ibmveth: Convert driver specific debug to netdev_dbg
[linux-2.6.git] / drivers / net / ibmveth.c
1 /**************************************************************************/
2 /*                                                                        */
3 /* IBM eServer i/pSeries Virtual Ethernet Device Driver                   */
4 /* Copyright (C) 2003 IBM Corp.                                           */
5 /*  Originally written by Dave Larson (larson1@us.ibm.com)                */
6 /*  Maintained by Santiago Leon (santil@us.ibm.com)                       */
7 /*                                                                        */
8 /*  This program is free software; you can redistribute it and/or modify  */
9 /*  it under the terms of the GNU General Public License as published by  */
10 /*  the Free Software Foundation; either version 2 of the License, or     */
11 /*  (at your option) any later version.                                   */
12 /*                                                                        */
13 /*  This program is distributed in the hope that it will be useful,       */
14 /*  but WITHOUT ANY WARRANTY; without even the implied warranty of        */
15 /*  MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the         */
16 /*  GNU General Public License for more details.                          */
17 /*                                                                        */
18 /*  You should have received a copy of the GNU General Public License     */
19 /*  along with this program; if not, write to the Free Software           */
20 /*  Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA  02111-1307  */
21 /*                                                                   USA  */
22 /*                                                                        */
23 /* This module contains the implementation of a virtual ethernet device   */
24 /* for use with IBM i/pSeries LPAR Linux.  It utilizes the logical LAN    */
25 /* option of the RS/6000 Platform Architechture to interface with virtual */
26 /* ethernet NICs that are presented to the partition by the hypervisor.   */
27 /*                                                                        */
28 /**************************************************************************/
29
30 #include <linux/module.h>
31 #include <linux/moduleparam.h>
32 #include <linux/types.h>
33 #include <linux/errno.h>
34 #include <linux/ioport.h>
35 #include <linux/dma-mapping.h>
36 #include <linux/kernel.h>
37 #include <linux/netdevice.h>
38 #include <linux/etherdevice.h>
39 #include <linux/skbuff.h>
40 #include <linux/init.h>
41 #include <linux/delay.h>
42 #include <linux/mm.h>
43 #include <linux/pm.h>
44 #include <linux/ethtool.h>
45 #include <linux/in.h>
46 #include <linux/ip.h>
47 #include <linux/ipv6.h>
48 #include <linux/slab.h>
49 #include <asm/hvcall.h>
50 #include <asm/atomic.h>
51 #include <asm/vio.h>
52 #include <asm/iommu.h>
53 #include <asm/uaccess.h>
54 #include <asm/firmware.h>
55
56 #include "ibmveth.h"
57
58 #undef DEBUG
59
60 #define ibmveth_printk(fmt, args...) \
61   printk(KERN_DEBUG "%s: " fmt, __FILE__, ## args)
62
63 #define ibmveth_error_printk(fmt, args...) \
64   printk(KERN_ERR "(%s:%3.3d ua:%x) ERROR: " fmt, __FILE__, __LINE__ , adapter->vdev->unit_address, ## args)
65
66 #ifdef DEBUG
67 #define ibmveth_assert(expr) \
68   if(!(expr)) {                                   \
69     printk(KERN_DEBUG "assertion failed (%s:%3.3d ua:%x): %s\n", __FILE__, __LINE__, adapter->vdev->unit_address, #expr); \
70     BUG(); \
71   }
72 #else
73 #define ibmveth_assert(expr)
74 #endif
75
76 static irqreturn_t ibmveth_interrupt(int irq, void *dev_instance);
77 static void ibmveth_rxq_harvest_buffer(struct ibmveth_adapter *adapter);
78 static unsigned long ibmveth_get_desired_dma(struct vio_dev *vdev);
79
80 static struct kobj_type ktype_veth_pool;
81
82
83 static const char ibmveth_driver_name[] = "ibmveth";
84 static const char ibmveth_driver_string[] = "IBM i/pSeries Virtual Ethernet Driver";
85 #define ibmveth_driver_version "1.03"
86
87 MODULE_AUTHOR("Santiago Leon <santil@us.ibm.com>");
88 MODULE_DESCRIPTION("IBM i/pSeries Virtual Ethernet Driver");
89 MODULE_LICENSE("GPL");
90 MODULE_VERSION(ibmveth_driver_version);
91
92 static unsigned int tx_copybreak __read_mostly = 128;
93 module_param(tx_copybreak, uint, 0644);
94 MODULE_PARM_DESC(tx_copybreak,
95         "Maximum size of packet that is copied to a new buffer on transmit");
96
97 static unsigned int rx_copybreak __read_mostly = 128;
98 module_param(rx_copybreak, uint, 0644);
99 MODULE_PARM_DESC(rx_copybreak,
100         "Maximum size of packet that is copied to a new buffer on receive");
101
102 static unsigned int rx_flush __read_mostly = 0;
103 module_param(rx_flush, uint, 0644);
104 MODULE_PARM_DESC(rx_flush, "Flush receive buffers before use");
105
106 struct ibmveth_stat {
107         char name[ETH_GSTRING_LEN];
108         int offset;
109 };
110
111 #define IBMVETH_STAT_OFF(stat) offsetof(struct ibmveth_adapter, stat)
112 #define IBMVETH_GET_STAT(a, off) *((u64 *)(((unsigned long)(a)) + off))
113
114 struct ibmveth_stat ibmveth_stats[] = {
115         { "replenish_task_cycles", IBMVETH_STAT_OFF(replenish_task_cycles) },
116         { "replenish_no_mem", IBMVETH_STAT_OFF(replenish_no_mem) },
117         { "replenish_add_buff_failure", IBMVETH_STAT_OFF(replenish_add_buff_failure) },
118         { "replenish_add_buff_success", IBMVETH_STAT_OFF(replenish_add_buff_success) },
119         { "rx_invalid_buffer", IBMVETH_STAT_OFF(rx_invalid_buffer) },
120         { "rx_no_buffer", IBMVETH_STAT_OFF(rx_no_buffer) },
121         { "tx_map_failed", IBMVETH_STAT_OFF(tx_map_failed) },
122         { "tx_send_failed", IBMVETH_STAT_OFF(tx_send_failed) },
123         { "fw_enabled_ipv4_csum", IBMVETH_STAT_OFF(fw_ipv4_csum_support) },
124         { "fw_enabled_ipv6_csum", IBMVETH_STAT_OFF(fw_ipv6_csum_support) },
125 };
126
127 /* simple methods of getting data from the current rxq entry */
128 static inline u32 ibmveth_rxq_flags(struct ibmveth_adapter *adapter)
129 {
130         return adapter->rx_queue.queue_addr[adapter->rx_queue.index].flags_off;
131 }
132
133 static inline int ibmveth_rxq_toggle(struct ibmveth_adapter *adapter)
134 {
135         return (ibmveth_rxq_flags(adapter) & IBMVETH_RXQ_TOGGLE) >> IBMVETH_RXQ_TOGGLE_SHIFT;
136 }
137
138 static inline int ibmveth_rxq_pending_buffer(struct ibmveth_adapter *adapter)
139 {
140         return (ibmveth_rxq_toggle(adapter) == adapter->rx_queue.toggle);
141 }
142
143 static inline int ibmveth_rxq_buffer_valid(struct ibmveth_adapter *adapter)
144 {
145         return (ibmveth_rxq_flags(adapter) & IBMVETH_RXQ_VALID);
146 }
147
148 static inline int ibmveth_rxq_frame_offset(struct ibmveth_adapter *adapter)
149 {
150         return (ibmveth_rxq_flags(adapter) & IBMVETH_RXQ_OFF_MASK);
151 }
152
153 static inline int ibmveth_rxq_frame_length(struct ibmveth_adapter *adapter)
154 {
155         return (adapter->rx_queue.queue_addr[adapter->rx_queue.index].length);
156 }
157
158 static inline int ibmveth_rxq_csum_good(struct ibmveth_adapter *adapter)
159 {
160         return (ibmveth_rxq_flags(adapter) & IBMVETH_RXQ_CSUM_GOOD);
161 }
162
163 /* setup the initial settings for a buffer pool */
164 static void ibmveth_init_buffer_pool(struct ibmveth_buff_pool *pool, u32 pool_index, u32 pool_size, u32 buff_size, u32 pool_active)
165 {
166         pool->size = pool_size;
167         pool->index = pool_index;
168         pool->buff_size = buff_size;
169         pool->threshold = pool_size * 7 / 8;
170         pool->active = pool_active;
171 }
172
173 /* allocate and setup an buffer pool - called during open */
174 static int ibmveth_alloc_buffer_pool(struct ibmveth_buff_pool *pool)
175 {
176         int i;
177
178         pool->free_map = kmalloc(sizeof(u16) * pool->size, GFP_KERNEL);
179
180         if(!pool->free_map) {
181                 return -1;
182         }
183
184         pool->dma_addr = kmalloc(sizeof(dma_addr_t) * pool->size, GFP_KERNEL);
185         if(!pool->dma_addr) {
186                 kfree(pool->free_map);
187                 pool->free_map = NULL;
188                 return -1;
189         }
190
191         pool->skbuff = kcalloc(pool->size, sizeof(void *), GFP_KERNEL);
192
193         if(!pool->skbuff) {
194                 kfree(pool->dma_addr);
195                 pool->dma_addr = NULL;
196
197                 kfree(pool->free_map);
198                 pool->free_map = NULL;
199                 return -1;
200         }
201
202         memset(pool->dma_addr, 0, sizeof(dma_addr_t) * pool->size);
203
204         for(i = 0; i < pool->size; ++i) {
205                 pool->free_map[i] = i;
206         }
207
208         atomic_set(&pool->available, 0);
209         pool->producer_index = 0;
210         pool->consumer_index = 0;
211
212         return 0;
213 }
214
215 static inline void ibmveth_flush_buffer(void *addr, unsigned long length)
216 {
217         unsigned long offset;
218
219         for (offset = 0; offset < length; offset += SMP_CACHE_BYTES)
220                 asm("dcbfl %0,%1" :: "b" (addr), "r" (offset));
221 }
222
223 /* replenish the buffers for a pool.  note that we don't need to
224  * skb_reserve these since they are used for incoming...
