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/*
296 * The final 8 bytes of the buffer list is a counter of frames dropped
297 * because there was not a buffer in the buffer list capable of holding
298 * the frame.
299 */
300static void ibmveth_update_rx_no_buffer(struct ibmveth_adapter *adapter)
301{
302	__be64 *p = adapter->buffer_list_addr + 4096 - 8;
303
304	adapter->rx_no_buffer = be64_to_cpup(p);
305}
306
307/* replenish routine */
308static void ibmveth_replenish_task(struct ibmveth_adapter *adapter)
309{
310	int i;
311
312	adapter->replenish_task_cycles++;
313
314	for (i = (IBMVETH_NUM_BUFF_POOLS - 1); i >= 0; i--) {
315		struct ibmveth_buff_pool *pool = &adapter->rx_buff_pool[i];
316
317		if (pool->active &&
318		    (atomic_read(&pool->available) < pool->threshold))
319			ibmveth_replenish_buffer_pool(adapter, pool);
320	}
321
322	ibmveth_update_rx_no_buffer(adapter);
323}
324
325/* empty and free ana buffer pool - also used to do cleanup in error paths */
326static void ibmveth_free_buffer_pool(struct ibmveth_adapter *adapter,
327				     struct ibmveth_buff_pool *pool)
328{
329	int i;
330
331	kfree(pool->free_map);
332	pool->free_map = NULL;
333
334	if (pool->skbuff && pool->dma_addr) {
335		for (i = 0; i < pool->size; ++i) {
336			struct sk_buff *skb = pool->skbuff[i];
337			if (skb) {
338				dma_unmap_single(&adapter->vdev->dev,
339						 pool->dma_addr[i],
340						 pool->buff_size,
341						 DMA_FROM_DEVICE);
342				dev_kfree_skb_any(skb);
343				pool->skbuff[i] = NULL;
344			}
345		}
346	}
347
348	if (pool->dma_addr) {
349		kfree(pool->dma_addr);
350		pool->dma_addr = NULL;
351	}
352
353	if (pool->skbuff) {
354		kfree(pool->skbuff);
355		pool->skbuff = NULL;
356	}
357}
358
359/* remove a buffer from a pool */
360static void ibmveth_remove_buffer_from_pool(struct ibmveth_adapter *adapter,
361					    u64 correlator)
362{
363	unsigned int pool  = correlator >> 32;
364	unsigned int index = correlator & 0xffffffffUL;
365	unsigned int free_index;
366	struct sk_buff *skb;
367
368	BUG_ON(pool >= IBMVETH_NUM_BUFF_POOLS);
369	BUG_ON(index >= adapter->rx_buff_pool[pool].size);
370
371	skb = adapter->rx_buff_pool[pool].skbuff[index];
372
373	BUG_ON(skb == NULL);
374
375	adapter->rx_buff_pool[pool].skbuff[index] = NULL;
376
377	dma_unmap_single(&adapter->vdev->dev,
378			 adapter->rx_buff_pool[pool].dma_addr[index],
379			 adapter->rx_buff_pool[pool].buff_size,
380			 DMA_FROM_DEVICE);
381
382	free_index = adapter->rx_buff_pool[pool].producer_index;
383	adapter->rx_buff_pool[pool].producer_index++;
384	if (adapter->rx_buff_pool[pool].producer_index >=
385	    adapter->rx_buff_pool[pool].size)
386		adapter->rx_buff_pool[pool].producer_index = 0;
387	adapter->rx_buff_pool[pool].free_map[free_index] = index;
388
389	mb();
390
391	atomic_dec(&(adapter->rx_buff_pool[pool].available));
392}
393
394/* get the current buffer on the rx queue */
395static inline struct sk_buff *ibmveth_rxq_get_buffer(struct ibmveth_adapter *adapter)
396{
397	u64 correlator = adapter->rx_queue.queue_addr[adapter->rx_queue.index].correlator;
398	unsigned int pool = correlator >> 32;
399	unsigned int index = correlator & 0xffffffffUL;
400
401	BUG_ON(pool >= IBMVETH_NUM_BUFF_POOLS);
402	BUG_ON(index >= adapter->rx_buff_pool[pool].size);
403
404	return adapter->rx_buff_pool[pool].skbuff[index];
405}
406
407/* recycle the current buffer on the rx queue */
408static int ibmveth_rxq_recycle_buffer(struct ibmveth_adapter *adapter)
409{
410	u32 q_index = adapter->rx_queue.index;
411	u64 correlator = adapter->rx_queue.queue_addr[q_index].correlator;
412	unsigned int pool = correlator >> 32;
413	unsigned int index = correlator & 0xffffffffUL;
414	union ibmveth_buf_desc desc;
415	unsigned long lpar_rc;
416	int ret = 1;
417
418	BUG_ON(pool >= IBMVETH_NUM_BUFF_POOLS);
419	BUG_ON(index >= adapter->rx_buff_pool[pool].size);
420
421	if (!adapter->rx_buff_pool[pool].active) {
422		ibmveth_rxq_harvest_buffer(adapter);
423		ibmveth_free_buffer_pool(adapter, &adapter->rx_buff_pool[pool]);
424		goto out;
425	}
426
427	desc.fields.flags_len = IBMVETH_BUF_VALID |
428		adapter->rx_buff_pool[pool].buff_size;
429	desc.fields.address = adapter->rx_buff_pool[pool].dma_addr[index];
430
431	lpar_rc = h_add_logical_lan_buffer(adapter->vdev->unit_address, desc.desc);
432
433	if (lpar_rc != H_SUCCESS) {
434		netdev_dbg(adapter->netdev, "h_add_logical_lan_buffer failed "
435			   "during recycle rc=%ld", lpar_rc);
436		ibmveth_remove_buffer_from_pool(adapter, adapter->rx_queue.queue_addr[adapter->rx_queue.index].correlator);
437		ret = 0;
438	}
439
440	if (++adapter->rx_queue.index == adapter->rx_queue.num_slots) {
441		adapter->rx_queue.index = 0;
442		adapter->rx_queue.toggle = !adapter->rx_queue.toggle;
443	}
444
445out:
446	return ret;
447}
448
449static void ibmveth_rxq_harvest_buffer(struct ibmveth_adapter *adapter)
450{
451	ibmveth_remove_buffer_from_pool(adapter, adapter->rx_queue.queue_addr[adapter->rx_queue.index].correlator);
452
453	if (++adapter->rx_queue.index == adapter->rx_queue.num_slots) {
454		adapter->rx_queue.index = 0;
455		adapter->rx_queue.toggle = !adapter->rx_queue.toggle;
456	}
457}
458
459static void ibmveth_cleanup(struct ibmveth_adapter *adapter)
460{
461	int i;
462	struct device *dev = &adapter->vdev->dev;
463
464	if (adapter->buffer_list_addr != NULL) {
465		if (!dma_mapping_error(dev, adapter->buffer_list_dma)) {
466			dma_unmap_single(dev, adapter->buffer_list_dma, 4096,
467					DMA_BIDIRECTIONAL);
468			adapter->buffer_list_dma = DMA_ERROR_CODE;
469		}
