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 <larson1@us.ibm.com>
   8 *	    Santiago Leon <santil@linux.vnet.ibm.com>
   9 *	    Brian King <brking@linux.vnet.ibm.com>
  10 *	    Robert Jennings <rcj@linux.vnet.ibm.com>
  11 *	    Anton Blanchard <anton@au.ibm.com>
  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
  41static irqreturn_t ibmveth_interrupt(int irq, void *dev_instance);
  42static void ibmveth_rxq_harvest_buffer(struct ibmveth_adapter *adapter);
 
  43static unsigned long ibmveth_get_desired_dma(struct vio_dev *vdev);
  44
  45static struct kobj_type ktype_veth_pool;
  46
  47
  48static const char ibmveth_driver_name[] = "ibmveth";
  49static const char ibmveth_driver_string[] = "IBM Power Virtual Ethernet Driver";
  50#define ibmveth_driver_version "1.06"
  51
  52MODULE_AUTHOR("Santiago Leon <santil@linux.vnet.ibm.com>");
  53MODULE_DESCRIPTION("IBM Power Virtual Ethernet Driver");
  54MODULE_LICENSE("GPL");
  55MODULE_VERSION(ibmveth_driver_version);
  56
  57static unsigned int tx_copybreak __read_mostly = 128;
  58module_param(tx_copybreak, uint, 0644);
  59MODULE_PARM_DESC(tx_copybreak,
  60	"Maximum size of packet that is copied to a new buffer on transmit");
  61
  62static unsigned int rx_copybreak __read_mostly = 128;
  63module_param(rx_copybreak, uint, 0644);
  64MODULE_PARM_DESC(rx_copybreak,
  65	"Maximum size of packet that is copied to a new buffer on receive");
  66
  67static unsigned int rx_flush __read_mostly = 0;
  68module_param(rx_flush, uint, 0644);
  69MODULE_PARM_DESC(rx_flush, "Flush receive buffers before use");
  70
  71static bool old_large_send __read_mostly;
  72module_param(old_large_send, bool, 0444);
  73MODULE_PARM_DESC(old_large_send,
  74	"Use old large send method on firmware that supports the new method");
  75
  76struct 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
  84static 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 */
 103static 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
 108static 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
 114static inline int ibmveth_rxq_pending_buffer(struct ibmveth_adapter *adapter)
 115{
 116	return ibmveth_rxq_toggle(adapter) == adapter->rx_queue.toggle;
 117}
 118
 119static inline int ibmveth_rxq_buffer_valid(struct ibmveth_adapter *adapter)
 120{
 121	return ibmveth_rxq_flags(adapter) & IBMVETH_RXQ_VALID;
 122}
 123
 124static inline int ibmveth_rxq_frame_offset(struct ibmveth_adapter *adapter)
 125{
 126	return ibmveth_rxq_flags(adapter) & IBMVETH_RXQ_OFF_MASK;
 127}
 128
 129static inline int ibmveth_rxq_large_packet(struct ibmveth_adapter *adapter)
 130{
 131	return ibmveth_rxq_flags(adapter) & IBMVETH_RXQ_LRG_PKT;
 132}
 133
 134static 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
 139static inline int ibmveth_rxq_csum_good(struct ibmveth_adapter *adapter)
 140{
 141	return ibmveth_rxq_flags(adapter) & IBMVETH_RXQ_CSUM_GOOD;
 142}
 143
 144static unsigned int ibmveth_real_max_tx_queues(void)
 145{
 146	unsigned int n_cpu = num_online_cpus();
 147
 148	return min(n_cpu, IBMVETH_MAX_QUEUES);
 149}
 150
 151/* setup the initial settings for a buffer pool */
 152static void ibmveth_init_buffer_pool(struct ibmveth_buff_pool *pool,
 153				     u32 pool_index, u32 pool_size,
 154				     u32 buff_size, u32 pool_active)
 155{
 156	pool->size = pool_size;
 157	pool->index = pool_index;
 158	pool->buff_size = buff_size;
 159	pool->threshold = pool_size * 7 / 8;
 160	pool->active = pool_active;
 161}
 162
 163/* allocate and setup an buffer pool - called during open */
 164static int ibmveth_alloc_buffer_pool(struct ibmveth_buff_pool *pool)
 165{
 166	int i;
 167
 168	pool->free_map = kmalloc_array(pool->size, sizeof(u16), GFP_KERNEL);
 169
 170	if (!pool->free_map)
 171		return -1;
 172
 173	pool->dma_addr = kcalloc(pool->size, sizeof(dma_addr_t), GFP_KERNEL);
 174	if (!pool->dma_addr) {
 175		kfree(pool->free_map);
 176		pool->free_map = NULL;
 177		return -1;
 178	}
 179
 180	pool->skbuff = kcalloc(pool->size, sizeof(void *), GFP_KERNEL);
 181
 182	if (!pool->skbuff) {
 183		kfree(pool->dma_addr);
 184		pool->dma_addr = NULL;
 185
 186		kfree(pool->free_map);
 187		pool->free_map = NULL;
 188		return -1;
 189	}
 190
 191	for (i = 0; i < pool->size; ++i)
 192		pool->free_map[i] = i;
 193
 194	atomic_set(&pool->available, 0);
 195	pool->producer_index = 0;
 196	pool->consumer_index = 0;
 197
 198	return 0;
 199}
 200
 201static inline void ibmveth_flush_buffer(void *addr, unsigned long length)
 202{
 203	unsigned long offset;
 204
 205	for (offset = 0; offset < length; offset += SMP_CACHE_BYTES)
 206		asm("dcbf %0,%1,1" :: "b" (addr), "r" (offset));
 207}
 208
 209/* replenish the buffers for a pool.  note that we don't need to
 210 * skb_reserve these since they are used for incoming...
 211 */
 212static void ibmveth_replenish_buffer_pool(struct ibmveth_adapter *adapter,
 213					  struct ibmveth_buff_pool *pool)
 214{
 215	u32 i;
 216	u32 count = pool->size - atomic_read(&pool->available);
 217	u32 buffers_added = 0;
 218	struct sk_buff *skb;
 219	unsigned int free_index, index;
 220	u64 correlator;
 221	unsigned long lpar_rc;
 222	dma_addr_t dma_addr;
 223
 224	mb();
 225
 226	for (i = 0; i < count; ++i) {
 227		union ibmveth_buf_desc desc;
 228
 
 
 
 
 
 
 
 
 
 
 229		skb = netdev_alloc_skb(adapter->netdev, pool->buff_size);
 230
 231		if (!skb) {
 232			netdev_dbg(adapter->netdev,
 233				   "replenish: unable to allocate skb\n");
 234			adapter->replenish_no_mem++;
 235			break;
 236		}
 237
 238		free_index = pool->consumer_index;
 239		pool->consumer_index++;
 240		if (pool->consumer_index >= pool->size)
 241			pool->consumer_index = 0;
 242		index = pool->free_map[free_index];
 243
 244		BUG_ON(index == IBM_VETH_INVALID_MAP);
 245		BUG_ON(pool->skbuff[index] != NULL);
 246
 247		dma_addr = dma_map_single(&adapter->vdev->dev, skb->data,
 248				pool->buff_size, DMA_FROM_DEVICE);
 249
 250		if (dma_mapping_error(&adapter->vdev->dev, dma_addr))
 251			goto failure;
 252
 253		pool->free_map[free_index] = IBM_VETH_INVALID_MAP;
 254		pool->dma_addr[index] = dma_addr;
 255		pool->skbuff[index] = skb;
 256
 257		correlator = ((u64)pool->index << 32) | index;
 258		*(u64 *)skb->data = correlator;
 259
 260		desc.fields.flags_len = IBMVETH_BUF_VALID | pool->buff_size;
 261		desc.fields.address = dma_addr;
 262
 263		if (rx_flush) {
 264			unsigned int len = min(pool->buff_size,
 265						adapter->netdev->mtu +
 266						IBMVETH_BUFF_OH);
 267			ibmveth_flush_buffer(skb->data, len);
 268		}
 
 
 
 
 
 
 
 
 269		lpar_rc = h_add_logical_lan_buffer(adapter->vdev->unit_address,
 270						   desc.desc);
 271
 272		if (lpar_rc != H_SUCCESS) {
 
 
 
 273			goto failure;
 274		} else {
 275			buffers_added++;
 276			adapter->replenish_add_buff_success++;
 277		}
 
 
 
 
 
 
 
 
 278	}
 279
 280	mb();
 281	atomic_add(buffers_added, &(pool->available));
 282	return;
 283
 284failure:
 285	pool->free_map[free_index] = index;
 286	pool->skbuff[index] = NULL;
 287	if (pool->consumer_index == 0)
 288		pool->consumer_index = pool->size - 1;
 289	else
 290		pool->consumer_index--;
 291	if (!dma_mapping_error(&adapter->vdev->dev, dma_addr))
 292		dma_unmap_single(&adapter->vdev->dev,
 293		                 pool->dma_addr[index], pool->buff_size,
 294		                 DMA_FROM_DEVICE);
 295	dev_kfree_skb_any(skb);
 
