss = 0;
1315
1316 restart_poll:
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
1334                         skb = ibmveth_rxq_get_buffer(adapter);
1335
1336                         /* if the large packet bit is set in the rx queue
1337                          * descriptor, the mss will be written by PHYP eight
1338                          * bytes from the start of the rx buffer, which is
1339                          * skb->data at this stage
1340                          */
1341                         if (lrg_pkt) {
1342                                 __be64 *rxmss = (__be64 *)(skb->data + 8);
1343
1344                                 mss = (u16)be64_to_cpu(*rxmss);
1345                         }
1346
1347                         new_skb = NULL;
1348                         if (length < rx_copybreak)
1349                                 new_skb = netdev_alloc_skb(netdev, length);
1350
1351                         if (new_skb) {
1352                                 skb_copy_to_linear_data(new_skb,
1353                                                         skb->data + offset,
1354                                                         length);
1355                                 if (rx_flush)
1356                                         ibmveth_flush_buffer(skb->data,
1357                                                 length + offset);
1358                                 if (!ibmveth_rxq_recycle_buffer(adapter))
1359                                         kfree_skb(skb);
1360                                 skb = new_skb;
1361                         } else {
1362                                 ibmveth_rxq_harvest_buffer(adapter);
1363                                 skb_reserve(skb, offset);
1364                         }
1365
1366                         skb_put(skb, length);
1367                         skb->protocol = eth_type_trans(skb, netdev);
1368
1369                         if (csum_good) {
1370                                 skb->ip_summed = CHECKSUM_UNNECESSARY;
1371                                 ibmveth_rx_csum_helper(skb, adapter);
1372                         }
1373
1374                         if (length > netdev->mtu + ETH_HLEN) {
1375                                 ibmveth_rx_mss_helper(skb, mss, lrg_pkt);
1376                                 adapter->rx_large_packets++;
1377                         }
1378
1379                         napi_gro_receive(napi, skb);    /* send it up */
1380
1381                         netdev->stats.rx_packets++;
1382                         netdev->stats.rx_bytes += length;
1383                         frames_processed++;
1384                 }
1385         }
1386
1387         ibmveth_replenish_task(adapter);
1388
1389         if (frames_processed < budget) {
1390                 napi_complete_done(napi, frames_processed);
1391
1392                 /* We think we are done - reenable interrupts,
1393                  * then check once more to make sure we are done.
1394                  */
1395                 lpar_rc = h_vio_signal(adapter->vdev->unit_address,
1396                                        VIO_IRQ_ENABLE);
1397
1398                 BUG_ON(lpar_rc != H_SUCCESS);
1399
1400                 if (ibmveth_rxq_pending_buffer(adapter) &&
1401                     napi_reschedule(napi)) {
1402                         lpar_rc = h_vio_signal(adapter->vdev->unit_address,
1403                                                VIO_IRQ_DISABLE);
1404                         goto restart_poll;
1405                 }
1406         }
1407
1408         return frames_processed;
1409 }
1410
1411 static irqreturn_t ibmveth_interrupt(int irq, void *dev_instance)
1412 {
1413         struct net_device *netdev = dev_instance;
1414         struct ibmveth_adapter *adapter = netdev_priv(netdev);
1415         unsigned long lpar_rc;
1416
1417         if (napi_schedule_prep(&adapter->napi)) {
1418                 lpar_rc = h_vio_signal(adapter->vdev->unit_address,
1419                                        VIO_IRQ_DISABLE);
1420                 BUG_ON(lpar_rc != H_SUCCESS);
1421                 __napi_schedule(&adapter->napi);
1422         }
1423         return IRQ_HANDLED;
1424 }
1425
1426 static void ibmveth_set_multicast_list(struct net_device *netdev)
1427 {
1428         struct ibmveth_adapter *adapter = netdev_priv(netdev);
1429         unsigned long lpar_rc;
1430
1431         if ((netdev->flags & IFF_PROMISC) ||
1432             (netdev_mc_count(netdev) > adapter->mcastFilterSize)) {
1433                 lpar_rc = h_multicast_ctrl(adapter->vdev->unit_address,
1434                                            IbmVethMcastEnableRecv |
1435                                            IbmVethMcastDisableFiltering,
1436                                            0);
1437                 if (lpar_rc != H_SUCCESS) {