470		free_page((unsigned long)adapter->buffer_list_addr);
471		adapter->buffer_list_addr = NULL;
472	}
473
474	if (adapter->filter_list_addr != NULL) {
475		if (!dma_mapping_error(dev, adapter->filter_list_dma)) {
476			dma_unmap_single(dev, adapter->filter_list_dma, 4096,
477					DMA_BIDIRECTIONAL);
478			adapter->filter_list_dma = DMA_ERROR_CODE;
479		}
480		free_page((unsigned long)adapter->filter_list_addr);
481		adapter->filter_list_addr = NULL;
482	}
483
484	if (adapter->rx_queue.queue_addr != NULL) {
485		dma_free_coherent(dev, adapter->rx_queue.queue_len,
486				  adapter->rx_queue.queue_addr,
487				  adapter->rx_queue.queue_dma);
488		adapter->rx_queue.queue_addr = NULL;
489	}
490
491	for (i = 0; i < IBMVETH_NUM_BUFF_POOLS; i++)
492		if (adapter->rx_buff_pool[i].active)
493			ibmveth_free_buffer_pool(adapter,
494						 &adapter->rx_buff_pool[i]);
495
496	if (adapter->bounce_buffer != NULL) {
497		if (!dma_mapping_error(dev, adapter->bounce_buffer_dma)) {
498			dma_unmap_single(&adapter->vdev->dev,
499					adapter->bounce_buffer_dma,
500					adapter->netdev->mtu + IBMVETH_BUFF_OH,
501					DMA_BIDIRECTIONAL);
502			adapter->bounce_buffer_dma = DMA_ERROR_CODE;
503		}
504		kfree(adapter->bounce_buffer);
505		adapter->bounce_buffer = NULL;
506	}
507}
508
509static int ibmveth_register_logical_lan(struct ibmveth_adapter *adapter,
510        union ibmveth_buf_desc rxq_desc, u64 mac_address)
511{
512	int rc, try_again = 1;
513
514	/*
515	 * After a kexec the adapter will still be open, so our attempt to
516	 * open it will fail. So if we get a failure we free the adapter and
517	 * try again, but only once.
518	 */
519retry:
520	rc = h_register_logical_lan(adapter->vdev->unit_address,
521				    adapter->buffer_list_dma, rxq_desc.desc,
522				    adapter->filter_list_dma, mac_address);
523
524	if (rc != H_SUCCESS && try_again) {
525		do {
526			rc = h_free_logical_lan(adapter->vdev->unit_address);
527		} while (H_IS_LONG_BUSY(rc) || (rc == H_BUSY));
528
529		try_again = 0;
530		goto retry;
531	}
532
533	return rc;
534}
535
536static u64 ibmveth_encode_mac_addr(u8 *mac)
537{
538	int i;
539	u64 encoded = 0;
540
541	for (i = 0; i < ETH_ALEN; i++)
542		encoded = (encoded << 8) | mac[i];
543
544	return encoded;
545}
546
547static int ibmveth_open(struct net_device *netdev)
548{
549	struct ibmveth_adapter *adapter = netdev_priv(netdev);
550	u64 mac_address;
551	int rxq_entries = 1;
552	unsigned long lpar_rc;
553	int rc;
554	union ibmveth_buf_desc rxq_desc;
555	int i;
556	struct device *dev;
557
558	netdev_dbg(netdev, "open starting\n");
559
560	napi_enable(&adapter->napi);
561
562	for(i = 0; i < IBMVETH_NUM_BUFF_POOLS; i++)
563		rxq_entries += adapter->rx_buff_pool[i].size;
564
565	adapter->buffer_list_addr = (void*) get_zeroed_page(GFP_KERNEL);
566	adapter->filter_list_addr = (void*) get_zeroed_page(GFP_KERNEL);
567
568	if (!adapter->buffer_list_addr || !adapter->filter_list_addr) {
569		netdev_err(netdev, "unable to allocate filter or buffer list "
570			   "pages\n");
571		rc = -ENOMEM;
572		goto err_out;
573	}
574
575	dev = &adapter->vdev->dev;
576
577	adapter->rx_queue.queue_len = sizeof(struct ibmveth_rx_q_entry) *
578						rxq_entries;
579	adapter->rx_queue.queue_addr =
580		dma_alloc_coherent(dev, adapter->rx_queue.queue_len,
581				   &adapter->rx_queue.queue_dma, GFP_KERNEL);
582	if (!adapter->rx_queue.queue_addr) {
583		rc = -ENOMEM;
584		goto err_out;
585	}
586
587	adapter->buffer_list_dma = dma_map_single(dev,
588			adapter->buffer_list_addr, 4096, DMA_BIDIRECTIONAL);
589	adapter->filter_list_dma = dma_map_single(dev,
590			adapter->filter_list_addr, 4096, DMA_BIDIRECTIONAL);
591
592	if ((dma_mapping_error(dev, adapter->buffer_list_dma)) ||
593	    (dma_mapping_error(dev, adapter->filter_list_dma))) {
594		netdev_err(netdev, "unable to map filter or buffer list "
595			   "pages\n");
596		rc = -ENOMEM;
597		goto err_out;
598	}
599
600	adapter->rx_queue.index = 0;
601	adapter->rx_queue.num_slots = rxq_entries;
602	adapter->rx_queue.toggle = 1;
603
604	mac_address = ibmveth_encode_mac_addr(netdev->dev_addr);
605
606	rxq_desc.fields.flags_len = IBMVETH_BUF_VALID |
607					adapter->rx_queue.queue_len;
608	rxq_desc.fields.address = adapter->rx_queue.queue_dma;
609
610	netdev_dbg(netdev, "buffer list @ 0x%p\n", adapter->buffer_list_addr);
611	netdev_dbg(netdev, "filter list @ 0x%p\n", adapter->filter_list_addr);
612	netdev_dbg(netdev, "receive q   @ 0x%p\n", adapter->rx_queue.queue_addr);
613
614	h_vio_signal(adapter->vdev->unit_address, VIO_IRQ_DISABLE);
615
616	lpar_rc = ibmveth_register_logical_lan(adapter, rxq_desc, mac_address);
617
618	if (lpar_rc != H_SUCCESS) {
619		netdev_err(netdev, "h_register_logical_lan failed with %ld\n",
620			   lpar_rc);
621		netdev_err(netdev, "buffer TCE:0x%llx filter TCE:0x%llx rxq "
622			   "desc:0x%llx MAC:0x%llx\n",
623				     adapter->buffer_list_dma,
624				     adapter->filter_list_dma,
625				     rxq_desc.desc,
626				     mac_address);
627		rc = -ENONET;
628		goto err_out;
629	}
630
631	for (i = 0; i < IBMVETH_NUM_BUFF_POOLS; i++) {
632		if (!adapter->rx_buff_pool[i].active)
633			continue;
634		if (ibmveth_alloc_buffer_pool(&adapter->rx_buff_pool[i])) {
635			netdev_err(netdev, "unable to alloc pool\n");
636			adapter->rx_buff_pool[i].active = 0;
637			rc = -ENOMEM;
638			goto err_out;
639		}
640	}
641
642	netdev_dbg(netdev, "registering irq 0x%x\n", netdev->irq);
643	rc = request_irq(netdev->irq, ibmveth_interrupt, 0, netdev->name,
644			 netdev);
645	if (rc != 0) {