225  */
226 static void ibmveth_replenish_buffer_pool(struct ibmveth_adapter *adapter, struct ibmveth_buff_pool *pool)
227 {
228         u32 i;
229         u32 count = pool->size - atomic_read(&pool->available);
230         u32 buffers_added = 0;
231         struct sk_buff *skb;
232         unsigned int free_index, index;
233         u64 correlator;
234         unsigned long lpar_rc;
235         dma_addr_t dma_addr;
236
237         mb();
238
239         for(i = 0; i < count; ++i) {
240                 union ibmveth_buf_desc desc;
241
242                 skb = netdev_alloc_skb(adapter->netdev, pool->buff_size);
243
244                 if(!skb) {
245                         netdev_dbg(adapter->netdev,
246                                    "replenish: unable to allocate skb\n");
247                         adapter->replenish_no_mem++;
248                         break;
249                 }
250
251                 free_index = pool->consumer_index;
252                 pool->consumer_index++;
253                 if (pool->consumer_index >= pool->size)
254                         pool->consumer_index = 0;
255                 index = pool->free_map[free_index];
256
257                 ibmveth_assert(index != IBM_VETH_INVALID_MAP);
258                 ibmveth_assert(pool->skbuff[index] == NULL);
259
260                 dma_addr = dma_map_single(&adapter->vdev->dev, skb->data,
261                                 pool->buff_size, DMA_FROM_DEVICE);
262
263                 if (dma_mapping_error(&adapter->vdev->dev, dma_addr))
264                         goto failure;
265
266                 pool->free_map[free_index] = IBM_VETH_INVALID_MAP;
267                 pool->dma_addr[index] = dma_addr;
268                 pool->skbuff[index] = skb;
269
270                 correlator = ((u64)pool->index << 32) | index;
271                 *(u64*)skb->data = correlator;
272
273                 desc.fields.flags_len = IBMVETH_BUF_VALID | pool->buff_size;
274                 desc.fields.address = dma_addr;
275
276                 if (rx_flush) {
277                         unsigned int len = min(pool->buff_size,
278                                                 adapter->netdev->mtu +
279                                                 IBMVETH_BUFF_OH);
280                         ibmveth_flush_buffer(skb->data, len);
281                 }
282                 lpar_rc = h_add_logical_lan_buffer(adapter->vdev->unit_address, desc.desc);
283
284                 if (lpar_rc != H_SUCCESS)
285                         goto failure;
286                 else {
287                         buffers_added++;
288                         adapter->replenish_add_buff_success++;
289                 }
290         }
291
292         mb();
293         atomic_add(buffers_added, &(pool->available));
294         return;
295
296 failure:
297         pool->free_map[free_index] = index;
298         pool->skbuff[index] = NULL;
299         if (pool->consumer_index == 0)
300                 pool->consumer_index = pool->size - 1;
301         else
302                 pool->consumer_index--;
303         if (!dma_mapping_error(&adapter->vdev->dev, dma_addr))
304                 dma_unmap_single(&adapter->vdev->dev,
305                                  pool->dma_addr[index], pool->buff_size,
306                                  DMA_FROM_DEVICE);
307         dev_kfree_skb_any(skb);
308         adapter->replenish_add_buff_failure++;
309
310         mb();
311         atomic_add(buffers_added, &(pool->available));
312 }
313
314 /* replenish routine */
315 static void ibmveth_replenish_task(struct ibmveth_adapter *adapter)
316 {
317         int i;
318
319         adapter->replenish_task_cycles++;
320
321         for (i = (IbmVethNumBufferPools - 1); i >= 0; i--) {
322                 struct ibmveth_buff_pool *pool = &adapter->rx_buff_pool[i];
323
324                 if (pool->active &&
325                     (atomic_read(&pool->available) < pool->threshold))
326                         ibmveth_replenish_buffer_pool(adapter, pool);
327         }
328
329         adapter->rx_no_buffer = *(u64*)(((char*)adapter->buffer_list_addr) + 4096 - 8);
330 }
331
332 /* empty and free ana buffer pool - also used to do cleanup in error paths */
333 static void ibmveth_free_buffer_pool(struct ibmveth_adapter *adapter, struct ibmveth_buff_pool *pool)
334 {
335         int i;
336
337         kfree(pool->free_map);
338         pool->free_map = NULL;
339
340         if(pool->skbuff && pool->dma_addr) {
341                 for(i = 0; i < pool->size; ++i) {
342                         struct sk_buff *skb = pool->skbuff[i];
343                         if(skb) {
344                                 dma_unmap_single(&adapter->vdev->dev,
345                                                  pool->dma_addr[i],
346                                                  pool->buff_size,
347                                                  DMA_FROM_DEVICE);
348                                 dev_kfree_skb_any(skb);
349                                 pool->skbuff[i] = NULL;
350                         }
351                 }
352         }
353
354         if(pool->dma_addr) {
355                 kfree(pool->dma_addr);
356                 pool->dma_addr = NULL;
357         }
358
359         if(pool->skbuff) {
360                 kfree(pool->skbuff);
361                 pool->skbuff = NULL;
362         }
363 }
364
365 /* remove a buffer from a pool */
366 static void ibmveth_remove_buffer_from_pool(struct ibmveth_adapter *adapter, u64 correlator)
367 {
368         unsigned int pool  = correlator >> 32;
369         unsigned int index = correlator & 0xffffffffUL;
370         unsigned int free_index;
371         struct sk_buff *skb;
372
373         ibmveth_assert(pool < IbmVethNumBufferPools);
374         ibmveth_assert(index < adapter->rx_buff_pool[pool].size);
375
376         skb = adapter->rx_buff_pool[pool].skbuff[index];
377
378         ibmveth_assert(skb != NULL);
379
380         adapter->rx_buff_pool[pool].skbuff[index] = NULL;
381
382         dma_unmap_single(&adapter->vdev->dev,
383                          adapter->rx_buff_pool[pool].dma_addr[index],
384                          adapter->rx_buff_pool[pool].buff_size,
385                          DMA_FROM_DEVICE);
386
387         free_index = adapter->rx_buff_pool[pool].producer_index;
388         adapter->rx_buff_pool[pool].producer_index++;
389         if (adapter->rx_buff_pool[pool].producer_index >=
390             adapter->rx_buff_pool[pool].size)
391                 adapter->rx_buff_pool[pool].producer_index = 0;
392         adapter->rx_buff_pool[pool].free_map[free_index] = index;
393
394         mb();
395
396         atomic_dec(&(adapter->rx_buff_pool[pool].available));
397 }
398
399 /* get the current buffer on the rx queue */
400 static inline struct sk_buff *ibmveth_rxq_get_buffer(struct ibmveth_adapter *adapter)
401 {
402         u64 correlator = adapter->rx_queue.queue_addr[adapter->rx_queue.index].correlator;
403         unsigned int pool = correlator >> 32;
404         unsigned int index = correlator & 0xffffffffUL;
405
406         ibmveth_assert(pool < IbmVethNumBufferPools);
407         ibmveth_assert(index < adapter->rx_buff_pool[pool].size);
408
409         return adapter->rx_buff_pool[pool].skbuff[index];
410 }
411
412 /* recycle the current buffer on the rx queue */
413 static void ibmveth_rxq_recycle_buffer(struct ibmveth_adapter *adapter)
414 {
415         u32 q_index = adapter->rx_queue.index;
416         u64 correlator = adapter->rx_queue.queue_addr[q_index].correlator;
417         unsigned int pool = correlator >> 32;
418         unsigned int index = correlator & 0xffffffffUL;
419         union ibmveth_buf_desc desc;
420         unsigned long lpar_rc;
421
422         ibmveth_assert(pool < IbmVethNumBufferPools);
423         ibmveth_assert(index < adapter->rx_buff_pool[pool].size);
424
425         if(!adapter->rx_buff_pool[pool].active) {
426                 ibmveth_rxq_harvest_buffer(adapter);
427                 ibmveth_free_buffer_pool(adapter, &adapter->rx_buff_pool[pool]);
428                 return;
