Linux Audio

Check our new training course

Linux kernel drivers training

May 6-19, 2025
Register
Loading...
Note: File does not exist in v3.1.
   1/*
   2 * IBM Power Virtual Ethernet Device Driver
   3 *
   4 * This program is free software; you can redistribute it and/or modify
   5 * it under the terms of the GNU General Public License as published by
   6 * the Free Software Foundation; either version 2 of the License, or
   7 * (at your option) any later version.
   8 *
   9 * This program is distributed in the hope that it will be useful,
  10 * but WITHOUT ANY WARRANTY; without even the implied warranty of
  11 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
  12 * GNU General Public License for more details.
  13 *
  14 * You should have received a copy of the GNU General Public License
  15 * along with this program; if not, see <http://www.gnu.org/licenses/>.
  16 *
  17 * Copyright (C) IBM Corporation, 2003, 2010
  18 *
  19 * Authors: Dave Larson <larson1@us.ibm.com>
  20 *	    Santiago Leon <santil@linux.vnet.ibm.com>
  21 *	    Brian King <brking@linux.vnet.ibm.com>
  22 *	    Robert Jennings <rcj@linux.vnet.ibm.com>
  23 *	    Anton Blanchard <anton@au.ibm.com>
  24 */
  25
  26#include <linux/module.h>
  27#include <linux/moduleparam.h>
  28#include <linux/types.h>
  29#include <linux/errno.h>
  30#include <linux/dma-mapping.h>
  31#include <linux/kernel.h>
  32#include <linux/netdevice.h>
  33#include <linux/etherdevice.h>
  34#include <linux/skbuff.h>
  35#include <linux/init.h>
  36#include <linux/interrupt.h>
  37#include <linux/mm.h>
  38#include <linux/pm.h>
  39#include <linux/ethtool.h>
  40#include <linux/in.h>
  41#include <linux/ip.h>
  42#include <linux/ipv6.h>
  43#include <linux/slab.h>
  44#include <asm/hvcall.h>
  45#include <linux/atomic.h>
  46#include <asm/vio.h>
  47#include <asm/iommu.h>
  48#include <asm/firmware.h>
  49
  50#include "ibmveth.h"
  51
  52static irqreturn_t ibmveth_interrupt(int irq, void *dev_instance);
  53static void ibmveth_rxq_harvest_buffer(struct ibmveth_adapter *adapter);
  54static unsigned long ibmveth_get_desired_dma(struct vio_dev *vdev);
  55
  56static struct kobj_type ktype_veth_pool;
  57
  58
  59static const char ibmveth_driver_name[] = "ibmveth";
  60static const char ibmveth_driver_string[] = "IBM Power Virtual Ethernet Driver";
  61#define ibmveth_driver_version "1.04"
  62
  63MODULE_AUTHOR("Santiago Leon <santil@linux.vnet.ibm.com>");
  64MODULE_DESCRIPTION("IBM Power Virtual Ethernet Driver");
  65MODULE_LICENSE("GPL");
  66MODULE_VERSION(ibmveth_driver_version);
  67
  68static unsigned int tx_copybreak __read_mostly = 128;
  69module_param(tx_copybreak, uint, 0644);
  70MODULE_PARM_DESC(tx_copybreak,
  71	"Maximum size of packet that is copied to a new buffer on transmit");
  72
  73static unsigned int rx_copybreak __read_mostly = 128;
  74module_param(rx_copybreak, uint, 0644);
  75MODULE_PARM_DESC(rx_copybreak,
  76	"Maximum size of packet that is copied to a new buffer on receive");
  77
  78static unsigned int rx_flush __read_mostly = 0;
  79module_param(rx_flush, uint, 0644);
  80MODULE_PARM_DESC(rx_flush, "Flush receive buffers before use");
  81
  82struct ibmveth_stat {
  83	char name[ETH_GSTRING_LEN];
  84	int offset;
  85};
  86
  87#define IBMVETH_STAT_OFF(stat) offsetof(struct ibmveth_adapter, stat)
  88#define IBMVETH_GET_STAT(a, off) *((u64 *)(((unsigned long)(a)) + off))
  89
  90struct ibmveth_stat ibmveth_stats[] = {
  91	{ "replenish_task_cycles", IBMVETH_STAT_OFF(replenish_task_cycles) },
  92	{ "replenish_no_mem", IBMVETH_STAT_OFF(replenish_no_mem) },
  93	{ "replenish_add_buff_failure",
  94			IBMVETH_STAT_OFF(replenish_add_buff_failure) },
  95	{ "replenish_add_buff_success",
  96			IBMVETH_STAT_OFF(replenish_add_buff_success) },
  97	{ "rx_invalid_buffer", IBMVETH_STAT_OFF(rx_invalid_buffer) },
  98	{ "rx_no_buffer", IBMVETH_STAT_OFF(rx_no_buffer) },
  99	{ "tx_map_failed", IBMVETH_STAT_OFF(tx_map_failed) },
 100	{ "tx_send_failed", IBMVETH_STAT_OFF(tx_send_failed) },
 101	{ "fw_enabled_ipv4_csum", IBMVETH_STAT_OFF(fw_ipv4_csum_support) },
 102	{ "fw_enabled_ipv6_csum", IBMVETH_STAT_OFF(fw_ipv6_csum_support) },
 103};
 104
 105/* simple methods of getting data from the current rxq entry */
 106static inline u32 ibmveth_rxq_flags(struct ibmveth_adapter *adapter)
 107{
 108	return be32_to_cpu(adapter->rx_queue.queue_addr[adapter->rx_queue.index].flags_off);
 109}
 110
 111static inline int ibmveth_rxq_toggle(struct ibmveth_adapter *adapter)
 112{
 113	return (ibmveth_rxq_flags(adapter) & IBMVETH_RXQ_TOGGLE) >>
 114			IBMVETH_RXQ_TOGGLE_SHIFT;
 115}
 116
 117static inline int ibmveth_rxq_pending_buffer(struct ibmveth_adapter *adapter)
 118{
 119	return ibmveth_rxq_toggle(adapter) == adapter->rx_queue.toggle;
 120}
 121
 122static inline int ibmveth_rxq_buffer_valid(struct ibmveth_adapter *adapter)
 123{
 124	return ibmveth_rxq_flags(adapter) & IBMVETH_RXQ_VALID;
 125}
 126
 127static inline int ibmveth_rxq_frame_offset(struct ibmveth_adapter *adapter)
 128{
 129	return ibmveth_rxq_flags(adapter) & IBMVETH_RXQ_OFF_MASK;
 130}
 131
 132static inline int ibmveth_rxq_frame_length(struct ibmveth_adapter *adapter)
 133{
 134	return be32_to_cpu(adapter->rx_queue.queue_addr[adapter->rx_queue.index].length);
 135}
 136
 137static inline int ibmveth_rxq_csum_good(struct ibmveth_adapter *adapter)
 138{
 139	return ibmveth_rxq_flags(adapter) & IBMVETH_RXQ_CSUM_GOOD;
 140}
 141
 142/* setup the initial settings for a buffer pool */
 143static void ibmveth_init_buffer_pool(struct ibmveth_buff_pool *pool,
 144				     u32 pool_index, u32 pool_size,
 145				     u32 buff_size, u32 pool_active)
 146{
 147	pool->size = pool_size;
 148	pool->index = pool_index;
 149	pool->buff_size = buff_size;
 150	pool->threshold = pool_size * 7 / 8;
 151	pool->active = pool_active;
 152}
 153
 154/* allocate and setup an buffer pool - called during open */
 155static int ibmveth_alloc_buffer_pool(struct ibmveth_buff_pool *pool)
 156{
 157	int i;
 158
 159	pool->free_map = kmalloc(sizeof(u16) * pool->size, GFP_KERNEL);
 160
 161	if (!pool->free_map)
 162		return -1;
 163
 164	pool->dma_addr = kmalloc(sizeof(dma_addr_t) * pool->size, GFP_KERNEL);
 165	if (!pool->dma_addr) {
 166		kfree(pool->free_map);
 167		pool->free_map = NULL;
 168		return -1;
 169	}
 170
 171	pool->skbuff = kcalloc(pool->size, sizeof(void *), GFP_KERNEL);
 172
 173	if (!pool->skbuff) {
 174		kfree(pool->dma_addr);
 175		pool->dma_addr = NULL;
 176
 177		kfree(pool->free_map);
 178		pool->free_map = NULL;
 179		return -1;
 180	}
 181
 182	memset(pool->dma_addr, 0, sizeof(dma_addr_t) * pool->size);
 183
 184	for (i = 0; i < pool->size; ++i)
 185		pool->free_map[i] = i;
 186
 187	atomic_set(&pool->available, 0);
 188	pool->producer_index = 0;
 189	pool->consumer_index = 0;
 190
 191	return 0;
 192}
 193
 194static inline void ibmveth_flush_buffer(void *addr, unsigned long length)
 195{
 196	unsigned long offset;
 197
 198	for (offset = 0; offset < length; offset += SMP_CACHE_BYTES)
 199		asm("dcbfl %0,%1" :: "b" (addr), "r" (offset));
 200}
 201
 202/* replenish the buffers for a pool.  note that we don't need to
 203 * skb_reserve these since they are used for incoming...