646		netdev_err(netdev, "unable to request irq 0x%x, rc %d\n",
647			   netdev->irq, rc);
648		do {
649			lpar_rc = h_free_logical_lan(adapter->vdev->unit_address);
650		} while (H_IS_LONG_BUSY(lpar_rc) || (lpar_rc == H_BUSY));
651
652		goto err_out;
653	}
654
655	adapter->bounce_buffer =
656	    kmalloc(netdev->mtu + IBMVETH_BUFF_OH, GFP_KERNEL);
657	if (!adapter->bounce_buffer) {
658		rc = -ENOMEM;
659		goto err_out_free_irq;
660	}
661	adapter->bounce_buffer_dma =
662	    dma_map_single(&adapter->vdev->dev, adapter->bounce_buffer,
663			   netdev->mtu + IBMVETH_BUFF_OH, DMA_BIDIRECTIONAL);
664	if (dma_mapping_error(dev, adapter->bounce_buffer_dma)) {
665		netdev_err(netdev, "unable to map bounce buffer\n");
666		rc = -ENOMEM;
667		goto err_out_free_irq;
668	}
669
670	netdev_dbg(netdev, "initial replenish cycle\n");
671	ibmveth_interrupt(netdev->irq, netdev);
672
673	netif_start_queue(netdev);
674
675	netdev_dbg(netdev, "open complete\n");
676
677	return 0;
678
679err_out_free_irq:
680	free_irq(netdev->irq, netdev);
681err_out:
682	ibmveth_cleanup(adapter);
683	napi_disable(&adapter->napi);
684	return rc;
685}
686
687static int ibmveth_close(struct net_device *netdev)
688{
689	struct ibmveth_adapter *adapter = netdev_priv(netdev);
690	long lpar_rc;
691
692	netdev_dbg(netdev, "close starting\n");
693
694	napi_disable(&adapter->napi);
695
696	if (!adapter->pool_config)
697		netif_stop_queue(netdev);
698
699	h_vio_signal(adapter->vdev->unit_address, VIO_IRQ_DISABLE);
700
701	do {
702		lpar_rc = h_free_logical_lan(adapter->vdev->unit_address);
703	} while (H_IS_LONG_BUSY(lpar_rc) || (lpar_rc == H_BUSY));
704
705	if (lpar_rc != H_SUCCESS) {
706		netdev_err(netdev, "h_free_logical_lan failed with %lx, "
707			   "continuing with close\n", lpar_rc);
708	}
709
710	free_irq(netdev->irq, netdev);
711
712	ibmveth_update_rx_no_buffer(adapter);
713
714	ibmveth_cleanup(adapter);
715
716	netdev_dbg(netdev, "close complete\n");
717
718	return 0;
719}
720
721static int netdev_get_settings(struct net_device *dev, struct ethtool_cmd *cmd)
722{
723	cmd->supported = (SUPPORTED_1000baseT_Full | SUPPORTED_Autoneg |
724				SUPPORTED_FIBRE);
725	cmd->advertising = (ADVERTISED_1000baseT_Full | ADVERTISED_Autoneg |
726				ADVERTISED_FIBRE);
727	ethtool_cmd_speed_set(cmd, SPEED_1000);
728	cmd->duplex = DUPLEX_FULL;
729	cmd->port = PORT_FIBRE;
730	cmd->phy_address = 0;
731	cmd->transceiver = XCVR_INTERNAL;
732	cmd->autoneg = AUTONEG_ENABLE;
733	cmd->maxtxpkt = 0;
734	cmd->maxrxpkt = 1;
735	return 0;
736}
737
738static void netdev_get_drvinfo(struct net_device *dev,
739			       struct ethtool_drvinfo *info)
740{
741	strlcpy(info->driver, ibmveth_driver_name, sizeof(info->driver));
742	strlcpy(info->version, ibmveth_driver_version, sizeof(info->version));
743}
744
745static netdev_features_t ibmveth_fix_features(struct net_device *dev,
746	netdev_features_t features)
747{
748	/*
749	 * Since the ibmveth firmware interface does not have the
750	 * concept of separate tx/rx checksum offload enable, if rx
751	 * checksum is disabled we also have to disable tx checksum
752	 * offload. Once we disable rx checksum offload, we are no
753	 * longer allowed to send tx buffers that are not properly
754	 * checksummed.
755	 */
756
757	if (!(features & NETIF_F_RXCSUM))
758		features &= ~NETIF_F_ALL_CSUM;
759
760	return features;
761}
762
763static int ibmveth_set_csum_offload(struct net_device *dev, u32 data)
764{
765	struct ibmveth_adapter *adapter = netdev_priv(dev);
766	unsigned long set_attr, clr_attr, ret_attr;
767	unsigned long set_attr6, clr_attr6;
768	long ret, ret4, ret6;
769	int rc1 = 0, rc2 = 0;
770	int restart = 0;
771
772	if (netif_running(dev)) {
773		restart = 1;
774		adapter->pool_config = 1;
775		ibmveth_close(dev);
776		adapter->pool_config = 0;
777	}
778
779	set_attr = 0;
780	clr_attr = 0;
781	set_attr6 = 0;
782	clr_attr6 = 0;
783
784	if (data) {
785		set_attr = IBMVETH_ILLAN_IPV4_TCP_CSUM;
786		set_attr6 = IBMVETH_ILLAN_IPV6_TCP_CSUM;
787	} else {
788		clr_attr = IBMVETH_ILLAN_IPV4_TCP_CSUM;
789		clr_attr6 = IBMVETH_ILLAN_IPV6_TCP_CSUM;
790	}
791
792	ret = h_illan_attributes(adapter->vdev->unit_address, 0, 0, &ret_attr);
793
794	if (ret == H_SUCCESS && !(ret_attr & IBMVETH_ILLAN_ACTIVE_TRUNK) &&
795	    !(ret_attr & IBMVETH_ILLAN_TRUNK_PRI_MASK) &&
796	    (ret_attr & IBMVETH_ILLAN_PADDED_PKT_CSUM)) {
797		ret4 = h_illan_attributes(adapter->vdev->unit_address, clr_attr,
798					 set_attr, &ret_attr);
799
800		if (ret4 != H_SUCCESS) {
801			netdev_err(dev, "unable to change IPv4 checksum "
802					"offload settings. %d rc=%ld\n",
803					data, ret4);
804
805			h_illan_attributes(adapter->vdev->unit_address,
806					   set_attr, clr_attr, &ret_attr);
807
808			if (data == 1)
809				dev->features &= ~NETIF_F_IP_CSUM;
810
811		} else {
812			adapter->fw_ipv4_csum_support = data;
813		}
814
815		ret6 = h_illan_attributes(adapter->vdev->unit_address,
816					 clr_attr6, set_attr6, &ret_attr);
817
818		if (ret6 != H_SUCCESS) {
819			netdev_err(dev, "unable to change IPv6 checksum "
820					"offload settings. %d rc=%ld\n",
821					data, ret6);
822
823			h_illan_attributes(adapter->vdev->unit_address,
824					   set_attr6, clr_attr6, &ret_attr);
825
826			if (data == 1)
827				dev->features &= ~NETIF_F_IPV6_CSUM;
828
829		} else
830			adapter->fw_ipv6_csum_support = data;
831
832		if (ret4 == H_SUCCESS || ret6 == H_SUCCESS)
833			adapter->rx_csum = data;
834		else
835			rc1 = -EIO;
836	} else {
837		rc1 = -EIO;
838		netdev_err(dev, "unable to change checksum offload settings."