429         }
430
431         desc.fields.flags_len = IBMVETH_BUF_VALID |
432                 adapter->rx_buff_pool[pool].buff_size;
433         desc.fields.address = adapter->rx_buff_pool[pool].dma_addr[index];
434
435         lpar_rc = h_add_logical_lan_buffer(adapter->vdev->unit_address, desc.desc);
436
437         if(lpar_rc != H_SUCCESS) {
438                 netdev_dbg(adapter->netdev, "h_add_logical_lan_buffer failed "
439                            "during recycle rc=%ld", lpar_rc);
440                 ibmveth_remove_buffer_from_pool(adapter, adapter->rx_queue.queue_addr[adapter->rx_queue.index].correlator);
441         }
442
443         if(++adapter->rx_queue.index == adapter->rx_queue.num_slots) {
444                 adapter->rx_queue.index = 0;
445                 adapter->rx_queue.toggle = !adapter->rx_queue.toggle;
446         }
447 }
448
449 static 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
459 static 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                 if (!dma_mapping_error(dev, adapter->rx_queue.queue_dma)) {
486                         dma_unmap_single(dev,
487                                         adapter->rx_queue.queue_dma,
488                                         adapter->rx_queue.queue_len,
489                                         DMA_BIDIRECTIONAL);
490                         adapter->rx_queue.queue_dma = DMA_ERROR_CODE;
491                 }
492                 kfree(adapter->rx_queue.queue_addr);
493                 adapter->rx_queue.queue_addr = NULL;
494         }
495
496         for(i = 0; i<IbmVethNumBufferPools; i++)
497                 if (adapter->rx_buff_pool[i].active)
498                         ibmveth_free_buffer_pool(adapter,
499                                                  &adapter->rx_buff_pool[i]);
500
501         if (adapter->bounce_buffer != NULL) {
502                 if (!dma_mapping_error(dev, adapter->bounce_buffer_dma)) {
503                         dma_unmap_single(&adapter->vdev->dev,
504                                         adapter->bounce_buffer_dma,
505                                         adapter->netdev->mtu + IBMVETH_BUFF_OH,
506                                         DMA_BIDIRECTIONAL);
507                         adapter->bounce_buffer_dma = DMA_ERROR_CODE;
508                 }
509                 kfree(adapter->bounce_buffer);
510                 adapter->bounce_buffer = NULL;
511         }
512 }
513
514 static int ibmveth_register_logical_lan(struct ibmveth_adapter *adapter,
515         union ibmveth_buf_desc rxq_desc, u64 mac_address)
516 {
517         int rc, try_again = 1;
518
519         /* After a kexec the adapter will still be open, so our attempt to
520         * open it will fail. So if we get a failure we free the adapter and
521         * try again, but only once. */
522 retry:
523         rc = h_register_logical_lan(adapter->vdev->unit_address,
524                                     adapter->buffer_list_dma, rxq_desc.desc,
525                                     adapter->filter_list_dma, mac_address);
526
527         if (rc != H_SUCCESS && try_again) {
528                 do {
529                         rc = h_free_logical_lan(adapter->vdev->unit_address);
530                 } while (H_IS_LONG_BUSY(rc) || (rc == H_BUSY));
531
532                 try_again = 0;
533                 goto retry;
534         }
535
536         return rc;
537 }
538
539 static int ibmveth_open(struct net_device *netdev)
540 {
541         struct ibmveth_adapter *adapter = netdev_priv(netdev);
542         u64 mac_address = 0;
543         int rxq_entries = 1;
544         unsigned long lpar_rc;
545         int rc;
546         union ibmveth_buf_desc rxq_desc;
547         int i;
548         struct device *dev;
549
550         netdev_dbg(netdev, "open starting\n");
551
552         napi_enable(&adapter->napi);
553
554         for(i = 0; i<IbmVethNumBufferPools; i++)
555                 rxq_entries += adapter->rx_buff_pool[i].size;
556
557         adapter->buffer_list_addr = (void*) get_zeroed_page(GFP_KERNEL);
558         adapter->filter_list_addr = (void*) get_zeroed_page(GFP_KERNEL);
559
560         if(!adapter->buffer_list_addr || !adapter->filter_list_addr) {
561                 ibmveth_error_printk("unable to allocate filter or buffer list pages\n");
562                 ibmveth_cleanup(adapter);
563                 napi_disable(&adapter->napi);
564                 return -ENOMEM;
565         }
566
567         adapter->rx_queue.queue_len = sizeof(struct ibmveth_rx_q_entry) * rxq_entries;
568         adapter->rx_queue.queue_addr = kmalloc(adapter->rx_queue.queue_len, GFP_KERNEL);
569
570         if(!adapter->rx_queue.queue_addr) {
571                 ibmveth_error_printk("unable to allocate rx queue pages\n");
572                 ibmveth_cleanup(adapter);
573                 napi_disable(&adapter->napi);
574                 return -ENOMEM;
575         }
576
577         dev = &adapter->vdev->dev;
578
579         adapter->buffer_list_dma = dma_map_single(dev,
580                         adapter->buffer_list_addr, 4096, DMA_BIDIRECTIONAL);
581         adapter->filter_list_dma = dma_map_single(dev,
582                         adapter->filter_list_addr, 4096, DMA_BIDIRECTIONAL);
583         adapter->rx_queue.queue_dma = dma_map_single(dev,
584                         adapter->rx_queue.queue_addr,
585                         adapter->rx_queue.queue_len, DMA_BIDIRECTIONAL);
586
587         if ((dma_mapping_error(dev, adapter->buffer_list_dma)) ||
588             (dma_mapping_error(dev, adapter->filter_list_dma)) ||
589             (dma_mapping_error(dev, adapter->rx_queue.queue_dma))) {
590                 ibmveth_error_printk("unable to map filter or buffer list pages\n");
591                 ibmveth_cleanup(adapter);
592                 napi_disable(&adapter->napi);
593                 return -ENOMEM;
594         }
595
596         adapter->rx_queue.index = 0;
597         adapter->rx_queue.num_slots = rxq_entries;
598         adapter->rx_queue.toggle = 1;
599
600         memcpy(&mac_address, netdev->dev_addr, netdev->addr_len);
601         mac_address = mac_address >> 16;
602
603         rxq_desc.fields.flags_len = IBMVETH_BUF_VALID | adapter->rx_queue.queue_len;
604         rxq_desc.fields.address = adapter->rx_queue.queue_dma;
605
606         netdev_dbg(netdev, "buffer list @ 0x%p\n", adapter->buffer_list_addr);
607         netdev_dbg(netdev, "filter list @ 0x%p\n", adapter->filter_list_addr);
608         netdev_dbg(netdev, "receive q   @ 0x%p\n", adapter->rx_queue.queue_addr);
609
610         h_vio_signal(adapter->vdev->unit_address, VIO_IRQ_DISABLE);
611
612         lpar_rc = ibmveth_register_logical_lan(adapter, rxq_desc, mac_address);
613
614         if(lpar_rc != H_SUCCESS) {
615                 ibmveth_error_printk("h_register_logical_lan failed with %ld\n", lpar_rc);
616                 ibmveth_error_printk("buffer TCE:0x%llx filter TCE:0x%llx rxq desc:0x%llx MAC:0x%llx\n",
617                                      adapter->buffer_list_dma,
618                                      adapter->filter_list_dma,
619                                      rxq_desc.desc,
620                                      mac_address);
621                 ibmveth_cleanup(adapter);
622                 napi_disable(&adapter->napi);
623                 return -ENONET;
624         }
625
626         for(i = 0; i<IbmVethNumBufferPools; i++) {
627                 if(!adapter->rx_buff_pool[i].active)
628                         continue;
629                 if (ibmveth_alloc_buffer_pool(&adapter->rx_buff_pool[i])) {
630                         ibmveth_error_printk("unable to alloc pool\n");
631                         adapter->rx_buff_pool[i].active = 0;
632                         ibmveth_cleanup(adapter);
633                         napi_disable(&adapter->napi);
634                         return -ENOMEM ;
635                 }
636         }
637
638         netdev_dbg(netdev, "registering irq 0x%x\n", netdev->irq);
639         if((rc = request_irq(netdev->irq, ibmveth_interrupt, 0, netdev->name, netdev)) != 0) {
