brintos

brintos / linux-shallow public Read only

0
0
Text · 51.9 KiB · b619a3e Raw
1950 lines · c
1// SPDX-License-Identifier: GPL-2.0-or-later2/*3 * IBM Power Virtual Ethernet Device Driver4 *5 * Copyright (C) IBM Corporation, 2003, 20106 *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,43				       bool reuse);44static unsigned long ibmveth_get_desired_dma(struct vio_dev *vdev);45 46static struct kobj_type ktype_veth_pool;47 48 49static const char ibmveth_driver_name[] = "ibmveth";50static const char ibmveth_driver_string[] = "IBM Power Virtual Ethernet Driver";51#define ibmveth_driver_version "1.06"52 53MODULE_AUTHOR("Santiago Leon <santil@linux.vnet.ibm.com>");54MODULE_DESCRIPTION("IBM Power Virtual Ethernet Driver");55MODULE_LICENSE("GPL");56MODULE_VERSION(ibmveth_driver_version);57 58static unsigned int tx_copybreak __read_mostly = 128;59module_param(tx_copybreak, uint, 0644);60MODULE_PARM_DESC(tx_copybreak,61	"Maximum size of packet that is copied to a new buffer on transmit");62 63static unsigned int rx_copybreak __read_mostly = 128;64module_param(rx_copybreak, uint, 0644);65MODULE_PARM_DESC(rx_copybreak,66	"Maximum size of packet that is copied to a new buffer on receive");67 68static unsigned int rx_flush __read_mostly = 0;69module_param(rx_flush, uint, 0644);70MODULE_PARM_DESC(rx_flush, "Flush receive buffers before use");71 72static bool old_large_send __read_mostly;73module_param(old_large_send, bool, 0444);74MODULE_PARM_DESC(old_large_send,75	"Use old large send method on firmware that supports the new method");76 77struct ibmveth_stat {78	char name[ETH_GSTRING_LEN];79	int offset;80};81 82#define IBMVETH_STAT_OFF(stat) offsetof(struct ibmveth_adapter, stat)83#define IBMVETH_GET_STAT(a, off) *((u64 *)(((unsigned long)(a)) + off))84 85static struct ibmveth_stat ibmveth_stats[] = {86	{ "replenish_task_cycles", IBMVETH_STAT_OFF(replenish_task_cycles) },87	{ "replenish_no_mem", IBMVETH_STAT_OFF(replenish_no_mem) },88	{ "replenish_add_buff_failure",89			IBMVETH_STAT_OFF(replenish_add_buff_failure) },90	{ "replenish_add_buff_success",91			IBMVETH_STAT_OFF(replenish_add_buff_success) },92	{ "rx_invalid_buffer", IBMVETH_STAT_OFF(rx_invalid_buffer) },93	{ "rx_no_buffer", IBMVETH_STAT_OFF(rx_no_buffer) },94	{ "tx_map_failed", IBMVETH_STAT_OFF(tx_map_failed) },95	{ "tx_send_failed", IBMVETH_STAT_OFF(tx_send_failed) },96	{ "fw_enabled_ipv4_csum", IBMVETH_STAT_OFF(fw_ipv4_csum_support) },97	{ "fw_enabled_ipv6_csum", IBMVETH_STAT_OFF(fw_ipv6_csum_support) },98	{ "tx_large_packets", IBMVETH_STAT_OFF(tx_large_packets) },99	{ "rx_large_packets", IBMVETH_STAT_OFF(rx_large_packets) },100	{ "fw_enabled_large_send", IBMVETH_STAT_OFF(fw_large_send_support) }101};102 103/* simple methods of getting data from the current rxq entry */104static inline u32 ibmveth_rxq_flags(struct ibmveth_adapter *adapter)105{106	return be32_to_cpu(adapter->rx_queue.queue_addr[adapter->rx_queue.index].flags_off);107}108 109static inline int ibmveth_rxq_toggle(struct ibmveth_adapter *adapter)110{111	return (ibmveth_rxq_flags(adapter) & IBMVETH_RXQ_TOGGLE) >>112			IBMVETH_RXQ_TOGGLE_SHIFT;113}114 115static inline int ibmveth_rxq_pending_buffer(struct ibmveth_adapter *adapter)116{117	return ibmveth_rxq_toggle(adapter) == adapter->rx_queue.toggle;118}119 120static inline int ibmveth_rxq_buffer_valid(struct ibmveth_adapter *adapter)121{122	return ibmveth_rxq_flags(adapter) & IBMVETH_RXQ_VALID;123}124 125static inline int ibmveth_rxq_frame_offset(struct ibmveth_adapter *adapter)126{127	return ibmveth_rxq_flags(adapter) & IBMVETH_RXQ_OFF_MASK;128}129 130static inline int ibmveth_rxq_large_packet(struct ibmveth_adapter *adapter)131{132	return ibmveth_rxq_flags(adapter) & IBMVETH_RXQ_LRG_PKT;133}134 135static inline int ibmveth_rxq_frame_length(struct ibmveth_adapter *adapter)136{137	return be32_to_cpu(adapter->rx_queue.queue_addr[adapter->rx_queue.index].length);138}139 140static inline int ibmveth_rxq_csum_good(struct ibmveth_adapter *adapter)141{142	return ibmveth_rxq_flags(adapter) & IBMVETH_RXQ_CSUM_GOOD;143}144 145static unsigned int ibmveth_real_max_tx_queues(void)146{147	unsigned int n_cpu = num_online_cpus();148 149	return min(n_cpu, IBMVETH_MAX_QUEUES);150}151 152/* setup the initial settings for a buffer pool */153static void ibmveth_init_buffer_pool(struct ibmveth_buff_pool *pool,154				     u32 pool_index, u32 pool_size,155				     u32 buff_size, u32 pool_active)156{157	pool->size = pool_size;158	pool->index = pool_index;159	pool->buff_size = buff_size;160	pool->threshold = pool_size * 7 / 8;161	pool->active = pool_active;162}163 164/* allocate and setup an buffer pool - called during open */165static int ibmveth_alloc_buffer_pool(struct ibmveth_buff_pool *pool)166{167	int i;168 169	pool->free_map = kmalloc_array(pool->size, sizeof(u16), GFP_KERNEL);170 171	if (!pool->free_map)172		return -1;173 174	pool->dma_addr = kcalloc(pool->size, sizeof(dma_addr_t), GFP_KERNEL);175	if (!pool->dma_addr) {176		kfree(pool->free_map);177		pool->free_map = NULL;178		return -1;179	}180 181	pool->skbuff = kcalloc(pool->size, sizeof(void *), GFP_KERNEL);182 183	if (!pool->skbuff) {184		kfree(pool->dma_addr);185		pool->dma_addr = NULL;186 187		kfree(pool->free_map);188		pool->free_map = NULL;189		return -1;190	}191 192	for (i = 0; i < pool->size; ++i)193		pool->free_map[i] = i;194 195	atomic_set(&pool->available, 0);196	pool->producer_index = 0;197	pool->consumer_index = 0;198 199	return 0;200}201 202static inline void ibmveth_flush_buffer(void *addr, unsigned long length)203{204	unsigned long offset;205 206	for (offset = 0; offset < length; offset += SMP_CACHE_BYTES)207		asm("dcbf %0,%1,1" :: "b" (addr), "r" (offset));208}209 210/* replenish the buffers for a pool.  note that we don't need to211 * skb_reserve these since they are used for incoming...212 */213static void ibmveth_replenish_buffer_pool(struct ibmveth_adapter *adapter,214					  struct ibmveth_buff_pool *pool)215{216	u32 i;217	u32 count = pool->size - atomic_read(&pool->available);218	u32 buffers_added = 0;219	struct sk_buff *skb;220	unsigned int free_index, index;221	u64 correlator;222	unsigned long lpar_rc;223	dma_addr_t dma_addr;224 225	mb();226 227	for (i = 0; i < count; ++i) {228		union ibmveth_buf_desc desc;229 230		free_index = pool->consumer_index;231		index = pool->free_map[free_index];232		skb = NULL;233 234		BUG_ON(index == IBM_VETH_INVALID_MAP);235 236		/* are we allocating a new buffer or recycling an old one */237		if (pool->skbuff[index])238			goto reuse;239 240		skb = netdev_alloc_skb(adapter->netdev, pool->buff_size);241 242		if (!skb) {243			netdev_dbg(adapter->netdev,244				   "replenish: unable to allocate skb\n");245			adapter->replenish_no_mem++;246			break;247		}248 249		dma_addr = dma_map_single(&adapter->vdev->dev, skb->data,250				pool->buff_size, DMA_FROM_DEVICE);251 252		if (dma_mapping_error(&adapter->vdev->dev, dma_addr))253			goto failure;254 255		pool->dma_addr[index] = dma_addr;256		pool->skbuff[index] = skb;257 258		if (rx_flush) {259			unsigned int len = min(pool->buff_size,260					       adapter->netdev->mtu +261					       IBMVETH_BUFF_OH);262			ibmveth_flush_buffer(skb->data, len);263		