839				     " %d rc=%ld ret_attr=%lx\n", data, ret,
840				     ret_attr);
841	}
842
843	if (restart)
844		rc2 = ibmveth_open(dev);
845
846	return rc1 ? rc1 : rc2;
847}
848
849static int ibmveth_set_features(struct net_device *dev,
850	netdev_features_t features)
851{
852	struct ibmveth_adapter *adapter = netdev_priv(dev);
853	int rx_csum = !!(features & NETIF_F_RXCSUM);
854	int rc;
855
856	if (rx_csum == adapter->rx_csum)
857		return 0;
858
859	rc = ibmveth_set_csum_offload(dev, rx_csum);
860	if (rc && !adapter->rx_csum)
861		dev->features = features & ~(NETIF_F_ALL_CSUM | NETIF_F_RXCSUM);
862
863	return rc;
864}
865
866static void ibmveth_get_strings(struct net_device *dev, u32 stringset, u8 *data)
867{
868	int i;
869
870	if (stringset != ETH_SS_STATS)
871		return;
872
873	for (i = 0; i < ARRAY_SIZE(ibmveth_stats); i++, data += ETH_GSTRING_LEN)
874		memcpy(data, ibmveth_stats[i].name, ETH_GSTRING_LEN);
875}
876
877static int ibmveth_get_sset_count(struct net_device *dev, int sset)
878{
879	switch (sset) {
880	case ETH_SS_STATS:
881		return ARRAY_SIZE(ibmveth_stats);
882	default:
883		return -EOPNOTSUPP;
884	}
885}
886
887static void ibmveth_get_ethtool_stats(struct net_device *dev,
888				      struct ethtool_stats *stats, u64 *data)
889{
890	int i;
891	struct ibmveth_adapter *adapter = netdev_priv(dev);
892
893	for (i = 0; i < ARRAY_SIZE(ibmveth_stats); i++)
894		data[i] = IBMVETH_GET_STAT(adapter, ibmveth_stats[i].offset);
895}
896
897static const struct ethtool_ops netdev_ethtool_ops = {
898	.get_drvinfo		= netdev_get_drvinfo,
899	.get_settings		= netdev_get_settings,
900	.get_link		= ethtool_op_get_link,
901	.get_strings		= ibmveth_get_strings,
902	.get_sset_count		= ibmveth_get_sset_count,
903	.get_ethtool_stats	= ibmveth_get_ethtool_stats,
904};
905
906static int ibmveth_ioctl(struct net_device *dev, struct ifreq *ifr, int cmd)
907{
908	return -EOPNOTSUPP;
909}
910
911#define page_offset(v) ((unsigned long)(v) & ((1 << 12) - 1))
912
913static int ibmveth_send(struct ibmveth_adapter *adapter,
914			union ibmveth_buf_desc *descs)
915{
916	unsigned long correlator;
917	unsigned int retry_count;
918	unsigned long ret;
919
920	/*
921	 * The retry count sets a maximum for the number of broadcast and
922	 * multicast destinations within the system.
923	 */
924	retry_count = 1024;
925	correlator = 0;
926	do {
927		ret = h_send_logical_lan(adapter->vdev->unit_address,
928					     descs[0].desc, descs[1].desc,
929					     descs[2].desc, descs[3].desc,
930					     descs[4].desc, descs[5].desc,
931					     correlator, &correlator);
932	} while ((ret == H_BUSY) && (retry_count--));
933
934	if (ret != H_SUCCESS && ret != H_DROPPED) {
935		netdev_err(adapter->netdev, "tx: h_send_logical_lan failed "
936			   "with rc=%ld\n", ret);
937		return 1;
938	}
939
940	return 0;
941}
942
943static netdev_tx_t ibmveth_start_xmit(struct sk_buff *skb,
944				      struct net_device *netdev)
945{
946	struct ibmveth_adapter *adapter = netdev_priv(netdev);
947	unsigned int desc_flags;
948	union ibmveth_buf_desc descs[6];
949	int last, i;
950	int force_bounce = 0;
951	dma_addr_t dma_addr;
952
953	/*
954	 * veth handles a maximum of 6 segments including the header, so
955	 * we have to linearize the skb if there are more than this.
956	 */
957	if (skb_shinfo(skb)->nr_frags > 5 && __skb_linearize(skb)) {
958		netdev->stats.tx_dropped++;
959		goto out;
960	}
961
962	/* veth can't checksum offload UDP */
963	if (skb->ip_summed == CHECKSUM_PARTIAL &&
964	    ((skb->protocol == htons(ETH_P_IP) &&
965	      ip_hdr(skb)->protocol != IPPROTO_TCP) ||
966	     (skb->protocol == htons(ETH_P_IPV6) &&
967	      ipv6_hdr(skb)->nexthdr != IPPROTO_TCP)) &&
968	    skb_checksum_help(skb)) {
969
970		netdev_err(netdev, "tx: failed to checksum packet\n");
971		netdev->stats.tx_dropped++;
972		goto out;
973	}
974
975	desc_flags = IBMVETH_BUF_VALID;
976
977	if (skb->ip_summed == CHECKSUM_PARTIAL) {
978		unsigned char *buf = skb_transport_header(skb) +
979						skb->csum_offset;
980
981		desc_flags |= (IBMVETH_BUF_NO_CSUM | IBMVETH_BUF_CSUM_GOOD);
982
983		/* Need to zero out the checksum */
984		buf[0] = 0;
985		buf[1] = 0;
986	}
987
988retry_bounce:
989	memset(descs, 0, sizeof(descs));
990
991	/*
992	 * If a linear packet is below the rx threshold then
993	 * copy it into the static bounce buffer. This avoids the
994	 * cost of a TCE insert and remove.