 296	adapter->replenish_add_buff_failure++;
 297
 298	mb();
 299	atomic_add(buffers_added, &(pool->available));
 300}
 301
 302/*
 303 * The final 8 bytes of the buffer list is a counter of frames dropped
 304 * because there was not a buffer in the buffer list capable of holding
 305 * the frame.
 306 */
 307static void ibmveth_update_rx_no_buffer(struct ibmveth_adapter *adapter)
 308{
 309	__be64 *p = adapter->buffer_list_addr + 4096 - 8;
 310
 311	adapter->rx_no_buffer = be64_to_cpup(p);
 312}
 313
 314/* replenish routine */
 315static void ibmveth_replenish_task(struct ibmveth_adapter *adapter)
 316{
 317	int i;
 318
 319	adapter->replenish_task_cycles++;
 320
 321	for (i = (IBMVETH_NUM_BUFF_POOLS - 1); i >= 0; i--) {
 322		struct ibmveth_buff_pool *pool = &adapter->rx_buff_pool[i];
 323
 324		if (pool->active &&
 325		    (atomic_read(&pool->available) < pool->threshold))
 326			ibmveth_replenish_buffer_pool(adapter, pool);
 327	}
 328
 329	ibmveth_update_rx_no_buffer(adapter);
 330}
 331
 332/* empty and free ana buffer pool - also used to do cleanup in error paths */
 333static void ibmveth_free_buffer_pool(struct ibmveth_adapter *adapter,
 334				     struct ibmveth_buff_pool *pool)
 335{
 336	int i;
 337
 338	kfree(pool->free_map);
 339	pool->free_map = NULL;
 340
 341	if (pool->skbuff && pool->dma_addr) {
 342		for (i = 0; i < pool->size; ++i) {
 343			struct sk_buff *skb = pool->skbuff[i];
 344			if (skb) {
 345				dma_unmap_single(&adapter->vdev->dev,
 346						 pool->dma_addr[i],
 347						 pool->buff_size,
 348						 DMA_FROM_DEVICE);
 349				dev_kfree_skb_any(skb);
 350				pool->skbuff[i] = NULL;
 351			}
 352		}
 353	}
 354
 355	if (pool->dma_addr) {
 356		kfree(pool->dma_addr);
 357		pool->dma_addr = NULL;
 358	}
 359
 360	if (pool->skbuff) {
 361		kfree(pool->skbuff);
 362		pool->skbuff = NULL;
 363	}
 364}
 365
 366/* remove a buffer from a pool */
 367static void ibmveth_remove_buffer_from_pool(struct ibmveth_adapter *adapter,
 368					    u64 correlator)
 369{
 370	unsigned int pool  = correlator >> 32;
 371	unsigned int index = correlator & 0xffffffffUL;
 372	unsigned int free_index;
 373	struct sk_buff *skb;
 374
 375	BUG_ON(pool >= IBMVETH_NUM_BUFF_POOLS);
 376	BUG_ON(index >= adapter->rx_buff_pool[pool].size);
 377
 378	skb = adapter->rx_buff_pool[pool].skbuff[index];
 379
 380	BUG_ON(skb == NULL);
 381
 382	adapter->rx_buff_pool[pool].skbuff[index] = NULL;
 
 
 
 
 
 
 
 
 383
 384	dma_unmap_single(&adapter->vdev->dev,
 385			 adapter->rx_buff_pool[pool].dma_addr[index],
 386			 adapter->rx_buff_pool[pool].buff_size,
 387			 DMA_FROM_DEVICE);
 