 204 */
 205static void ibmveth_replenish_buffer_pool(struct ibmveth_adapter *adapter,
 206					  struct ibmveth_buff_pool *pool)
 207{
 208	u32 i;
 209	u32 count = pool->size - atomic_read(&pool->available);
 210	u32 buffers_added = 0;
 211	struct sk_buff *skb;
 212	unsigned int free_index, index;
 213	u64 correlator;
 214	unsigned long lpar_rc;
 215	dma_addr_t dma_addr;
 216
 217	mb();
 218
 219	for (i = 0; i < count; ++i) {
 220		union ibmveth_buf_desc desc;
 221
 222		skb = netdev_alloc_skb(adapter->netdev, pool->buff_size);
 223
 224		if (!skb) {
 225			netdev_dbg(adapter->netdev,
 226				   "replenish: unable to allocate skb\n");
 227			adapter->replenish_no_mem++;
 228			break;
 229		}
 230
 231		free_index = pool->consumer_index;
 232		pool->consumer_index++;
 233		if (pool->consumer_index >= pool->size)
 234			pool->consumer_index = 0;
 235		index = pool->free_map[free_index];
 236
 237		BUG_ON(index == IBM_VETH_INVALID_MAP);
 238		BUG_ON(pool->skbuff[index] != NULL);
 239
 240		dma_addr = dma_map_single(&adapter->vdev->dev, skb->data,
 241				pool->buff_size, DMA_FROM_DEVICE);
 242
 243		if (dma_mapping_error(&adapter->vdev->dev, dma_addr))
 244			goto failure;
 245
 246		pool->free_map[free_index] = IBM_VETH_INVALID_MAP;
 247		pool->dma_addr[index] = dma_addr;
 248		pool->skbuff[index] = skb;
 249
 250		correlator = ((u64)pool->index << 32) | index;
 251		*(u64 *)skb->data = correlator;
 252
 253		desc.fields.flags_len = IBMVETH_BUF_VALID | pool->buff_size;
 254		desc.fields.address = dma_addr;
 255
 256		if (rx_flush) {
 257			unsigned int len = min(pool->buff_size,
 258						adapter->netdev->mtu +
 259						IBMVETH_BUFF_OH);
 260			ibmveth_flush_buffer(skb->data, len);
 261		}
 262		lpar_rc = h_add_logical_lan_buffer(adapter->vdev->unit_address,
 263						   desc.desc);
 264
 265		if (lpar_rc != H_SUCCESS) {
 266			goto failure;
 267		} else {
 268			buffers_added++;
 269			adapter->replenish_add_buff_success++;
 270		}
 271	}
 272
 273	mb();
 274	atomic_add(buffers_added, &(pool->available));
 275	return;
 276
 277failure:
 278	pool->free_map[free_index] = index;
 279	pool->skbuff[index] = NULL;
 280	if (pool->consumer_index == 0)
 281		pool->consumer_index = pool->size - 1;
 282	else
 283		pool->consumer_index--;
 284	if (!dma_mapping_error(&adapter->vdev->dev, dma_addr))
 285		dma_unmap_single(&adapter->vdev->dev,
 286		                 pool->dma_addr[index], pool->buff_size,
 287		                 DMA_FROM_DEVICE);
 288	dev_kfree_skb_any(skb);
 289	adapter->replenish_add_buff_failure++;
 290
 291	mb();
 292	atomic_add(buffers_added, &(pool->available));
 293}
 294
 295/* replenish routine */
 296static void ibmveth_replenish_task(struct ibmveth_adapter *adapter)
 297{
 298	int i;
 299
 300	adapter->replenish_task_cycles++;
 301
 302	for (i = (IBMVETH_NUM_BUFF_POOLS - 1); i >= 0; i--) {
 303		struct ibmveth_buff_pool *pool = &adapter->rx_buff_pool[i];
 304
 305		if (pool->active &&
 306		    (atomic_read(&pool->available) < pool->threshold))
 307			ibmveth_replenish_buffer_pool(adapter, pool);
 308	}
 309
 310	adapter->rx_no_buffer = *(u64 *)(((char*)adapter->buffer_list_addr) +
 311						4096 - 8);
 312}
 313
 314/* empty and free ana buffer pool - also used to do cleanup in error paths */
 315static void ibmveth_free_buffer_pool(struct ibmveth_adapter *adapter,
 316				     struct ibmveth_buff_pool *pool)
 317{
 318	int i;
 319
 320	kfree(pool->free_map);
 321	pool->free_map = NULL;
 322
 323	if (pool->skbuff && pool->dma_addr) {
 324		for (i = 0; i < pool->size; ++i) {
 325			struct sk_buff *skb = pool->skbuff[i];
 326			if (skb) {
 327				dma_unmap_single(&adapter->vdev->dev,
 328						 pool->dma_addr[i],
 329						 pool->buff_size,
 330						 DMA_FROM_DEVICE);
 331				dev_kfree_skb_any(skb);
 332				pool->skbuff[i] = NULL;
 333			}
 334		}
 335	}
 336
 337	if (pool->dma_addr) {
 338		kfree(pool->dma_addr);
 339		pool->dma_addr = NULL;
 340	}
 341
 342	if (pool->skbuff) {
 343		kfree(pool->skbuff);
 344		pool->skbuff = NULL;
 345	}
 346}
 347
 348/* remove a buffer from a pool */
 349static void ibmveth_remove_buffer_from_pool(struct ibmveth_adapter *adapter,
 350					    u64 correlator)
 351{
 352	unsigned int pool  = correlator >> 32;
 353	unsigned int index = correlator & 0xffffffffUL;
 354	unsigned int free_index;
 355	struct sk_buff *skb;
 356
 357	BUG_ON(pool >= IBMVETH_NUM_BUFF_POOLS);
 358	BUG_ON(index >= adapter->rx_buff_pool[pool].size);
 359
 360	skb = adapter->rx_buff_pool[pool].skbuff[index];
 361
 362	BUG_ON(skb == NULL);
 363
 364	adapter->rx_buff_pool[pool].skbuff[index] = NULL;
 365
 366	dma_unmap_single(&adapter->vdev->dev,
 367			 adapter->rx_buff_pool[pool].dma_addr[index],
 368			 adapter->rx_buff_pool[pool].buff_size,
 369			 DMA_FROM_DEVICE);
 370
 371	free_index = adapter->rx_buff_pool[pool].producer_index;
 372	adapter->rx_buff_pool[pool].producer_index++;
 373	if (adapter->rx_buff_pool[pool].producer_index >=
 374	    adapter->rx_buff_pool[pool].size)
 375		adapter->rx_buff_pool[pool].producer_index = 0;
 376	adapter->rx_buff_pool[pool].free_map[free_index] = index;
 377
 378	mb();
 379
 380	atomic_dec(&(adapter->rx_buff_pool[pool].available));
 381}
 382
 383/* get the current buffer on the rx queue */
 384static inline struct sk_buff *ibmveth_rxq_get_buffer(struct ibmveth_adapter *adapter)
 385{
 386	u64 correlator = adapter->rx_queue.queue_addr[adapter->rx_queue.index].correlator;
 387	unsigned int pool = correlator >> 32;
 388	unsigned int index = correlator & 0xffffffffUL;
 389
 390	BUG_ON(pool >= IBMVETH_NUM_BUFF_POOLS);
 391	BUG_ON(index >= adapter->rx_buff_pool[pool].size);
 392
 393	return adapter->rx_buff_pool[pool].skbuff[index];
 394}
 395
 396/* recycle the current buffer on the rx queue */
 397static int ibmveth_rxq_recycle_buffer(struct ibmveth_adapter *adapter)
 398{
 399	u32 q_index = adapter->rx_queue.index;
 400	u64 correlator = adapter->rx_queue.queue_addr[q_index].correlator;
 401	unsigned int pool = correlator >> 32;
 402	unsigned int index = correlator & 0xffffffffUL;
 403	union ibmveth_buf_desc desc;
 404	unsigned long lpar_rc;
 405	int ret = 1;
 406
 407	BUG_ON(pool >= IBMVETH_NUM_BUFF_POOLS);
 408	BUG_ON(index >= adapter->rx_buff_pool[pool].size);