}264reuse:265		dma_addr = pool->dma_addr[index];266		desc.fields.flags_len = IBMVETH_BUF_VALID | pool->buff_size;267		desc.fields.address = dma_addr;268 269		correlator = ((u64)pool->index << 32) | index;270		*(u64 *)pool->skbuff[index]->data = correlator;271 272		lpar_rc = h_add_logical_lan_buffer(adapter->vdev->unit_address,273						   desc.desc);274 275		if (lpar_rc != H_SUCCESS) {276			netdev_warn(adapter->netdev,277				    "%sadd_logical_lan failed %lu\n",278				    skb ? "" : "When recycling: ", lpar_rc);279			goto failure;280		}281 282		pool->free_map[free_index] = IBM_VETH_INVALID_MAP;283		pool->consumer_index++;284		if (pool->consumer_index >= pool->size)285			pool->consumer_index = 0;286 287		buffers_added++;288		adapter->replenish_add_buff_success++;289	}290 291	mb();292	atomic_add(buffers_added, &(pool->available));293	return;294 295failure:296 297	if (dma_addr && !dma_mapping_error(&adapter->vdev->dev, dma_addr))298		dma_unmap_single(&adapter->vdev->dev,299		                 pool->dma_addr[index], pool->buff_size,300		                 DMA_FROM_DEVICE);301	dev_kfree_skb_any(pool->skbuff[index]);302	pool->skbuff[index] = NULL;303	adapter->replenish_add_buff_failure++;304 305	mb();306	atomic_add(buffers_added, &(pool->available));307}308 309/*310 * The final 8 bytes of the buffer list is a counter of frames dropped311 * because there was not a buffer in the buffer list capable of holding312 * the frame.313 */314static void ibmveth_update_rx_no_buffer(struct ibmveth_adapter *adapter)315{316	__be64 *p = adapter->buffer_list_addr + 4096 - 8;317 318	adapter->rx_no_buffer = be64_to_cpup(p);319}320 321/* replenish routine */322static void ibmveth_replenish_task(struct ibmveth_adapter *adapter)323{324	int i;325 326	adapter->replenish_task_cycles++;327 328	for (i = (IBMVETH_NUM_BUFF_POOLS - 1); i >= 0; i--) {329		struct ibmveth_buff_pool *pool = &adapter->rx_buff_pool[i];330 331		if (pool->active &&332		    (atomic_read(&pool->available) < pool->threshold))333			ibmveth_replenish_buffer_pool(adapter, pool);334	}335 336	ibmveth_update_rx_no_buffer(adapter);337}338 339/* empty and free ana buffer pool - also used to do cleanup in error paths */340static void ibmveth_free_buffer_pool(struct ibmveth_adapter *adapter,341				     struct ibmveth_buff_pool *pool)342{343	int i;344 345	kfree(pool->free_map);346	pool->free_map = NULL;347 348	if (pool->skbuff && pool->dma_addr) {349		for (i = 0; i < pool->size; ++i) {350			struct sk_buff *skb = pool->skbuff[i];351			if (skb) {352				dma_unmap_single(&adapter->vdev->dev,353						 pool->dma_addr[i],354						 pool->buff_size,355						 DMA_FROM_DEVICE);356				dev_kfree_skb_any(skb);357				pool->skbuff[i] = NULL;358			}359		}360	}361 362	if (pool->dma_addr) {363		kfree(pool->dma_addr);364		pool->dma_addr = NULL;365	}366 367	if (pool->skbuff) {368		kfree(pool->skbuff);369		pool->skbuff = NULL;370	}371}372 373/* remove a buffer from a pool */374static void ibmveth_remove_buffer_from_pool(struct ibmveth_adapter *adapter,375					    u64 correlator, bool reuse)376{377	unsigned int pool  = correlator >> 32;378	unsigned int index = correlator & 0xffffffffUL;379	unsigned int free_index;380	struct sk_buff *skb;381 382	BUG_ON(pool >= IBMVETH_NUM_BUFF_POOLS);383	BUG_ON(index >= adapter->rx_buff_pool[pool].size);384 385	skb = adapter->rx_buff_pool[pool].skbuff[index];386	BUG_ON(skb == NULL);387 388	/* if we are going to reuse the buffer then keep the pointers around389	 * but mark index as available. replenish will see the skb pointer and390	 * assume it is to be recycled.391	 */392	if (!reuse) {393		/* remove the skb pointer to mark free. actual freeing is done394		 * by upper level networking after gro_recieve395		 */396		adapter->rx_buff_pool[pool].skbuff[index] = NULL;397 398		dma_unmap_single(&adapter->vdev->dev,399				 adapter->rx_buff_pool[pool].dma_addr[index],400				 adapter->rx_buff_pool[pool].buff_size,401				 DMA_FROM_DEVICE);402	}403 404	free_index = adapter->rx_buff_pool[pool].producer_index;405	adapter->rx_buff_pool[pool].producer_index++;406	if (adapter->rx_buff_pool[pool].producer_index >=407	    adapter->rx_buff_pool[pool].size)408		adapter->rx_buff_pool[pool].producer_index = 0;409	adapter->rx_buff_pool[pool].free_map[free_index] = index;410 411	mb();412 413	atomic_dec(&(adapter->rx_buff_pool[pool].available));414}415 416/* get the current buffer on the rx queue */417static inline struct sk_buff *ibmveth_rxq_get_buffer(struct ibmveth_adapter *adapter)418{419	u64 correlator = adapter->rx_queue.queue_addr[adapter->rx_queue.index].correlator;420	unsigned int pool = correlator >> 32;421	unsigned int index = correlator & 0xffffffffUL;422 423	BUG_ON(pool >= IBMVETH_NUM_BUFF_POOLS);424	BUG_ON(index >= adapter->rx_buff_pool[pool].size);425 426	return adapter->rx_buff_pool[pool].skbuff[index];427}428 429static void ibmveth_rxq_harvest_buffer(struct ibmveth_adapter *adapter,430				       bool reuse)431{432	u64 cor;433 434	cor = adapter->rx_queue.queue_addr[adapter->rx_queue.index].correlator;435	ibmveth_remove_buffer_from_pool(adapter, cor, reuse);436 437	if (++adapter->rx_queue.index == adapter->rx_queue.num_slots) {438		adapter->rx_queue.index = 0;439		adapter->rx_queue.toggle = !adapter->rx_queue.toggle;440	}441}442 443static void ibmveth_free_tx_ltb(struct ibmveth_adapter *adapter, int idx)444{445	dma_unmap_single(&adapter->vdev->dev, adapter->tx_ltb_dma[idx],446			 adapter->tx_ltb_size, DMA_TO_DEVICE);447	kfree(adapter->tx_ltb_ptr[idx]);448	adapter->tx_ltb_ptr[idx] = NULL;449}450 451static int ibmveth_allocate_tx_ltb(struct ibmveth_adapter *adapter, int idx)452{453	adapter->tx_ltb_ptr[idx] = kzalloc(adapter->tx_ltb_size,454					   GFP_KERNEL);455	if (!adapter->tx_ltb_ptr[idx]) {456		netdev_err(adapter->netdev,457			   "unable to allocate tx long term buffer\n");458		return -ENOMEM;459	}460	adapter->tx_ltb_dma[idx] = dma_map_single(&adapter->vdev->dev,461						  adapter->tx_ltb_ptr[idx],462						  adapter->tx_ltb_size,463						  DMA_TO_DEVICE);464	if (dma_mapping_error(&adapter->vdev->dev, adapter->tx_ltb_dma[idx])) {465		netdev_err(adapter->netdev,466			   "unable to DMA map tx long term buffer\n");467		kfree(adapter->tx_ltb_ptr[idx]);468		adapter->tx_ltb_ptr[idx] = NULL;469		return -ENOMEM;470	}471 472	return 0;473}474 475static int ibmveth_register_logical_lan(struct ibmveth_adapter *adapter,476        union ibmveth_buf_desc rxq_desc, u64 mac_address)477{478	int rc, try_again = 1;479 480	/*481	 * After a kexec the adapter will still be open, so our attempt to482	 * open it will fail. So if we get a failure we free the adapter and483	 * try again, but only once.484	 */485retry:486	rc = h_register_logical_lan(adapter->vdev->unit_address,487				    adapter->buffer_list_dma, rxq_desc.desc,488				    adapter->filter_list_dma, mac_address);489 490	if (rc != H_SUCCESS && try_again) {491		do {492			rc = h_free_logical_lan(adapter->vdev->unit_address);493		} while (H_IS_LONG_BUSY(rc) || (rc == H_BUSY));494 495		try_again = 0;496		goto retry;497	}498 499	return rc;500}501 502static int ibmveth_open(struct net_device *netdev)503{504	struct ibmveth_adapter *adapter = netdev_priv(netdev);505	u64 