1438                         netdev_err(netdev, "h_multicast_ctrl rc=%ld when "
1439                                    "entering promisc mode\n", lpar_rc);
1440                 }
1441         } else {
1442                 struct netdev_hw_addr *ha;
1443                 /* clear the filter table & disable filtering */
1444                 lpar_rc = h_multicast_ctrl(adapter->vdev->unit_address,
1445                                            IbmVethMcastEnableRecv |
1446                                            IbmVethMcastDisableFiltering |
1447                                            IbmVethMcastClearFilterTable,
1448                                            0);
1449                 if (lpar_rc != H_SUCCESS) {
1450                         netdev_err(netdev, "h_multicast_ctrl rc=%ld when "
1451                                    "attempting to clear filter table\n",
1452                                    lpar_rc);
1453                 }
1454                 /* add the addresses to the filter table */
1455                 netdev_for_each_mc_addr(ha, netdev) {
1456                         /* add the multicast address to the filter table */
1457                         u64 mcast_addr;
1458                         mcast_addr = ibmveth_encode_mac_addr(ha->addr);
1459                         lpar_rc = h_multicast_ctrl(adapter->vdev->unit_address,
1460                                                    IbmVethMcastAddFilter,
1461                                                    mcast_addr);
1462                         if (lpar_rc != H_SUCCESS) {
1463                                 netdev_err(netdev, "h_multicast_ctrl rc=%ld "
1464                                            "when adding an entry to the filter "
1465                                            "table\n", lpar_rc);
1466                         }
1467                 }
1468
1469                 /* re-enable filtering */
1470                 lpar_rc = h_multicast_ctrl(adapter->vdev->unit_address,
1471                                            IbmVethMcastEnableFiltering,
1472                                            0);
1473                 if (lpar_rc != H_SUCCESS) {
1474                         netdev_err(netdev, "h_multicast_ctrl rc=%ld when "
1475                                    "enabling filtering\n", lpar_rc);
1476                 }
1477         }
1478 }
1479
1480 static int ibmveth_change_mtu(struct net_device *dev, int new_mtu)
1481 {
1482         struct ibmveth_adapter *adapter = netdev_priv(dev);
1483         struct vio_dev *viodev = adapter->vdev;
1484         int new_mtu_oh = new_mtu + IBMVETH_BUFF_OH;
1485         int i, rc;
1486         int need_restart = 0;
1487
1488         for (i = 0; i < IBMVETH_NUM_BUFF_POOLS; i++)
1489                 if (new_mtu_oh <= adapter->rx_buff_pool[i].buff_size)
1490                         break;
1491
1492         if (i == IBMVETH_NUM_BUFF_POOLS)
1493                 return -EINVAL;
1494
1495         /* Deactivate all the buffer pools so that the next loop can activate
1496            only the buffer pools necessary to hold the new MTU */
1497         if (netif_running(adapter->netdev)) {
1498                 need_restart = 1;
1499                 adapter->pool_config = 1;
1500                 ibmveth_close(adapter->netdev);
1501                 adapter->pool_config = 0;
1502         }
1503
1504         /* Look for an active buffer pool that can hold the new MTU */
1505         for (i = 0; i < IBMVETH_NUM_BUFF_POOLS; i++) {
1506                 adapter->rx_buff_pool[i].active = 1;
1507
1508                 if (new_mtu_oh <= adapter->rx_buff_pool[i].buff_size) {
1509                         dev->mtu = new_mtu;
1510                         vio_cmo_set_dev_desired(viodev,
1511                                                 ibmveth_get_desired_dma
1512                                                 (viodev));
1513                         if (need_restart) {
1514                                 return ibmveth_open(adapter->netdev);
1515                         }
1516                         return 0;
1517                 }
1518         }
1519
1520         if (need_restart && (rc = ibmveth_open(adapter->netdev)))
1521                 return rc;
1522
1523         return -EINVAL;
1524 }
1525
1526 #ifdef CONFIG_NET_POLL_CONTROLLER
1527 static void ibmveth_poll_controller(struct net_device *dev)
1528 {
1529         ibmveth_replenish_task(netdev_priv(dev));
1530         ibmveth_interrupt(dev->irq, dev);
1531 }
1532 #endif
1533
1534 /**
1535  * ibmveth_get_desired_dma - Calculate IO memory desired by the driver
1536  *
1537  * @vdev: struct vio_dev for the device whose desired IO mem is to be returned
1538  *
1539  * Return value:
1540  *      Number of bytes of IO data the driver will need to perform well.