 409
 410	if (!adapter->rx_buff_pool[pool].active) {
 411		ibmveth_rxq_harvest_buffer(adapter);
 412		ibmveth_free_buffer_pool(adapter, &adapter->rx_buff_pool[pool]);
 413		goto out;
 414	}
 415
 416	desc.fields.flags_len = IBMVETH_BUF_VALID |
 417		adapter->rx_buff_pool[pool].buff_size;
 418	desc.fields.address = adapter->rx_buff_pool[pool].dma_addr[index];
 419
 420	lpar_rc = h_add_logical_lan_buffer(adapter->vdev->unit_address, desc.desc);
 421
 422	if (lpar_rc != H_SUCCESS) {
 423		netdev_dbg(adapter->netdev, "h_add_logical_lan_buffer failed "
 424			   "during recycle rc=%ld", lpar_rc);
 425		ibmveth_remove_buffer_from_pool(adapter, adapter->rx_queue.queue_addr[adapter->rx_queue.index].correlator);
 426		ret = 0;
 427	}
 428
 429	if (++adapter->rx_queue.index == adapter->rx_queue.num_slots) {
 430		adapter->rx_queue.index = 0;
 431		adapter->rx_queue.toggle = !adapter->rx_queue.toggle;
 432	}
 433
 434out:
 435	return ret;
 436}
 437
 438static void ibmveth_rxq_harvest_buffer(struct ibmveth_adapter *adapter)
 439{
 440	ibmveth_remove_buffer_from_pool(adapter, adapter->rx_queue.queue_addr[adapter->rx_queue.index].correlator);
 441
 442	if (++adapter->rx_queue.index == adapter->rx_queue.num_slots) {
 443		adapter->rx_queue.index = 0;
 444		adapter->rx_queue.toggle = !adapter->rx_queue.toggle;
 445	}
 446}
 447
 448static void ibmveth_cleanup(struct ibmveth_adapter *adapter)
 449{
 450	int i;
 451	struct device *dev = &adapter->vdev->dev;
 452
 453	if (adapter->buffer_list_addr != NULL) {
 454		if (!dma_mapping_error(dev, adapter->buffer_list_dma)) {
 455			dma_unmap_single(dev, adapter->buffer_list_dma, 4096,
 456					DMA_BIDIRECTIONAL);
 457			adapter->buffer_list_dma = DMA_ERROR_CODE;
 458		}
 459		free_page((unsigned long)adapter->buffer_list_addr);
 460		adapter->buffer_list_addr = NULL;
 461	}
 462
 463	if (adapter->filter_list_addr != NULL) {
 464		if (!dma_mapping_error(dev, adapter->filter_list_dma)) {
 465			dma_unmap_single(dev, adapter->filter_list_dma, 4096,
 466					DMA_BIDIRECTIONAL);
 467			adapter->filter_list_dma = DMA_ERROR_CODE;
 468		}
 469		free_page((unsigned long)adapter->filter_list_addr);
 470		adapter->filter_list_addr = NULL;
 471	}
 472
 473	if (adapter->rx_queue.queue_addr != NULL) {
 474		dma_free_coherent(dev, adapter->rx_queue.queue_len,
 475				  adapter->rx_queue.queue_addr,
 476				  adapter->rx_queue.queue_dma);
 477		adapter->rx_queue.queue_addr = NULL;
 478	}
 479
 480	for (i = 0; i < IBMVETH_NUM_BUFF_POOLS; i++)
 481		if (adapter->rx_buff_pool[i].active)
 482			ibmveth_free_buffer_pool(adapter,
 483						 &adapter->rx_buff_pool[i]);
 484
 485	if (adapter->bounce_buffer != NULL) {
 486		if (!dma_mapping_error(dev, adapter->bounce_buffer_dma)) {
 487			dma_unmap_single(&adapter->vdev->dev,
 488					adapter->bounce_buffer_dma,
 489					adapter->netdev->mtu + IBMVETH_BUFF_OH,
 490					DMA_BIDIRECTIONAL);
 491			adapter->bounce_buffer_dma = DMA_ERROR_CODE;
 492		}
 493		kfree(adapter->bounce_buffer);
 494		adapter->bounce_buffer = NULL;
 495	}
 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 u64 ibmveth_encode_mac_addr(u8 *mac)
 526{
 527	int i;
 528	u64 encoded = 0;
 529
 530	for (i = 0; i < ETH_ALEN; i++)
 531		encoded = (encoded << 8) | mac[i];
 532
 533	return encoded;
 534}
 535
 536static int ibmveth_open(struct net_device *netdev)
 537{
 538	struct ibmveth_adapter *adapter = netdev_priv(netdev);
 539	u64 mac_address;
 540	int rxq_entries = 1;
 541	unsigned long lpar_rc;
 542	int rc;
 543	union ibmveth_buf_desc rxq_desc;
 544	int i;
 545	struct device *dev;
 546
 547	netdev_dbg(netdev, "open starting\n");
 548
 549	napi_enable(&adapter->napi);
 550
 551	for(i = 0; i < IBMVETH_NUM_BUFF_POOLS; i++)
 552		rxq_entries += adapter->rx_buff_pool[i].size;
 553
 554	adapter->buffer_list_addr = (void*) get_zeroed_page(GFP_KERNEL);
 555	adapter->filter_list_addr = (void*) get_zeroed_page(GFP_KERNEL);
 556
 557	if (!adapter->buffer_list_addr || !adapter->filter_list_addr) {
 558		netdev_err(netdev, "unable to allocate filter or buffer list "
 559			   "pages\n");
 560		rc = -ENOMEM;
 561		goto err_out;
 562	}
 563
 564	dev = &adapter->vdev->dev;
 565
 566	adapter->rx_queue.queue_len = sizeof(struct ibmveth_rx_q_entry) *
 567						rxq_entries;
 568	adapter->rx_queue.queue_addr =
 569		dma_alloc_coherent(dev, adapter->rx_queue.queue_len,
 570				   &adapter->rx_queue.queue_dma, GFP_KERNEL);
 571	if (!adapter->rx_queue.queue_addr) {
 572		rc = -ENOMEM;
 573		goto err_out;
 574	}
 575
 576	adapter->buffer_list_dma = dma_map_single(dev,
 577			adapter->buffer_list_addr, 4096, DMA_BIDIRECTIONAL);
 578	adapter->filter_list_dma = dma_map_single(dev,
 579			adapter->filter_list_addr, 4096, DMA_BIDIRECTIONAL);
 580
 581	if ((dma_mapping_error(dev, adapter->buffer_list_dma)) ||
 582	    (dma_mapping_error(dev, adapter->filter_list_dma))) {
 583		netdev_err(netdev, "unable to map filter or buffer list "
 584			   "pages\n");
 585		rc = -ENOMEM;
 586		goto err_out;
 587	}
 588
 589	adapter->rx_queue.index = 0;
 590	adapter->rx_queue.num_slots = rxq_entries;
 591	adapter->rx_queue.toggle = 1;
 592
 593	mac_address = ibmveth_encode_mac_addr(netdev->dev_addr);
 594
 595	rxq_desc.fields.flags_len = IBMVETH_BUF_VALID |
 596					adapter->rx_queue.queue_len;
 597	rxq_desc.fields.address = adapter->rx_queue.queue_dma;
 598
 599	netdev_dbg(netdev, "buffer list @ 0x%p\n", adapter->buffer_list_addr);
 600	netdev_dbg(netdev, "filter list @ 0x%p\n", adapter->filter_list_addr);
 601	netdev_dbg(netdev, "receive q   @ 0x%p\n", adapter->rx_queue.queue_addr);
 602
 603	h_vio_signal(adapter->vdev->unit_address, VIO_IRQ_DISABLE);
 604
 605	lpar_rc = ibmveth_register_logical_lan(adapter, rxq_desc, mac_address);
 606
 607	if (lpar_rc != H_SUCCESS) {
 608		netdev_err(netdev, "h_register_logical_lan failed with %ld\n",
 609			   lpar_rc);
 610		netdev_err(netdev, "buffer TCE:0x%llx filter TCE:0x%llx rxq "
 611			   "desc:0x%llx MAC:0x%llx\n",
 612				     adapter->buffer_list_dma,
 613				     adapter->filter_list_dma,
 614				     rxq_desc.desc,
 615				     mac_address);
 616		rc = -ENONET;
 617		goto err_out;
 618	}
 619
 620	for (i = 0; i < IBMVETH_NUM_BUFF_POOLS; i++) {
 621		if (!adapter->rx_buff_pool[i].active)