640                 ibmveth_error_printk("unable to request irq 0x%x, rc %d\n", netdev->irq, rc);
641                 do {
642                         rc = h_free_logical_lan(adapter->vdev->unit_address);
643                 } while (H_IS_LONG_BUSY(rc) || (rc == H_BUSY));
644
645                 ibmveth_cleanup(adapter);
646                 napi_disable(&adapter->napi);
647                 return rc;
648         }
649
650         adapter->bounce_buffer =
651             kmalloc(netdev->mtu + IBMVETH_BUFF_OH, GFP_KERNEL);
652         if (!adapter->bounce_buffer) {
653                 ibmveth_error_printk("unable to allocate bounce buffer\n");
654                 ibmveth_cleanup(adapter);
655                 napi_disable(&adapter->napi);
656                 return -ENOMEM;
657         }
658         adapter->bounce_buffer_dma =
659             dma_map_single(&adapter->vdev->dev, adapter->bounce_buffer,
660                            netdev->mtu + IBMVETH_BUFF_OH, DMA_BIDIRECTIONAL);
661         if (dma_mapping_error(dev, adapter->bounce_buffer_dma)) {
662                 ibmveth_error_printk("unable to map bounce buffer\n");
663                 ibmveth_cleanup(adapter);
664                 napi_disable(&adapter->napi);
665                 return -ENOMEM;
666         }
667
668         netdev_dbg(netdev, "initial replenish cycle\n");
669         ibmveth_interrupt(netdev->irq, netdev);
670
671         netif_start_queue(netdev);
672
673         netdev_dbg(netdev, "open complete\n");
674
675         return 0;
676 }
677
678 static int ibmveth_close(struct net_device *netdev)
679 {
680         struct ibmveth_adapter *adapter = netdev_priv(netdev);
681         long lpar_rc;
682
683         netdev_dbg(netdev, "close starting\n");
684
685         napi_disable(&adapter->napi);
686
687         if (!adapter->pool_config)
688                 netif_stop_queue(netdev);
689
690         h_vio_signal(adapter->vdev->unit_address, VIO_IRQ_DISABLE);
691
692         do {
693                 lpar_rc = h_free_logical_lan(adapter->vdev->unit_address);
694         } while (H_IS_LONG_BUSY(lpar_rc) || (lpar_rc == H_BUSY));
695
696         if(lpar_rc != H_SUCCESS)
697         {
698                 ibmveth_error_printk("h_free_logical_lan failed with %lx, continuing with close\n",
699                                      lpar_rc);
700         }
701
702         free_irq(netdev->irq, netdev);
703
704         adapter->rx_no_buffer = *(u64*)(((char*)adapter->buffer_list_addr) + 4096 - 8);
705
706         ibmveth_cleanup(adapter);
707
708         netdev_dbg(netdev, "close complete\n");
709
710         return 0;
711 }
712
713 static int netdev_get_settings(struct net_device *dev, struct ethtool_cmd *cmd) {
714         cmd->supported = (SUPPORTED_1000baseT_Full | SUPPORTED_Autoneg | SUPPORTED_FIBRE);
715         cmd->advertising = (ADVERTISED_1000baseT_Full | ADVERTISED_Autoneg | ADVERTISED_FIBRE);
716         cmd->speed = SPEED_1000;
717         cmd->duplex = DUPLEX_FULL;
718         cmd->port = PORT_FIBRE;
719         cmd->phy_address = 0;
720         cmd->transceiver = XCVR_INTERNAL;
721         cmd->autoneg = AUTONEG_ENABLE;
722         cmd->maxtxpkt = 0;
723         cmd->maxrxpkt = 1;
724         return 0;
725 }
726
727 static void netdev_get_drvinfo (struct net_device *dev, struct ethtool_drvinfo *info) {
728         strncpy(info->driver, ibmveth_driver_name, sizeof(info->driver) - 1);
729         strncpy(info->version, ibmveth_driver_version, sizeof(info->version) - 1);
730 }
731
732 static u32 netdev_get_link(struct net_device *dev) {
733         return 1;
734 }
735
736 static void ibmveth_set_rx_csum_flags(struct net_device *dev, u32 data)
737 {
738         struct ibmveth_adapter *adapter = netdev_priv(dev);
739
740         if (data)
741                 adapter->rx_csum = 1;
742         else {
743                 /*
744                  * Since the ibmveth firmware interface does not have the concept of
745                  * separate tx/rx checksum offload enable, if rx checksum is disabled
746                  * we also have to disable tx checksum offload. Once we disable rx
747                  * checksum offload, we are no longer allowed to send tx buffers that
748                  * are not properly checksummed.
749                  */
750                 adapter->rx_csum = 0;
751                 dev->features &= ~NETIF_F_IP_CSUM;
752                 dev->features &= ~NETIF_F_IPV6_CSUM;
753         }
754 }
755
756 static void ibmveth_set_tx_csum_flags(struct net_device *dev, u32 data)
757 {
758         struct ibmveth_adapter *adapter = netdev_priv(dev);
759
760         if (data) {
761                 if (adapter->fw_ipv4_csum_support)
762                         dev->features |= NETIF_F_IP_CSUM;
763                 if (adapter->fw_ipv6_csum_support)
764                         dev->features |= NETIF_F_IPV6_CSUM;
765                 adapter->rx_csum = 1;
766         } else {
767                 dev->features &= ~NETIF_F_IP_CSUM;
768                 dev->features &= ~NETIF_F_IPV6_CSUM;
769         }
770 }
771
772 static int ibmveth_set_csum_offload(struct net_device *dev, u32 data,
773                                     void (*done) (struct net_device *, u32))
774 {
775         struct ibmveth_adapter *adapter = netdev_priv(dev);
776         unsigned long set_attr, clr_attr, ret_attr;
777         unsigned long set_attr6, clr_attr6;
778         long ret, ret6;
779         int rc1 = 0, rc2 = 0;
780         int restart = 0;
781
782         if (netif_running(dev)) {
783                 restart = 1;
784                 adapter->pool_config = 1;
785                 ibmveth_close(dev);
786                 adapter->pool_config = 0;
787         }
788
789         set_attr = 0;
790         clr_attr = 0;
791
792         if (data) {
793                 set_attr = IBMVETH_ILLAN_IPV4_TCP_CSUM;
794                 set_attr6 = IBMVETH_ILLAN_IPV6_TCP_CSUM;
795         } else {
796                 clr_attr = IBMVETH_ILLAN_IPV4_TCP_CSUM;
797                 clr_attr6 = IBMVETH_ILLAN_IPV6_TCP_CSUM;
798         }
799
800         ret = h_illan_attributes(adapter->vdev->unit_address, 0, 0, &ret_attr);
801
802         if (ret == H_SUCCESS && !(ret_attr & IBMVETH_ILLAN_ACTIVE_TRUNK) &&
803             !(ret_attr & IBMVETH_ILLAN_TRUNK_PRI_MASK) &&
804             (ret_attr & IBMVETH_ILLAN_PADDED_PKT_CSUM)) {
805                 ret = h_illan_attributes(adapter->vdev->unit_address, clr_attr,
806                                          set_attr, &ret_attr);
807
808                 if (ret != H_SUCCESS) {
809                         ibmveth_error_printk("unable to change IPv4 checksum "
810                                              "offload settings. %d rc=%ld\n",
811                                              data, ret);
812
813                         ret = h_illan_attributes(adapter->vdev->unit_address,
814                                                  set_attr, clr_attr, &ret_attr);
815                 } else
816                         adapter->fw_ipv4_csum_support = data;
817
818                 ret6 = h_illan_attributes(adapter->vdev->unit_address,
819                                          clr_attr6, set_attr6, &ret_attr);
820
821                 if (ret6 != H_SUCCESS) {
822                         ibmveth_error_printk("unable to change IPv6 checksum "
823                                              "offload settings. %d rc=%ld\n",
824                                              data, ret);
825
826                         ret = h_illan_attributes(adapter->vdev->unit_address,
827                                                  set_attr6, clr_attr6,
828                                                  &ret_attr);
829                 } else
830                         adapter->fw_ipv6_csum_support = data;
831
832                 if (ret == H_SUCCESS || ret6 == H_SUCCESS)
833                         done(dev, data);
834                 else
835                         rc1 = -EIO;
836         } else {
837                 rc1 = -EIO;
838                 ibmveth_error_printk("unable to change checksum offload settings."