995	 */
996	if (force_bounce || (!skb_is_nonlinear(skb) &&
997				(skb->len < tx_copybreak))) {
998		skb_copy_from_linear_data(skb, adapter->bounce_buffer,
999					  skb->len);
1000
1001		descs[0].fields.flags_len = desc_flags | skb->len;
1002		descs[0].fields.address = adapter->bounce_buffer_dma;
1003
1004		if (ibmveth_send(adapter, descs)) {
1005			adapter->tx_send_failed++;
1006			netdev->stats.tx_dropped++;
1007		} else {
1008			netdev->stats.tx_packets++;
1009			netdev->stats.tx_bytes += skb->len;
1010		}
1011
1012		goto out;
1013	}
1014
1015	/* Map the header */
1016	dma_addr = dma_map_single(&adapter->vdev->dev, skb->data,
1017				  skb_headlen(skb), DMA_TO_DEVICE);
1018	if (dma_mapping_error(&adapter->vdev->dev, dma_addr))
1019		goto map_failed;
1020
1021	descs[0].fields.flags_len = desc_flags | skb_headlen(skb);
1022	descs[0].fields.address = dma_addr;
1023
1024	/* Map the frags */
1025	for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
1026		const skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
1027
1028		dma_addr = skb_frag_dma_map(&adapter->vdev->dev, frag, 0,
1029					    skb_frag_size(frag), DMA_TO_DEVICE);
1030
1031		if (dma_mapping_error(&adapter->vdev->dev, dma_addr))
1032			goto map_failed_frags;
1033
1034		descs[i+1].fields.flags_len = desc_flags | skb_frag_size(frag);
1035		descs[i+1].fields.address = dma_addr;
1036	}
1037
1038	if (ibmveth_send(adapter, descs)) {
1039		adapter->tx_send_failed++;
1040		netdev->stats.tx_dropped++;
1041	} else {
1042		netdev->stats.tx_packets++;
1043		netdev->stats.tx_bytes += skb->len;
1044	}
1045
1046	dma_unmap_single(&adapter->vdev->dev,
1047			 descs[0].fields.address,
1048			 descs[0].fields.flags_len & IBMVETH_BUF_LEN_MASK,
1049			 DMA_TO_DEVICE);
1050
1051	for (i = 1; i < skb_shinfo(skb)->nr_frags + 1; i++)
1052		dma_unmap_page(&adapter->vdev->dev, descs[i].fields.address,
1053			       descs[i].fields.flags_len & IBMVETH_BUF_LEN_MASK,
1054			       DMA_TO_DEVICE);
1055
1056out:
1057	dev_consume_skb_any(skb);
1058	return NETDEV_TX_OK;
1059
1060map_failed_frags:
1061	last = i+1;
1062	for (i = 0; i < last; i++)
1063		dma_unmap_page(&adapter->vdev->dev, descs[i].fields.address,
1064			       descs[i].fields.flags_len & IBMVETH_BUF_LEN_MASK,
1065			       DMA_TO_DEVICE);
1066
1067map_failed:
1068	if (!firmware_has_feature(FW_FEATURE_CMO))
1069		netdev_err(netdev, "tx: unable to map xmit buffer\n");
1070	adapter->tx_map_failed++;
1071	skb_linearize(skb);
1072	force_bounce = 1;
1073	goto retry_bounce;
1074}
1075
1076static int ibmveth_poll(struct napi_struct *napi, int budget)
1077{
1078	struct ibmveth_adapter *adapter =
1079			container_of(napi, struct ibmveth_adapter, napi);
1080	struct net_device *netdev = adapter->netdev;
1081	int frames_processed = 0;
1082	unsigned long lpar_rc;
1083
1084restart_poll:
1085	while (frames_processed < budget) {
1086		if (!ibmveth_rxq_pending_buffer(adapter))
1087			break;
1088
1089		smp_rmb();
1090		if (!ibmveth_rxq_buffer_valid(adapter)) {
1091			wmb(); /* suggested by larson1 */
1092			adapter->rx_invalid_buffer++;
1093			netdev_dbg(netdev, "recycling invalid buffer\n");
1094			ibmveth_rxq_recycle_buffer(adapter);
1095		} else {
1096			struct sk_buff *skb, *new_skb;
1097			int length = ibmveth_rxq_frame_length(adapter);
1098			int offset = ibmveth_rxq_frame_offset(adapter);
1099			int csum_good = ibmveth_rxq_csum_good(adapter);
1100
1101			skb = ibmveth_rxq_get_buffer(adapter);
1102
1103			new_skb = NULL;
1104			if (length < rx_copybreak)
1105				new_skb = netdev_alloc_skb(netdev, length);
1106
1107			if (new_skb) {
1108				skb_copy_to_linear_data(new_skb,
1109							skb->data + offset,
1110							length);
1111				if (rx_flush)
1112					ibmveth_flush_buffer(skb->data,
1113						length + offset);
1114				if (!ibmveth_rxq_recycle_buffer(adapter))
1115					kfree_skb(skb);
1116				skb = new_skb;
1117			} else {
1118				ibmveth_rxq_harvest_buffer(adapter);
1119				skb_reserve(skb, offset);
1120			}
1121
1122			skb_put(skb, length);
1123			skb->protocol = eth_type_trans(skb, netdev);
1124
1125			if (csum_good)
1126				skb->ip_summed = CHECKSUM_UNNECESSARY;
1127
1128			netif_receive_skb(skb);	/* send it up */
1129
1130			netdev->stats.rx_packets++;
1131			netdev->stats.rx_bytes += length;
1132			frames_processed++;
1133		}
1134	}
1135
1136	ibmveth_replenish_task(adapter);
1137
1138	if (frames_processed < budget) {
1139		napi_complete(napi);
1140
1141		/* We think we are done - reenable interrupts,
1142		 * then check once more to make sure we are done.