 622			continue;
 623		if (ibmveth_alloc_buffer_pool(&adapter->rx_buff_pool[i])) {
 624			netdev_err(netdev, "unable to alloc pool\n");
 625			adapter->rx_buff_pool[i].active = 0;
 626			rc = -ENOMEM;
 627			goto err_out;
 628		}
 629	}
 630
 631	netdev_dbg(netdev, "registering irq 0x%x\n", netdev->irq);
 632	rc = request_irq(netdev->irq, ibmveth_interrupt, 0, netdev->name,
 633			 netdev);
 634	if (rc != 0) {
 635		netdev_err(netdev, "unable to request irq 0x%x, rc %d\n",
 636			   netdev->irq, rc);
 637		do {
 638			lpar_rc = h_free_logical_lan(adapter->vdev->unit_address);
 639		} while (H_IS_LONG_BUSY(lpar_rc) || (lpar_rc == H_BUSY));
 640
 641		goto err_out;
 642	}
 643
 644	adapter->bounce_buffer =
 645	    kmalloc(netdev->mtu + IBMVETH_BUFF_OH, GFP_KERNEL);
 646	if (!adapter->bounce_buffer) {
 647		rc = -ENOMEM;
 648		goto err_out_free_irq;
 649	}
 650	adapter->bounce_buffer_dma =
 651	    dma_map_single(&adapter->vdev->dev, adapter->bounce_buffer,
 652			   netdev->mtu + IBMVETH_BUFF_OH, DMA_BIDIRECTIONAL);
 653	if (dma_mapping_error(dev, adapter->bounce_buffer_dma)) {
 654		netdev_err(netdev, "unable to map bounce buffer\n");
 655		rc = -ENOMEM;
 656		goto err_out_free_irq;
 657	}
 658
 659	netdev_dbg(netdev, "initial replenish cycle\n");
 660	ibmveth_interrupt(netdev->irq, netdev);
 661
 662	netif_start_queue(netdev);
 663
 664	netdev_dbg(netdev, "open complete\n");
 665
 666	return 0;
 667
 668err_out_free_irq:
 669	free_irq(netdev->irq, netdev);
 670err_out:
 671	ibmveth_cleanup(adapter);
 672	napi_disable(&adapter->napi);
 673	return rc;
 674}
 675
 676static int ibmveth_close(struct net_device *netdev)
 677{
 678	struct ibmveth_adapter *adapter = netdev_priv(netdev);
 679	long lpar_rc;
 680
 681	netdev_dbg(netdev, "close starting\n");
 682
 683	napi_disable(&adapter->napi);
 684
 685	if (!adapter->pool_config)
 686		netif_stop_queue(netdev);
 687
 688	h_vio_signal(adapter->vdev->unit_address, VIO_IRQ_DISABLE);
 689
 690	do {
 691		lpar_rc = h_free_logical_lan(adapter->vdev->unit_address);
 692	} while (H_IS_LONG_BUSY(lpar_rc) || (lpar_rc == H_BUSY));
 693
 694	if (lpar_rc != H_SUCCESS) {
 695		netdev_err(netdev, "h_free_logical_lan failed with %lx, "
 696			   "continuing with close\n", lpar_rc);
 697	}
 698
 699	free_irq(netdev->irq, netdev);
 700
 701	adapter->rx_no_buffer = *(u64 *)(((char *)adapter->buffer_list_addr) +
 702						4096 - 8);
 703
 704	ibmveth_cleanup(adapter);
 705
 706	netdev_dbg(netdev, "close complete\n");
 707
 708	return 0;
 709}
 710
 711static int netdev_get_settings(struct net_device *dev, struct ethtool_cmd *cmd)
 712{
 713	cmd->supported = (SUPPORTED_1000baseT_Full | SUPPORTED_Autoneg |
 714				SUPPORTED_FIBRE);
 715	cmd->advertising = (ADVERTISED_1000baseT_Full | ADVERTISED_Autoneg |
 716				ADVERTISED_FIBRE);
 717	ethtool_cmd_speed_set(cmd, SPEED_1000);
 718	cmd->duplex = DUPLEX_FULL;
 719	cmd->port = PORT_FIBRE;
 720	cmd->phy_address = 0;
 721	cmd->transceiver = XCVR_INTERNAL;
 722	cmd->autoneg = AUTONEG_ENABLE;
 723	cmd->maxtxpkt = 0;
 724	cmd->maxrxpkt = 1;
 725	return 0;
 726}
 727
 728static void netdev_get_drvinfo(struct net_device *dev,
 729			       struct ethtool_drvinfo *info)
 730{
 731	strlcpy(info->driver, ibmveth_driver_name, sizeof(info->driver));
 732	strlcpy(info->version, ibmveth_driver_version, sizeof(info->version));
 733}
 734
 735static netdev_features_t ibmveth_fix_features(struct net_device *dev,
 736	netdev_features_t features)
 737{
 738	/*
 739	 * Since the ibmveth firmware interface does not have the
 740	 * concept of separate tx/rx checksum offload enable, if rx
 741	 * checksum is disabled we also have to disable tx checksum
 742	 * offload. Once we disable rx checksum offload, we are no
 743	 * longer allowed to send tx buffers that are not properly
 744	 * checksummed.
 745	 */
 746
 747	if (!(features & NETIF_F_RXCSUM))
 748		features &= ~NETIF_F_ALL_CSUM;
 749
 750	return features;
 751}
 752
 753static int ibmveth_set_csum_offload(struct net_device *dev, u32 data)
 754{
 755	struct ibmveth_adapter *adapter = netdev_priv(dev);
 756	unsigned long set_attr, clr_attr, ret_attr;
 757	unsigned long set_attr6, clr_attr6;
 758	long ret, ret4, ret6;
 759	int rc1 = 0, rc2 = 0;
 760	int restart = 0;
 761
 762	if (netif_running(dev)) {
 763		restart = 1;
 764		adapter->pool_config = 1;
 765		ibmveth_close(dev);
 766		adapter->pool_config = 0;
 767	}
 768
 769	set_attr = 0;
 770	clr_attr = 0;
 771	set_attr6 = 0;
 772	clr_attr6 = 0;
 773
 774	if (data) {
 775		set_attr = IBMVETH_ILLAN_IPV4_TCP_CSUM;
 776		set_attr6 = IBMVETH_ILLAN_IPV6_TCP_CSUM;
 777	} else {
 778		clr_attr = IBMVETH_ILLAN_IPV4_TCP_CSUM;
 779		clr_attr6 = IBMVETH_ILLAN_IPV6_TCP_CSUM;
 780	}
 781
 782	ret = h_illan_attributes(adapter->vdev->unit_address, 0, 0, &ret_attr);
 783
 784	if (ret == H_SUCCESS && !(ret_attr & IBMVETH_ILLAN_ACTIVE_TRUNK) &&
 785	    !(ret_attr & IBMVETH_ILLAN_TRUNK_PRI_MASK) &&
 786	    (ret_attr & IBMVETH_ILLAN_PADDED_PKT_CSUM)) {
 787		ret4 = h_illan_attributes(adapter->vdev->unit_address, clr_attr,
 788					 set_attr, &ret_attr);
 789
 790		if (ret4 != H_SUCCESS) {
 791			netdev_err(dev, "unable to change IPv4 checksum "
 792					"offload settings. %d rc=%ld\n",
 793					data, ret4);
 794
 795			h_illan_attributes(adapter->vdev->unit_address,
 796					   set_attr, clr_attr, &ret_attr);
 797
 798			if (data == 1)
 799				dev->features &= ~NETIF_F_IP_CSUM;
 800
 801		} else {
 802			adapter->fw_ipv4_csum_support = data;
 803		}
 804
 805		ret6 = h_illan_attributes(adapter->vdev->unit_address,
 806					 clr_attr6, set_attr6, &ret_attr);
 807
 808		if (ret6 != H_SUCCESS) {
 809			netdev_err(dev, "unable to change IPv6 checksum "
 810					"offload settings. %d rc=%ld\n",
 811					data, ret6);
 812
 813			h_illan_attributes(adapter->vdev->unit_address,
 814					   set_attr6, clr_attr6, &ret_attr);
 815
 816			if (data == 1)
 817				dev->features &= ~NETIF_F_IPV6_CSUM;
 818
 819		} else
 820			adapter->fw_ipv6_csum_support = data;
 821
 822		if (ret4 == H_SUCCESS || ret6 == H_SUCCESS)
 823			adapter->rx_csum = data;
 824		else
 825			rc1 = -EIO;
 826	} else {
 827		rc1 = -EIO;
 828		netdev_err(dev, "unable to change checksum offload settings."