839                                      " %d rc=%ld ret_attr=%lx\n", data, ret, ret_attr);
840         }
841
842         if (restart)
843                 rc2 = ibmveth_open(dev);
844
845         return rc1 ? rc1 : rc2;
846 }
847
848 static int ibmveth_set_rx_csum(struct net_device *dev, u32 data)
849 {
850         struct ibmveth_adapter *adapter = netdev_priv(dev);
851
852         if ((data && adapter->rx_csum) || (!data && !adapter->rx_csum))
853                 return 0;
854
855         return ibmveth_set_csum_offload(dev, data, ibmveth_set_rx_csum_flags);
856 }
857
858 static int ibmveth_set_tx_csum(struct net_device *dev, u32 data)
859 {
860         struct ibmveth_adapter *adapter = netdev_priv(dev);
861         int rc = 0;
862
863         if (data && (dev->features & (NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM)))
864                 return 0;
865         if (!data && !(dev->features & (NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM)))
866                 return 0;
867
868         if (data && !adapter->rx_csum)
869                 rc = ibmveth_set_csum_offload(dev, data, ibmveth_set_tx_csum_flags);
870         else
871                 ibmveth_set_tx_csum_flags(dev, data);
872
873         return rc;
874 }
875
876 static u32 ibmveth_get_rx_csum(struct net_device *dev)
877 {
878         struct ibmveth_adapter *adapter = netdev_priv(dev);
879         return adapter->rx_csum;
880 }
881
882 static void ibmveth_get_strings(struct net_device *dev, u32 stringset, u8 *data)
883 {
884         int i;
885
886         if (stringset != ETH_SS_STATS)
887                 return;
888
889         for (i = 0; i < ARRAY_SIZE(ibmveth_stats); i++, data += ETH_GSTRING_LEN)
890                 memcpy(data, ibmveth_stats[i].name, ETH_GSTRING_LEN);
891 }
892
893 static int ibmveth_get_sset_count(struct net_device *dev, int sset)
894 {
895         switch (sset) {
896         case ETH_SS_STATS:
897                 return ARRAY_SIZE(ibmveth_stats);
898         default:
899                 return -EOPNOTSUPP;
900         }
901 }
902
903 static void ibmveth_get_ethtool_stats(struct net_device *dev,
904                                       struct ethtool_stats *stats, u64 *data)
905 {
906         int i;
907         struct ibmveth_adapter *adapter = netdev_priv(dev);
908
909         for (i = 0; i < ARRAY_SIZE(ibmveth_stats); i++)
910                 data[i] = IBMVETH_GET_STAT(adapter, ibmveth_stats[i].offset);
911 }
912
913 static const struct ethtool_ops netdev_ethtool_ops = {
914         .get_drvinfo            = netdev_get_drvinfo,
915         .get_settings           = netdev_get_settings,
916         .get_link               = netdev_get_link,
917         .set_tx_csum            = ibmveth_set_tx_csum,
918         .get_rx_csum            = ibmveth_get_rx_csum,
919         .set_rx_csum            = ibmveth_set_rx_csum,
920         .get_strings            = ibmveth_get_strings,
921         .get_sset_count         = ibmveth_get_sset_count,
922         .get_ethtool_stats      = ibmveth_get_ethtool_stats,
923         .set_sg                 = ethtool_op_set_sg,
924 };
925
926 static int ibmveth_ioctl(struct net_device *dev, struct ifreq *ifr, int cmd)
927 {
928         return -EOPNOTSUPP;
929 }
930
931 #define page_offset(v) ((unsigned long)(v) & ((1 << 12) - 1))
932
933 static int ibmveth_send(struct ibmveth_adapter *adapter,
934                         union ibmveth_buf_desc *descs)
935 {
936         unsigned long correlator;
937         unsigned int retry_count;
938         unsigned long ret;
939
940         /*
941          * The retry count sets a maximum for the number of broadcast and
942          * multicast destinations within the system.
943          */
944         retry_count = 1024;
945         correlator = 0;
946         do {
947                 ret = h_send_logical_lan(adapter->vdev->unit_address,
948                                              descs[0].desc, descs[1].desc,
949                                              descs[2].desc, descs[3].desc,
950                                              descs[4].desc, descs[5].desc,
951                                              correlator, &correlator);
952         } while ((ret == H_BUSY) && (retry_count--));
953
954         if (ret != H_SUCCESS && ret != H_DROPPED) {
955                 ibmveth_error_printk("tx: h_send_logical_lan failed with "
956                                      "rc=%ld\n", ret);
957                 return 1;
958         }
959
960         return 0;
961 }
962
963 static netdev_tx_t ibmveth_start_xmit(struct sk_buff *skb,
964                                       struct net_device *netdev)
965 {
966         struct ibmveth_adapter *adapter = netdev_priv(netdev);
967         unsigned int desc_flags;
968         union ibmveth_buf_desc descs[6];
969         int last, i;
970         int force_bounce = 0;
971
972         /*
973          * veth handles a maximum of 6 segments including the header, so
974          * we have to linearize the skb if there are more than this.
975          */
976         if (skb_shinfo(skb)->nr_frags > 5 && __skb_linearize(skb)) {
977                 netdev->stats.tx_dropped++;
978                 goto out;
979         }
980
981         /* veth can't checksum offload UDP */
982         if (skb->ip_summed == CHECKSUM_PARTIAL &&
983             ((skb->protocol == htons(ETH_P_IP) &&
984               ip_hdr(skb)->protocol != IPPROTO_TCP) ||
985              (skb->protocol == htons(ETH_P_IPV6) &&
986               ipv6_hdr(skb)->nexthdr != IPPROTO_TCP)) &&
987             skb_checksum_help(skb)) {
988
989                 ibmveth_error_printk("tx: failed to checksum packet\n");
990                 netdev->stats.tx_dropped++;
991                 goto out;
992         }
993
994         desc_flags = IBMVETH_BUF_VALID;
995
996         if (skb->ip_summed == CHECKSUM_PARTIAL) {
997                 unsigned char *buf = skb_transport_header(skb) +
998                                                 skb->csum_offset;
999
1000                 desc_flags |= (IBMVETH_BUF_NO_CSUM | IBMVETH_BUF_CSUM_GOOD);
1001
1002                 /* Need to zero out the checksum */
1003                 buf[0] = 0;
1004                 buf[1] = 0;
1005         }
1006
1007 retry_bounce:
1008         memset(descs, 0, sizeof(descs));
1009
1010         /*
1011          * If a linear packet is below the rx threshold then
1012          * copy it into the static bounce buffer. This avoids the
1013          * cost of a TCE insert and remove.