 388
 389	free_index = adapter->rx_buff_pool[pool].producer_index;
 390	adapter->rx_buff_pool[pool].producer_index++;
 391	if (adapter->rx_buff_pool[pool].producer_index >=
 392	    adapter->rx_buff_pool[pool].size)
 393		adapter->rx_buff_pool[pool].producer_index = 0;
 394	adapter->rx_buff_pool[pool].free_map[free_index] = index;
 395
 396	mb();
 397
 398	atomic_dec(&(adapter->rx_buff_pool[pool].available));
 399}
 400
 401/* get the current buffer on the rx queue */
 402static inline struct sk_buff *ibmveth_rxq_get_buffer(struct ibmveth_adapter *adapter)
 403{
 404	u64 correlator = adapter->rx_queue.queue_addr[adapter->rx_queue.index].correlator;
 405	unsigned int pool = correlator >> 32;
 406	unsigned int index = correlator & 0xffffffffUL;
 407
 408	BUG_ON(pool >= IBMVETH_NUM_BUFF_POOLS);
 409	BUG_ON(index >= adapter->rx_buff_pool[pool].size);
 410
 411	return adapter->rx_buff_pool[pool].skbuff[index];
 412}
 413
 414/* recycle the current buffer on the rx queue */
 415static int ibmveth_rxq_recycle_buffer(struct ibmveth_adapter *adapter)
 416{
 417	u32 q_index = adapter->rx_queue.index;
 418	u64 correlator = adapter->rx_queue.queue_addr[q_index].correlator;
 419	unsigned int pool = correlator >> 32;
 420	unsigned int index = correlator & 0xffffffffUL;
 421	union ibmveth_buf_desc desc;
 422	unsigned long lpar_rc;
 423	int ret = 1;
 424
 425	BUG_ON(pool >= IBMVETH_NUM_BUFF_POOLS);
 426	BUG_ON(index >= adapter->rx_buff_pool[pool].size);
 427
 428	if (!adapter->rx_buff_pool[pool].active) {
 429		ibmveth_rxq_harvest_buffer(adapter);
 430		ibmveth_free_buffer_pool(adapter, &adapter->rx_buff_pool[pool]);
 431		goto out;
 432	}
 433
 434	desc.fields.flags_len = IBMVETH_BUF_VALID |
 435		adapter->rx_buff_pool[pool].buff_size;
 436	desc.fields.address = adapter->rx_buff_pool[pool].dma_addr[index];
 437
 438	lpar_rc = h_add_logical_lan_buffer(adapter->vdev->unit_address, desc.desc);
 439
 440	if (lpar_rc != H_SUCCESS) {
 441		netdev_dbg(adapter->netdev, "h_add_logical_lan_buffer failed "
 442			   "during recycle rc=%ld", lpar_rc);
 443		ibmveth_remove_buffer_from_pool(adapter, adapter->rx_queue.queue_addr[adapter->rx_queue.index].correlator);
 444		ret = 0;
 445	}
 446
 447	if (++adapter->rx_queue.index == adapter->rx_queue.num_slots) {
 448		adapter->rx_queue.index = 0;
 449		adapter->rx_queue.toggle = !adapter->rx_queue.toggle;
 450	}
 451
 452out:
 453	return ret;
 454}
 455
 456static void ibmveth_rxq_harvest_buffer(struct ibmveth_adapter *adapter)
 457{
 458	ibmveth_remove_buffer_from_pool(adapter, adapter->rx_queue.queue_addr[adapter->rx_queue.index].correlator);
 459
 460	if (++adapter->rx_queue.index == adapter->rx_queue.num_slots) {
 461		adapter->rx_queue.index = 0;
 462		adapter->rx_queue.toggle = !adapter->rx_queue.toggle;
 463	}
 464}
 465
 466static void ibmveth_free_tx_ltb(struct ibmveth_adapter *adapter, int idx)
 467{
 468	dma_unmap_single(&adapter->vdev->dev, adapter->tx_ltb_dma[idx],
 469			 adapter->tx_ltb_size, DMA_TO_DEVICE);
 470	kfree(adapter->tx_ltb_ptr[idx]);
 471	adapter->tx_ltb_ptr[idx] = NULL;
 472}
 473
 474static int ibmveth_allocate_tx_ltb(struct ibmveth_adapter *adapter, int idx)
 475{
 476	adapter->tx_ltb_ptr[idx] = kzalloc(adapter->tx_ltb_size,
 477					   GFP_KERNEL);
 478	if (!adapter->tx_ltb_ptr[idx]) {
 479		netdev_err(adapter->netdev,
 480			   "unable to allocate tx long term buffer\n");
 481		return -ENOMEM;
 482	}
 483	adapter->tx_ltb_dma[idx] = dma_map_single(&adapter->vdev->dev,
 484						  adapter->tx_ltb_ptr[idx],
 485						  adapter->tx_ltb_size,
 486						  DMA_TO_DEVICE);
 487	if (dma_mapping_error(&adapter->vdev->dev, adapter->tx_ltb_dma[idx])) {
 488		netdev_err(adapter->netdev,
 489			   "unable to DMA map tx long term buffer\n");
 490		kfree(adapter->tx_ltb_ptr[idx]);
 491		adapter->tx_ltb_ptr[idx] = NULL;
 492		return -ENOMEM;
 493	}
 494
 495	return 0;
 496}
 497
 498static int ibmveth_register_logical_lan(struct ibmveth_adapter *adapter,
 499        union ibmveth_buf_desc rxq_desc, u64 mac_address)
 500{
 501	int rc, try_again = 1;
 502
 503	/*
 504	 * After a kexec the adapter will still be open, so our attempt to
 505	 * open it will fail. So if we get a failure we free the adapter and
 506	 * try again, but only once.
 507	 */
 508retry:
 509	rc = h_register_logical_lan(adapter->vdev->unit_address,
 510				    adapter->buffer_list_dma, rxq_desc.desc,
 511				    adapter->filter_list_dma, mac_address);
 512
 513	if (rc != H_SUCCESS && try_again) {
 514		do {
 515			rc = h_free_logical_lan(adapter->vdev->unit_address);
 516		} while (H_IS_LONG_BUSY(rc) || (rc == H_BUSY));
 517
 518		try_again = 0;
 519		goto retry;
 520	}
 521
 522	return rc;
 523}
 524
 525static int ibmveth_open(struct net_device *netdev)
 526{
 527	struct ibmveth_adapter *adapter = netdev_priv(netdev);
 528	u64 mac_address;
 529	int rxq_entries = 1;
 530	unsigned long lpar_rc;
 531	int rc;
 532	union ibmveth_buf_desc rxq_desc;
 533	int i;
 534	struct device *dev;
 535
 536	netdev_dbg(netdev, "open starting\n");
 537
 538	napi_enable(&adapter->napi);
 539
 540	for(i = 0; i < IBMVETH_NUM_BUFF_POOLS; i++)
 541		rxq_entries += adapter->rx_buff_pool[i].size;
 542
 543	rc = -ENOMEM;
 544	adapter->buffer_list_addr = (void*) get_zeroed_page(GFP_KERNEL);
 545	if (!adapter->buffer_list_addr) {
 546		netdev_err(netdev, "unable to allocate list pages\n");
 547		goto out;
 548	}
 549
 550	adapter->filter_list_addr = (void*) get_zeroed_page(GFP_KERNEL);
 551	if (!adapter->filter_list_addr) {
 552		netdev_err(netdev, "unable to allocate filter pages\n");
 553		goto out_free_buffer_list;
 554	}
 555
 556	dev = &adapter->vdev->dev;
 557
 558	adapter->rx_queue.queue_len = sizeof(struct ibmveth_rx_q_entry) *
 559						rxq_entries;
 560	adapter->rx_queue.queue_addr =
 561		dma_alloc_coherent(dev, adapter->rx_queue.queue_len,
 562				   &adapter->rx_queue.queue_dma, GFP_KERNEL);
 563	if (!adapter->rx_queue.queue_addr)
 564		goto out_free_filter_list;
 565
 566	adapter->buffer_list_dma = dma_map_single(dev,
 567			adapter->buffer_list_addr, 4096, DMA_BIDIRECTIONAL);
 568	if (dma_mapping_error(dev, adapter->buffer_list_dma)) {
 569		netdev_err(netdev, "unable to map buffer list pages\n");
 570		goto out_free_queue_mem;
 571	}
 572
 573	adapter->filter_list_dma = dma_map_single(dev,
 574			adapter->filter_list_addr, 4096, DMA_BIDIRECTIONAL);
 575	if (dma_mapping_error(dev, adapter->filter_list_dma)) {
 576		netdev_err(netdev, "unable to map filter list pages\n");
 577		goto out_unmap_buffer_list;
 578	}
 579
 580	for (i = 0; i < netdev->real_num_tx_queues; i++) {
 581		if (ibmveth_allocate_tx_ltb(adapter, i))
 582			goto out_free_tx_ltb;
 583	}
 584
 585	adapter->rx_queue.index = 0;
 586	adapter->rx_queue.num_slots = rxq_entries;
 587	adapter->rx_queue.toggle = 1;
 588
 589	mac_address = ether_addr_to_u64(netdev->dev_addr);
 590
 591	rxq_desc.fields.flags_len = IBMVETH_BUF_VALID |
 592					adapter->rx_queue.queue_len;
 593	rxq_desc.fields.address = adapter->rx_queue.queue_dma;
 594
 595	netdev_dbg(netdev, "buffer list @ 0x%p\n", adapter->buffer_list_addr);
 596	netdev_dbg(netdev, "filter list @ 0x%p\n", adapter->filter_list_addr);
 597	netdev_dbg(netdev, "receive q   @ 0x%p\n", adapter->rx_queue.queue_addr);
 598
 599	h_vio_signal(adapter->vdev->unit_address, VIO_IRQ_DISABLE);
 600
 601	lpar_rc = ibmveth_register_logical_lan(adapter, rxq_desc, mac_address);
 602
 603	if (lpar_rc != H_SUCCESS) {
 604		netdev_err(netdev, "h_register_logical_lan failed with %ld\n",
 605			   lpar_rc);
 606		netdev_err(netdev, "buffer TCE:0x%llx filter TCE:0x%llx rxq "
 607			   "desc:0x%llx MAC:0x%llx\n",
 608				     adapter->buffer_list_dma,
 609				     adapter->filter_list_dma,
 610				     rxq_desc.desc,
 611				     mac_address);
 612		rc = -ENONET;
 613		goto out_unmap_filter_list;
 614	}
 615
 616	for (i = 0; i < IBMVETH_NUM_BUFF_POOLS; i++) {
 617		if (!adapter->rx_buff_pool[i].active)
 618			continue;
 619		if (ibmveth_alloc_buffer_pool(&adapter->rx_buff_pool[i])) {
 620			netdev_err(netdev, "unable to alloc pool\n");
 621			adapter->rx_buff_pool[i].active = 0;
 622			rc = -ENOMEM;
 623			goto out_free_buffer_pools;
 624		}
 625	}
 626
 627	netdev_dbg(netdev, "registering irq 0x%x\n", netdev->irq);
 628	rc = request_irq(netdev->irq, ibmveth_interrupt, 0, netdev->name,
 629			 netdev);
 630	if (rc != 0) {
 631		netdev_err(netdev, "unable to request irq 0x%x, rc %d\n",
 632			   netdev->irq, rc);
 633		do {
 634			lpar_rc = h_free_logical_lan(adapter->vdev->unit_address);
 635		} while (H_IS_LONG_BUSY(lpar_rc) || (lpar_rc == H_BUSY));
 636
 637		goto out_free_buffer_pools;
 638	}
 639
 640	rc = -ENOMEM;
 641
 642	netdev_dbg(netdev, "initial replenish cycle\n");
 643	ibmveth_interrupt(netdev->irq, netdev);
 644
 645	netif_tx_start_all_queues(netdev);
 646
 647	netdev_dbg(netdev, "open complete\n");
 648
 649	return 0;
 650
 651out_free_buffer_pools:
 652	while (--i >= 0) {
 653		if (adapter->rx_buff_pool[i].active)
 654			ibmveth_free_buffer_pool(adapter,
 655						 &adapter->rx_buff_pool[i]);
 656	}
 657out_unmap_filter_list:
 658	dma_unmap_single(dev, adapter->filter_list_dma, 4096,
 659			 DMA_BIDIRECTIONAL);
 660
 661out_free_tx_ltb:
 662	while (--i >= 0) {
 663		ibmveth_free_tx_ltb(adapter, i);
 664	}
 665
 666out_unmap_buffer_list:
 667	dma_unmap_single(dev, adapter->buffer_list_dma, 4096,
 668			 DMA_BIDIRECTIONAL);
 669out_free_queue_mem:
 670	dma_free_coherent(dev, adapter->rx_queue.queue_len,
 671			  adapter->rx_queue.queue_addr,
 672			  adapter->rx_queue.queue_dma);
 673out_free_filter_list:
 674	free_page((unsigned long)adapter->filter_list_addr);
 675out_free_buffer_list:
 676	free_page((unsigned long)adapter->buffer_list_addr);
 677out:
 678	napi_disable(&adapter->napi);
 679	return rc;
 680}
 681
 682static int ibmveth_close(struct net_device *netdev)
 683{
 684	struct ibmveth_adapter *adapter = netdev_priv(netdev);
 685	struct device *dev = &adapter->vdev->dev;
 686	long lpar_rc;
 687	int i;
 688