1143		 */
1144		lpar_rc = h_vio_signal(adapter->vdev->unit_address,
1145				       VIO_IRQ_ENABLE);
1146
1147		BUG_ON(lpar_rc != H_SUCCESS);
1148
1149		if (ibmveth_rxq_pending_buffer(adapter) &&
1150		    napi_reschedule(napi)) {
1151			lpar_rc = h_vio_signal(adapter->vdev->unit_address,
1152					       VIO_IRQ_DISABLE);
1153			goto restart_poll;
1154		}
1155	}
1156
1157	return frames_processed;
1158}
1159
1160static irqreturn_t ibmveth_interrupt(int irq, void *dev_instance)
1161{
1162	struct net_device *netdev = dev_instance;
1163	struct ibmveth_adapter *adapter = netdev_priv(netdev);
1164	unsigned long lpar_rc;
1165
1166	if (napi_schedule_prep(&adapter->napi)) {
1167		lpar_rc = h_vio_signal(adapter->vdev->unit_address,
1168				       VIO_IRQ_DISABLE);
1169		BUG_ON(lpar_rc != H_SUCCESS);
1170		__napi_schedule(&adapter->napi);
1171	}
1172	return IRQ_HANDLED;
1173}
1174
1175static void ibmveth_set_multicast_list(struct net_device *netdev)
1176{
1177	struct ibmveth_adapter *adapter = netdev_priv(netdev);
1178	unsigned long lpar_rc;
1179
1180	if ((netdev->flags & IFF_PROMISC) ||
1181	    (netdev_mc_count(netdev) > adapter->mcastFilterSize)) {
1182		lpar_rc = h_multicast_ctrl(adapter->vdev->unit_address,
1183					   IbmVethMcastEnableRecv |
1184					   IbmVethMcastDisableFiltering,
1185					   0);
1186		if (lpar_rc != H_SUCCESS) {
1187			netdev_err(netdev, "h_multicast_ctrl rc=%ld when "
1188				   "entering promisc mode\n", lpar_rc);
1189		}
1190	} else {
1191		struct netdev_hw_addr *ha;
1192		/* clear the filter table & disable filtering */
1193		lpar_rc = h_multicast_ctrl(adapter->vdev->unit_address,
1194					   IbmVethMcastEnableRecv |
1195					   IbmVethMcastDisableFiltering |
1196					   IbmVethMcastClearFilterTable,
1197					   0);
1198		if (lpar_rc != H_SUCCESS) {
1199			netdev_err(netdev, "h_multicast_ctrl rc=%ld when "
1200				   "attempting to clear filter table\n",
1201				   lpar_rc);
1202		}
1203		/* add the addresses to the filter table */
1204		netdev_for_each_mc_addr(ha, netdev) {
1205			/* add the multicast address to the filter table */
1206			u64 mcast_addr;
1207			mcast_addr = ibmveth_encode_mac_addr(ha->addr);
1208			lpar_rc = h_multicast_ctrl(adapter->vdev->unit_address,
1209						   IbmVethMcastAddFilter,
1210						   mcast_addr);
1211			if (lpar_rc != H_SUCCESS) {
1212				netdev_err(netdev, "h_multicast_ctrl rc=%ld "
1213					   "when adding an entry to the filter "
1214					   "table\n", lpar_rc);
1215			}
1216		}
1217
1218		/* re-enable filtering */
1219		lpar_rc = h_multicast_ctrl(adapter->vdev->unit_address,
1220					   IbmVethMcastEnableFiltering,
1221					   0);
1222		if (lpar_rc != H_SUCCESS) {
1223			netdev_err(netdev, "h_multicast_ctrl rc=%ld when "
1224				   "enabling filtering\n", lpar_rc);
1225		}
1226	}
1227}
1228
1229static int ibmveth_change_mtu(struct net_device *dev, int new_mtu)
1230{
1231	struct ibmveth_adapter *adapter = netdev_priv(dev);
1232	struct vio_dev *viodev = adapter->vdev;
1233	int new_mtu_oh = new_mtu + IBMVETH_BUFF_OH;
1234	int i, rc;
1235	int need_restart = 0;
1236
1237	if (new_mtu < IBMVETH_MIN_MTU)
1238		return -EINVAL;
1239
1240	for (i = 0; i < IBMVETH_NUM_BUFF_POOLS; i++)
1241		if (new_mtu_oh <= adapter->rx_buff_pool[i].buff_size)
1242			break;
1243
1244	if (i == IBMVETH_NUM_BUFF_POOLS)
1245		return -EINVAL;
1246
1247	/* Deactivate all the buffer pools so that the next loop can activate
1248	   only the buffer pools necessary to hold the new MTU */
1249	if (netif_running(adapter->netdev)) {
1250		need_restart = 1;
1251		adapter->pool_config = 1;
1252		ibmveth_close(adapter->netdev);
1253		adapter->pool_config = 0;
1254	}
1255
1256	/* Look for an active buffer pool that can hold the new MTU */
1257	for (i = 0; i < IBMVETH_NUM_BUFF_POOLS; i++) {
1258		adapter->rx_buff_pool[i].active = 1;
1259
1260		if (new_mtu_oh <= adapter->rx_buff_pool[i].buff_size) {
1261			dev->mtu = new_mtu;
1262			vio_cmo_set_dev_desired(viodev,
1263						ibmveth_get_desired_dma
1264						(viodev));
1265			if (need_restart) {
1266				return ibmveth_open(adapter->netdev);
1267			}
1268			return 0;
1269		}
1270	}
1271
1272	if (need_restart && (rc = ibmveth_open(adapter->netdev)))
1273		return rc;
1274
1275	return -EINVAL;
1276}
1277
1278#ifdef CONFIG_NET_POLL_CONTROLLER
1279static void ibmveth_poll_controller(struct net_device *dev)
1280{
1281	ibmveth_replenish_task(netdev_priv(dev));
1282	ibmveth_interrupt(dev->irq, dev);
1283}
1284#endif
1285
1286/**
1287 * ibmveth_get_desired_dma - Calculate IO memory desired by the driver
1288 *
1289 * @vdev: struct vio_dev for the device whose desired IO mem is to be returned
1290 *
1291 * Return value:
1292 *	Number of bytes of IO data the driver will need to perform well.
1293 */
1294static unsigned long ibmveth_get_desired_dma(struct vio_dev *vdev)
1295{
1296	struct net_device *netdev = dev_get_drvdata(&vdev->dev);
1297	struct ibmveth_adapter *adapter;
1298	struct iommu_table *tbl;
1299	unsigned long ret;
1300	int i;
1301	int rxqentries = 1;
1302
1303	tbl = get_iommu_table_base(&vdev->dev);
1304
1305	/* netdev inits at probe time along with the structures we need below*/
1306	if (netdev == NULL)
1307		return IOMMU_PAGE_ALIGN(IBMVETH_IO_ENTITLEMENT_DEFAULT, tbl);
1308
1309	adapter = netdev_priv(netdev);
1310
1311	ret = IBMVETH_BUFF_LIST_SIZE + IBMVETH_FILT_LIST_SIZE;
1312	ret += IOMMU_PAGE_ALIGN(netdev->mtu, tbl);
1313
1314	for (i = 0; i < IBMVETH_NUM_BUFF_POOLS; i++) {
1315		/* add the size of the active receive buffers */
1316		if (adapter->rx_buff_pool[i].active)