mac_address;506	int rxq_entries = 1;507	unsigned long lpar_rc;508	int rc;509	union ibmveth_buf_desc rxq_desc;510	int i;511	struct device *dev;512 513	netdev_dbg(netdev, "open starting\n");514 515	napi_enable(&adapter->napi);516 517	for(i = 0; i < IBMVETH_NUM_BUFF_POOLS; i++)518		rxq_entries += adapter->rx_buff_pool[i].size;519 520	rc = -ENOMEM;521	adapter->buffer_list_addr = (void*) get_zeroed_page(GFP_KERNEL);522	if (!adapter->buffer_list_addr) {523		netdev_err(netdev, "unable to allocate list pages\n");524		goto out;525	}526 527	adapter->filter_list_addr = (void*) get_zeroed_page(GFP_KERNEL);528	if (!adapter->filter_list_addr) {529		netdev_err(netdev, "unable to allocate filter pages\n");530		goto out_free_buffer_list;531	}532 533	dev = &adapter->vdev->dev;534 535	adapter->rx_queue.queue_len = sizeof(struct ibmveth_rx_q_entry) *536						rxq_entries;537	adapter->rx_queue.queue_addr =538		dma_alloc_coherent(dev, adapter->rx_queue.queue_len,539				   &adapter->rx_queue.queue_dma, GFP_KERNEL);540	if (!adapter->rx_queue.queue_addr)541		goto out_free_filter_list;542 543	adapter->buffer_list_dma = dma_map_single(dev,544			adapter->buffer_list_addr, 4096, DMA_BIDIRECTIONAL);545	if (dma_mapping_error(dev, adapter->buffer_list_dma)) {546		netdev_err(netdev, "unable to map buffer list pages\n");547		goto out_free_queue_mem;548	}549 550	adapter->filter_list_dma = dma_map_single(dev,551			adapter->filter_list_addr, 4096, DMA_BIDIRECTIONAL);552	if (dma_mapping_error(dev, adapter->filter_list_dma)) {553		netdev_err(netdev, "unable to map filter list pages\n");554		goto out_unmap_buffer_list;555	}556 557	for (i = 0; i < netdev->real_num_tx_queues; i++) {558		if (ibmveth_allocate_tx_ltb(adapter, i))559			goto out_free_tx_ltb;560	}561 562	adapter->rx_queue.index = 0;563	adapter->rx_queue.num_slots = rxq_entries;564	adapter->rx_queue.toggle = 1;565 566	mac_address = ether_addr_to_u64(netdev->dev_addr);567 568	rxq_desc.fields.flags_len = IBMVETH_BUF_VALID |569					adapter->rx_queue.queue_len;570	rxq_desc.fields.address = adapter->rx_queue.queue_dma;571 572	netdev_dbg(netdev, "buffer list @ 0x%p\n", adapter->buffer_list_addr);573	netdev_dbg(netdev, "filter list @ 0x%p\n", adapter->filter_list_addr);574	netdev_dbg(netdev, "receive q   @ 0x%p\n", adapter->rx_queue.queue_addr);575 576	h_vio_signal(adapter->vdev->unit_address, VIO_IRQ_DISABLE);577 578	lpar_rc = ibmveth_register_logical_lan(adapter, rxq_desc, mac_address);579 580	if (lpar_rc != H_SUCCESS) {581		netdev_err(netdev, "h_register_logical_lan failed with %ld\n",582			   lpar_rc);583		netdev_err(netdev, "buffer TCE:0x%llx filter TCE:0x%llx rxq "584			   "desc:0x%llx MAC:0x%llx\n",585				     adapter->buffer_list_dma,586				     adapter->filter_list_dma,587				     rxq_desc.desc,588				     mac_address);589		rc = -ENONET;590		goto out_unmap_filter_list;591	}592 593	for (i = 0; i < IBMVETH_NUM_BUFF_POOLS; i++) {594		if (!adapter->rx_buff_pool[i].active)595			continue;596		if (ibmveth_alloc_buffer_pool(&adapter->rx_buff_pool[i])) {597			netdev_err(netdev, "unable to alloc pool\n");598			adapter->rx_buff_pool[i].active = 0;599			rc = -ENOMEM;600			goto out_free_buffer_pools;601		}602	}603 604	netdev_dbg(netdev, "registering irq 0x%x\n", netdev->irq);605	rc = request_irq(netdev->irq, ibmveth_interrupt, 0, netdev->name,606			 netdev);607	if (rc != 0) {608		netdev_err(netdev, "unable to request irq 0x%x, rc %d\n",609			   netdev->irq, rc);610		do {611			lpar_rc = h_free_logical_lan(adapter->vdev->unit_address);612		} while (H_IS_LONG_BUSY(lpar_rc) || (lpar_rc == H_BUSY));613 614		goto out_free_buffer_pools;615	}616 617	rc = -ENOMEM;618 619	netdev_dbg(netdev, "initial replenish cycle\n");620	ibmveth_interrupt(netdev->irq, netdev);621 622	netif_tx_start_all_queues(netdev);623 624	netdev_dbg(netdev, "open complete\n");625 626	return 0;627 628out_free_buffer_pools:629	while (--i >= 0) {630		if (adapter->rx_buff_pool[i].active)631			ibmveth_free_buffer_pool(adapter,632						 &adapter->rx_buff_pool[i]);633	}634out_unmap_filter_list:635	dma_unmap_single(dev, adapter->filter_list_dma, 4096,636			 DMA_BIDIRECTIONAL);637 638out_free_tx_ltb:639	while (--i >= 0) {640		ibmveth_free_tx_ltb(adapter, i);641	}642 643out_unmap_buffer_list:644	dma_unmap_single(dev, adapter->buffer_list_dma, 4096,645			 DMA_BIDIRECTIONAL);646out_free_queue_mem:647	dma_free_coherent(dev, adapter->rx_queue.queue_len,648			  adapter->rx_queue.queue_addr,649			  adapter->rx_queue.queue_dma);650out_free_filter_list:651	free_page((unsigned long)adapter->filter_list_addr);652out_free_buffer_list:653	free_page((unsigned long)adapter->buffer_list_addr);654out:655	napi_disable(&adapter->napi);656	return rc;657}658 659static int ibmveth_close(struct net_device *netdev)660{661	struct ibmveth_adapter *adapter = netdev_priv(netdev);662	struct device *dev = &adapter->vdev->dev;663	long lpar_rc;664	int i;665 666	netdev_dbg(netdev, "close starting\n");667 668	napi_disable(&adapter->napi);669 670	netif_tx_stop_all_queues(netdev);671 672	h_vio_signal(adapter->vdev->unit_address, VIO_IRQ_DISABLE);673 674	do {675		lpar_rc = h_free_logical_lan(adapter->vdev->unit_address);676	} while (H_IS_LONG_BUSY(lpar_rc) || (lpar_rc == H_BUSY));677 678	if (lpar_rc != H_SUCCESS) {679		netdev_err(netdev, "h_free_logical_lan failed with %lx, "680			   "continuing with close\n", lpar_rc);681	}682 683	free_irq(netdev->irq, netdev);684 685	ibmveth_update_rx_no_buffer(adapter);686 687	dma_unmap_single(dev, adapter->buffer_list_dma, 4096,688			 DMA_BIDIRECTIONAL);689	free_page((unsigned long)adapter->buffer_list_addr);690 691	dma_unmap_single(dev, adapter->filter_list_dma, 4096,692			 DMA_BIDIRECTIONAL);693	free_page((unsigned long)adapter->filter_list_addr);694 695	dma_free_coherent(dev, adapter->rx_queue.queue_len,696			  adapter->rx_queue.queue_addr,697			  adapter->rx_queue.queue_dma);698 699	for (i = 0; i < IBMVETH_NUM_BUFF_POOLS; i++)700		if (adapter->rx_buff_pool[i].active)701			ibmveth_free_buffer_pool(adapter,702						 &adapter->rx_buff_pool[i]);703 704	for (i = 0; i < netdev->real_num_tx_queues; i++)705		ibmveth_free_tx_ltb(adapter, i);706 707	netdev_dbg(netdev, "close complete\n");708 709	return 0;710}711 712static int ibmveth_set_link_ksettings(struct net_device *dev,713				      const struct ethtool_link_ksettings *cmd)714{715	struct ibmveth_adapter *adapter = netdev_priv(dev);716 717	return ethtool_virtdev_set_link_ksettings(dev, cmd,718						  &adapter->speed,719						  &adapter->duplex);720}721 722static int ibmveth_get_link_ksettings(struct net_device *dev,723				      struct ethtool_link_ksettings *cmd)724{725	struct ibmveth_adapter *adapter = netdev_priv(dev);726 727	cmd->base.speed = adapter->speed;728	cmd->base.duplex = adapter->duplex;729	cmd->base.port = PORT_OTHER;730 731	return 0;732}733 734static void ibmveth_init_link_settings(struct net_device *dev)735{736	struct ibmveth_adapter *adapter = netdev_priv(dev);737 738	adapter->speed = SPEED_1000;739	adapter->duplex = DUPLEX_FULL;740}741 742static void netdev_get_drvinfo(struct net_device *dev,743			       struct ethtool_drvinfo *info)744{745	strscpy(info->driver, ibmveth_driver_name, sizeof(info->driver));746	strscpy(info->version, ibmveth_driver_version, sizeof(info->version));747}748 749static netdev_features_t ibmveth_fix_features(struct net_device *dev,750	netdev_features_t features)751{752	/*753	 * Since the ibmveth firmware interface does not have the754	 * concept of separate tx/rx checksum offload enable, if rx755	 * checksum is disabled we also have to disable tx checksum756	 * offload. Once we disable rx checksum offload, we are no757	 * longer allowed to send tx buffers that are not properly758	 * checksummed.759	 */760 761	if (!