 689	netdev_dbg(netdev, "close starting\n");
 690
 691	napi_disable(&adapter->napi);
 692
 693	netif_tx_stop_all_queues(netdev);
 694
 695	h_vio_signal(adapter->vdev->unit_address, VIO_IRQ_DISABLE);
 696
 697	do {
 698		lpar_rc = h_free_logical_lan(adapter->vdev->unit_address);
 699	} while (H_IS_LONG_BUSY(lpar_rc) || (lpar_rc == H_BUSY));
 700
 701	if (lpar_rc != H_SUCCESS) {
 702		netdev_err(netdev, "h_free_logical_lan failed with %lx, "
 703			   "continuing with close\n", lpar_rc);
 704	}
 705
 706	free_irq(netdev->irq, netdev);
 707
 708	ibmveth_update_rx_no_buffer(adapter);
 709
 710	dma_unmap_single(dev, adapter->buffer_list_dma, 4096,
 711			 DMA_BIDIRECTIONAL);
 712	free_page((unsigned long)adapter->buffer_list_addr);
 713
 714	dma_unmap_single(dev, adapter->filter_list_dma, 4096,
 715			 DMA_BIDIRECTIONAL);
 716	free_page((unsigned long)adapter->filter_list_addr);
 717
 718	dma_free_coherent(dev, adapter->rx_queue.queue_len,
 719			  adapter->rx_queue.queue_addr,
 720			  adapter->rx_queue.queue_dma);
 721
 722	for (i = 0; i < IBMVETH_NUM_BUFF_POOLS; i++)
 723		if (adapter->rx_buff_pool[i].active)
 724			ibmveth_free_buffer_pool(adapter,
 725						 &adapter->rx_buff_pool[i]);
 726
 727	for (i = 0; i < netdev->real_num_tx_queues; i++)
 728		ibmveth_free_tx_ltb(adapter, i);
 729
 730	netdev_dbg(netdev, "close complete\n");
 731
 732	return 0;
 733}
 734
 735static int ibmveth_set_link_ksettings(struct net_device *dev,
 736				      const struct ethtool_link_ksettings *cmd)
 737{
 738	struct ibmveth_adapter *adapter = netdev_priv(dev);
 739
 740	return ethtool_virtdev_set_link_ksettings(dev, cmd,
 741						  &adapter->speed,
 742						  &adapter->duplex);
 743}
 744
 745static int ibmveth_get_link_ksettings(struct net_device *dev,
 746				      struct ethtool_link_ksettings *cmd)
 747{
 748	struct ibmveth_adapter *adapter = netdev_priv(dev);
 749
 750	cmd->base.speed = adapter->speed;
 751	cmd->base.duplex = adapter->duplex;
 752	cmd->base.port = PORT_OTHER;
 753
 754	return 0;
 755}
 756
 757static void ibmveth_init_link_settings(struct net_device *dev)
 758{
 759	struct ibmveth_adapter *adapter = netdev_priv(dev);
 760
 761	adapter->speed = SPEED_1000;
 762	adapter->duplex = DUPLEX_FULL;
 763}
 764
 765static void netdev_get_drvinfo(struct net_device *dev,
 766			       struct ethtool_drvinfo *info)
 767{
 768	strscpy(info->driver, ibmveth_driver_name, sizeof(info->driver));
 769	strscpy(info->version, ibmveth_driver_version, sizeof(info->version));
 770}
 771
 772static netdev_features_t ibmveth_fix_features(struct net_device *dev,
 773	netdev_features_t features)
 774{
 775	/*
 776	 * Since the ibmveth firmware interface does not have the
 777	 * concept of separate tx/rx checksum offload enable, if rx
 778	 * checksum is disabled we also have to disable tx checksum
 779	 * offload. Once we disable rx checksum offload, we are no
 780	 * longer allowed to send tx buffers that are not properly
 781	 * checksummed.
 782	 */
 783
 784	if (!(features & NETIF_F_RXCSUM))
 785		features &= ~NETIF_F_CSUM_MASK;
 786
 787	return features;
 788}
 789
 790static int ibmveth_set_csum_offload(struct net_device *dev, u32 data)
 791{
 792	struct ibmveth_adapter *adapter = netdev_priv(dev);
 793	unsigned long set_attr, clr_attr, ret_attr;
 794	unsigned long set_attr6, clr_attr6;
 795	long ret, ret4, ret6;
 796	int rc1 = 0, rc2 = 0;
 797	int restart = 0;
 798
 799	if (netif_running(dev)) {
 800		restart = 1;
 801		ibmveth_close(dev);
 802	}
 803
 804	set_attr = 0;
 805	clr_attr = 0;
 806	set_attr6 = 0;
 807	clr_attr6 = 0;
 808
 809	if (data) {
 810		set_attr = IBMVETH_ILLAN_IPV4_TCP_CSUM;
 811		set_attr6 = IBMVETH_ILLAN_IPV6_TCP_CSUM;
 812	} else {
 813		clr_attr = IBMVETH_ILLAN_IPV4_TCP_CSUM;
 814		clr_attr6 = IBMVETH_ILLAN_IPV6_TCP_CSUM;
 815	}
 816
 817	ret = h_illan_attributes(adapter->vdev->unit_address, 0, 0, &ret_attr);
 818
 819	if (ret == H_SUCCESS &&
 820	    (ret_attr & IBMVETH_ILLAN_PADDED_PKT_CSUM)) {
 821		ret4 = h_illan_attributes(adapter->vdev->unit_address, clr_attr,
 822					 set_attr, &ret_attr);
 823
 824		if (ret4 != H_SUCCESS) {
 825			netdev_err(dev, "unable to change IPv4 checksum "
 826					"offload settings. %d rc=%ld\n",
 827					data, ret4);
 828
 829			h_illan_attributes(adapter->vdev->unit_address,
 830					   set_attr, clr_attr, &ret_attr);
 831
 832			if (data == 1)
 833				dev->features &= ~NETIF_F_IP_CSUM;
 834
 835		} else {
 836			adapter->fw_ipv4_csum_support = data;
 837		}
 838
 839		ret6 = h_illan_attributes(adapter->vdev->unit_address,
 840					 clr_attr6, set_attr6, &ret_attr);
 841
 842		if (ret6 != H_SUCCESS) {
 843			netdev_err(dev, "unable to change IPv6 checksum "
 844					"offload settings. %d rc=%ld\n",
 845					data, ret6);
 846
 847			h_illan_attributes(adapter->vdev->unit_address,
 848					   set_attr6, clr_attr6, &ret_attr);
 849
 850			if (data == 1)
 851				dev->features &= ~NETIF_F_IPV6_CSUM;
 852
 853		} else
 854			adapter->fw_ipv6_csum_support = data;
 855
 856		if (ret4 == H_SUCCESS || ret6 == H_SUCCESS)
 857			adapter->rx_csum = data;
 858		else
 859			rc1 = -EIO;
 860	} else {
 861		rc1 = -EIO;
 862		netdev_err(dev, "unable to change checksum offload settings."
 863				     " %d rc=%ld ret_attr=%lx\n", data, ret,
 864				     ret_attr);
 865	}
 866
 867	if (restart)
 868		rc2 = ibmveth_open(dev);
 869
 870	return rc1 ? rc1 : rc2;
 871}
 872
 873static int ibmveth_set_tso(struct net_device *dev, u32 data)
 874{
 875	struct ibmveth_adapter *adapter = netdev_priv(dev);
 876	unsigned long set_attr, clr_attr, ret_attr;
 877	long ret1, ret2;
 878	int rc1 = 0, rc2 = 0;
 879	int restart = 0;
 880
 881	if (netif_running(dev)) {
 882		restart = 1;
 883		ibmveth_close(dev);
 884	}
 885
 886	set_attr = 0;
 887	clr_attr = 0;
 888
 889	if (data)
 890		set_attr = IBMVETH_ILLAN_LRG_SR_ENABLED;
 891	else
 892		clr_attr = IBMVETH_ILLAN_LRG_SR_ENABLED;
 893
 894	ret1 = h_illan_attributes(adapter->vdev->unit_address, 0, 0, &ret_attr);
 895
 896	if (ret1 == H_SUCCESS && (ret_attr & IBMVETH_ILLAN_LRG_SND_SUPPORT) &&
 897	    !old_large_send) {
 898		ret2 = h_illan_attributes(adapter->vdev->unit_address, clr_attr,
 899					  set_attr, &ret_attr);
 900
 901		if (ret2 != H_SUCCESS) {
 902			netdev_err(dev, "unable to change tso settings. %d rc=%ld\n",
 903				   data, ret2);
 904
 905			h_illan_attributes(adapter->vdev->unit_address,
 906					   set_attr, clr_attr, &ret_attr);
 907
 908			if (data == 1)
 909				dev->features &= ~(NETIF_F_TSO | NETIF_F_TSO6);
 910			rc1 = -EIO;
 911
 912		} else {
 913			adapter->fw_large_send_support = data;
 914			adapter->large_send = data;
 915		}
 916	} else {
 917		/* Older firmware version of large send offload does not
 918		 * support tcp6/ipv6
 919		 */
 920		if (data == 1) {
 921			dev->features &= ~NETIF_F_TSO6;
 922			netdev_info(dev, "TSO feature requires all partitions to have updated driver");
 923		}
 924		adapter->large_send = data;
 925	}
 926
 927	if (restart)
 928		rc2 = ibmveth_open(dev);
 929
 930	return rc1 ? rc1 : rc2;
 931}
 932
 933static int ibmveth_set_features(struct net_device *dev,
 934	netdev_features_t features)
 935{
 936	struct ibmveth_adapter *adapter = netdev_priv(dev);
 937	int rx_csum = !!(features & NETIF_F_RXCSUM);
 938	int large_send = !!(features & (NETIF_F_TSO | NETIF_F_TSO6));
 939	int rc1 = 0, rc2 = 0;
 940
 941	if (rx_csum != adapter->rx_csum) {
 942		rc1 = ibmveth_set_csum_offload(dev, rx_csum);
 943		if (rc1 && !adapter->rx_csum)
 944			dev->features =
 945				features & ~(NETIF_F_CSUM_MASK |
 946					     NETIF_F_RXCSUM);
 947	}
 948
 949	if (large_send != adapter->large_send) {
 950		rc2 = ibmveth_set_tso(dev, large_send);
 951		if (rc2 && !adapter->large_send)
 952			dev->features =
 953				features & ~(NETIF_F_TSO | NETIF_F_TSO6);
 954	}
 955
 956	return rc1 ? rc1 : rc2;
 957}
 958
 959static void ibmveth_get_strings(struct net_device *dev, u32 stringset, u8 *data)
 960{
 961	int i;
 962
 963	if (stringset != ETH_SS_STATS)
 964		return;
 965
 966	for (i = 0; i < ARRAY_SIZE(ibmveth_stats); i++, data += ETH_GSTRING_LEN)
 967		memcpy(data, ibmveth_stats[i].name, ETH_GSTRING_LEN);
 968}
 969
 970static int ibmveth_get_sset_count(struct net_device *dev, int sset)
 971{
 972	switch (sset) {
 973	case ETH_SS_STATS:
 974		return ARRAY_SIZE(ibmveth_stats);
 975	default:
 976		return -EOPNOTSUPP;
 977	}
 978}
 979
 980static void ibmveth_get_ethtool_stats(struct net_device *dev,
 981				      struct ethtool_stats *stats, u64 *data)
 982{
 983	int i;
 984	struct ibmveth_adapter *adapter = netdev_priv(dev);
 985
 986	for (i = 0; i < ARRAY_SIZE(ibmveth_stats); i++)
 987		data[i] = IBMVETH_GET_STAT(adapter, ibmveth_stats[i].offset);
 988}
 989
 990static void ibmveth_get_channels(struct net_device *netdev,
 991				 struct ethtool_channels *channels)
 992{
 993	channels->max_tx = ibmveth_real_max_tx_queues();
 994	channels->tx_count = netdev->real_num_tx_queues;
 995
 996	channels->max_rx = netdev->real_num_rx_queues;
 997	channels->rx_count = netdev->real_num_rx_queues;
 998}
 999
1000static int ibmveth_set_channels(struct net_device *netdev,
1001				struct ethtool_channels *channels)
1002{
1003	struct ibmveth_adapter *adapter = netdev_priv(netdev);
1004	unsigned int old = netdev->real_num_tx_queues,
1005		     goal = channels->tx_count;
1006	int rc, i;
1007
1008	/* If ndo_open has not been called yet then don't allocate, just set
1009	 * desired netdev_queue's and return
1010	 */
1011	if (!(netdev->flags & IFF_UP))
1012		return netif_set_real_num_tx_queues(netdev, goal);
1013
1014	/* We have IBMVETH_MAX_QUEUES netdev_queue's allocated
1015	 * but we may need to alloc/free the ltb's.
1016	 */
1017	netif_tx_stop_all_queues(netdev);
1018
1019	/* Allocate any queue that we need */
1020	for (i = old; i < goal; i++) {
1021		if (adapter->tx_ltb_ptr[i])
1022			continue;
1023
1024		rc = ibmveth_allocate_tx_ltb(adapter, i);
1025		if (!rc)
1026			continue;
1027
1028		/* if something goes wrong, free everything we just allocated */