 829				     " %d rc=%ld ret_attr=%lx\n", data, ret,
 830				     ret_attr);
 831	}
 832
 833	if (restart)
 834		rc2 = ibmveth_open(dev);
 835
 836	return rc1 ? rc1 : rc2;
 837}
 838
 839static int ibmveth_set_features(struct net_device *dev,
 840	netdev_features_t features)
 841{
 842	struct ibmveth_adapter *adapter = netdev_priv(dev);
 843	int rx_csum = !!(features & NETIF_F_RXCSUM);
 844	int rc;
 845
 846	if (rx_csum == adapter->rx_csum)
 847		return 0;
 848
 849	rc = ibmveth_set_csum_offload(dev, rx_csum);
 850	if (rc && !adapter->rx_csum)
 851		dev->features = features & ~(NETIF_F_ALL_CSUM | NETIF_F_RXCSUM);
 852
 853	return rc;
 854}
 855
 856static void ibmveth_get_strings(struct net_device *dev, u32 stringset, u8 *data)
 857{
 858	int i;
 859
 860	if (stringset != ETH_SS_STATS)
 861		return;
 862
 863	for (i = 0; i < ARRAY_SIZE(ibmveth_stats); i++, data += ETH_GSTRING_LEN)
 864		memcpy(data, ibmveth_stats[i].name, ETH_GSTRING_LEN);
 865}
 866
 867static int ibmveth_get_sset_count(struct net_device *dev, int sset)
 868{
 869	switch (sset) {
 870	case ETH_SS_STATS:
 871		return ARRAY_SIZE(ibmveth_stats);
 872	default:
 873		return -EOPNOTSUPP;
 874	}
 875}
 876
 877static void ibmveth_get_ethtool_stats(struct net_device *dev,
 878				      struct ethtool_stats *stats, u64 *data)
 879{
 880	int i;
 881	struct ibmveth_adapter *adapter = netdev_priv(dev);
 882
 883	for (i = 0; i < ARRAY_SIZE(ibmveth_stats); i++)
 884		data[i] = IBMVETH_GET_STAT(adapter, ibmveth_stats[i].offset);
 885}
 886
 887static const struct ethtool_ops netdev_ethtool_ops = {
 888	.get_drvinfo		= netdev_get_drvinfo,
 889	.get_settings		= netdev_get_settings,
 890	.get_link		= ethtool_op_get_link,
 891	.get_strings		= ibmveth_get_strings,
 892	.get_sset_count		= ibmveth_get_sset_count,
 893	.get_ethtool_stats	= ibmveth_get_ethtool_stats,
 894};
 895
 896static int ibmveth_ioctl(struct net_device *dev, struct ifreq *ifr, int cmd)
 897{
 898	return -EOPNOTSUPP;
 899}
 900
 901#define page_offset(v) ((unsigned long)(v) & ((1 << 12) - 1))
 902
 903static int ibmveth_send(struct ibmveth_adapter *adapter,
 904			union ibmveth_buf_desc *descs)
 905{
 906	unsigned long correlator;
 907	unsigned int retry_count;
 908	unsigned long ret;
 909
 910	/*
 911	 * The retry count sets a maximum for the number of broadcast and
 912	 * multicast destinations within the system.
 913	 */
 914	retry_count = 1024;
 915	correlator = 0;
 916	do {
 917		ret = h_send_logical_lan(adapter->vdev->unit_address,
 918					     descs[0].desc, descs[1].desc,
 919					     descs[2].desc, descs[3].desc,
 920					     descs[4].desc, descs[5].desc,
 921					     correlator, &correlator);
 922	} while ((ret == H_BUSY) && (retry_count--));
 923
 924	if (ret != H_SUCCESS && ret != H_DROPPED) {
 925		netdev_err(adapter->netdev, "tx: h_send_logical_lan failed "
 926			   "with rc=%ld\n", ret);
 927		return 1;
 928	}
 929
 930	return 0;
 931}
 932
 933static netdev_tx_t ibmveth_start_xmit(struct sk_buff *skb,
 934				      struct net_device *netdev)
 935{
 936	struct ibmveth_adapter *adapter = netdev_priv(netdev);
 937	unsigned int desc_flags;
 938	union ibmveth_buf_desc descs[6];
 939	int last, i;
 940	int force_bounce = 0;
 941	dma_addr_t dma_addr;
 942
 943	/*
 944	 * veth handles a maximum of 6 segments including the header, so
 945	 * we have to linearize the skb if there are more than this.
 946	 */
 947	if (skb_shinfo(skb)->nr_frags > 5 && __skb_linearize(skb)) {
 948		netdev->stats.tx_dropped++;
 949		goto out;
 950	}
 951
 952	/* veth can't checksum offload UDP */
 953	if (skb->ip_summed == CHECKSUM_PARTIAL &&
 954	    ((skb->protocol == htons(ETH_P_IP) &&
 955	      ip_hdr(skb)->protocol != IPPROTO_TCP) ||
 956	     (skb->protocol == htons(ETH_P_IPV6) &&
 957	      ipv6_hdr(skb)->nexthdr != IPPROTO_TCP)) &&
 958	    skb_checksum_help(skb)) {
 959
 960		netdev_err(netdev, "tx: failed to checksum packet\n");
 961		netdev->stats.tx_dropped++;
 962		goto out;
 963	}
 964
 965	desc_flags = IBMVETH_BUF_VALID;
 966
 967	if (skb->ip_summed == CHECKSUM_PARTIAL) {
 968		unsigned char *buf = skb_transport_header(skb) +
 969						skb->csum_offset;
 970
 971		desc_flags |= (IBMVETH_BUF_NO_CSUM | IBMVETH_BUF_CSUM_GOOD);
 972
 973		/* Need to zero out the checksum */
 974		buf[0] = 0;
 975		buf[1] = 0;
 976	}
 977
 978retry_bounce:
 979	memset(descs, 0, sizeof(descs));
 980
 981	/*
 982	 * If a linear packet is below the rx threshold then
 983	 * copy it into the static bounce buffer. This avoids the
 984	 * cost of a TCE insert and remove.