1014          */
1015         if (force_bounce || (!skb_is_nonlinear(skb) &&
1016                                 (skb->len < tx_copybreak))) {
1017                 skb_copy_from_linear_data(skb, adapter->bounce_buffer,
1018                                           skb->len);
1019
1020                 descs[0].fields.flags_len = desc_flags | skb->len;
1021                 descs[0].fields.address = adapter->bounce_buffer_dma;
1022
1023                 if (ibmveth_send(adapter, descs)) {
1024                         adapter->tx_send_failed++;
1025                         netdev->stats.tx_dropped++;
1026                 } else {
1027                         netdev->stats.tx_packets++;
1028                         netdev->stats.tx_bytes += skb->len;
1029                 }
1030
1031                 goto out;
1032         }
1033
1034         /* Map the header */
1035         descs[0].fields.address = dma_map_single(&adapter->vdev->dev, skb->data,
1036                                                  skb_headlen(skb),
1037                                                  DMA_TO_DEVICE);
1038         if (dma_mapping_error(&adapter->vdev->dev, descs[0].fields.address))
1039                 goto map_failed;
1040
1041         descs[0].fields.flags_len = desc_flags | skb_headlen(skb);
1042
1043         /* Map the frags */
1044         for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
1045                 unsigned long dma_addr;
1046                 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
1047
1048                 dma_addr = dma_map_page(&adapter->vdev->dev, frag->page,
1049                                         frag->page_offset, frag->size,
1050                                         DMA_TO_DEVICE);
1051
1052                 if (dma_mapping_error(&adapter->vdev->dev, dma_addr))
1053                         goto map_failed_frags;
1054
1055                 descs[i+1].fields.flags_len = desc_flags | frag->size;
1056                 descs[i+1].fields.address = dma_addr;
1057         }
1058
1059         if (ibmveth_send(adapter, descs)) {
1060                 adapter->tx_send_failed++;
1061                 netdev->stats.tx_dropped++;
1062         } else {
1063                 netdev->stats.tx_packets++;
1064                 netdev->stats.tx_bytes += skb->len;
1065         }
1066
1067         for (i = 0; i < skb_shinfo(skb)->nr_frags + 1; i++)
1068                 dma_unmap_page(&adapter->vdev->dev, descs[i].fields.address,
1069                                descs[i].fields.flags_len & IBMVETH_BUF_LEN_MASK,
1070                                DMA_TO_DEVICE);
1071
1072 out:
1073         dev_kfree_skb(skb);
1074         return NETDEV_TX_OK;
1075
1076 map_failed_frags:
1077         last = i+1;
1078         for (i = 0; i < last; i++)
1079                 dma_unmap_page(&adapter->vdev->dev, descs[i].fields.address,
1080                                descs[i].fields.flags_len & IBMVETH_BUF_LEN_MASK,
1081                                DMA_TO_DEVICE);
1082
1083 map_failed:
1084         if (!firmware_has_feature(FW_FEATURE_CMO))
1085                 ibmveth_error_printk("tx: unable to map xmit buffer\n");
1086         adapter->tx_map_failed++;
1087         skb_linearize(skb);
1088         force_bounce = 1;
1089         goto retry_bounce;
1090 }
1091
1092 static int ibmveth_poll(struct napi_struct *napi, int budget)
1093 {
1094         struct ibmveth_adapter *adapter = container_of(napi, struct ibmveth_adapter, napi);
1095         struct net_device *netdev = adapter->netdev;
1096         int frames_processed = 0;
1097         unsigned long lpar_rc;
1098
1099  restart_poll:
1100         do {
1101                 if (!ibmveth_rxq_pending_buffer(adapter))
1102                         break;
1103
1104                 smp_rmb();
1105                 if (!ibmveth_rxq_buffer_valid(adapter)) {
1106                         wmb(); /* suggested by larson1 */
1107                         adapter->rx_invalid_buffer++;
1108                         netdev_dbg(netdev, "recycling invalid buffer\n");
1109                         ibmveth_rxq_recycle_buffer(adapter);
1110                 } else {
1111                         struct sk_buff *skb, *new_skb;
1112                         int length = ibmveth_rxq_frame_length(adapter);
1113                         int offset = ibmveth_rxq_frame_offset(adapter);
1114                         int csum_good = ibmveth_rxq_csum_good(adapter);
1115
1116                         skb = ibmveth_rxq_get_buffer(adapter);
1117
1118                         new_skb = NULL;
1119                         if (length < rx_copybreak)
1120                                 new_skb = netdev_alloc_skb(netdev, length);
1121
1122                         if (new_skb) {
1123                                 skb_copy_to_linear_data(new_skb,
1124                                                         skb->data + offset,
1125                                                         length);
1126                                 if (rx_flush)
1127                                         ibmveth_flush_buffer(skb->data,
1128                                                 length + offset);
1129                                 skb = new_skb;
1130                                 ibmveth_rxq_recycle_buffer(adapter);
1131                         } else {
1132                                 ibmveth_rxq_harvest_buffer(adapter);
1133                                 skb_reserve(skb, offset);
1134                         }
1135
1136                         skb_put(skb, length);
1137                         skb->protocol = eth_type_trans(skb, netdev);
1138
1139                         if (csum_good)
1140                                 skb->ip_summed = CHECKSUM_UNNECESSARY;
1141
1142                         netif_receive_skb(skb); /* send it up */
1143
1144                         netdev->stats.rx_packets++;
1145                         netdev->stats.rx_bytes += length;
1146                         frames_processed++;
1147                 }
1148         } while (frames_processed < budget);
1149
1150         ibmveth_replenish_task(adapter);
1151
1152         if (frames_processed < budget) {
1153                 /* We think we are done - reenable interrupts,
1154                  * then check once more to make sure we are done.
1155                  */
1156                 lpar_rc = h_vio_signal(adapter->vdev->unit_address,
1157                                        VIO_IRQ_ENABLE);
1158
1159                 ibmveth_assert(lpar_rc == H_SUCCESS);
1160
1161                 napi_complete(napi);
1162
1163                 if (ibmveth_rxq_pending_buffer(adapter) &&
1164                     napi_reschedule(napi)) {
1165                         lpar_rc = h_vio_signal(adapter->vdev->unit_address,
1166                                                VIO_IRQ_DISABLE);
1167                         goto restart_poll;
1168                 }
1169         }
1170
1171         return frames_processed;
1172 }
1173
1174 static irqreturn_t ibmveth_interrupt(int irq, void *dev_instance)
1175 {
1176         struct net_device *netdev = dev_instance;
1177         struct ibmveth_adapter *adapter = netdev_priv(netdev);
1178         unsigned long lpar_rc;
1179
1180         if (napi_schedule_prep(&adapter->napi)) {
1181                 lpar_rc = h_vio_signal(adapter->vdev->unit_address,
1182                                        VIO_IRQ_DISABLE);
1183                 ibmveth_assert(lpar_rc == H_SUCCESS);
1184                 __napi_schedule(&adapter->napi);
1185         }
1186         return IRQ_HANDLED;
1187 }
1188
1189 static void ibmveth_set_multicast_list(struct net_device *netdev)
1190 {
1191         struct ibmveth_adapter *adapter = netdev_priv(netdev);
1192         unsigned long lpar_rc;
1193
1194         if ((netdev->flags & IFF_PROMISC) ||
1195             (netdev_mc_count(netdev) > adapter->mcastFilterSize)) {
1196                 lpar_rc = h_multicast_ctrl(adapter->vdev->unit_address,
1197                                            IbmVethMcastEnableRecv |
1198                                            IbmVethMcastDisableFiltering,
1199                                            0);
1200                 if(lpar_rc != H_SUCCESS) {
1201                         ibmveth_error_printk("h_multicast_ctrl rc=%ld when entering promisc mode\n", lpar_rc);
1202                 }
1203         } else {
1204                 struct netdev_hw_addr *ha;
1205                 /* clear the filter table & disable filtering */