1029		netdev_err(netdev, "Failed to allocate more tx queues, returning to %d queues\n",
1030			   old);
1031		goal = old;
1032		old = i;
1033		break;
1034	}
1035	rc = netif_set_real_num_tx_queues(netdev, goal);
1036	if (rc) {
1037		netdev_err(netdev, "Failed to set real tx queues, returning to %d queues\n",
1038			   old);
1039		goal = old;
1040		old = i;
1041	}
1042	/* Free any that are no longer needed */
1043	for (i = old; i > goal; i--) {
1044		if (adapter->tx_ltb_ptr[i - 1])
1045			ibmveth_free_tx_ltb(adapter, i - 1);
1046	}
1047
1048	netif_tx_wake_all_queues(netdev);
1049
1050	return rc;
1051}
1052
1053static const struct ethtool_ops netdev_ethtool_ops = {
1054	.get_drvinfo		         = netdev_get_drvinfo,
1055	.get_link		         = ethtool_op_get_link,
1056	.get_strings		         = ibmveth_get_strings,
1057	.get_sset_count		         = ibmveth_get_sset_count,
1058	.get_ethtool_stats	         = ibmveth_get_ethtool_stats,
1059	.get_link_ksettings	         = ibmveth_get_link_ksettings,
1060	.set_link_ksettings              = ibmveth_set_link_ksettings,
1061	.get_channels			 = ibmveth_get_channels,
1062	.set_channels			 = ibmveth_set_channels
1063};
1064
1065static int ibmveth_ioctl(struct net_device *dev, struct ifreq *ifr, int cmd)
1066{
1067	return -EOPNOTSUPP;
1068}
1069
1070static int ibmveth_send(struct ibmveth_adapter *adapter,
1071			unsigned long desc, unsigned long mss)
1072{
1073	unsigned long correlator;
1074	unsigned int retry_count;
1075	unsigned long ret;
1076
1077	/*
1078	 * The retry count sets a maximum for the number of broadcast and
1079	 * multicast destinations within the system.
1080	 */
1081	retry_count = 1024;
1082	correlator = 0;
1083	do {
1084		ret = h_send_logical_lan(adapter->vdev->unit_address, desc,
1085					 correlator, &correlator, mss,
1086					 adapter->fw_large_send_support);
1087	} while ((ret == H_BUSY) && (retry_count--));
1088
1089	if (ret != H_SUCCESS && ret != H_DROPPED) {
1090		netdev_err(adapter->netdev, "tx: h_send_logical_lan failed "
1091			   "with rc=%ld\n", ret);
1092		return 1;
1093	}
1094
1095	return 0;
1096}
1097
1098static int ibmveth_is_packet_unsupported(struct sk_buff *skb,
1099					 struct net_device *netdev)
1100{
1101	struct ethhdr *ether_header;
1102	int ret = 0;
1103
1104	ether_header = eth_hdr(skb);
1105
1106	if (ether_addr_equal(ether_header->h_dest, netdev->dev_addr)) {
1107		netdev_dbg(netdev, "veth doesn't support loopback packets, dropping packet.\n");
1108		netdev->stats.tx_dropped++;
1109		ret = -EOPNOTSUPP;
1110	}
1111
1112	return ret;
1113}
1114
1115static netdev_tx_t ibmveth_start_xmit(struct sk_buff *skb,
1116				      struct net_device *netdev)
1117{
1118	struct ibmveth_adapter *adapter = netdev_priv(netdev);
1119	unsigned int desc_flags, total_bytes;
1120	union ibmveth_buf_desc desc;
1121	int i, queue_num = skb_get_queue_mapping(skb);
1122	unsigned long mss = 0;
1123
1124	if (ibmveth_is_packet_unsupported(skb, netdev))
1125		goto out;
1126	/* veth can't checksum offload UDP */
1127	if (skb->ip_summed == CHECKSUM_PARTIAL &&
1128	    ((skb->protocol == htons(ETH_P_IP) &&
1129	      ip_hdr(skb)->protocol != IPPROTO_TCP) ||
1130	     (skb->protocol == htons(ETH_P_IPV6) &&
1131	      ipv6_hdr(skb)->nexthdr != IPPROTO_TCP)) &&
1132	    skb_checksum_help(skb)) {
1133
1134		netdev_err(netdev, "tx: failed to checksum packet\n");
1135		netdev->stats.tx_dropped++;
1136		goto out;
1137	}
1138
1139	desc_flags = IBMVETH_BUF_VALID;
1140
1141	if (skb->ip_summed == CHECKSUM_PARTIAL) {
1142		unsigned char *buf = skb_transport_header(skb) +
1143						skb->csum_offset;
1144
1145		desc_flags |= (IBMVETH_BUF_NO_CSUM | IBMVETH_BUF_CSUM_GOOD);
1146
1147		/* Need to zero out the checksum */
1148		buf[0] = 0;
1149		buf[1] = 0;
1150
1151		if (skb_is_gso(skb) && adapter->fw_large_send_support)
1152			desc_flags |= IBMVETH_BUF_LRG_SND;
1153	}
1154
1155	if (skb->ip_summed == CHECKSUM_PARTIAL && skb_is_gso(skb)) {
1156		if (adapter->fw_large_send_support) {
1157			mss = (unsigned long)skb_shinfo(skb)->gso_size;
1158			adapter->tx_large_packets++;
1159		} else if (!skb_is_gso_v6(skb)) {
1160			/* Put -1 in the IP checksum to tell phyp it
1161			 * is a largesend packet. Put the mss in
1162			 * the TCP checksum.
1163			 */
1164			ip_hdr(skb)->check = 0xffff;
1165			tcp_hdr(skb)->check =
1166				cpu_to_be16(skb_shinfo(skb)->gso_size);
1167			adapter->tx_large_packets++;
1168		}
1169	}
1170
1171	/* Copy header into mapped buffer */
1172	if (unlikely(skb->len > adapter->tx_ltb_size)) {
1173		netdev_err(adapter->netdev, "tx: packet size (%u) exceeds ltb (%u)\n",
1174			   skb->len, adapter->tx_ltb_size);
1175		netdev->stats.tx_dropped++;
1176		goto out;
1177	}
1178	memcpy(adapter->tx_ltb_ptr[queue_num], skb->data, skb_headlen(skb));
1179	total_bytes = skb_headlen(skb);
1180	/* Copy frags into mapped buffers */
1181	for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
1182		const skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
1183
1184		memcpy(adapter->tx_ltb_ptr[queue_num] + total_bytes,
1185		       skb_frag_address_safe(frag), skb_frag_size(frag));
1186		total_bytes += skb_frag_size(frag);
1187	}
1188
1189	if (unlikely(total_bytes != skb->len)) {
1190		netdev_err(adapter->netdev, "tx: incorrect packet len copied into ltb (%u != %u)\n",
1191			   skb->len, total_bytes);
1192		netdev->stats.tx_dropped++;
1193		goto out;
1194	}
1195	desc.fields.flags_len = desc_flags | skb->len;
1196	desc.fields.address = adapter->tx_ltb_dma[queue_num];
1197	/* finish writing to long_term_buff before VIOS accessing it */
1198	dma_wmb();
1199
1200	if (ibmveth_send(adapter, desc.desc, mss)) {
1201		adapter->tx_send_failed++;
1202		netdev->stats.tx_dropped++;
1203	} else {
1204		netdev->stats.tx_packets++;
1205		netdev->stats.tx_bytes += skb->len;
1206	}
1207
1208out:
1209	dev_consume_skb_any(skb);
1210	return NETDEV_TX_OK;
1211
1212
1213}
1214
1215static void ibmveth_rx_mss_helper(struct sk_buff *skb, u16 mss, int lrg_pkt)
1216{
1217	struct tcphdr *tcph;
1218	int offset = 0;
1219	int hdr_len;
1220
1221	/* only TCP packets will be aggregated */
1222	if (skb->protocol == htons(ETH_P_IP)) {
1223		struct iphdr *iph = (struct iphdr *)skb->data;
1224
1225		if (iph->protocol == IPPROTO_TCP) {
1226			offset = iph->ihl * 4;
1227			skb_shinfo(skb)->gso_type = SKB_GSO_TCPV4;
1228		} else {
1229			return;
1230		}
1231	} else if (skb->protocol == htons(ETH_P_IPV6)) {
1232		struct ipv6hdr *iph6 = (struct ipv6hdr *)skb->data;
1233
1234		if (iph6->nexthdr == IPPROTO_TCP) {
1235			offset = sizeof(struct ipv6hdr);
1236			skb_shinfo(skb)->gso_type = SKB_GSO_TCPV6;
1237		} else {
1238			return;
1239		}
1240	} else {
1241		return;
1242	}
1243	/* if mss is not set through Large Packet bit/mss in rx buffer,
1244	 * expect that the mss will be written to the tcp header checksum.
1245	 */
1246	tcph = (struct tcphdr *)(skb->data + offset);
1247	if (lrg_pkt) {
1248		skb_shinfo(skb)->gso_size = mss;
1249	} else if (offset) {
1250		skb_shinfo(skb)->gso_size = ntohs(tcph->check);
1251		tcph->check = 0;
1252	}
1253
1254	if (skb_shinfo(skb)->gso_size) {
1255		hdr_len = offset + tcph->doff * 4;
1256		skb_shinfo(skb)->gso_segs =
1257				DIV_ROUND_UP(skb->len - hdr_len,
1258					     skb_shinfo(skb)->gso_size);
1259	}
1260}
1261
1262static void ibmveth_rx_csum_helper(struct sk_buff *skb,
1263				   struct ibmveth_adapter *adapter)
1264{
1265	struct iphdr *iph = NULL;
1266	struct ipv6hdr *iph6 = NULL;
1267	__be16 skb_proto = 0;
1268	u16 iphlen = 0;
1269	u16 iph_proto = 0;
1270	u16 tcphdrlen = 0;
1271
1272	skb_proto = be16_to_cpu(skb->protocol);
1273
1274	if (skb_proto == ETH_P_IP) {
1275		iph = (struct iphdr *)skb->data;
1276
1277		/* If the IP checksum is not offloaded and if the packet
1278		 *  is large send, the checksum must be rebuilt.
1279		 */
1280		if (iph->check == 0xffff) {
1281			iph->check = 0;
1282			iph->check = ip_fast_csum((unsigned char *)iph,
1283						  iph->ihl);
1284		}
1285
1286		iphlen = iph->ihl * 4;
1287		iph_proto = iph->protocol;
1288	} else if (skb_proto == ETH_P_IPV6) {
1289		iph6 = (struct ipv6hdr *)skb->data;
1290		iphlen = sizeof(struct ipv6hdr);
1291		iph_proto = iph6->nexthdr;
1292	}
1293
1294	/* When CSO is enabled the TCP checksum may have be set to NULL by
1295	 * the sender given that we zeroed out TCP checksum field in
1296	 * transmit path (refer ibmveth_start_xmit routine). In this case set
1297	 * up CHECKSUM_PARTIAL. If the packet is forwarded, the checksum will
1298	 * then be recalculated by the destination NIC (CSO must be enabled
1299	 * on the destination NIC).
1300	 *
1301	 * In an OVS environment, when a flow is not cached, specifically for a
1302	 * new TCP connection, the first packet information is passed up to
1303	 * the user space for finding a flow. During this process, OVS computes
1304	 * checksum on the first packet when CHECKSUM_PARTIAL flag is set.
1305	 *
1306	 * So, re-compute TCP pseudo header checksum.
1307	 */
1308
1309	if (iph_proto == IPPROTO_TCP) {
1310		struct tcphdr *tcph = (struct tcphdr *)(skb->data + iphlen);
1311
1312		if (tcph->check == 0x0000) {
1313			/* Recompute TCP pseudo header checksum  */
1314			tcphdrlen = skb->len - iphlen;
1315			if (skb_proto == ETH_P_IP)
1316				tcph->check =
1317				 ~csum_tcpudp_magic(iph->saddr,
1318				iph->daddr, tcphdrlen, iph_proto, 0);
1319			else if (skb_proto == ETH_P_IPV6)
1320				tcph->check =
1321				 ~csum_ipv6_magic(&iph6->saddr,
1322				&iph6->daddr, tcphdrlen, iph_proto, 0);
1323			/* Setup SKB fields for checksum offload */
1324			skb_partial_csum_set(skb, iphlen,
1325					     offsetof(struct tcphdr, check));
1326			skb_reset_network_header(skb);
1327		}
1328	}
1329}
1330
1331static int ibmveth_poll(struct napi_struct *napi, int budget)
1332{
1333	struct ibmveth_adapter *adapter =
1334			container_of(napi, struct ibmveth_adapter, napi);
1335	struct net_device *netdev = adapter->netdev;
1336	int frames_processed = 0;
1337	unsigned long lpar_rc;
1338	u16 mss = 0;
1339
 