 985	 */
 986	if (force_bounce || (!skb_is_nonlinear(skb) &&
 987				(skb->len < tx_copybreak))) {
 988		skb_copy_from_linear_data(skb, adapter->bounce_buffer,
 989					  skb->len);
 990
 991		descs[0].fields.flags_len = desc_flags | skb->len;
 992		descs[0].fields.address = adapter->bounce_buffer_dma;
 993
 994		if (ibmveth_send(adapter, descs)) {
 995			adapter->tx_send_failed++;
 996			netdev->stats.tx_dropped++;
 997		} else {
 998			netdev->stats.tx_packets++;
 999			netdev->stats.tx_bytes += skb->len;
1000		}
1001
1002		goto out;
1003	}
1004
1005	/* Map the header */
1006	dma_addr = dma_map_single(&adapter->vdev->dev, skb->data,
1007				  skb_headlen(skb), DMA_TO_DEVICE);
1008	if (dma_mapping_error(&adapter->vdev->dev, dma_addr))
1009		goto map_failed;
1010
1011	descs[0].fields.flags_len = desc_flags | skb_headlen(skb);
1012	descs[0].fields.address = dma_addr;
1013
1014	/* Map the frags */
1015	for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
1016		const skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
1017
1018		dma_addr = skb_frag_dma_map(&adapter->vdev->dev, frag, 0,
1019					    skb_frag_size(frag), DMA_TO_DEVICE);
1020
1021		if (dma_mapping_error(&adapter->vdev->dev, dma_addr))
1022			goto map_failed_frags;
1023
1024		descs[i+1].fields.flags_len = desc_flags | skb_frag_size(frag);
1025		descs[i+1].fields.address = dma_addr;
1026	}
1027
1028	if (ibmveth_send(adapter, descs)) {
1029		adapter->tx_send_failed++;
1030		netdev->stats.tx_dropped++;
1031	} else {
1032		netdev->stats.tx_packets++;
1033		netdev->stats.tx_bytes += skb->len;
1034	}
1035
1036	dma_unmap_single(&adapter->vdev->dev,
1037			 descs[0].fields.address,
1038			 descs[0].fields.flags_len & IBMVETH_BUF_LEN_MASK,
1039			 DMA_TO_DEVICE);
1040
1041	for (i = 1; i < skb_shinfo(skb)->nr_frags + 1; i++)
1042		dma_unmap_page(&adapter->vdev->dev, descs[i].fields.address,
1043			       descs[i].fields.flags_len & IBMVETH_BUF_LEN_MASK,
1044			       DMA_TO_DEVICE);
1045
1046out:
1047	dev_consume_skb_any(skb);
1048	return NETDEV_TX_OK;
1049
1050map_failed_frags:
1051	last = i+1;
1052	for (i = 0; i < last; i++)
1053		dma_unmap_page(&adapter->vdev->dev, descs[i].fields.address,
1054			       descs[i].fields.flags_len & IBMVETH_BUF_LEN_MASK,
1055			       DMA_TO_DEVICE);
1056
1057map_failed:
1058	if (!firmware_has_feature(FW_FEATURE_CMO))
1059		netdev_err(netdev, "tx: unable to map xmit buffer\n");
1060	adapter->tx_map_failed++;
1061	skb_linearize(skb);
1062	force_bounce = 1;
1063	goto retry_bounce;
1064}
1065
1066static int ibmveth_poll(struct napi_struct *napi, int budget)
1067{
1068	struct ibmveth_adapter *adapter =
1069			container_of(napi, struct ibmveth_adapter, napi);
1070	struct net_device *netdev = adapter->netdev;
1071	int frames_processed = 0;
1072	unsigned long lpar_rc;
1073
1074restart_poll:
1075	while (frames_processed < budget) {
1076		if (!ibmveth_rxq_pending_buffer(adapter))
1077			break;
1078
1079		smp_rmb();
1080		if (!ibmveth_rxq_buffer_valid(adapter)) {
1081			wmb(); /* suggested by larson1 */
1082			adapter->rx_invalid_buffer++;
1083			netdev_dbg(netdev, "recycling invalid buffer\n");
1084			ibmveth_rxq_recycle_buffer(adapter);
1085		} else {
1086			struct sk_buff *skb, *new_skb;
1087			int length = ibmveth_rxq_frame_length(adapter);
1088			int offset = ibmveth_rxq_frame_offset(adapter);
1089			int csum_good = ibmveth_rxq_csum_good(adapter);
1090
1091			skb = ibmveth_rxq_get_buffer(adapter);
1092
1093			new_skb = NULL;
1094			if (length < rx_copybreak)
1095				new_skb = netdev_alloc_skb(netdev, length);
1096
1097			if (new_skb) {
1098				skb_copy_to_linear_data(new_skb,
1099							skb->data + offset,
1100							length);
1101				if (rx_flush)
1102					ibmveth_flush_buffer(skb->data,
1103						length + offset);
1104				if (!ibmveth_rxq_recycle_buffer(adapter))
1105					kfree_skb(skb);
1106				skb = new_skb;
1107			} else {
1108				ibmveth_rxq_harvest_buffer(adapter);
1109				skb_reserve(skb, offset);
1110			}
1111
1112			skb_put(skb, length);
1113			skb->protocol = eth_type_trans(skb, netdev);
1114
1115			if (csum_good)
1116				skb->ip_summed = CHECKSUM_UNNECESSARY;
1117
1118			netif_receive_skb(skb);	/* send it up */
1119
1120			netdev->stats.rx_packets++;
1121			netdev->stats.rx_bytes += length;
1122			frames_processed++;
1123		}
1124	}
1125
1126	ibmveth_replenish_task(adapter);
1127
1128	if (frames_processed < budget) {
1129		/* We think we are done - reenable interrupts,
1130		 * then check once more to make sure we are done.
1131		 */
1132		lpar_rc = h_vio_signal(adapter->vdev->unit_address,
1133				       VIO_IRQ_ENABLE);
1134
1135		BUG_ON(lpar_rc != H_SUCCESS);
1136
1137		napi_complete(napi);
1138
1139		if (ibmveth_rxq_pending_buffer(adapter) &&
1140		    napi_reschedule(napi)) {
1141			lpar_rc = h_vio_signal(adapter->vdev->unit_address,
1142					       VIO_IRQ_DISABLE);
1143			goto restart_poll;
1144		}
1145	}
1146
1147	return frames_processed;
1148}
1149
1150static irqreturn_t ibmveth_interrupt(int irq, void *dev_instance)
1151{
1152	struct net_device *netdev = dev_instance;
1153	struct ibmveth_adapter *adapter = netdev_priv(netdev);
1154	unsigned long lpar_rc;
1155
1156	if (napi_schedule_prep(&adapter->napi)) {
1157		lpar_rc = h_vio_signal(adapter->vdev->unit_address,
1158				       VIO_IRQ_DISABLE);
1159		BUG_ON(lpar_rc != H_SUCCESS);
1160		__napi_schedule(&adapter->napi);
1161	}
1162	return IRQ_HANDLED;
1163}
1164
1165static void ibmveth_set_multicast_list(struct net_device *netdev)
1166{
1167	struct ibmveth_adapter *adapter = netdev_priv(netdev);
1168	unsigned long lpar_rc;
1169
1170	if ((netdev->flags & IFF_PROMISC) ||
1171	    (netdev_mc_count(netdev) > adapter->mcastFilterSize)) {
1172		lpar_rc = h_multicast_ctrl(adapter->vdev->unit_address,
1173					   IbmVethMcastEnableRecv |
1174					   IbmVethMcastDisableFiltering,
1175					   0);
1176		if (lpar_rc != H_SUCCESS) {
1177			netdev_err(netdev, "h_multicast_ctrl rc=%ld when "
1178				   "entering promisc mode\n", lpar_rc);
1179		}
1180	} else {
1181		struct netdev_hw_addr *ha;
1182		/* clear the filter table & disable filtering */
1183		lpar_rc = h_multicast_ctrl(adapter->vdev->unit_address,
1184					   IbmVethMcastEnableRecv |
1185					   IbmVethMcastDisableFiltering |
1186					   IbmVethMcastClearFilterTable,
1187					   0);
1188		if (lpar_rc != H_SUCCESS) {
1189			netdev_err(netdev, "h_multicast_ctrl rc=%ld when "
1190				   "attempting to clear filter table\n",
1191				   lpar_rc);
1192		}
1193		/* add the addresses to the filter table */
1194		netdev_for_each_mc_addr(ha, netdev) {