1206                 lpar_rc = h_multicast_ctrl(adapter->vdev->unit_address,
1207                                            IbmVethMcastEnableRecv |
1208                                            IbmVethMcastDisableFiltering |
1209                                            IbmVethMcastClearFilterTable,
1210                                            0);
1211                 if(lpar_rc != H_SUCCESS) {
1212                         ibmveth_error_printk("h_multicast_ctrl rc=%ld when attempting to clear filter table\n", lpar_rc);
1213                 }
1214                 /* add the addresses to the filter table */
1215                 netdev_for_each_mc_addr(ha, netdev) {
1216                         // add the multicast address to the filter table
1217                         unsigned long mcast_addr = 0;
1218                         memcpy(((char *)&mcast_addr)+2, ha->addr, 6);
1219                         lpar_rc = h_multicast_ctrl(adapter->vdev->unit_address,
1220                                                    IbmVethMcastAddFilter,
1221                                                    mcast_addr);
1222                         if(lpar_rc != H_SUCCESS) {
1223                                 ibmveth_error_printk("h_multicast_ctrl rc=%ld when adding an entry to the filter table\n", lpar_rc);
1224                         }
1225                 }
1226
1227                 /* re-enable filtering */
1228                 lpar_rc = h_multicast_ctrl(adapter->vdev->unit_address,
1229                                            IbmVethMcastEnableFiltering,
1230                                            0);
1231                 if(lpar_rc != H_SUCCESS) {
1232                         ibmveth_error_printk("h_multicast_ctrl rc=%ld when enabling filtering\n", lpar_rc);
1233                 }
1234         }
1235 }
1236
1237 static int ibmveth_change_mtu(struct net_device *dev, int new_mtu)
1238 {
1239         struct ibmveth_adapter *adapter = netdev_priv(dev);
1240         struct vio_dev *viodev = adapter->vdev;
1241         int new_mtu_oh = new_mtu + IBMVETH_BUFF_OH;
1242         int i, rc;
1243         int need_restart = 0;
1244
1245         if (new_mtu < IBMVETH_MAX_MTU)
1246                 return -EINVAL;
1247
1248         for (i = 0; i < IbmVethNumBufferPools; i++)
1249                 if (new_mtu_oh < adapter->rx_buff_pool[i].buff_size)
1250                         break;
1251
1252         if (i == IbmVethNumBufferPools)
1253                 return -EINVAL;
1254
1255         /* Deactivate all the buffer pools so that the next loop can activate
1256            only the buffer pools necessary to hold the new MTU */
1257         if (netif_running(adapter->netdev)) {
1258                 need_restart = 1;
1259                 adapter->pool_config = 1;
1260                 ibmveth_close(adapter->netdev);
1261                 adapter->pool_config = 0;
1262         }
1263
1264         /* Look for an active buffer pool that can hold the new MTU */
1265         for(i = 0; i<IbmVethNumBufferPools; i++) {
1266                 adapter->rx_buff_pool[i].active = 1;
1267
1268                 if (new_mtu_oh < adapter->rx_buff_pool[i].buff_size) {
1269                         dev->mtu = new_mtu;
1270                         vio_cmo_set_dev_desired(viodev,
1271                                                 ibmveth_get_desired_dma
1272                                                 (viodev));
1273                         if (need_restart) {
1274                                 return ibmveth_open(adapter->netdev);
1275                         }
1276                         return 0;
1277                 }
1278         }
1279
1280         if (need_restart && (rc = ibmveth_open(adapter->netdev)))
1281                 return rc;
1282
1283         return -EINVAL;
1284 }
1285
1286 #ifdef CONFIG_NET_POLL_CONTROLLER
1287 static void ibmveth_poll_controller(struct net_device *dev)
1288 {
1289         ibmveth_replenish_task(netdev_priv(dev));
1290         ibmveth_interrupt(dev->irq, dev);
1291 }
1292 #endif
1293
1294 /**
1295  * ibmveth_get_desired_dma - Calculate IO memory desired by the driver
1296  *
1297  * @vdev: struct vio_dev for the device whose desired IO mem is to be returned
1298  *
1299  * Return value:
1300  *      Number of bytes of IO data the driver will need to perform well.
1301  */
1302 static unsigned long ibmveth_get_desired_dma(struct vio_dev *vdev)
1303 {
1304         struct net_device *netdev = dev_get_drvdata(&vdev->dev);
1305         struct ibmveth_adapter *adapter;
1306         unsigned long ret;
1307         int i;
1308         int rxqentries = 1;
1309
1310         /* netdev inits at probe time along with the structures we need below*/
1311         if (netdev == NULL)
1312                 return IOMMU_PAGE_ALIGN(IBMVETH_IO_ENTITLEMENT_DEFAULT);
1313
1314         adapter = netdev_priv(netdev);
1315
1316         ret = IBMVETH_BUFF_LIST_SIZE + IBMVETH_FILT_LIST_SIZE;
1317         ret += IOMMU_PAGE_ALIGN(netdev->mtu);
1318
1319         for (i = 0; i < IbmVethNumBufferPools; i++) {
1320                 /* add the size of the active receive buffers */
1321                 if (adapter->rx_buff_pool[i].active)
1322                         ret +=
1323                             adapter->rx_buff_pool[i].size *
1324                             IOMMU_PAGE_ALIGN(adapter->rx_buff_pool[i].
1325                                     buff_size);
1326                 rxqentries += adapter->rx_buff_pool[i].size;
1327         }
1328         /* add the size of the receive queue entries */
1329         ret += IOMMU_PAGE_ALIGN(rxqentries * sizeof(struct ibmveth_rx_q_entry));
1330
1331         return ret;
1332 }
1333
1334 static const struct net_device_ops ibmveth_netdev_ops = {
1335         .ndo_open               = ibmveth_open,
1336         .ndo_stop               = ibmveth_close,
1337         .ndo_start_xmit         = ibmveth_start_xmit,
1338         .ndo_set_multicast_list = ibmveth_set_multicast_list,
1339         .ndo_do_ioctl           = ibmveth_ioctl,
1340         .ndo_change_mtu         = ibmveth_change_mtu,
1341         .ndo_validate_addr      = eth_validate_addr,
1342         .ndo_set_mac_address    = eth_mac_addr,
1343 #ifdef CONFIG_NET_POLL_CONTROLLER
1344         .ndo_poll_controller    = ibmveth_poll_controller,
1345 #endif
1346 };
1347
1348 static int __devinit ibmveth_probe(struct vio_dev *dev, const struct vio_device_id *id)
1349 {
1350         int rc, i;
1351         struct net_device *netdev;
1352         struct ibmveth_adapter *adapter;
1353
1354         unsigned char *mac_addr_p;
1355         unsigned int *mcastFilterSize_p;
1356
1357
1358         dev_dbg(&dev->dev, "entering ibmveth_probe for UA 0x%x\n",
1359                 dev->unit_address);
1360
1361         mac_addr_p = (unsigned char *) vio_get_attribute(dev,
1362                                                 VETH_MAC_ADDR, NULL);
1363         if(!mac_addr_p) {
1364                 printk(KERN_ERR "(%s:%3.3d) ERROR: Can't find VETH_MAC_ADDR "
1365                                 "attribute\n", __FILE__, __LINE__);
1366                 return 0;
1367         }
1368
1369         mcastFilterSize_p = (unsigned int *) vio_get_attribute(dev,
1370                                                 VETH_MCAST_FILTER_SIZE, NULL);
1371         if(!mcastFilterSize_p) {
1372                 printk(KERN_ERR "(%s:%3.3d) ERROR: Can't find "
1373                                 "VETH_MCAST_FILTER_SIZE attribute\n",
1374                                 __FILE__, __LINE__);
1375                 return 0;
1376         }
1377
1378         netdev = alloc_etherdev(sizeof(struct ibmveth_adapter));
1379
1380         if(!netdev)
1381                 return -ENOMEM;
1382
1383         adapter = netdev_priv(netdev);
1384         dev_set_drvdata(&dev->dev, netdev);
1385
1386         adapter->vdev = dev;
1387         adapter->netdev = netdev;
1388         adapter->mcastFilterSize= *mcastFilterSize_p;
1389         adapter->pool_config = 0;
1390
1391         netif_napi_add(netdev, &adapter->napi, ibmveth_poll, 16);
1392
1393         /*      Some older boxes running PHYP non-natively have an OF that
1394                 returns a 8-byte local-mac-address field (and the first
1395                 2 bytes have to be ignored) while newer boxes' OF return
1396                 a 6-byte field. Note that IEEE 1275 specifies that
1397                 local-mac-address must be a 6-byte field.