1340	while (frames_processed < budget) {
1341		if (!ibmveth_rxq_pending_buffer(adapter))
1342			break;
1343
1344		smp_rmb();
1345		if (!ibmveth_rxq_buffer_valid(adapter)) {
1346			wmb(); /* suggested by larson1 */
1347			adapter->rx_invalid_buffer++;
1348			netdev_dbg(netdev, "recycling invalid buffer\n");
1349			ibmveth_rxq_recycle_buffer(adapter);
1350		} else {
1351			struct sk_buff *skb, *new_skb;
1352			int length = ibmveth_rxq_frame_length(adapter);
1353			int offset = ibmveth_rxq_frame_offset(adapter);
1354			int csum_good = ibmveth_rxq_csum_good(adapter);
1355			int lrg_pkt = ibmveth_rxq_large_packet(adapter);
1356			__sum16 iph_check = 0;
1357
1358			skb = ibmveth_rxq_get_buffer(adapter);
1359
1360			/* if the large packet bit is set in the rx queue
1361			 * descriptor, the mss will be written by PHYP eight
1362			 * bytes from the start of the rx buffer, which is
1363			 * skb->data at this stage
1364			 */
1365			if (lrg_pkt) {
1366				__be64 *rxmss = (__be64 *)(skb->data + 8);
1367
1368				mss = (u16)be64_to_cpu(*rxmss);
1369			}
1370
1371			new_skb = NULL;
1372			if (length < rx_copybreak)
1373				new_skb = netdev_alloc_skb(netdev, length);
1374
1375			if (new_skb) {
1376				skb_copy_to_linear_data(new_skb,
1377							skb->data + offset,
1378							length);
1379				if (rx_flush)
1380					ibmveth_flush_buffer(skb->data,
1381						length + offset);
1382				if (!ibmveth_rxq_recycle_buffer(adapter))
1383					kfree_skb(skb);
1384				skb = new_skb;
1385			} else {
1386				ibmveth_rxq_harvest_buffer(adapter);
1387				skb_reserve(skb, offset);
1388			}
1389
1390			skb_put(skb, length);
1391			skb->protocol = eth_type_trans(skb, netdev);
1392
1393			/* PHYP without PLSO support places a -1 in the ip
1394			 * checksum for large send frames.
1395			 */
1396			if (skb->protocol == cpu_to_be16(ETH_P_IP)) {
1397				struct iphdr *iph = (struct iphdr *)skb->data;
1398
1399				iph_check = iph->check;
1400			}
1401
1402			if ((length > netdev->mtu + ETH_HLEN) ||
1403			    lrg_pkt || iph_check == 0xffff) {
1404				ibmveth_rx_mss_helper(skb, mss, lrg_pkt);
1405				adapter->rx_large_packets++;
1406			}
1407
1408			if (csum_good) {
1409				skb->ip_summed = CHECKSUM_UNNECESSARY;
1410				ibmveth_rx_csum_helper(skb, adapter);
1411			}
1412
1413			napi_gro_receive(napi, skb);	/* send it up */
1414
1415			netdev->stats.rx_packets++;
1416			netdev->stats.rx_bytes += length;
1417			frames_processed++;
1418		}
1419	}
1420
1421	ibmveth_replenish_task(adapter);
1422
1423	if (frames_processed < budget) {
1424		napi_complete_done(napi, frames_processed);
1425
1426		/* We think we are done - reenable interrupts,
1427		 * then check once more to make sure we are done.
1428		 */
1429		lpar_rc = h_vio_signal(adapter->vdev->unit_address,
1430				       VIO_IRQ_ENABLE);
1431
1432		BUG_ON(lpar_rc != H_SUCCESS);
 