1195			/* add the multicast address to the filter table */
1196			u64 mcast_addr;
1197			mcast_addr = ibmveth_encode_mac_addr(ha->addr);
1198			lpar_rc = h_multicast_ctrl(adapter->vdev->unit_address,
1199						   IbmVethMcastAddFilter,
1200						   mcast_addr);
1201			if (lpar_rc != H_SUCCESS) {
1202				netdev_err(netdev, "h_multicast_ctrl rc=%ld "
1203					   "when adding an entry to the filter "
1204					   "table\n", lpar_rc);
1205			}
1206		}
1207
1208		/* re-enable filtering */
1209		lpar_rc = h_multicast_ctrl(adapter->vdev->unit_address,
1210					   IbmVethMcastEnableFiltering,
1211					   0);
1212		if (lpar_rc != H_SUCCESS) {
1213			netdev_err(netdev, "h_multicast_ctrl rc=%ld when "
1214				   "enabling filtering\n", lpar_rc);
1215		}
1216	}
1217}
1218
1219static int ibmveth_change_mtu(struct net_device *dev, int new_mtu)
1220{
1221	struct ibmveth_adapter *adapter = netdev_priv(dev);
1222	struct vio_dev *viodev = adapter->vdev;
1223	int new_mtu_oh = new_mtu + IBMVETH_BUFF_OH;
1224	int i, rc;
1225	int need_restart = 0;
1226
1227	if (new_mtu < IBMVETH_MIN_MTU)
1228		return -EINVAL;
1229
1230	for (i = 0; i < IBMVETH_NUM_BUFF_POOLS; i++)
1231		if (new_mtu_oh < adapter->rx_buff_pool[i].buff_size)
1232			break;
1233
1234	if (i == IBMVETH_NUM_BUFF_POOLS)
1235		return -EINVAL;
1236
1237	/* Deactivate all the buffer pools so that the next loop can activate
1238	   only the buffer pools necessary to hold the new MTU */
1239	if (netif_running(adapter->netdev)) {
1240		need_restart = 1;
1241		adapter->pool_config = 1;
1242		ibmveth_close(adapter->netdev);
1243		adapter->pool_config = 0;
1244	}
1245
1246	/* Look for an active buffer pool that can hold the new MTU */
1247	for (i = 0; i < IBMVETH_NUM_BUFF_POOLS; i++) {
1248		adapter->rx_buff_pool[i].active = 1;
1249
1250		if (new_mtu_oh < adapter->rx_buff_pool[i].buff_size) {
1251			dev->mtu = new_mtu;
1252			vio_cmo_set_dev_desired(viodev,
1253						ibmveth_get_desired_dma
1254						(viodev));
1255			if (need_restart) {
1256				return ibmveth_open(adapter->netdev);
1257			}
1258			return 0;
1259		}
1260	}
1261
1262	if (need_restart && (rc = ibmveth_open(adapter->netdev)))
1263		return rc;
1264
1265	return -EINVAL;
1266}
1267
1268#ifdef CONFIG_NET_POLL_CONTROLLER
1269static void ibmveth_poll_controller(struct net_device *dev)
1270{
1271	ibmveth_replenish_task(netdev_priv(dev));
1272	ibmveth_interrupt(dev->irq, dev);
1273}
1274#endif
1275
1276/**
1277 * ibmveth_get_desired_dma - Calculate IO memory desired by the driver
1278 *
1279 * @vdev: struct vio_dev for the device whose desired IO mem is to be returned
1280 *
1281 * Return value:
1282 *	Number of bytes of IO data the driver will need to perform well.
1283 */
1284static unsigned long ibmveth_get_desired_dma(struct vio_dev *vdev)
1285{
1286	struct net_device *netdev = dev_get_drvdata(&vdev->dev);
1287	struct ibmveth_adapter *adapter;
1288	struct iommu_table *tbl;
1289	unsigned long ret;
1290	int i;
1291	int rxqentries = 1;
1292
1293	tbl = get_iommu_table_base(&vdev->dev);
1294
1295	/* netdev inits at probe time along with the structures we need below*/
1296	if (netdev == NULL)
1297		return IOMMU_PAGE_ALIGN(IBMVETH_IO_ENTITLEMENT_DEFAULT, tbl);
1298
1299	adapter = netdev_priv(netdev);
1300
1301	ret = IBMVETH_BUFF_LIST_SIZE + IBMVETH_FILT_LIST_SIZE;
1302	ret += IOMMU_PAGE_ALIGN(netdev->mtu, tbl);
1303
1304	for (i = 0; i < IBMVETH_NUM_BUFF_POOLS; i++) {
1305		/* add the size of the active receive buffers */
1306		if (adapter->rx_buff_pool[i].active)
1307			ret +=
1308			    adapter->rx_buff_pool[i].size *
1309			    IOMMU_PAGE_ALIGN(adapter->rx_buff_pool[i].
1310					     buff_size, tbl);
1311		rxqentries += adapter->rx_buff_pool[i].size;
1312	}
1313	/* add the size of the receive queue entries */
1314	ret += IOMMU_PAGE_ALIGN(
1315		rxqentries * sizeof(struct ibmveth_rx_q_entry), tbl);
1316
1317	return ret;
1318}
1319
1320static const struct net_device_ops ibmveth_netdev_ops = {
1321	.ndo_open		= ibmveth_open,
1322	.ndo_stop		= ibmveth_close,
1323	.ndo_start_xmit		= ibmveth_start_xmit,
1324	.ndo_set_rx_mode	= ibmveth_set_multicast_list,
1325	.ndo_do_ioctl		= ibmveth_ioctl,
1326	.ndo_change_mtu		= ibmveth_change_mtu,
1327	.ndo_fix_features	= ibmveth_fix_features,
1328	.ndo_set_features	= ibmveth_set_features,
1329	.ndo_validate_addr	= eth_validate_addr,
1330	.ndo_set_mac_address	= eth_mac_addr,
1331#ifdef CONFIG_NET_POLL_CONTROLLER
1332	.ndo_poll_controller	= ibmveth_poll_controller,
1333#endif
1334};
1335
1336static int ibmveth_probe(struct vio_dev *dev, const struct vio_device_id *id)
1337{
1338	int rc, i, mac_len;
1339	struct net_device *netdev;
1340	struct ibmveth_adapter *adapter;
1341	unsigned char *mac_addr_p;
1342	unsigned int *mcastFilterSize_p;
1343
1344	dev_dbg(&dev->dev, "entering ibmveth_probe for UA 0x%x\n",
1345		dev->unit_address);
1346
1347	mac_addr_p = (unsigned char *)vio_get_attribute(dev, VETH_MAC_ADDR,
1348							&mac_len);
1349	if (!mac_addr_p) {
1350		dev_err(&dev->dev, "Can't find VETH_MAC_ADDR attribute\n");
1351		return -EINVAL;
1352	}
1353	/* Workaround for old/broken pHyp */
1354	if (mac_len == 8)
1355		mac_addr_p += 2;
1356	else if (mac_len != 6) {
1357		dev_err(&dev->dev, "VETH_MAC_ADDR attribute wrong len %d\n",
1358			mac_len);
1359		return -EINVAL;
1360	}
1361
1362	mcastFilterSize_p = (unsigned int *)vio_get_attribute(dev,
1363						VETH_MCAST_FILTER_SIZE, NULL);
1364	if (!mcastFilterSize_p) {
1365		dev_err(&dev->dev, "Can't find VETH_MCAST_FILTER_SIZE "
1366			"attribute\n");
1367		return -EINVAL;
1368	}
1369
1370	netdev = alloc_etherdev(sizeof(struct ibmveth_adapter));
1371
1372	if (!netdev)
1373		return -ENOMEM;
1374
1375	adapter = netdev_priv(netdev);
1376	dev_set_drvdata(&dev->dev, netdev);
1377
1378	adapter->vdev = dev;
1379	adapter->netdev = netdev;
1380	adapter->mcastFilterSize = *mcastFilterSize_p;
1381	adapter->pool_config = 0;
1382
1383	netif_napi_add(netdev, &adapter->napi, ibmveth_poll, 16);
1384
1385	netdev->irq = dev->irq;
1386	netdev->netdev_ops = &ibmveth_netdev_ops;
1387	netdev->ethtool_ops = &netdev_ethtool_ops;
1388	SET_NETDEV_DEV(netdev, &dev->dev);
1389	netdev->hw_features = NETIF_F_SG | NETIF_F_RXCSUM |
1390		NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM;
1391	netdev->features |= netdev->hw_features;
1392
1393	memcpy(netdev->dev_addr, mac_addr_p, ETH_ALEN);
1394
1395	for (i = 0; i < IBMVETH_NUM_BUFF_POOLS; i++) {
1396		struct kobject *kobj = &adapter->rx_buff_pool[i].kobj;
1397		int error;
1398
1399		ibmveth_init_buffer_pool(&adapter->rx_buff_pool[i], i,
1400					 pool_count[i], pool_size[i],
1401					 pool_active[i]);