(features & NETIF_F_RXCSUM))762		features &= ~NETIF_F_CSUM_MASK;763 764	return features;765}766 767static int ibmveth_set_csum_offload(struct net_device *dev, u32 data)768{769	struct ibmveth_adapter *adapter = netdev_priv(dev);770	unsigned long set_attr, clr_attr, ret_attr;771	unsigned long set_attr6, clr_attr6;772	long ret, ret4, ret6;773	int rc1 = 0, rc2 = 0;774	int restart = 0;775 776	if (netif_running(dev)) {777		restart = 1;778		ibmveth_close(dev);779	}780 781	set_attr = 0;782	clr_attr = 0;783	set_attr6 = 0;784	clr_attr6 = 0;785 786	if (data) {787		set_attr = IBMVETH_ILLAN_IPV4_TCP_CSUM;788		set_attr6 = IBMVETH_ILLAN_IPV6_TCP_CSUM;789	} else {790		clr_attr = IBMVETH_ILLAN_IPV4_TCP_CSUM;791		clr_attr6 = IBMVETH_ILLAN_IPV6_TCP_CSUM;792	}793 794	ret = h_illan_attributes(adapter->vdev->unit_address, 0, 0, &ret_attr);795 796	if (ret == H_SUCCESS &&797	    (ret_attr & IBMVETH_ILLAN_PADDED_PKT_CSUM)) {798		ret4 = h_illan_attributes(adapter->vdev->unit_address, clr_attr,799					 set_attr, &ret_attr);800 801		if (ret4 != H_SUCCESS) {802			netdev_err(dev, "unable to change IPv4 checksum "803					"offload settings. %d rc=%ld\n",804					data, ret4);805 806			h_illan_attributes(adapter->vdev->unit_address,807					   set_attr, clr_attr, &ret_attr);808 809			if (data == 1)810				dev->features &= ~NETIF_F_IP_CSUM;811 812		} else {813			adapter->fw_ipv4_csum_support = data;814		}815 816		ret6 = h_illan_attributes(adapter->vdev->unit_address,817					 clr_attr6, set_attr6, &ret_attr);818 819		if (ret6 != H_SUCCESS) {820			netdev_err(dev, "unable to change IPv6 checksum "821					"offload settings. %d rc=%ld\n",822					data, ret6);823 824			h_illan_attributes(adapter->vdev->unit_address,825					   set_attr6, clr_attr6, &ret_attr);826 827			if (data == 1)828				dev->features &= ~NETIF_F_IPV6_CSUM;829 830		} else831			adapter->fw_ipv6_csum_support = data;832 833		if (ret4 == H_SUCCESS || ret6 == H_SUCCESS)834			adapter->rx_csum = data;835		else836			rc1 = -EIO;837	} else {838		rc1 = -EIO;839		netdev_err(dev, "unable to change checksum offload settings."840				     " %d rc=%ld ret_attr=%lx\n", data, ret,841				     ret_attr);842	}843 844	if (restart)845		rc2 = ibmveth_open(dev);846 847	return rc1 ? rc1 : rc2;848}849 850static int ibmveth_set_tso(struct net_device *dev, u32 data)851{852	struct ibmveth_adapter *adapter = netdev_priv(dev);853	unsigned long set_attr, clr_attr, ret_attr;854	long ret1, ret2;855	int rc1 = 0, rc2 = 0;856	int restart = 0;857 858	if (netif_running(dev)) {859		restart = 1;860		ibmveth_close(dev);861	}862 863	set_attr = 0;864	clr_attr = 0;865 866	if (data)867		set_attr = IBMVETH_ILLAN_LRG_SR_ENABLED;868	else869		clr_attr = IBMVETH_ILLAN_LRG_SR_ENABLED;870 871	ret1 = h_illan_attributes(adapter->vdev->unit_address, 0, 0, &ret_attr);872 873	if (ret1 == H_SUCCESS && (ret_attr & IBMVETH_ILLAN_LRG_SND_SUPPORT) &&874	    !old_large_send) {875		ret2 = h_illan_attributes(adapter->vdev->unit_address, clr_attr,876					  set_attr, &ret_attr);877 878		if (ret2 != H_SUCCESS) {879			netdev_err(dev, "unable to change tso settings. %d rc=%ld\n",880				   data, ret2);881 882			h_illan_attributes(adapter->vdev->unit_address,883					   set_attr, clr_attr, &ret_attr);884 885			if (data == 1)886				dev->features &= ~(NETIF_F_TSO | NETIF_F_TSO6);887			rc1 = -EIO;888 889		} else {890			adapter->fw_large_send_support = data;891			adapter->large_send = data;892		}893	} else {894		/* Older firmware version of large send offload does not895		 * support tcp6/ipv6896		 */897		if (data == 1) {898			dev->features &= ~NETIF_F_TSO6;899			netdev_info(dev, "TSO feature requires all partitions to have updated driver");900		}901		adapter->large_send = data;902	}903 904	if (restart)905		rc2 = ibmveth_open(dev);906 907	return rc1 ? rc1 : rc2;908}909 910static int ibmveth_set_features(struct net_device *dev,911	netdev_features_t features)912{913	struct ibmveth_adapter *adapter = netdev_priv(dev);914	int rx_csum = !!(features & NETIF_F_RXCSUM);915	int large_send = !!(features & (NETIF_F_TSO | NETIF_F_TSO6));916	int rc1 = 0, rc2 = 0;917 918	if (rx_csum != adapter->rx_csum) {919		rc1 = ibmveth_set_csum_offload(dev, rx_csum);920		if (rc1 && !adapter->rx_csum)921			dev->features =922				features & ~(NETIF_F_CSUM_MASK |923					     NETIF_F_RXCSUM);924	}925 926	if (large_send != adapter->large_send) {927		rc2 = ibmveth_set_tso(dev, large_send);928		if (rc2 && !adapter->large_send)929			dev->features =930				features & ~(NETIF_F_TSO | NETIF_F_TSO6);931	}932 933	return rc1 ? rc1 : rc2;934}935 936static void ibmveth_get_strings(struct net_device *dev, u32 stringset, u8 *data)937{938	int i;939 940	if (stringset != ETH_SS_STATS)941		return;942 943	for (i = 0; i < ARRAY_SIZE(ibmveth_stats); i++, data += ETH_GSTRING_LEN)944		memcpy(data, ibmveth_stats[i].name, ETH_GSTRING_LEN);945}946 947static int ibmveth_get_sset_count(struct net_device *dev, int sset)948{949	switch (sset) {950	case ETH_SS_STATS:951		return ARRAY_SIZE(ibmveth_stats);952	default:953		return -EOPNOTSUPP;954	}955}956 957static void ibmveth_get_ethtool_stats(struct net_device *dev,958				      struct ethtool_stats *stats, u64 *data)959{960	int i;961	struct ibmveth_adapter *adapter = netdev_priv(dev);962 963	for (i = 0; i < ARRAY_SIZE(ibmveth_stats); i++)964		data[i] = IBMVETH_GET_STAT(adapter, ibmveth_stats[i].offset);965}966 967static void ibmveth_get_channels(struct net_device *netdev,968				 struct ethtool_channels *channels)969{970	channels->max_tx = ibmveth_real_max_tx_queues();971	channels->tx_count = netdev->real_num_tx_queues;972 973	channels->max_rx = netdev->real_num_rx_queues;974	channels->rx_count = netdev->real_num_rx_queues;975}976 977static int ibmveth_set_channels(struct net_device *netdev,978				struct ethtool_channels *channels)979{980	struct ibmveth_adapter *adapter = netdev_priv(netdev);981	unsigned int old = netdev->real_num_tx_queues,982		     goal = channels->tx_count;983	int rc, i;984 985	/* If ndo_open has not been called yet then don't allocate, just set986	 * desired netdev_queue's and return987	 */988	if (!