1317			ret +=
1318			    adapter->rx_buff_pool[i].size *
1319			    IOMMU_PAGE_ALIGN(adapter->rx_buff_pool[i].
1320					     buff_size, tbl);
1321		rxqentries += adapter->rx_buff_pool[i].size;
1322	}
1323	/* add the size of the receive queue entries */
1324	ret += IOMMU_PAGE_ALIGN(
1325		rxqentries * sizeof(struct ibmveth_rx_q_entry), tbl);
1326
1327	return ret;
1328}
1329
1330static int ibmveth_set_mac_addr(struct net_device *dev, void *p)
1331{
1332	struct ibmveth_adapter *adapter = netdev_priv(dev);
1333	struct sockaddr *addr = p;
1334	u64 mac_address;
1335	int rc;
1336
1337	if (!is_valid_ether_addr(addr->sa_data))
1338		return -EADDRNOTAVAIL;
1339
1340	mac_address = ibmveth_encode_mac_addr(addr->sa_data);
1341	rc = h_change_logical_lan_mac(adapter->vdev->unit_address, mac_address);
1342	if (rc) {
1343		netdev_err(adapter->netdev, "h_change_logical_lan_mac failed with rc=%d\n", rc);
1344		return rc;
1345	}
1346
1347	ether_addr_copy(dev->dev_addr, addr->sa_data);
1348
1349	return 0;
1350}
1351
1352static const struct net_device_ops ibmveth_netdev_ops = {
1353	.ndo_open		= ibmveth_open,
1354	.ndo_stop		= ibmveth_close,
1355	.ndo_start_xmit		= ibmveth_start_xmit,
1356	.ndo_set_rx_mode	= ibmveth_set_multicast_list,
1357	.ndo_do_ioctl		= ibmveth_ioctl,
1358	.ndo_change_mtu		= ibmveth_change_mtu,
1359	.ndo_fix_features	= ibmveth_fix_features,
1360	.ndo_set_features	= ibmveth_set_features,
1361	.ndo_validate_addr	= eth_validate_addr,
1362	.ndo_set_mac_address    = ibmveth_set_mac_addr,
1363#ifdef CONFIG_NET_POLL_CONTROLLER
1364	.ndo_poll_controller	= ibmveth_poll_controller,
1365#endif
1366};
1367
1368static int ibmveth_probe(struct vio_dev *dev, const struct vio_device_id *id)
1369{
1370	int rc, i, mac_len;
1371	struct net_device *netdev;
1372	struct ibmveth_adapter *adapter;
1373	unsigned char *mac_addr_p;
1374	unsigned int *mcastFilterSize_p;
1375
1376	dev_dbg(&dev->dev, "entering ibmveth_probe for UA 0x%x\n",
1377		dev->unit_address);
1378
1379	mac_addr_p = (unsigned char *)vio_get_attribute(dev, VETH_MAC_ADDR,
1380							&mac_len);
1381	if (!mac_addr_p) {
1382		dev_err(&dev->dev, "Can't find VETH_MAC_ADDR attribute\n");
1383		return -EINVAL;
1384	}
1385	/* Workaround for old/broken pHyp */
1386	if (mac_len == 8)
1387		mac_addr_p += 2;
1388	else if (mac_len != 6) {
1389		dev_err(&dev->dev, "VETH_MAC_ADDR attribute wrong len %d\n",
1390			mac_len);
1391		return -EINVAL;
1392	}
1393
1394	mcastFilterSize_p = (unsigned int *)vio_get_attribute(dev,
1395						VETH_MCAST_FILTER_SIZE, NULL);
1396	if (!mcastFilterSize_p) {
1397		dev_err(&dev->dev, "Can't find VETH_MCAST_FILTER_SIZE "
1398			"attribute\n");
1399		return -EINVAL;
1400	}
1401
1402	netdev = alloc_etherdev(sizeof(struct ibmveth_adapter));
1403
1404	if (!netdev)
1405		return -ENOMEM;
1406
1407	adapter = netdev_priv(netdev);
1408	dev_set_drvdata(&dev->dev, netdev);
1409
1410	adapter->vdev = dev;
1411	adapter->netdev = netdev;
1412	adapter->mcastFilterSize = *mcastFilterSize_p;
1413	adapter->pool_config = 0;
1414
1415	netif_napi_add(netdev, &adapter->napi, ibmveth_poll, 16);
1416
1417	netdev->irq = dev->irq;
1418	netdev->netdev_ops = &ibmveth_netdev_ops;
1419	netdev->ethtool_ops = &netdev_ethtool_ops;
1420	SET_NETDEV_DEV(netdev, &dev->dev);
1421	netdev->hw_features = NETIF_F_SG | NETIF_F_RXCSUM |
1422		NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM;
1423	netdev->features |= netdev->hw_features;
1424
1425	memcpy(netdev->dev_addr, mac_addr_p, ETH_ALEN);
1426
1427	for (i = 0; i < IBMVETH_NUM_BUFF_POOLS; i++) {
1428		struct kobject *kobj = &adapter->rx_buff_pool[i].kobj;
1429		int error;
1430
1431		ibmveth_init_buffer_pool(&adapter->rx_buff_pool[i], i,
1432					 pool_count[i], pool_size[i],
1433					 pool_active[i]);
1434		error = kobject_init_and_add(kobj, &ktype_veth_pool,
1435					     &dev->dev.kobj, "pool%d", i);
1436		if (!error)
1437			kobject_uevent(kobj, KOBJ_ADD);
1438	}
1439
1440	netdev_dbg(netdev, "adapter @ 0x%p\n", adapter);
1441
1442	adapter->buffer_list_dma = DMA_ERROR_CODE;
1443	adapter->filter_list_dma = DMA_ERROR_CODE;
1444	adapter->rx_queue.queue_dma = DMA_ERROR_CODE;
1445
1446	netdev_dbg(netdev, "registering netdev...\n");
1447
1448	ibmveth_set_features(netdev, netdev->features);
1449
1450	rc = register_netdev(netdev);
1451
1452	if (rc) {
1453		netdev_dbg(netdev, "failed to register netdev rc=%d\n", rc);
1454		free_netdev(netdev);
1455		return rc;
1456	}
1457
1458	netdev_dbg(netdev, "registered\n");
1459
1460	return 0;
1461}
1462
1463static int ibmveth_remove(struct vio_dev *dev)
1464{
1465	struct net_device *netdev = dev_get_drvdata(&dev->dev);
1466	struct ibmveth_adapter *adapter = netdev_priv(netdev);
1467	int i;
1468
1469	for (i = 0; i < IBMVETH_NUM_BUFF_POOLS; i++)
1470		kobject_put(&adapter->rx_buff_pool[i].kobj);
1471
1472	unregister_netdev(netdev);
1473
1474	free_netdev(netdev);
1475	dev_set_drvdata(&dev->dev, NULL);
1476
1477	return 0;
1478}
1479
1480static struct attribute veth_active_attr;
1481static struct attribute veth_num_attr;
1482static struct attribute veth_size_attr;
1483
1484static ssize_t veth_pool_show(struct kobject *kobj,
1485			      struct attribute *attr, char *buf)