1541  */
1542 static unsigned long ibmveth_get_desired_dma(struct vio_dev *vdev)
1543 {
1544         struct net_device *netdev = dev_get_drvdata(&vdev->dev);
1545         struct ibmveth_adapter *adapter;
1546         struct iommu_table *tbl;
1547         unsigned long ret;
1548         int i;
1549         int rxqentries = 1;
1550
1551         tbl = get_iommu_table_base(&vdev->dev);
1552
1553         /* netdev inits at probe time along with the structures we need below*/
1554         if (netdev == NULL)
1555                 return IOMMU_PAGE_ALIGN(IBMVETH_IO_ENTITLEMENT_DEFAULT, tbl);
1556
1557         adapter = netdev_priv(netdev);
1558
1559         ret = IBMVETH_BUFF_LIST_SIZE + IBMVETH_FILT_LIST_SIZE;
1560         ret += IOMMU_PAGE_ALIGN(netdev->mtu, tbl);
1561
1562         for (i = 0; i < IBMVETH_NUM_BUFF_POOLS; i++) {
1563                 /* add the size of the active receive buffers */
1564                 if (adapter->rx_buff_pool[i].active)
1565                         ret +=
1566                             adapter->rx_buff_pool[i].size *
1567                             IOMMU_PAGE_ALIGN(adapter->rx_buff_pool[i].
1568                                              buff_size, tbl);
1569                 rxqentries += adapter->rx_buff_pool[i].size;
1570         }
1571         /* add the size of the receive queue entries */
1572         ret += IOMMU_PAGE_ALIGN(
1573                 rxqentries * sizeof(struct ibmveth_rx_q_entry), tbl);
1574
1575         return ret;
1576 }
1577
1578 static int ibmveth_set_mac_addr(struct net_device *dev, void *p)
1579 {
1580         struct ibmveth_adapter *adapter = netdev_priv(dev);
1581         struct sockaddr *addr = p;
1582         u64 mac_address;
1583         int rc;
1584
1585         if (!is_valid_ether_addr(addr->sa_data))
1586                 return -EADDRNOTAVAIL;
1587
1588         mac_address = ibmveth_encode_mac_addr(addr->sa_data);
1589         rc = h_change_logical_lan_mac(adapter->vdev->unit_address, mac_address);
1590         if (rc) {
1591                 netdev_err(adapter->netdev, "h_change_logical_lan_mac failed with rc=%d\n", rc);
1592                 return rc;
1593         }
1594
1595         ether_addr_copy(dev->dev_addr, addr->sa_data);
1596
1597         return 0;
1598 }
1599
1600 static const struct net_device_ops ibmveth_netdev_ops = {
1601         .ndo_open               = ibmveth_open,
1602         .ndo_stop               = ibmveth_close,
1603         .ndo_start_xmit         = ibmveth_start_xmit,
1604         .ndo_set_rx_mode        = ibmveth_set_multicast_list,
1605         .ndo_do_ioctl           = ibmveth_ioctl,
1606         .ndo_change_mtu         = ibmveth_change_mtu,
1607         .ndo_fix_features       = ibmveth_fix_features,
1608         .ndo_set_features       = ibmveth_set_features,
1609         .ndo_validate_addr      = eth_validate_addr,
1610         .ndo_set_mac_address    = ibmveth_set_mac_addr,
1611 #ifdef CONFIG_NET_POLL_CONTROLLER
1612         .ndo_poll_controller    = ibmveth_poll_controller,
1613 #endif
1614 };
1615
1616 static int ibmveth_probe(struct vio_dev *dev, const struct vio_device_id *id)
1617 {
1618         int rc, i, mac_len;
1619         struct net_device *netdev;
1620         struct ibmveth_adapter *adapter;
1621         unsigned char *mac_addr_p;
1622         unsigned int *mcastFilterSize_p;
1623         long ret;
1624         unsigned long ret_attr;
1625
1626         dev_dbg(&dev->dev, "entering ibmveth_probe for UA 0x%x\n",
1627                 dev->unit_address);
1628
1629         mac_addr_p = (unsigned char *)vio_get_attribute(dev, VETH_MAC_ADDR,
1630                                                         &mac_len);
1631         if (!mac_addr_p) {
1632                 dev_err(&dev->dev, "Can't find VETH_MAC_ADDR attribute\n");
1633                 return -EINVAL;
1634         }
1635         /* Workaround for old/broken pHyp */
1636         if (mac_len == 8)
1637                 mac_addr_p += 2;
1638         else if (mac_len != 6) {
1639                 dev_err(&dev->dev, "VETH_MAC_ADDR attribute wrong len %d\n",
1640                         mac_len);