(netdev->flags & IFF_UP))989		return netif_set_real_num_tx_queues(netdev, goal);990 991	/* We have IBMVETH_MAX_QUEUES netdev_queue's allocated992	 * but we may need to alloc/free the ltb's.993	 */994	netif_tx_stop_all_queues(netdev);995 996	/* Allocate any queue that we need */997	for (i = old; i < goal; i++) {998		if (adapter->tx_ltb_ptr[i])999			continue;1000 1001		rc = ibmveth_allocate_tx_ltb(adapter, i);1002		if (!rc)1003			continue;1004 1005		/* if something goes wrong, free everything we just allocated */1006		netdev_err(netdev, "Failed to allocate more tx queues, returning to %d queues\n",1007			   old);1008		goal = old;1009		old = i;1010		break;1011	}1012	rc = netif_set_real_num_tx_queues(netdev, goal);1013	if (rc) {1014		netdev_err(netdev, "Failed to set real tx queues, returning to %d queues\n",1015			   old);1016		goal = old;1017		old = i;1018	}1019	/* Free any that are no longer needed */1020	for (i = old; i > goal; i--) {1021		if (adapter->tx_ltb_ptr[i - 1])1022			ibmveth_free_tx_ltb(adapter, i - 1);1023	}1024 1025	netif_tx_wake_all_queues(netdev);1026 1027	return rc;1028}1029 1030static const struct ethtool_ops netdev_ethtool_ops = {1031	.get_drvinfo		         = netdev_get_drvinfo,1032	.get_link		         = ethtool_op_get_link,1033	.get_strings		         = ibmveth_get_strings,1034	.get_sset_count		         = ibmveth_get_sset_count,1035	.get_ethtool_stats	         = ibmveth_get_ethtool_stats,1036	.get_link_ksettings	         = ibmveth_get_link_ksettings,1037	.set_link_ksettings              = ibmveth_set_link_ksettings,1038	.get_channels			 = ibmveth_get_channels,1039	.set_channels			 = ibmveth_set_channels1040};1041 1042static int ibmveth_ioctl(struct net_device *dev, struct ifreq *ifr, int cmd)1043{1044	return -EOPNOTSUPP;1045}1046 1047static int ibmveth_send(struct ibmveth_adapter *adapter,1048			unsigned long desc, unsigned long mss)1049{1050	unsigned long correlator;1051	unsigned int retry_count;1052	unsigned long ret;1053 1054	/*1055	 * The retry count sets a maximum for the number of broadcast and1056	 * multicast destinations within the system.1057	 */1058	retry_count = 1024;1059	correlator = 0;1060	do {1061		ret = h_send_logical_lan(adapter->vdev->unit_address, desc,1062					 correlator, &correlator, mss,1063					 adapter->fw_large_send_support);1064	} while ((ret == H_BUSY) && (retry_count--));1065 1066	if (ret != H_SUCCESS && ret != H_DROPPED) {1067		netdev_err(adapter->netdev, "tx: h_send_logical_lan failed "1068			   "with rc=%ld\n", ret);1069		return 1;1070	}1071 1072	return 0;1073}1074 1075static int ibmveth_is_packet_unsupported(struct sk_buff *skb,1076					 struct net_device *netdev)1077{1078	struct ethhdr *ether_header;1079	int ret = 0;1080 1081	ether_header = eth_hdr(skb);1082 1083	if (ether_addr_equal(ether_header->h_dest, netdev->dev_addr)) {1084		netdev_dbg(netdev, "veth doesn't support loopback packets, dropping packet.\n");1085		netdev->stats.tx_dropped++;1086		ret = -EOPNOTSUPP;1087	}1088 1089	return ret;1090}1091 1092static netdev_tx_t ibmveth_start_xmit(struct sk_buff *skb,1093				      struct net_device *netdev)1094{1095	struct ibmveth_adapter *adapter = netdev_priv(netdev);1096	unsigned int desc_flags, total_bytes;1097	union ibmveth_buf_desc desc;1098	int i, queue_num = skb_get_queue_mapping(skb);1099	unsigned long mss = 0;1100 1101	if (ibmveth_is_packet_unsupported(skb, netdev))1102		goto out;1103	/* veth can't checksum offload UDP */1104	if (skb->ip_summed == CHECKSUM_PARTIAL &&1105	    ((skb->protocol == htons(ETH_P_IP) &&1106	      ip_hdr(skb)->protocol != IPPROTO_TCP) ||1107	     (skb->protocol == htons(ETH_P_IPV6) &&1108	      ipv6_hdr(skb)->nexthdr != IPPROTO_TCP)) &&1109	    skb_checksum_help(skb)) {1110 1111		netdev_err(netdev, "tx: failed to checksum packet\n");1112		netdev->stats.tx_dropped++;1113		goto out;1114	}1115 1116	desc_flags = IBMVETH_BUF_VALID;1117 1118	if (skb->ip_summed == CHECKSUM_PARTIAL) {1119		unsigned char *buf = skb_transport_header(skb) +1120						skb->csum_offset;1121 1122		desc_flags |= (IBMVETH_BUF_NO_CSUM | IBMVETH_BUF_CSUM_GOOD);1123 1124		/* Need to zero out the checksum */1125		buf[0] = 0;1126		buf[1] = 0;1127 1128		if (skb_is_gso(skb) && adapter->fw_large_send_support)1129			desc_flags |= IBMVETH_BUF_LRG_SND;1130	}1131 1132	if (skb->ip_summed == CHECKSUM_PARTIAL && skb_is_gso(skb)) {1133		if (adapter->fw_large_send_support) {1134			mss = (unsigned long)skb_shinfo(skb)->gso_size;1135			adapter->tx_large_packets++;1136		} else if (!skb_is_gso_v6(skb)) {1137			/* Put -1 in the IP checksum to tell phyp it1138			 * is a largesend packet. Put the mss in1139			 * the TCP checksum.1140			 */1141			ip_hdr(skb)->check = 0xffff;1142			tcp_hdr(skb)->check =1143				cpu_to_be16(skb_shinfo(skb)->gso_size);1144			adapter->tx_large_packets++;1145		}1146	}1147 1148	/* Copy header into mapped buffer */1149	if (unlikely(skb->len > adapter->tx_ltb_size)) {1150		netdev_err(adapter->netdev, "tx: packet size (%u) exceeds ltb (%u)\n",1151			   skb->len, adapter->tx_ltb_size);1152		netdev->stats.tx_dropped++;1153		goto out;1154	}1155	memcpy(adapter->tx_ltb_ptr[queue_num], skb->data, skb_headlen(skb));1156	total_bytes = skb_headlen(skb);1157	/* Copy frags into mapped buffers */1158	for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {1159		const skb_frag_t *frag = &skb_shinfo(skb)->frags[i];1160 1161		memcpy(adapter->tx_ltb_ptr[queue_num] + total_bytes,1162		       skb_frag_address_safe(frag), skb_frag_size(frag));1163		total_bytes += skb_frag_size(frag);1164	}1165 1166	if (unlikely(total_bytes != skb->len)) {1167		netdev_err(adapter->netdev, "tx: incorrect packet len copied into ltb (%u != %u)\n",1168			   skb->len, total_bytes);1169		netdev->stats.tx_dropped++;1170		goto out;1171	}1172	desc.fields.flags_len = desc_flags | skb->len;1173	desc.fields.address = adapter->tx_ltb_dma[queue_num];1174	/* finish writing to long_term_buff before VIOS accessing it */1175	dma_wmb();1176 1177	if (ibmveth_send(adapter, desc.desc, mss)) {1178		adapter->tx_send_failed++;1179		netdev->stats.tx_dropped++;1180	} else {1181		netdev->stats.tx_packets++;1182		netdev->stats.tx_bytes += skb->len;1183	}1184 1185out:1186	dev_consume_skb_any(skb);1187	return NETDEV_TX_OK;1188 1189 1190}1191 1192static void ibmveth_rx_mss_helper(struct sk_buff *skb, u16 mss, int lrg_pkt)1193{1194	struct tcphdr *tcph;1195	int offset = 0;1196	int hdr_len;1197 1198	/* only TCP packets will be aggregated */1199	if (skb->protocol == htons(ETH_P_IP)) {1200		struct iphdr *iph = (struct iphdr *)skb->data;1201 1202		if (iph->protocol == IPPROTO_TCP) {1203			offset = iph->ihl * 4;1204			skb_shinfo(skb)->gso_type = SKB_GSO_TCPV4;1205		} else {1206			return;1207		}1208	} else if (skb->protocol == htons(ETH_P_IPV6)) {1209		struct ipv6hdr *iph6 = (struct ipv6hdr *)skb->data;1210 1211		if (iph6->nexthdr == IPPROTO_TCP) {1212			offset = sizeof(struct ipv6hdr);1213			skb_shinfo(skb)->gso_type = SKB_GSO_TCPV6;1214		} else {1215			return;1216		}1217	} else {1218		return;1219	}1220	/* if mss is not set through Large Packet bit/mss in rx buffer,1221	 * expect that the mss will be written to the tcp header checksum.1222	 */1223	tcph = (struct tcphdr *)(skb->data + offset);1224	if (lrg_pkt) {1225		skb_shinfo(skb)->gso_size = mss;1226	} else if (offset) {1227		skb_shinfo(skb)->gso_size = ntohs(tcph->check);1228		tcph->check = 0;1229	}1230 1231	if (skb_shinfo(skb)->gso_size) {1232		hdr_len = offset + tcph->doff * 4;1233		skb_shinfo(skb)->gso_segs =1234				DIV_ROUND_UP(skb->len - hdr_len,1235					     skb_shinfo(skb)->gso_size);1236	}1237}1238 1239static void ibmveth_rx_csum_helper(struct sk_buff *skb,1240				   struct ibmveth_adapter *adapter)1241{1242	struct iphdr *iph = NULL;1243	struct ipv6hdr *iph6 = NULL;1244	__be16 