1486{
1487	struct ibmveth_buff_pool *pool = container_of(kobj,
1488						      struct ibmveth_buff_pool,
1489						      kobj);
1490
1491	if (attr == &veth_active_attr)
1492		return sprintf(buf, "%d\n", pool->active);
1493	else if (attr == &veth_num_attr)
1494		return sprintf(buf, "%d\n", pool->size);
1495	else if (attr == &veth_size_attr)
1496		return sprintf(buf, "%d\n", pool->buff_size);
1497	return 0;
1498}
1499
1500static ssize_t veth_pool_store(struct kobject *kobj, struct attribute *attr,
1501			       const char *buf, size_t count)
1502{
1503	struct ibmveth_buff_pool *pool = container_of(kobj,
1504						      struct ibmveth_buff_pool,
1505						      kobj);
1506	struct net_device *netdev = dev_get_drvdata(
1507	    container_of(kobj->parent, struct device, kobj));
1508	struct ibmveth_adapter *adapter = netdev_priv(netdev);
1509	long value = simple_strtol(buf, NULL, 10);
1510	long rc;
1511
1512	if (attr == &veth_active_attr) {
1513		if (value && !pool->active) {
1514			if (netif_running(netdev)) {
1515				if (ibmveth_alloc_buffer_pool(pool)) {
1516					netdev_err(netdev,
1517						   "unable to alloc pool\n");
1518					return -ENOMEM;
1519				}
1520				pool->active = 1;
1521				adapter->pool_config = 1;
1522				ibmveth_close(netdev);
1523				adapter->pool_config = 0;
1524				if ((rc = ibmveth_open(netdev)))
1525					return rc;
1526			} else {
1527				pool->active = 1;
1528			}
1529		} else if (!value && pool->active) {
1530			int mtu = netdev->mtu + IBMVETH_BUFF_OH;
1531			int i;
1532			/* Make sure there is a buffer pool with buffers that
1533			   can hold a packet of the size of the MTU */
1534			for (i = 0; i < IBMVETH_NUM_BUFF_POOLS; i++) {
1535				if (pool == &adapter->rx_buff_pool[i])
1536					continue;
1537				if (!adapter->rx_buff_pool[i].active)
1538					continue;
1539				if (mtu <= adapter->rx_buff_pool[i].buff_size)
1540					break;
1541			}
1542
1543			if (i == IBMVETH_NUM_BUFF_POOLS) {
1544				netdev_err(netdev, "no active pool >= MTU\n");
1545				return -EPERM;
1546			}
1547
1548			if (netif_running(netdev)) {
1549				adapter->pool_config = 1;
1550				ibmveth_close(netdev);
1551				pool->active = 0;
1552				adapter->pool_config = 0;
1553				if ((rc = ibmveth_open(netdev)))
1554					return rc;
1555			}
1556			pool->active = 0;
1557		}
1558	} else if (attr == &veth_num_attr) {
1559		if (value <= 0 || value > IBMVETH_MAX_POOL_COUNT) {
1560			return -EINVAL;
1561		} else {
1562			if (netif_running(netdev)) {
1563				adapter->pool_config = 1;
1564				ibmveth_close(netdev);
1565				adapter->pool_config = 0;
1566				pool->size = value;
1567				if ((rc = ibmveth_open(netdev)))
1568					return rc;
1569			} else {
1570				pool->size = value;
1571			}
1572		}
1573	} else if (attr == &veth_size_attr) {
1574		if (value <= IBMVETH_BUFF_OH || value > IBMVETH_MAX_BUF_SIZE) {
1575			return -EINVAL;
1576		} else {
1577			if (netif_running(netdev)) {
1578				adapter->pool_config = 1;
1579				ibmveth_close(netdev);
1580				adapter->pool_config = 0;
1581				pool->buff_size = value;
1582				if ((rc = ibmveth_open(netdev)))
1583					return rc;
1584			} else {
1585				pool->buff_size = value;
1586			}
1587		}
1588	}
1589
1590	/* kick the interrupt handler to allocate/deallocate pools */
1591	ibmveth_interrupt(netdev->irq, netdev);
1592	return count;
1593}
1594
1595
1596#define ATTR(_name, _mode)				\
1597	struct attribute veth_##_name##_attr = {	\
1598	.name = __stringify(_name), .mode = _mode,	\
1599	};
1600
1601static ATTR(active, 0644);
1602static ATTR(num, 0644);
1603static ATTR(size, 0644);
1604
1605static struct attribute *veth_pool_attrs[] = {
1606	&veth_active_attr,
1607	&veth_num_attr,
1608	&veth_size_attr,
1609	NULL,
1610};
1611
1612static const struct sysfs_ops veth_pool_ops = {
1613	.show   = veth_pool_show,
1614	.store  = veth_pool_store,
1615};
1616
1617static struct kobj_type ktype_veth_pool = {
1618	.release        = NULL,
1619	.sysfs_ops      = &veth_pool_ops,
1620	.default_attrs  = veth_pool_attrs,
1621};
1622
1623static int ibmveth_resume(struct device *dev)
1624{
1625	struct net_device *netdev = dev_get_drvdata(dev);
1626	ibmveth_interrupt(netdev->irq, netdev);
1627	return 0;
1628}
1629
1630static struct vio_device_id ibmveth_device_table[] = {
1631	{ "network", "IBM,l-lan"},
1632	{ "", "" }
1633};
1634MODULE_DEVICE_TABLE(vio, ibmveth_device_table);
1635
1636static struct dev_pm_ops ibmveth_pm_ops = {
1637	.resume = ibmveth_resume
1638};
1639
1640static struct vio_driver ibmveth_driver = {
1641	.id_table	= ibmveth_device_table,
1642	.probe		= ibmveth_probe,
1643	.remove		= ibmveth_remove,
1644	.get_desired_dma = ibmveth_get_desired_dma,
1645	.name		= ibmveth_driver_name,
1646	.pm		= &ibmveth_pm_ops,
1647};
1648
1649static int __init ibmveth_module_init(void)
1650{
1651	printk(KERN_DEBUG "%s: %s %s\n", ibmveth_driver_name,
1652	       ibmveth_driver_string, ibmveth_driver_version);
1653
1654	return vio_register_driver(&ibmveth_driver);
1655}
1656
1657static void __exit ibmveth_module_exit(void)
1658{
1659	vio_unregister_driver(&ibmveth_driver);
1660}
1661
1662module_init(ibmveth_module_init);
1663module_exit(ibmveth_module_exit);
1664