1641                 return -EINVAL;
1642         }
1643
1644         mcastFilterSize_p = (unsigned int *)vio_get_attribute(dev,
1645                                                 VETH_MCAST_FILTER_SIZE, NULL);
1646         if (!mcastFilterSize_p) {
1647                 dev_err(&dev->dev, "Can't find VETH_MCAST_FILTER_SIZE "
1648                         "attribute\n");
1649                 return -EINVAL;
1650         }
1651
1652         netdev = alloc_etherdev(sizeof(struct ibmveth_adapter));
1653
1654         if (!netdev)
1655                 return -ENOMEM;
1656
1657         adapter = netdev_priv(netdev);
1658         dev_set_drvdata(&dev->dev, netdev);
1659
1660         adapter->vdev = dev;
1661         adapter->netdev = netdev;
1662         adapter->mcastFilterSize = *mcastFilterSize_p;
1663         adapter->pool_config = 0;
1664
1665         netif_napi_add(netdev, &adapter->napi, ibmveth_poll, 16);
1666
1667         netdev->irq = dev->irq;
1668         netdev->netdev_ops = &ibmveth_netdev_ops;
1669         netdev->ethtool_ops = &netdev_ethtool_ops;
1670         SET_NETDEV_DEV(netdev, &dev->dev);
1671         netdev->hw_features = NETIF_F_SG;
1672         if (vio_get_attribute(dev, "ibm,illan-options", NULL) != NULL) {
1673                 netdev->hw_features |= NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM |
1674                                        NETIF_F_RXCSUM;
1675         }
1676
1677         netdev->features |= netdev->hw_features;
1678
1679         ret = h_illan_attributes(adapter->vdev->unit_address, 0, 0, &ret_attr);
1680
1681         /* If running older firmware, TSO should not be enabled by default */
1682         if (ret == H_SUCCESS && (ret_attr & IBMVETH_ILLAN_LRG_SND_SUPPORT) &&
1683             !old_large_send) {
1684                 netdev->hw_features |= NETIF_F_TSO | NETIF_F_TSO6;
1685                 netdev->features |= netdev->hw_features;
1686         } else {
1687                 netdev->hw_features |= NETIF_F_TSO;
1688         }
1689
1690         adapter->is_active_trunk = false;
1691         if (ret == H_SUCCESS && (ret_attr & IBMVETH_ILLAN_ACTIVE_TRUNK)) {
1692                 adapter->is_active_trunk = true;
1693                 netdev->hw_features |= NETIF_F_FRAGLIST;
1694                 netdev->features |= NETIF_F_FRAGLIST;
1695         }
1696
1697         netdev->min_mtu = IBMVETH_MIN_MTU;
1698         netdev->max_mtu = ETH_MAX_MTU;
1699
1700         memcpy(netdev->dev_addr, mac_addr_p, ETH_ALEN);
1701
1702         if (firmware_has_feature(FW_FEATURE_CMO))
1703                 memcpy(pool_count, pool_count_cmo, sizeof(pool_count));
1704
1705         for (i = 0; i < IBMVETH_NUM_BUFF_POOLS; i++) {
1706                 struct kobject *kobj = &adapter->rx_buff_pool[i].kobj;
1707                 int error;
1708
1709                 ibmveth_init_buffer_pool(&adapter->rx_buff_pool[i], i,
1710                                          pool_count[i], pool_size[i],
1711                                          pool_active[i]);
1712                 error = kobject_init_and_add(kobj, &ktype_veth_pool,
1713                                              &dev->dev.kobj, "pool%d", i);
1714                 if (!error)
1715                         kobject_uevent(kobj, KOBJ_ADD);
1716         }
1717
1718         netdev_dbg(netdev, "adapter @ 0x%p\n", adapter);
1719         netdev_dbg(netdev, "registering netdev...\n");
1720
1721         ibmveth_set_features(netdev, netdev->features);
1722
1723         rc = register_netdev(netdev);
1724
1725         if (rc) {
1726                 netdev_dbg(netdev, "failed to register netdev rc=%d\n", rc);
1727                 free_netdev(netdev);
1728                 return rc;
1729         }
1730
1731         netdev_dbg(netdev, "registered\n");
1732
1733         return 0;
1734 }
1735
1736 static int ibmveth_remove(struct vio_dev *dev)
1737 {
1738         struct net_device *netdev = dev_get_drvdata(&dev->dev);
1739         struct ibmveth_adapter *adapter = netdev_priv(netdev);
1740         int i;
1741
1742         for (i = 0; i < IBMVETH_NUM_BUFF_POOLS; i++)