 
 
 
1433
1434		if (ibmveth_rxq_pending_buffer(adapter) &&
1435		    napi_schedule(napi)) {
1436			lpar_rc = h_vio_signal(adapter->vdev->unit_address,
1437					       VIO_IRQ_DISABLE);
1438		}
1439	}
1440
 
1441	return frames_processed;
1442}
1443
1444static irqreturn_t ibmveth_interrupt(int irq, void *dev_instance)
1445{
1446	struct net_device *netdev = dev_instance;
1447	struct ibmveth_adapter *adapter = netdev_priv(netdev);
1448	unsigned long lpar_rc;
1449
1450	if (napi_schedule_prep(&adapter->napi)) {
1451		lpar_rc = h_vio_signal(adapter->vdev->unit_address,
1452				       VIO_IRQ_DISABLE);
1453		BUG_ON(lpar_rc != H_SUCCESS);
1454		__napi_schedule(&adapter->napi);
1455	}
1456	return IRQ_HANDLED;
1457}
1458
1459static void ibmveth_set_multicast_list(struct net_device *netdev)
1460{
1461	struct ibmveth_adapter *adapter = netdev_priv(netdev);
1462	unsigned long lpar_rc;
1463
1464	if ((netdev->flags & IFF_PROMISC) ||
1465	    (netdev_mc_count(netdev) > adapter->mcastFilterSize)) {
1466		lpar_rc = h_multicast_ctrl(adapter->vdev->unit_address,
1467					   IbmVethMcastEnableRecv |
1468					   IbmVethMcastDisableFiltering,
1469					   0);
1470		if (lpar_rc != H_SUCCESS) {
1471			netdev_err(netdev, "h_multicast_ctrl rc=%ld when "
1472				   "entering promisc mode\n", lpar_rc);
1473		}
1474	} else {
1475		struct netdev_hw_addr *ha;
1476		/* clear the filter table & disable filtering */
1477		lpar_rc = h_multicast_ctrl(adapter->vdev->unit_address,
1478					   IbmVethMcastEnableRecv |
1479					   IbmVethMcastDisableFiltering |
1480					   IbmVethMcastClearFilterTable,
1481					   0);
1482		if (lpar_rc != H_SUCCESS) {
1483			netdev_err(netdev, "h_multicast_ctrl rc=%ld when "
1484				   "attempting to clear filter table\n",
1485				   lpar_rc);
1486		}
1487		/* add the addresses to the filter table */
1488		netdev_for_each_mc_addr(ha, netdev) {
1489			/* add the multicast address to the filter table */
1490			u64 mcast_addr;
1491			mcast_addr = ether_addr_to_u64(ha->addr);
1492			lpar_rc = h_multicast_ctrl(adapter->vdev->unit_address,
1493						   IbmVethMcastAddFilter,
1494						   mcast_addr);
1495			if (lpar_rc != H_SUCCESS) {
1496				netdev_err(netdev, "h_multicast_ctrl rc=%ld "
1497					   "when adding an entry to the filter "
1498					   "table\n", lpar_rc);
1499			}
1500		}
1501
1502		/* re-enable filtering */
1503		lpar_rc = h_multicast_ctrl(adapter->vdev->unit_address,
1504					   IbmVethMcastEnableFiltering,
1505					   0);
1506		if (lpar_rc != H_SUCCESS) {
1507			netdev_err(netdev, "h_multicast_ctrl rc=%ld when "
1508				   "enabling filtering\n", lpar_rc);
1509		}
1510	}
1511}
1512
1513static int ibmveth_change_mtu(struct net_device *dev, int new_mtu)
1514{
1515	struct ibmveth_adapter *adapter = netdev_priv(dev);
1516	struct vio_dev *viodev = adapter->vdev;
1517	int new_mtu_oh = new_mtu + IBMVETH_BUFF_OH;
1518	int i, rc;
1519	int need_restart = 0;
1520
1521	for (i = 0; i < IBMVETH_NUM_BUFF_POOLS; i++)
1522		if (new_mtu_oh <= adapter->rx_buff_pool[i].buff_size)
1523			break;
1524
1525	if (i == IBMVETH_NUM_BUFF_POOLS)
1526		return -EINVAL;
1527
1528	/* Deactivate all the buffer pools so that the next loop can activate
1529	   only the buffer pools necessary to hold the new MTU */
1530	if (netif_running(adapter->netdev)) {
1531		need_restart = 1;
1532		ibmveth_close(adapter->netdev);
1533	}
1534
1535	/* Look for an active buffer pool that can hold the new MTU */
1536	for (i = 0; i < IBMVETH_NUM_BUFF_POOLS; i++) {
1537		adapter->rx_buff_pool[i].active = 1;
1538
1539		if (new_mtu_oh <= adapter->rx_buff_pool[i].buff_size) {
1540			dev->mtu = new_mtu;
1541			vio_cmo_set_dev_desired(viodev,
1542						ibmveth_get_desired_dma
1543						(viodev));
1544			if (need_restart) {
1545				return ibmveth_open(adapter->netdev);
1546			}
1547			return 0;
1548		}
1549	}
1550
1551	if (need_restart && (rc = ibmveth_open(adapter->netdev)))
1552		return rc;
1553
1554	return -EINVAL;
1555}
1556
1557#ifdef CONFIG_NET_POLL_CONTROLLER
1558static void ibmveth_poll_controller(struct net_device *dev)
1559{
1560	ibmveth_replenish_task(netdev_priv(dev));
1561	ibmveth_interrupt(dev->irq, dev);
1562}
1563#endif
1564
1565/**
1566 * ibmveth_get_desired_dma - Calculate IO memory desired by the driver
1567 *
1568 * @vdev: struct vio_dev for the device whose desired IO mem is to be returned
1569 *
1570 * Return value:
1571 *	Number of bytes of IO data the driver will need to perform well.
1572 */
1573static unsigned long ibmveth_get_desired_dma(struct vio_dev *vdev)
1574{
1575	struct net_device *netdev = dev_get_drvdata(&vdev->dev);
1576	struct ibmveth_adapter *adapter;
1577	struct iommu_table *tbl;
1578	unsigned long ret;
1579	int i;
1580	int rxqentries = 1;
1581
1582	tbl = get_iommu_table_base(&vdev->dev);
1583
1584	/* netdev inits at probe time along with the structures we need below*/
1585	if (netdev == NULL)
1586		return IOMMU_PAGE_ALIGN(IBMVETH_IO_ENTITLEMENT_DEFAULT, tbl);
1587
1588	adapter = netdev_priv(netdev);
1589
1590	ret = IBMVETH_BUFF_LIST_SIZE + IBMVETH_FILT_LIST_SIZE;
1591	ret += IOMMU_PAGE_ALIGN(netdev->mtu, tbl);
1592	/* add size of mapped tx buffers */
1593	ret += IOMMU_PAGE_ALIGN(IBMVETH_MAX_TX_BUF_SIZE, tbl);
1594
1595	for (i = 0; i < IBMVETH_NUM_BUFF_POOLS; i++) {
1596		/* add the size of the active receive buffers */
1597		if (adapter->rx_buff_pool[i].active)
1598			ret +=
1599			    adapter->rx_buff_pool[i].size *
1600			    IOMMU_PAGE_ALIGN(adapter->rx_buff_pool[i].
1601					     buff_size, tbl);
1602		rxqentries += adapter->rx_buff_pool[i].size;
1603	}
1604	/* add the size of the receive queue entries */
1605	ret += IOMMU_PAGE_ALIGN(
1606		rxqentries * sizeof(struct ibmveth_rx_q_entry), tbl);
1607
1608	return ret;
1609}
1610
1611static int ibmveth_set_mac_addr(struct net_device *dev, void *p)
1612{
1613	struct ibmveth_adapter *adapter = netdev_priv(dev);
1614	struct sockaddr *addr = p;
1615	u64 mac_address;
1616	int rc;
1617
1618	if (!is_valid_ether_addr(addr->sa_data))
1619		return -EADDRNOTAVAIL;
1620
1621	mac_address = ether_addr_to_u64(addr->sa_data);
1622	rc = h_change_logical_lan_mac(adapter->vdev->unit_address, mac_address);
1623	if (rc) {
1624		netdev_err(adapter->netdev, "h_change_logical_lan_mac failed with rc=%d\n", rc);
1625		return rc;
1626	}
1627
1628	eth_hw_addr_set(dev, addr->sa_data);
1629
1630	return 0;
1631}
1632
1633static const struct net_device_ops ibmveth_netdev_ops = {
1634	.ndo_open		= ibmveth_open,
1635	.ndo_stop		= ibmveth_close,
1636	.ndo_start_xmit		= ibmveth_start_xmit,
1637	.ndo_set_rx_mode	= ibmveth_set_multicast_list,
1638	.ndo_eth_ioctl		= ibmveth_ioctl,
1639	.ndo_change_mtu		= ibmveth_change_mtu,
1640	.ndo_fix_features	= ibmveth_fix_features,
1641	.ndo_set_features	= ibmveth_set_features,
1642	.ndo_validate_addr	= eth_validate_addr,
1643	.ndo_set_mac_address    = ibmveth_set_mac_addr,
1644#ifdef CONFIG_NET_POLL_CONTROLLER
1645	.ndo_poll_controller	= ibmveth_poll_controller,
1646#endif
1647};
1648
1649static int ibmveth_probe(struct vio_dev *dev, const struct vio_device_id *id)
1650{
1651	int rc, i, mac_len;
1652	struct net_device *netdev;
1653	struct ibmveth_adapter *adapter;
1654	unsigned char *mac_addr_p;
1655	__be32 *mcastFilterSize_p;
1656	long ret;
1657	unsigned long ret_attr;
1658
1659	dev_dbg(&dev->dev, "entering ibmveth_probe for UA 0x%x\n",
1660		dev->unit_address);
1661
1662	mac_addr_p = (unsigned char *)vio_get_attribute(dev, VETH_MAC_ADDR,
1663							&mac_len);
1664	if (!mac_addr_p) {
1665		dev_err(&dev->dev, "Can't find VETH_MAC_ADDR attribute\n");
1666		return -EINVAL;
1667	}
1668	/* Workaround for old/broken pHyp */
1669	if (mac_len == 8)
1670		mac_addr_p += 2;
1671	else if (mac_len != 6) {
1672		dev_err(&dev->dev, "VETH_MAC_ADDR attribute wrong len %d\n",
1673			mac_len);
1674		return -EINVAL;
1675	}
1676
1677	mcastFilterSize_p = (__be32 *)vio_get_attribute(dev,
1678							VETH_MCAST_FILTER_SIZE,
1679							NULL);
1680	if (!mcastFilterSize_p) {
1681		dev_err(&dev->dev, "Can't find VETH_MCAST_FILTER_SIZE "
1682			"attribute\n");
1683		return -EINVAL;
1684	}
1685
1686	netdev = alloc_etherdev_mqs(sizeof(struct ibmveth_adapter), IBMVETH_MAX_QUEUES, 1);
1687	if (!netdev)
1688		return -ENOMEM;
1689
1690	adapter = netdev_priv(netdev);
1691	dev_set_drvdata(&dev->dev, netdev);
1692
1693	adapter->vdev = dev;
1694	adapter->netdev = netdev;
1695	adapter->mcastFilterSize = be32_to_cpu(*mcastFilterSize_p);