1402		error = kobject_init_and_add(kobj, &ktype_veth_pool,
1403					     &dev->dev.kobj, "pool%d", i);
1404		if (!error)
1405			kobject_uevent(kobj, KOBJ_ADD);
1406	}
1407
1408	netdev_dbg(netdev, "adapter @ 0x%p\n", adapter);
1409
1410	adapter->buffer_list_dma = DMA_ERROR_CODE;
1411	adapter->filter_list_dma = DMA_ERROR_CODE;
1412	adapter->rx_queue.queue_dma = DMA_ERROR_CODE;
1413
1414	netdev_dbg(netdev, "registering netdev...\n");
1415
1416	ibmveth_set_features(netdev, netdev->features);
1417
1418	rc = register_netdev(netdev);
1419
1420	if (rc) {
1421		netdev_dbg(netdev, "failed to register netdev rc=%d\n", rc);
1422		free_netdev(netdev);
1423		return rc;
1424	}
1425
1426	netdev_dbg(netdev, "registered\n");
1427
1428	return 0;
1429}
1430
1431static int ibmveth_remove(struct vio_dev *dev)
1432{
1433	struct net_device *netdev = dev_get_drvdata(&dev->dev);
1434	struct ibmveth_adapter *adapter = netdev_priv(netdev);
1435	int i;
1436
1437	for (i = 0; i < IBMVETH_NUM_BUFF_POOLS; i++)
1438		kobject_put(&adapter->rx_buff_pool[i].kobj);
1439
1440	unregister_netdev(netdev);
1441
1442	free_netdev(netdev);
1443	dev_set_drvdata(&dev->dev, NULL);
1444
1445	return 0;
1446}
1447
1448static struct attribute veth_active_attr;
1449static struct attribute veth_num_attr;
1450static struct attribute veth_size_attr;
1451
1452static ssize_t veth_pool_show(struct kobject *kobj,
1453			      struct attribute *attr, char *buf)
1454{
1455	struct ibmveth_buff_pool *pool = container_of(kobj,
1456						      struct ibmveth_buff_pool,
1457						      kobj);
1458
1459	if (attr == &veth_active_attr)
1460		return sprintf(buf, "%d\n", pool->active);
1461	else if (attr == &veth_num_attr)
1462		return sprintf(buf, "%d\n", pool->size);
1463	else if (attr == &veth_size_attr)
1464		return sprintf(buf, "%d\n", pool->buff_size);
1465	return 0;
1466}
1467
1468static ssize_t veth_pool_store(struct kobject *kobj, struct attribute *attr,
1469			       const char *buf, size_t count)
1470{
1471	struct ibmveth_buff_pool *pool = container_of(kobj,
1472						      struct ibmveth_buff_pool,
1473						      kobj);
1474	struct net_device *netdev = dev_get_drvdata(
1475	    container_of(kobj->parent, struct device, kobj));
1476	struct ibmveth_adapter *adapter = netdev_priv(netdev);
1477	long value = simple_strtol(buf, NULL, 10);
1478	long rc;
1479
1480	if (attr == &veth_active_attr) {
1481		if (value && !pool->active) {
1482			if (netif_running(netdev)) {
1483				if (ibmveth_alloc_buffer_pool(pool)) {
1484					netdev_err(netdev,
1485						   "unable to alloc pool\n");
1486					return -ENOMEM;
1487				}
1488				pool->active = 1;
1489				adapter->pool_config = 1;
1490				ibmveth_close(netdev);
1491				adapter->pool_config = 0;
1492				if ((rc = ibmveth_open(netdev)))
1493					return rc;
1494			} else {
1495				pool->active = 1;
1496			}
1497		} else if (!value && pool->active) {
1498			int mtu = netdev->mtu + IBMVETH_BUFF_OH;
1499			int i;
1500			/* Make sure there is a buffer pool with buffers that
1501			   can hold a packet of the size of the MTU */
1502			for (i = 0; i < IBMVETH_NUM_BUFF_POOLS; i++) {
1503				if (pool == &adapter->rx_buff_pool[i])
1504					continue;
1505				if (!adapter->rx_buff_pool[i].active)
1506					continue;
1507				if (mtu <= adapter->rx_buff_pool[i].buff_size)
1508					break;
1509			}
1510
1511			if (i == IBMVETH_NUM_BUFF_POOLS) {
1512				netdev_err(netdev, "no active pool >= MTU\n");
1513				return -EPERM;
1514			}
1515
1516			if (netif_running(netdev)) {
1517				adapter->pool_config = 1;
1518				ibmveth_close(netdev);
1519				pool->active = 0;
1520				adapter->pool_config = 0;
1521				if ((rc = ibmveth_open(netdev)))
1522					return rc;
1523			}
1524			pool->active = 0;
1525		}
1526	} else if (attr == &veth_num_attr) {
1527		if (value <= 0 || value > IBMVETH_MAX_POOL_COUNT) {
1528			return -EINVAL;
1529		} else {
1530			if (netif_running(netdev)) {
1531				adapter->pool_config = 1;
1532				ibmveth_close(netdev);
1533				adapter->pool_config = 0;
1534				pool->size = value;
1535				if ((rc = ibmveth_open(netdev)))
1536					return rc;
1537			} else {
1538				pool->size = value;
1539			}
1540		}
1541	} else if (attr == &veth_size_attr) {
1542		if (value <= IBMVETH_BUFF_OH || value > IBMVETH_MAX_BUF_SIZE) {
1543			return -EINVAL;
1544		} else {
1545			if (netif_running(netdev)) {
1546				adapter->pool_config = 1;
1547				ibmveth_close(netdev);
1548				adapter->pool_config = 0;
1549				pool->buff_size = value;
1550				if ((rc = ibmveth_open(netdev)))
1551					return rc;
1552			} else {
1553				pool->buff_size = value;
1554			}
1555		}
1556	}
1557
1558	/* kick the interrupt handler to allocate/deallocate pools */
1559	ibmveth_interrupt(netdev->irq, netdev);
1560	return count;
1561}
1562
1563
1564#define ATTR(_name, _mode)				\
1565	struct attribute veth_##_name##_attr = {	\
1566	.name = __stringify(_name), .mode = _mode,	\
1567	};
1568
1569static ATTR(active, 0644);
1570static ATTR(num, 0644);
1571static ATTR(size, 0644);
1572
1573static struct attribute *veth_pool_attrs[] = {
1574	&veth_active_attr,
1575	&veth_num_attr,
1576	&veth_size_attr,
1577	NULL,
1578};
1579
1580static const struct sysfs_ops veth_pool_ops = {
1581	.show   = veth_pool_show,
1582	.store  = veth_pool_store,
1583};
1584
1585static struct kobj_type ktype_veth_pool = {
1586	.release        = NULL,
1587	.sysfs_ops      = &veth_pool_ops,
1588	.default_attrs  = veth_pool_attrs,
1589};
1590
1591static int ibmveth_resume(struct device *dev)
1592{
1593	struct net_device *netdev = dev_get_drvdata(dev);
1594	ibmveth_interrupt(netdev->irq, netdev);
1595	return 0;
1596}
1597
1598static struct vio_device_id ibmveth_device_table[] = {
1599	{ "network", "IBM,l-lan"},
1600	{ "", "" }
1601};
1602MODULE_DEVICE_TABLE(vio, ibmveth_device_table);
1603
1604static struct dev_pm_ops ibmveth_pm_ops = {
1605	.resume = ibmveth_resume
1606};
1607
1608static struct vio_driver ibmveth_driver = {
1609	.id_table	= ibmveth_device_table,
1610	.probe		= ibmveth_probe,
1611	.remove		= ibmveth_remove,
1612	.get_desired_dma = ibmveth_get_desired_dma,
1613	.name		= ibmveth_driver_name,
1614	.pm		= &ibmveth_pm_ops,
1615};
1616
1617static int __init ibmveth_module_init(void)
1618{
1619	printk(KERN_DEBUG "%s: %s %s\n", ibmveth_driver_name,
1620	       ibmveth_driver_string, ibmveth_driver_version);
1621
1622	return vio_register_driver(&ibmveth_driver);
1623}
1624
1625static void __exit ibmveth_module_exit(void)
1626{
1627	vio_unregister_driver(&ibmveth_driver);
1628}
1629
1630module_init(ibmveth_module_init);
1631module_exit(ibmveth_module_exit);