1743                 kobject_put(&adapter->rx_buff_pool[i].kobj);
1744
1745         unregister_netdev(netdev);
1746
1747         free_netdev(netdev);
1748         dev_set_drvdata(&dev->dev, NULL);
1749
1750         return 0;
1751 }
1752
1753 static struct attribute veth_active_attr;
1754 static struct attribute veth_num_attr;
1755 static struct attribute veth_size_attr;
1756
1757 static ssize_t veth_pool_show(struct kobject *kobj,
1758                               struct attribute *attr, char *buf)
1759 {
1760         struct ibmveth_buff_pool *pool = container_of(kobj,
1761                                                       struct ibmveth_buff_pool,
1762                                                       kobj);
1763
1764         if (attr == &veth_active_attr)
1765                 return sprintf(buf, "%d\n", pool->active);
1766         else if (attr == &veth_num_attr)
1767                 return sprintf(buf, "%d\n", pool->size);
1768         else if (attr == &veth_size_attr)
1769                 return sprintf(buf, "%d\n", pool->buff_size);
1770         return 0;
1771 }
1772
1773 static ssize_t veth_pool_store(struct kobject *kobj, struct attribute *attr,
1774                                const char *buf, size_t count)
1775 {
1776         struct ibmveth_buff_pool *pool = container_of(kobj,
1777                                                       struct ibmveth_buff_pool,
1778                                                       kobj);
1779         struct net_device *netdev = dev_get_drvdata(
1780             container_of(kobj->parent, struct device, kobj));
1781         struct ibmveth_adapter *adapter = netdev_priv(netdev);
1782         long value = simple_strtol(buf, NULL, 10);
1783         long rc;
1784
1785         if (attr == &veth_active_attr) {
1786                 if (value && !pool->active) {
1787                         if (netif_running(netdev)) {
1788                                 if (ibmveth_alloc_buffer_pool(pool)) {
1789                                         netdev_err(netdev,
1790                                                    "unable to alloc pool\n");
1791                                         return -ENOMEM;
1792                                 }
1793                                 pool->active = 1;
1794                                 adapter->pool_config = 1;
1795                                 ibmveth_close(netdev);
1796                                 adapter->pool_config = 0;
1797                                 if ((rc = ibmveth_open(netdev)))
1798                                         return rc;
1799                         } else {
1800                                 pool->active = 1;
1801                         }
1802                 } else if (!value && pool->active) {
1803                         int mtu = netdev->mtu + IBMVETH_BUFF_OH;
1804                         int i;
1805                         /* Make sure there is a buffer pool with buffers that
1806                            can hold a packet of the size of the MTU */
1807                         for (i = 0; i < IBMVETH_NUM_BUFF_POOLS; i++) {
1808                                 if (pool == &adapter->rx_buff_pool[i])
1809                                         continue;
1810                                 if (!adapter->rx_buff_pool[i].active)
1811                                         continue;
1812                                 if (mtu <= adapter->rx_buff_pool[i].buff_size)
1813                                         break;
1814                         }
1815
1816                         if (i == IBMVETH_NUM_BUFF_POOLS) {
1817                                 netdev_err(netdev, "no active pool >= MTU\n");
1818                                 return -EPERM;
1819                         }
1820
1821                         if (netif_running(netdev)) {
1822                                 adapter->pool_config = 1;
1823                                 ibmveth_close(netdev);
1824                                 pool->active = 0;
1825                                 adapter->pool_config = 0;
1826                                 if ((rc = ibmveth_open(netdev)))