1696	ibmveth_init_link_settings(netdev);
1697
1698	netif_napi_add_weight(netdev, &adapter->napi, ibmveth_poll, 16);
1699
1700	netdev->irq = dev->irq;
1701	netdev->netdev_ops = &ibmveth_netdev_ops;
1702	netdev->ethtool_ops = &netdev_ethtool_ops;
1703	SET_NETDEV_DEV(netdev, &dev->dev);
1704	netdev->hw_features = NETIF_F_SG;
1705	if (vio_get_attribute(dev, "ibm,illan-options", NULL) != NULL) {
1706		netdev->hw_features |= NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM |
1707				       NETIF_F_RXCSUM;
1708	}
1709
1710	netdev->features |= netdev->hw_features;
1711
1712	ret = h_illan_attributes(adapter->vdev->unit_address, 0, 0, &ret_attr);
1713
1714	/* If running older firmware, TSO should not be enabled by default */
1715	if (ret == H_SUCCESS && (ret_attr & IBMVETH_ILLAN_LRG_SND_SUPPORT) &&
1716	    !old_large_send) {
1717		netdev->hw_features |= NETIF_F_TSO | NETIF_F_TSO6;
1718		netdev->features |= netdev->hw_features;
1719	} else {
1720		netdev->hw_features |= NETIF_F_TSO;
1721	}
1722
1723	adapter->is_active_trunk = false;
1724	if (ret == H_SUCCESS && (ret_attr & IBMVETH_ILLAN_ACTIVE_TRUNK)) {
1725		adapter->is_active_trunk = true;
1726		netdev->hw_features |= NETIF_F_FRAGLIST;
1727		netdev->features |= NETIF_F_FRAGLIST;
1728	}
1729
1730	netdev->min_mtu = IBMVETH_MIN_MTU;
1731	netdev->max_mtu = ETH_MAX_MTU - IBMVETH_BUFF_OH;
1732
1733	eth_hw_addr_set(netdev, mac_addr_p);
1734
1735	if (firmware_has_feature(FW_FEATURE_CMO))
1736		memcpy(pool_count, pool_count_cmo, sizeof(pool_count));
1737
1738	for (i = 0; i < IBMVETH_NUM_BUFF_POOLS; i++) {
1739		struct kobject *kobj = &adapter->rx_buff_pool[i].kobj;
1740		int error;
1741
1742		ibmveth_init_buffer_pool(&adapter->rx_buff_pool[i], i,
1743					 pool_count[i], pool_size[i],
1744					 pool_active[i]);
1745		error = kobject_init_and_add(kobj, &ktype_veth_pool,
1746					     &dev->dev.kobj, "pool%d", i);
1747		if (!error)
1748			kobject_uevent(kobj, KOBJ_ADD);
1749	}
1750
1751	rc = netif_set_real_num_tx_queues(netdev, min(num_online_cpus(),
1752						      IBMVETH_DEFAULT_QUEUES));
1753	if (rc) {
1754		netdev_dbg(netdev, "failed to set number of tx queues rc=%d\n",
1755			   rc);
1756		free_netdev(netdev);
1757		return rc;
1758	}
1759	adapter->tx_ltb_size = PAGE_ALIGN(IBMVETH_MAX_TX_BUF_SIZE);
1760	for (i = 0; i < IBMVETH_MAX_QUEUES; i++)
1761		adapter->tx_ltb_ptr[i] = NULL;
1762
1763	netdev_dbg(netdev, "adapter @ 0x%p\n", adapter);
1764	netdev_dbg(netdev, "registering netdev...\n");
1765
1766	ibmveth_set_features(netdev, netdev->features);
1767
1768	rc = register_netdev(netdev);
1769
1770	if (rc) {
1771		netdev_dbg(netdev, "failed to register netdev rc=%d\n", rc);
1772		free_netdev(netdev);
1773		return rc;
1774	}
1775
1776	netdev_dbg(netdev, "registered\n");
1777
1778	return 0;
1779}
1780
1781static void ibmveth_remove(struct vio_dev *dev)
1782{
1783	struct net_device *netdev = dev_get_drvdata(&dev->dev);
1784	struct ibmveth_adapter *adapter = netdev_priv(netdev);
1785	int i;
1786
1787	for (i = 0; i < IBMVETH_NUM_BUFF_POOLS; i++)
1788		kobject_put(&adapter->rx_buff_pool[i].kobj);
1789
1790	unregister_netdev(netdev);
1791
1792	free_netdev(netdev);
1793	dev_set_drvdata(&dev->dev, NULL);
1794}
1795
1796static struct attribute veth_active_attr;
1797static struct attribute veth_num_attr;
1798static struct attribute veth_size_attr;
1799
1800static ssize_t veth_pool_show(struct kobject *kobj,
1801			      struct attribute *attr, char *buf)
1802{
1803	struct ibmveth_buff_pool *pool = container_of(kobj,
1804						      struct ibmveth_buff_pool,
1805						      kobj);
1806
1807	if (attr == &veth_active_attr)
1808		return sprintf(buf, "%d\n", pool->active);
1809	else if (attr == &veth_num_attr)
1810		return sprintf(buf, "%d\n", pool->size);
1811	else if (attr == &veth_size_attr)
1812		return sprintf(buf, "%d\n", pool->buff_size);
1813	return 0;
1814}
1815
1816static ssize_t veth_pool_store(struct kobject *kobj, struct attribute *attr,
1817			       const char *buf, size_t count)
1818{
1819	struct ibmveth_buff_pool *pool = container_of(kobj,
1820						      struct ibmveth_buff_pool,
1821						      kobj);
1822	struct net_device *netdev = dev_get_drvdata(kobj_to_dev(kobj->parent));
1823	struct ibmveth_adapter *adapter = netdev_priv(netdev);
1824	long value = simple_strtol(buf, NULL, 10);
1825	long rc;
1826
1827	if (attr == &veth_active_attr) {
1828		if (value && !pool->active) {
1829			if (netif_running(netdev)) {
1830				if (ibmveth_alloc_buffer_pool(pool)) {
1831					netdev_err(netdev,
1832						   "unable to alloc pool\n");
1833					return -ENOMEM;
1834				}
1835				pool->active = 1;
1836				ibmveth_close(netdev);
1837				if ((rc = ibmveth_open(netdev)))
1838					return rc;
1839			} else {
1840				pool->active = 1;
1841			}
1842		} else if (!value && pool->active) {
1843			int mtu = netdev->mtu + IBMVETH_BUFF_OH;
1844			int i;
1845			/* Make sure there is a buffer pool with buffers that
1846			   can hold a packet of the size of the MTU */
1847			for (i = 0; i < IBMVETH_NUM_BUFF_POOLS; i++) {
1848				if (pool == &adapter->rx_buff_pool[i])
1849					continue;
1850				if (!adapter->rx_buff_pool[i].active)
1851					continue;
1852				if (mtu <= adapter->rx_buff_pool[i].buff_size)
1853					break;
1854			}
1855
1856			if (i == IBMVETH_NUM_BUFF_POOLS) {
1857				netdev_err(netdev, "no active pool >= MTU\n");
1858				return -EPERM;
1859			}
1860
1861			if (netif_running(netdev)) {
1862				ibmveth_close(netdev);
1863				pool->active = 0;
1864				if ((rc = ibmveth_open(netdev)))
1865					return rc;
1866			}
1867			pool->active = 0;
1868		}
1869	} else if (attr == &veth_num_attr) {
1870		if (value <= 0 || value > IBMVETH_MAX_POOL_COUNT) {
1871			return -EINVAL;
1872		} else {
1873			if (netif_running(netdev)) {
1874				ibmveth_close(netdev);
1875				pool->size = value;
1876				if ((rc = ibmveth_open(netdev)))
1877					return rc;
1878			} else {
1879				pool->size = value;
1880			}
1881		}
1882	} else if (attr == &veth_size_attr) {
1883		if (value <= IBMVETH_BUFF_OH || value > IBMVETH_MAX_BUF_SIZE) {
1884			return -EINVAL;
1885		} else {
1886			if (netif_running(netdev)) {
1887				ibmveth_close(netdev);
1888				pool->buff_size = value;
1889				if ((rc = ibmveth_open(netdev)))
1890					return rc;
1891			} else {
1892				pool->buff_size = value;
1893			}
1894		}
1895	}
1896
1897	/* kick the interrupt handler to allocate/deallocate pools */
1898	ibmveth_interrupt(netdev->irq, netdev);
1899	return count;
1900}
1901
1902
1903#define ATTR(_name, _mode)				\
1904	struct attribute veth_##_name##_attr = {	\
1905	.name = __stringify(_name), .mode = _mode,	\
1906	};
1907
1908static ATTR(active, 0644);
1909static ATTR(num, 0644);
1910static ATTR(size, 0644);
1911
1912static struct attribute *veth_pool_attrs[] = {
1913	&veth_active_attr,
1914	&veth_num_attr,
1915	&veth_size_attr,
1916	NULL,
1917};
1918ATTRIBUTE_GROUPS(veth_pool);
1919
1920static const struct sysfs_ops veth_pool_ops = {
1921	.show   = veth_pool_show,
1922	.store  = veth_pool_store,
1923};
1924
1925static struct kobj_type ktype_veth_pool = {
1926	.release        = NULL,
1927	.sysfs_ops      = &veth_pool_ops,
1928	.default_groups = veth_pool_groups,
1929};
1930
1931static int ibmveth_resume(struct device *dev)
1932{
1933	struct net_device *netdev = dev_get_drvdata(dev);
1934	ibmveth_interrupt(netdev->irq, netdev);
1935	return 0;
1936}
1937
1938static const struct vio_device_id ibmveth_device_table[] = {
1939	{ "network", "IBM,l-lan"},
1940	{ "", "" }
1941};
1942MODULE_DEVICE_TABLE(vio, ibmveth_device_table);
1943
1944static const struct dev_pm_ops ibmveth_pm_ops = {
1945	.resume = ibmveth_resume
1946};
1947
1948static struct vio_driver ibmveth_driver = {
1949	.id_table	= ibmveth_device_table,
1950	.probe		= ibmveth_probe,
1951	.remove		= ibmveth_remove,
1952	.get_desired_dma = ibmveth_get_desired_dma,
1953	.name		= ibmveth_driver_name,
1954	.pm		= &ibmveth_pm_ops,
1955};
1956
1957static int __init ibmveth_module_init(void)
1958{
1959	printk(KERN_DEBUG "%s: %s %s\n", ibmveth_driver_name,
1960	       ibmveth_driver_string, ibmveth_driver_version);
1961
1962	return vio_register_driver(&ibmveth_driver);
1963}
1964
1965static void __exit ibmveth_module_exit(void)
1966{
1967	vio_unregister_driver(&ibmveth_driver);
1968}
1969
1970module_init(ibmveth_module_init);
1971module_exit(ibmveth_module_exit);