skb_proto = 0;1245	u16 iphlen = 0;1246	u16 iph_proto = 0;1247	u16 tcphdrlen = 0;1248 1249	skb_proto = be16_to_cpu(skb->protocol);1250 1251	if (skb_proto == ETH_P_IP) {1252		iph = (struct iphdr *)skb->data;1253 1254		/* If the IP checksum is not offloaded and if the packet1255		 *  is large send, the checksum must be rebuilt.1256		 */1257		if (iph->check == 0xffff) {1258			iph->check = 0;1259			iph->check = ip_fast_csum((unsigned char *)iph,1260						  iph->ihl);1261		}1262 1263		iphlen = iph->ihl * 4;1264		iph_proto = iph->protocol;1265	} else if (skb_proto == ETH_P_IPV6) {1266		iph6 = (struct ipv6hdr *)skb->data;1267		iphlen = sizeof(struct ipv6hdr);1268		iph_proto = iph6->nexthdr;1269	}1270 1271	/* When CSO is enabled the TCP checksum may have be set to NULL by1272	 * the sender given that we zeroed out TCP checksum field in1273	 * transmit path (refer ibmveth_start_xmit routine). In this case set1274	 * up CHECKSUM_PARTIAL. If the packet is forwarded, the checksum will1275	 * then be recalculated by the destination NIC (CSO must be enabled1276	 * on the destination NIC).1277	 *1278	 * In an OVS environment, when a flow is not cached, specifically for a1279	 * new TCP connection, the first packet information is passed up to1280	 * the user space for finding a flow. During this process, OVS computes1281	 * checksum on the first packet when CHECKSUM_PARTIAL flag is set.1282	 *1283	 * So, re-compute TCP pseudo header checksum.1284	 */1285 1286	if (iph_proto == IPPROTO_TCP) {1287		struct tcphdr *tcph = (struct tcphdr *)(skb->data + iphlen);1288 1289		if (tcph->check == 0x0000) {1290			/* Recompute TCP pseudo header checksum  */1291			tcphdrlen = skb->len - iphlen;1292			if (skb_proto == ETH_P_IP)1293				tcph->check =1294				 ~csum_tcpudp_magic(iph->saddr,1295				iph->daddr, tcphdrlen, iph_proto, 0);1296			else if (skb_proto == ETH_P_IPV6)1297				tcph->check =1298				 ~csum_ipv6_magic(&iph6->saddr,1299				&iph6->daddr, tcphdrlen, iph_proto, 0);1300			/* Setup SKB fields for checksum offload */1301			skb_partial_csum_set(skb, iphlen,1302					     offsetof(struct tcphdr, check));1303			skb_reset_network_header(skb);1304		}1305	}1306}1307 1308static int ibmveth_poll(struct napi_struct *napi, int budget)1309{1310	struct ibmveth_adapter *adapter =1311			container_of(napi, struct ibmveth_adapter, napi);1312	struct net_device *netdev = adapter->netdev;1313	int frames_processed = 0;1314	unsigned long lpar_rc;1315	u16 mss = 0;1316 1317restart_poll:1318	while (frames_processed < budget) {1319		if (!ibmveth_rxq_pending_buffer(adapter))1320			break;1321 1322		smp_rmb();1323		if (!ibmveth_rxq_buffer_valid(adapter)) {1324			wmb(); /* suggested by larson1 */1325			adapter->rx_invalid_buffer++;1326			netdev_dbg(netdev, "recycling invalid buffer\n");1327			ibmveth_rxq_harvest_buffer(adapter, true);1328		} else {1329			struct sk_buff *skb, *new_skb;1330			int length = ibmveth_rxq_frame_length(adapter);1331			int offset = ibmveth_rxq_frame_offset(adapter);1332			int csum_good = ibmveth_rxq_csum_good(adapter);1333			int lrg_pkt = ibmveth_rxq_large_packet(adapter);1334			__sum16 iph_check = 0;1335 1336			skb = ibmveth_rxq_get_buffer(adapter);1337 1338			/* if the large packet bit is set in the rx queue1339			 * descriptor, the mss will be written by PHYP eight1340			 * bytes from the start of the rx buffer, which is1341			 * skb->data at this stage1342			 */1343			if (lrg_pkt) {1344				__be64 *rxmss = (__be64 *)(skb->data + 8);1345 1346				mss = (u16)be64_to_cpu(*rxmss);1347			}1348 1349			new_skb = NULL;1350			if (length < rx_copybreak)1351				new_skb = netdev_alloc_skb(netdev, length);1352 1353			if (new_skb) {1354				skb_copy_to_linear_data(new_skb,1355							skb->data + offset,1356							length);1357				if (rx_flush)1358					ibmveth_flush_buffer(skb->data,1359						length + offset);1360				ibmveth_rxq_harvest_buffer(adapter, true);1361				skb = new_skb;1362			} else {1363				ibmveth_rxq_harvest_buffer(adapter, false);1364				skb_reserve(skb, offset);1365			}1366 1367			skb_put(skb, length);1368			skb->protocol = eth_type_trans(skb, netdev);1369 1370			/* PHYP without PLSO support places a -1 in the ip1371			 * checksum for large send frames.1372			 */1373			if (skb->protocol == cpu_to_be16(ETH_P_IP)) {1374				struct iphdr *iph = (struct iphdr *)skb->data;1375 1376				iph_check = iph->check;1377			}1378 1379			if ((length > netdev->mtu + ETH_HLEN) ||1380			    lrg_pkt || iph_check == 0xffff) {1381				ibmveth_rx_mss_helper(skb, mss, lrg_pkt);1382				adapter->rx_large_packets++;1383			}1384 1385			if (csum_good) {1386				skb->ip_summed = CHECKSUM_UNNECESSARY;1387				ibmveth_rx_csum_helper(skb, adapter);1388			}1389 1390			napi_gro_receive(napi, skb);	/* send it up */1391 1392			netdev->stats.rx_packets++;1393			netdev->stats.rx_bytes += length;1394			frames_processed++;1395		}1396	}1397 1398	ibmveth_replenish_task(adapter);1399 1400	if (frames_processed == budget)1401		goto out;1402 1403	if (!napi_complete_done(napi, frames_processed))1404		goto out;1405 1406	/* We think we are done - reenable interrupts,1407	 * then check once more to make sure we are done.1408	 */1409	lpar_rc = h_vio_signal(adapter->vdev->unit_address, VIO_IRQ_ENABLE);1410	BUG_ON(lpar_rc != H_SUCCESS);1411 1412	if (ibmveth_rxq_pending_buffer(adapter) && napi_schedule(napi)) {1413		lpar_rc = h_vio_signal(adapter->vdev->unit_address,1414				       VIO_IRQ_DISABLE);1415		goto restart_poll;1416	}1417 1418out:1419	return frames_processed;1420}1421 1422static irqreturn_t ibmveth_interrupt(int irq, void *dev_instance)1423{1424	struct net_device *netdev = dev_instance;1425	struct ibmveth_adapter *adapter = netdev_priv(netdev);1426	unsigned long lpar_rc;1427 1428	if (napi_schedule_prep(&adapter->napi)) {1429		lpar_rc = h_vio_signal(adapter->vdev->unit_address,1430				       VIO_IRQ_DISABLE);1431		BUG_ON(lpar_rc != H_SUCCESS);1432		__napi_schedule(&adapter->napi);1433	}1434	return IRQ_HANDLED;1435}1436 1437static void ibmveth_set_multicast_list(struct net_device *netdev)1438{1439	struct ibmveth_adapter *adapter = netdev_priv(netdev);1440	unsigned long lpar_rc;1441 1442	if ((netdev->flags & IFF_PROMISC) ||1443	    (netdev_mc_count(netdev) > adapter->mcastFilterSize)) {1444		lpar_rc = h_multicast_ctrl(adapter->vdev->unit_address,1445					   IbmVethMcastEnableRecv |1446					   IbmVethMcastDisableFiltering,1447					   0);1448		if (lpar_rc != H_SUCCESS) {1449			netdev_err(netdev, "h_multicast_ctrl rc=%ld when "1450				   "entering promisc mode\n", lpar_rc);1451		}1452	} else {1453		struct netdev_hw_addr *ha;1454		/* clear the filter table & disable filtering */1455		lpar_rc = h_multicast_ctrl(adapter->vdev->unit_address,1456					   IbmVethMcastEnableRecv |1457					   IbmVethMcastDisableFiltering |1458					   IbmVethMcastClearFilterTable,1459					   0);1460		if (lpar_rc != H_SUCCESS) {1461			netdev_err(netdev, "h_multicast_ctrl rc=%ld when "1462				   "attempting to clear filter table\n",1463				   lpar_rc);1464		}1465		/* add the addresses to the filter table */1466		netdev_for_each_mc_addr(ha, netdev) {1467			/* add the multicast address to the filter table */1468			u64 mcast_addr;1469			mcast_addr = ether_addr_to_u64(ha->addr);1470			lpar_rc = h_multicast_ctrl(adapter->vdev->unit_address,1471						   IbmVethMcastAddFilter,1472						   mcast_addr);1473			if (lpar_rc != H_SUCCESS) {1474				netdev_err(netdev, "h_multicast_ctrl rc=%ld "1475					   "when adding an entry to the filter "1476					   "table\n", lpar_rc);1477			}1478		}1479 1480		/* re-enable filtering */1481		lpar_rc = h_multicast_ctrl(adapter->vdev->unit_address,1482					   IbmVethMcastEnableFiltering,1483					   0);1484		if (lpar_rc != H_SUCCESS) {1485			netdev_err(netdev, "h_multicast_ctrl rc=%ld when "1486				   "enabling filtering\n", lpar_rc);1487		}1488	}1489}1490 1491static int ibmveth_change_mtu(struct net_device *dev, int new_mtu)1492{1493	struct ibmveth_adapter *adapter = netdev_priv(dev);1494	struct vio_dev *viodev = adapter->vdev;1495	int new_mtu_oh = new_mtu + IBMVETH_BUFF_OH;1496	int i, rc;1497	int need_restart = 0;1498 1499	for (i = 0; i < IBMVETH_NUM_BUFF_POOLS; i++)1500		if (new_mtu_oh <= adapter->rx_buff_pool[i].buff_size)1501			break;1502 1503	if (i == IBMVETH_NUM_BUFF_POOLS)1504		return -EINVAL;1505 1506	/* Deactivate all the buffer pools so that the next loop can activate1507	   only the buffer pools necessary to hold the new MTU */1508	if (netif_running(adapter->netdev)) {1509		need_restart = 1;1510		ibmveth_close(adapter->netdev);1511	}1512 1513	/* Look for an active buffer pool that can hold the new MTU */1514	for (i = 0; i < IBMVETH_NUM_BUFF_POOLS; i++) {1515		adapter->rx_buff_pool[i].active = 1;1516 1517		if (new_mtu_oh <= adapter->rx_buff_pool[i].buff_size) {1518			WRITE_ONCE(dev->mtu, new_mtu);1519			vio_cmo_set_dev_desired(viodev,1520						ibmveth_get_desired_dma1521						(viodev));1522			if (need_restart) {1523				return ibmveth_open(adapter->netdev);1524			}1525			return 0;1526		}1527	}1528 1529	if (need_restart && (rc = ibmveth_open(adapter->netdev)))1530		return rc;1531 1532	return -EINVAL;1533}1534 1535#ifdef CONFIG_NET_POLL_CONTROLLER1536static void ibmveth_poll_controller(struct net_device *dev)1537{1538	ibmveth_replenish_task(netdev_priv(dev));1539	ibmveth_interrupt(dev->irq, dev);1540}1541#endif1542 1543/**1544 * ibmveth_get_desired_dma - Calculate IO memory desired by the driver1545 *1546 * @vdev: struct vio_dev for the device whose desired IO mem is to be returned1547 *1548 * Return value:1549 *	Number of bytes of IO data the driver will need to perform well.1550 */1551static unsigned long ibmveth_get_desired_dma(struct vio_dev *vdev)1552{1553	struct net_device *netdev = dev_get_drvdata(&vdev->dev);1554	struct ibmveth_adapter *adapter;1555	struct iommu_table *tbl;1556	unsigned long ret;1557	int i;1558	int rxqentries = 1;1559 1560	tbl = get_iommu_table_base(&vdev->dev);1561 1562	/* netdev inits at probe time along with the structures we need below*/1563	if (netdev == NULL)1564		return IOMMU_PAGE_ALIGN(IBMVETH_IO_ENTITLEMENT_DEFAULT, tbl);1565 1566	adapter = netdev_priv(netdev);1567 1568	ret = IBMVETH_BUFF_LIST_SIZE + IBMVETH_FILT_LIST_SIZE;1569	ret += IOMMU_PAGE_ALIGN(netdev->mtu, tbl);1570	/* add size of mapped tx buffers */1571	ret += IOMMU_PAGE_ALIGN(IBMVETH_MAX_TX_BUF_SIZE, tbl);1572 1573	for (i = 0; i < IBMVETH_NUM_BUFF_POOLS; i++) {1574		/* add the size of the active receive buffers */1575		if (adapter->rx_buff_pool[i].active)1576			ret +=1577			    adapter->rx_buff_pool[i].size *1578			    IOMMU_PAGE_ALIGN(adapter->rx_buff_pool[i].1579					     buff_size, tbl);1580		rxqentries += adapter->rx_buff_pool[i].size;1581	}1582	/* add the size of the receive queue entries */1583	ret += IOMMU_PAGE_ALIGN(1584		rxqentries * sizeof(struct ibmveth_rx_q_entry), tbl);1585 1586	return ret;1587}1588 1589static int ibmveth_set_mac_addr(struct net_device *dev, void *p)1590{1591	struct ibmveth_adapter *adapter = netdev_priv(dev);1592	struct sockaddr *addr = p;1593	u64 mac_address;1594	int rc;1595 1596	if (!is_valid_ether_addr(addr->sa_data))1597		return -EADDRNOTAVAIL;1598 1599	mac_address = ether_addr_to_u64(addr->sa_data);1600	rc = h_change_logical_lan_mac(adapter->vdev->unit_address, mac_address);1601	if (rc) {1602		netdev_err(adapter->netdev, "h_change_logical_lan_mac failed with rc=%d\n", rc);1603		return rc;1604	}1605 1606	eth_hw_addr_set(dev, addr->sa_data);1607 1608	return 0;1609}1610 1611static const struct net_device_ops ibmveth_netdev_ops = {1612	.ndo_open		= ibmveth_open,1613	.ndo_stop		= ibmveth_close,1614	.ndo_start_xmit		= ibmveth_start_xmit,1615	.ndo_set_rx_mode	= ibmveth_set_multicast_list,1616	.ndo_eth_ioctl		= ibmveth_ioctl,1617	.ndo_change_mtu		= ibmveth_change_mtu,1618	.ndo_fix_features	= ibmveth_fix_features,1619	.ndo_set_features	= ibmveth_set_features,1620	.ndo_validate_addr	= eth_validate_addr,1621	.ndo_set_mac_address    = ibmveth_set_mac_addr,1622#ifdef CONFIG_NET_POLL_CONTROLLER1623	.ndo_poll_controller	= ibmveth_poll_controller,1624#endif1625};1626 1627static int ibmveth_probe(struct vio_dev *dev, const struct vio_device_id *id)1628{1629	int rc, i, mac_len;1630	struct net_device *netdev;1631	struct ibmveth_adapter *adapter;1632	unsigned char *mac_addr_p;1633	__be32 *mcastFilterSize_p;1634	long ret;1635	unsigned long ret_attr;1636 1637	dev_dbg(&dev->dev, "entering ibmveth_probe for UA 0x%x\n",1638		dev->unit_address);1639 1640	mac_addr_p = (unsigned char *)vio_get_attribute(dev, VETH_MAC_ADDR,1641							&mac_len);1642	if (!mac_addr_p) {1643		dev_err(&dev->dev, "Can't find VETH_MAC_ADDR attribute\n");1644		return -EINVAL;1645	}1646	/* Workaround for old/broken pHyp */1647	if (mac_len == 8)1648		mac_addr_p += 2;1649	else if (mac_len != 6) {1650		dev_err(&dev->dev, "VETH_MAC_ADDR attribute wrong len %d\n",1651			mac_len);1652		return -EINVAL;1653	}1654 1655	mcastFilterSize_p = (__be32 *)vio_get_attribute(dev,1656							VETH_MCAST_FILTER_SIZE,1657							NULL);1658	if (!mcastFilterSize_p) {1659		dev_err(&dev->dev, "Can't find VETH_MCAST_FILTER_SIZE "1660			"attribute\n");1661		return -EINVAL;1662	}1663 1664	netdev = alloc_etherdev_mqs(sizeof(struct ibmveth_adapter), IBMVETH_MAX_QUEUES, 1);1665	if (!netdev)1666		return -ENOMEM;1667 1668	adapter = netdev_priv(netdev);1669	dev_set_drvdata(&dev->dev, netdev);1670 1671	adapter->vdev = dev;1672	adapter->netdev = netdev;1673	adapter->mcastFilterSize = be32_to_cpu(*mcastFilterSize_p);1674	ibmveth_init_link_settings(netdev);1675 1676	netif_napi_add_weight(netdev, &adapter->napi, ibmveth_poll, 16);1677 1678	netdev->irq = dev->irq;1679	netdev->netdev_ops = &ibmveth_netdev_ops;1680	netdev->ethtool_ops = &netdev_ethtool_ops;1681	SET_NETDEV_DEV(netdev, &dev->dev);1682	netdev->hw_features = NETIF_F_SG;1683	if (vio_get_attribute(dev, "ibm,illan-options", NULL) != NULL) {1684		netdev->hw_features |= NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM |1685				       NETIF_F_RXCSUM;1686	}1687 1688	netdev->features |= netdev->hw_features;1689 1690	ret = h_illan_attributes(adapter->vdev->unit_address, 0, 0, &ret_attr);1691 1692	/* If running older firmware, TSO should not be enabled by default */1693	if (ret == H_SUCCESS && (ret_attr & IBMVETH_ILLAN_LRG_SND_SUPPORT) &&1694	    !old_large_send) {1695		netdev->hw_features |= NETIF_F_TSO | NETIF_F_TSO6;1696		netdev->features |= netdev->hw_features;1697	