1827                                         return rc;
1828                         }
1829                         pool->active = 0;
1830                 }
1831         } else if (attr == &veth_num_attr) {
1832                 if (value <= 0 || value > IBMVETH_MAX_POOL_COUNT) {
1833                         return -EINVAL;
1834                 } else {
1835                         if (netif_running(netdev)) {
1836                                 adapter->pool_config = 1;
1837                                 ibmveth_close(netdev);
1838                                 adapter->pool_config = 0;
1839                                 pool->size = value;
1840                                 if ((rc = ibmveth_open(netdev)))
1841                                         return rc;
1842                         } else {
1843                                 pool->size = value;
1844                         }
1845                 }
1846         } else if (attr == &veth_size_attr) {
1847                 if (value <= IBMVETH_BUFF_OH || value > IBMVETH_MAX_BUF_SIZE) {
1848                         return -EINVAL;
1849                 } else {
1850                         if (netif_running(netdev)) {
1851                                 adapter->pool_config = 1;
1852                                 ibmveth_close(netdev);
1853                                 adapter->pool_config = 0;
1854                                 pool->buff_size = value;
1855                                 if ((rc = ibmveth_open(netdev)))
1856                                         return rc;
1857                         } else {
1858                                 pool->buff_size = value;
1859                         }
1860                 }
1861         }
1862
1863         /* kick the interrupt handler to allocate/deallocate pools */
1864         ibmveth_interrupt(netdev->irq, netdev);
1865         return count;
1866 }
1867
1868
1869 #define ATTR(_name, _mode)                              \
1870         struct attribute veth_##_name##_attr = {        \
1871         .name = __stringify(_name), .mode = _mode,      \
1872         };
1873
1874 static ATTR(active, 0644);
1875 static ATTR(num, 0644);
1876 static ATTR(size, 0644);
1877
1878 static struct attribute *veth_pool_attrs[] = {
1879         &veth_active_attr,
1880         &veth_num_attr,
1881         &veth_size_attr,
1882         NULL,
1883 };
1884
1885 static const struct sysfs_ops veth_pool_ops = {
1886         .show   = veth_pool_show,
1887         .store  = veth_pool_store,
1888 };
1889
1890 static struct kobj_type ktype_veth_pool = {
1891         .release        = NULL,
1892         .sysfs_ops      = &veth_pool_ops,
1893         .default_attrs  = veth_pool_attrs,
1894 };
1895
1896 static int ibmveth_resume(struct device *dev)
1897 {
1898         struct net_device *netdev = dev_get_drvdata(dev);
1899         ibmveth_interrupt(netdev->irq, netdev);
1900         return 0;
1901 }
1902
1903 static const struct vio_device_id ibmveth_device_table[] = {
1904         { "network", "IBM,l-lan"},
1905         { "", "" }
1906 };
1907 MODULE_DEVICE_TABLE(vio, ibmveth_device_table);
1908
1909 static const struct dev_pm_ops ibmveth_pm_ops = {
1910         .resume = ibmveth_resume
1911 };
1912
1913 static struct vio_driver ibmveth_driver = {
1914         .id_table       = ibmveth_device_table,
1915         .probe          = ibmveth_probe,
1916         .remove         = ibmveth_remove,
1917         .get_desired_dma = ibmveth_get_desired_dma,
1918         .name           = ibmveth_driver_name,
1919         .pm             = &ibmveth_pm_ops,
1920 };
1921
1922 static int __init ibmveth_module_init(void)
1923 {
1924         printk(KERN_DEBUG "%s: %s %s\n", ibmveth_driver_name,
1925                ibmveth_driver_string, ibmveth_driver_version);
1926
1927         return vio_register_driver(&ibmveth_driver);
1928 }
1929
1930 static void __exit ibmveth_module_exit(void)
1931 {
1932         vio_unregister_driver(&ibmveth_driver);
1933 }
1934
1935 module_init(ibmveth_module_init);
1936 module_exit(ibmveth_module_exit);