1398                 The RPA doc specifies that the first byte must be 10b, so
1399                 we'll just look for it to solve this 8 vs. 6 byte field issue */
1400
1401         if ((*mac_addr_p & 0x3) != 0x02)
1402                 mac_addr_p += 2;
1403
1404         adapter->mac_addr = 0;
1405         memcpy(&adapter->mac_addr, mac_addr_p, 6);
1406
1407         netdev->irq = dev->irq;
1408         netdev->netdev_ops = &ibmveth_netdev_ops;
1409         netdev->ethtool_ops = &netdev_ethtool_ops;
1410         SET_NETDEV_DEV(netdev, &dev->dev);
1411         netdev->features |= NETIF_F_SG;
1412
1413         memcpy(netdev->dev_addr, &adapter->mac_addr, netdev->addr_len);
1414
1415         for(i = 0; i<IbmVethNumBufferPools; i++) {
1416                 struct kobject *kobj = &adapter->rx_buff_pool[i].kobj;
1417                 int error;
1418
1419                 ibmveth_init_buffer_pool(&adapter->rx_buff_pool[i], i,
1420                                          pool_count[i], pool_size[i],
1421                                          pool_active[i]);
1422                 error = kobject_init_and_add(kobj, &ktype_veth_pool,
1423                                              &dev->dev.kobj, "pool%d", i);
1424                 if (!error)
1425                         kobject_uevent(kobj, KOBJ_ADD);
1426         }
1427
1428         netdev_dbg(netdev, "adapter @ 0x%p\n", adapter);
1429
1430         adapter->buffer_list_dma = DMA_ERROR_CODE;
1431         adapter->filter_list_dma = DMA_ERROR_CODE;
1432         adapter->rx_queue.queue_dma = DMA_ERROR_CODE;
1433
1434         netdev_dbg(netdev, "registering netdev...\n");
1435
1436         ibmveth_set_csum_offload(netdev, 1, ibmveth_set_tx_csum_flags);
1437
1438         rc = register_netdev(netdev);
1439
1440         if(rc) {
1441                 netdev_dbg(netdev, "failed to register netdev rc=%d\n", rc);
1442                 free_netdev(netdev);
1443                 return rc;
1444         }
1445
1446         netdev_dbg(netdev, "registered\n");
1447
1448         return 0;
1449 }
1450
1451 static int __devexit ibmveth_remove(struct vio_dev *dev)
1452 {
1453         struct net_device *netdev = dev_get_drvdata(&dev->dev);
1454         struct ibmveth_adapter *adapter = netdev_priv(netdev);
1455         int i;
1456
1457         for(i = 0; i<IbmVethNumBufferPools; i++)
1458                 kobject_put(&adapter->rx_buff_pool[i].kobj);
1459
1460         unregister_netdev(netdev);
1461
1462         free_netdev(netdev);
1463         dev_set_drvdata(&dev->dev, NULL);
1464
1465         return 0;
1466 }
1467
1468 static struct attribute veth_active_attr;
1469 static struct attribute veth_num_attr;
1470 static struct attribute veth_size_attr;
1471
1472 static ssize_t veth_pool_show(struct kobject * kobj,
1473                               struct attribute * attr, char * buf)
1474 {
1475         struct ibmveth_buff_pool *pool = container_of(kobj,
1476                                                       struct ibmveth_buff_pool,
1477                                                       kobj);
1478
1479         if (attr == &veth_active_attr)
1480                 return sprintf(buf, "%d\n", pool->active);
1481         else if (attr == &veth_num_attr)
1482                 return sprintf(buf, "%d\n", pool->size);
1483         else if (attr == &veth_size_attr)
1484                 return sprintf(buf, "%d\n", pool->buff_size);
1485         return 0;
1486 }
1487
1488 static ssize_t veth_pool_store(struct kobject * kobj, struct attribute * attr,
1489 const char * buf, size_t count)
1490 {
1491         struct ibmveth_buff_pool *pool = container_of(kobj,
1492                                                       struct ibmveth_buff_pool,
1493                                                       kobj);
1494         struct net_device *netdev = dev_get_drvdata(
1495             container_of(kobj->parent, struct device, kobj));
1496         struct ibmveth_adapter *adapter = netdev_priv(netdev);
1497         long value = simple_strtol(buf, NULL, 10);
1498         long rc;
1499
1500         if (attr == &veth_active_attr) {
1501                 if (value && !pool->active) {
1502                         if (netif_running(netdev)) {
1503                                 if(ibmveth_alloc_buffer_pool(pool)) {
1504                                         ibmveth_error_printk("unable to alloc pool\n");
1505                                         return -ENOMEM;
1506                                 }
1507                                 pool->active = 1;
1508                                 adapter->pool_config = 1;
1509                                 ibmveth_close(netdev);
1510                                 adapter->pool_config = 0;
1511                                 if ((rc = ibmveth_open(netdev)))
1512                                         return rc;
1513                         } else
1514                                 pool->active = 1;
1515                 } else if (!value && pool->active) {
1516                         int mtu = netdev->mtu + IBMVETH_BUFF_OH;
1517                         int i;
1518                         /* Make sure there is a buffer pool with buffers that
1519                            can hold a packet of the size of the MTU */
1520                         for (i = 0; i < IbmVethNumBufferPools; i++) {
1521                                 if (pool == &adapter->rx_buff_pool[i])
1522                                         continue;
1523                                 if (!adapter->rx_buff_pool[i].active)
1524                                         continue;
1525                                 if (mtu <= adapter->rx_buff_pool[i].buff_size)
1526                                         break;
1527                         }
1528
1529                         if (i == IbmVethNumBufferPools) {
1530                                 ibmveth_error_printk("no active pool >= MTU\n");
1531                                 return -EPERM;
1532                         }
1533
1534                         if (netif_running(netdev)) {
1535                                 adapter->pool_config = 1;
1536                                 ibmveth_close(netdev);
1537                                 pool->active = 0;
1538                                 adapter->pool_config = 0;
1539                                 if ((rc = ibmveth_open(netdev)))
1540                                         return rc;
1541                         }
1542                         pool->active = 0;
1543                 }
1544         } else if (attr == &veth_num_attr) {
1545                 if (value <= 0 || value > IBMVETH_MAX_POOL_COUNT)
1546                         return -EINVAL;
1547                 else {
1548                         if (netif_running(netdev)) {
1549                                 adapter->pool_config = 1;
1550                                 ibmveth_close(netdev);
1551                                 adapter->pool_config = 0;
1552                                 pool->size = value;
1553                                 if ((rc = ibmveth_open(netdev)))
1554                                         return rc;
1555                         } else
1556                                 pool->size = value;
1557                 }
1558         } else if (attr == &veth_size_attr) {
1559                 if (value <= IBMVETH_BUFF_OH || value > IBMVETH_MAX_BUF_SIZE)
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->buff_size = value;
1567                                 if ((rc = ibmveth_open(netdev)))
1568                                         return rc;
1569                         } else
1570                                 pool->buff_size = value;
1571                 }
1572         }
1573
1574         /* kick the interrupt handler to allocate/deallocate pools */
1575         ibmveth_interrupt(netdev->irq, netdev);
1576         return count;
1577 }
1578
1579
1580 #define ATTR(_name, _mode)      \
1581         struct attribute veth_##_name##_attr = {               \
1582         .name = __stringify(_name), .mode = _mode, \
1583         };
1584
1585 static ATTR(active, 0644);
1586 static ATTR(num, 0644);
1587 static ATTR(size, 0644);
1588
1589 static struct attribute * veth_pool_attrs[] = {
1590         &veth_active_attr,
1591         &veth_num_attr,
1592         &veth_size_attr,
1593         NULL,
1594 };
1595
1596 static const struct sysfs_ops veth_pool_ops = {
1597         .show   = veth_pool_show,
1598         .store  = veth_pool_store,
1599 };
1600
1601 static struct kobj_type ktype_veth_pool = {
1602         .release        = NULL,
1603         .sysfs_ops      = &veth_pool_ops,
1604         .default_attrs  = veth_pool_attrs,
1605 };
1606
1607 static int ibmveth_resume(struct device *dev)
1608 {
1609         struct net_device *netdev = dev_get_drvdata(dev);
1610         ibmveth_interrupt(netdev->irq, netdev);
1611         return 0;
1612 }
1613
1614 static struct vio_device_id ibmveth_device_table[] __devinitdata= {
1615         { "network", "IBM,l-lan"},
1616         { "", "" }
1617 };
1618 MODULE_DEVICE_TABLE(vio, ibmveth_device_table);
1619
1620 static struct dev_pm_ops ibmveth_pm_ops = {
1621         .resume = ibmveth_resume
1622 };
1623
1624 static struct vio_driver ibmveth_driver = {
1625         .id_table       = ibmveth_device_table,
1626         .probe          = ibmveth_probe,
1627         .remove         = ibmveth_remove,
1628         .get_desired_dma = ibmveth_get_desired_dma,
1629         .driver         = {
1630                 .name   = ibmveth_driver_name,
1631                 .owner  = THIS_MODULE,
1632                 .pm = &ibmveth_pm_ops,
1633         }
1634 };
1635
1636 static int __init ibmveth_module_init(void)
1637 {
1638         ibmveth_printk("%s: %s %s\n", ibmveth_driver_name, ibmveth_driver_string, ibmveth_driver_version);
1639
1640         return vio_register_driver(&ibmveth_driver);
1641 }
1642
1643 static void __exit ibmveth_module_exit(void)
1644 {
1645         vio_unregister_driver(&ibmveth_driver);
1646 }
1647
1648 module_init(ibmveth_module_init);
1649 module_exit(ibmveth_module_exit);