} else {1698		netdev->hw_features |= NETIF_F_TSO;1699	}1700 1701	adapter->is_active_trunk = false;1702	if (ret == H_SUCCESS && (ret_attr & IBMVETH_ILLAN_ACTIVE_TRUNK)) {1703		adapter->is_active_trunk = true;1704		netdev->hw_features |= NETIF_F_FRAGLIST;1705		netdev->features |= NETIF_F_FRAGLIST;1706	}1707 1708	netdev->min_mtu = IBMVETH_MIN_MTU;1709	netdev->max_mtu = ETH_MAX_MTU - IBMVETH_BUFF_OH;1710 1711	eth_hw_addr_set(netdev, mac_addr_p);1712 1713	if (firmware_has_feature(FW_FEATURE_CMO))1714		memcpy(pool_count, pool_count_cmo, sizeof(pool_count));1715 1716	for (i = 0; i < IBMVETH_NUM_BUFF_POOLS; i++) {1717		struct kobject *kobj = &adapter->rx_buff_pool[i].kobj;1718		int error;1719 1720		ibmveth_init_buffer_pool(&adapter->rx_buff_pool[i], i,1721					 pool_count[i], pool_size[i],1722					 pool_active[i]);1723		error = kobject_init_and_add(kobj, &ktype_veth_pool,1724					     &dev->dev.kobj, "pool%d", i);1725		if (!error)1726			kobject_uevent(kobj, KOBJ_ADD);1727	}1728 1729	rc = netif_set_real_num_tx_queues(netdev, min(num_online_cpus(),1730						      IBMVETH_DEFAULT_QUEUES));1731	if (rc) {1732		netdev_dbg(netdev, "failed to set number of tx queues rc=%d\n",1733			   rc);1734		free_netdev(netdev);1735		return rc;1736	}1737	adapter->tx_ltb_size = PAGE_ALIGN(IBMVETH_MAX_TX_BUF_SIZE);1738	for (i = 0; i < IBMVETH_MAX_QUEUES; i++)1739		adapter->tx_ltb_ptr[i] = NULL;1740 1741	netdev_dbg(netdev, "adapter @ 0x%p\n", adapter);1742	netdev_dbg(netdev, "registering netdev...\n");1743 1744	ibmveth_set_features(netdev, netdev->features);1745 1746	rc = register_netdev(netdev);1747 1748	if (rc) {1749		netdev_dbg(netdev, "failed to register netdev rc=%d\n", rc);1750		free_netdev(netdev);1751		return rc;1752	}1753 1754	netdev_dbg(netdev, "registered\n");1755 1756	return 0;1757}1758 1759static void ibmveth_remove(struct vio_dev *dev)1760{1761	struct net_device *netdev = dev_get_drvdata(&dev->dev);1762	struct ibmveth_adapter *adapter = netdev_priv(netdev);1763	int i;1764 1765	for (i = 0; i < IBMVETH_NUM_BUFF_POOLS; i++)1766		kobject_put(&adapter->rx_buff_pool[i].kobj);1767 1768	unregister_netdev(netdev);1769 1770	free_netdev(netdev);1771	dev_set_drvdata(&dev->dev, NULL);1772}1773 1774static struct attribute veth_active_attr;1775static struct attribute veth_num_attr;1776static struct attribute veth_size_attr;1777 1778static ssize_t veth_pool_show(struct kobject *kobj,1779			      struct attribute *attr, char *buf)1780{1781	struct ibmveth_buff_pool *pool = container_of(kobj,1782						      struct ibmveth_buff_pool,1783						      kobj);1784 1785	if (attr == &veth_active_attr)1786		return sprintf(buf, "%d\n", pool->active);1787	else if (attr == &veth_num_attr)1788		return sprintf(buf, "%d\n", pool->size);1789	else if (attr == &veth_size_attr)1790		return sprintf(buf, "%d\n", pool->buff_size);1791	return 0;1792}1793 1794static ssize_t veth_pool_store(struct kobject *kobj, struct attribute *attr,1795			       const char *buf, size_t count)1796{1797	struct ibmveth_buff_pool *pool = container_of(kobj,1798						      struct ibmveth_buff_pool,1799						      kobj);1800	struct net_device *netdev = dev_get_drvdata(kobj_to_dev(kobj->parent));1801	struct ibmveth_adapter *adapter = netdev_priv(netdev);1802	long value = simple_strtol(buf, NULL, 10);1803	long rc;1804 1805	if (attr == &veth_active_attr) {1806		if (value && !pool->active) {1807			if (netif_running(netdev)) {1808				if (ibmveth_alloc_buffer_pool(pool)) {1809					netdev_err(netdev,1810						   "unable to alloc pool\n");1811					return -ENOMEM;1812				}1813				pool->active = 1;1814				ibmveth_close(netdev);1815				if ((rc = ibmveth_open(netdev)))1816					return rc;1817			} else {1818				pool->active = 1;1819			}1820		} else if (!value && pool->active) {1821			int mtu = netdev->mtu + IBMVETH_BUFF_OH;1822			int i;1823			/* Make sure there is a buffer pool with buffers that1824			   can hold a packet of the size of the MTU */1825			for (i = 0; i < IBMVETH_NUM_BUFF_POOLS; i++) {1826				if (pool == &adapter->rx_buff_pool[i])1827					continue;1828				if (!adapter->rx_buff_pool[i].active)1829					continue;1830				if (mtu <= adapter->rx_buff_pool[i].buff_size)1831					break;1832			}1833 1834			if (i == IBMVETH_NUM_BUFF_POOLS) {1835				netdev_err(netdev, "no active pool >= MTU\n");1836				return -EPERM;1837			}1838 1839			if (netif_running(netdev)) {1840				ibmveth_close(netdev);1841				pool->active = 0;1842				if ((rc = ibmveth_open(netdev)))1843					return rc;1844			}1845			pool->active = 0;1846		}1847	} else if (attr == &veth_num_attr) {1848		if (value <= 0 || value > IBMVETH_MAX_POOL_COUNT) {1849			return -EINVAL;1850		} else {1851			if (netif_running(netdev)) {1852				ibmveth_close(netdev);1853				pool->size = value;1854				if ((rc = ibmveth_open(netdev)))1855					return rc;1856			} else {1857				pool->size = value;1858			}1859		}1860	} else if (attr == &veth_size_attr) {1861		if (value <= IBMVETH_BUFF_OH || value > IBMVETH_MAX_BUF_SIZE) {1862			return -EINVAL;1863		} else {1864			if (netif_running(netdev)) {1865				ibmveth_close(netdev);1866				pool->buff_size = value;1867				if ((rc = ibmveth_open(netdev)))1868					return rc;1869			} else {1870				pool->buff_size = value;1871			}1872		}1873	}1874 1875	/* kick the interrupt handler to allocate/deallocate pools */1876	ibmveth_interrupt(netdev->irq, netdev);1877	return count;1878}1879 1880 1881#define ATTR(_name, _mode)				\1882	struct attribute veth_##_name##_attr = {	\1883	.name = __stringify(_name), .mode = _mode,	\1884	};1885 1886static ATTR(active, 0644);1887static ATTR(num, 0644);1888static ATTR(size, 0644);1889 1890static struct attribute *veth_pool_attrs[] = {1891	&veth_active_attr,1892	&veth_num_attr,1893	&veth_size_attr,1894	NULL,1895};1896ATTRIBUTE_GROUPS(veth_pool);1897 1898static const struct sysfs_ops veth_pool_ops = {1899	.show   = veth_pool_show,1900	.store  = veth_pool_store,1901};1902 1903static struct kobj_type ktype_veth_pool = {1904	.release        = NULL,1905	.sysfs_ops      = &veth_pool_ops,1906	.default_groups = veth_pool_groups,1907};1908 1909static int ibmveth_resume(struct device *dev)1910{1911	struct net_device *netdev = dev_get_drvdata(dev);1912	ibmveth_interrupt(netdev->irq, netdev);1913	return 0;1914}1915 1916static const struct vio_device_id ibmveth_device_table[] = {1917	{ "network", "IBM,l-lan"},1918	{ "", "" }1919};1920MODULE_DEVICE_TABLE(vio, ibmveth_device_table);1921 1922static const struct dev_pm_ops ibmveth_pm_ops = {1923	.resume = ibmveth_resume1924};1925 1926static struct vio_driver ibmveth_driver = {1927	.id_table	= ibmveth_device_table,1928	.probe		= ibmveth_probe,1929	.remove		= ibmveth_remove,1930	.get_desired_dma = ibmveth_get_desired_dma,1931	.name		= ibmveth_driver_name,1932	.pm		= &ibmveth_pm_ops,1933};1934 1935static int __init ibmveth_module_init(void)1936{1937	printk(KERN_DEBUG "%s: %s %s\n", ibmveth_driver_name,1938	       ibmveth_driver_string, ibmveth_driver_version);1939 1940	return vio_register_driver(&ibmveth_driver);1941}1942 1943static void __exit ibmveth_module_exit(void)1944{1945	vio_unregister_driver(&ibmveth_driver);1946}1947 1948module_init(ibmveth_module_init);1949module_exit(ibmveth_module_exit);1950