ibmveth: Convert driver specific error functions to netdev_err
[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 #ifdef DEBUG
61 #define ibmveth_assert(expr) \
62   if(!(expr)) {                                   \
63     printk(KERN_DEBUG "assertion failed (%s:%3.3d ua:%x): %s\n", __FILE__, __LINE__, adapter->vdev->unit_address, #expr); \
64     BUG(); \
65   }
66 #else
67 #define ibmveth_assert(expr)
68 #endif
69
70 static irqreturn_t ibmveth_interrupt(int irq, void *dev_instance);
71 static void ibmveth_rxq_harvest_buffer(struct ibmveth_adapter *adapter);
72 static unsigned long ibmveth_get_desired_dma(struct vio_dev *vdev);
73
74 static struct kobj_type ktype_veth_pool;
75
76
77 static const char ibmveth_driver_name[] = "ibmveth";
78 static const char ibmveth_driver_string[] = "IBM i/pSeries Virtual Ethernet Driver";
79 #define ibmveth_driver_version "1.03"
80
81 MODULE_AUTHOR("Santiago Leon <santil@us.ibm.com>");
82 MODULE_DESCRIPTION("IBM i/pSeries Virtual Ethernet Driver");
83 MODULE_LICENSE("GPL");
84 MODULE_VERSION(ibmveth_driver_version);
85
86 static unsigned int tx_copybreak __read_mostly = 128;
87 module_param(tx_copybreak, uint, 0644);
88 MODULE_PARM_DESC(tx_copybreak,
89         "Maximum size of packet that is copied to a new buffer on transmit");
90
91 static unsigned int rx_copybreak __read_mostly = 128;
92 module_param(rx_copybreak, uint, 0644);
93 MODULE_PARM_DESC(rx_copybreak,
94         "Maximum size of packet that is copied to a new buffer on receive");
95
96 static unsigned int rx_flush __read_mostly = 0;
97 module_param(rx_flush, uint, 0644);
98 MODULE_PARM_DESC(rx_flush, "Flush receive buffers before use");
99
100 struct ibmveth_stat {
101         char name[ETH_GSTRING_LEN];
102         int offset;
103 };
104
105 #define IBMVETH_STAT_OFF(stat) offsetof(struct ibmveth_adapter, stat)
106 #define IBMVETH_GET_STAT(a, off) *((u64 *)(((unsigned long)(a)) + off))
107
108 struct ibmveth_stat ibmveth_stats[] = {
109         { "replenish_task_cycles", IBMVETH_STAT_OFF(replenish_task_cycles) },
110         { "replenish_no_mem", IBMVETH_STAT_OFF(replenish_no_mem) },
111         { "replenish_add_buff_failure", IBMVETH_STAT_OFF(replenish_add_buff_failure) },
112         { "replenish_add_buff_success", IBMVETH_STAT_OFF(replenish_add_buff_success) },
113         { "rx_invalid_buffer", IBMVETH_STAT_OFF(rx_invalid_buffer) },
114         { "rx_no_buffer", IBMVETH_STAT_OFF(rx_no_buffer) },
115         { "tx_map_failed", IBMVETH_STAT_OFF(tx_map_failed) },
116         { "tx_send_failed", IBMVETH_STAT_OFF(tx_send_failed) },
117         { "fw_enabled_ipv4_csum", IBMVETH_STAT_OFF(fw_ipv4_csum_support) },
118         { "fw_enabled_ipv6_csum", IBMVETH_STAT_OFF(fw_ipv6_csum_support) },
119 };
120
121 /* simple methods of getting data from the current rxq entry */
122 static inline u32 ibmveth_rxq_flags(struct ibmveth_adapter *adapter)
123 {
124         return adapter->rx_queue.queue_addr[adapter->rx_queue.index].flags_off;
125 }
126
127 static inline int ibmveth_rxq_toggle(struct ibmveth_adapter *adapter)
128 {
129         return (ibmveth_rxq_flags(adapter) & IBMVETH_RXQ_TOGGLE) >> IBMVETH_RXQ_TOGGLE_SHIFT;
130 }
131
132 static inline int ibmveth_rxq_pending_buffer(struct ibmveth_adapter *adapter)
133 {
134         return (ibmveth_rxq_toggle(adapter) == adapter->rx_queue.toggle);
135 }
136
137 static inline int ibmveth_rxq_buffer_valid(struct ibmveth_adapter *adapter)
138 {
139         return (ibmveth_rxq_flags(adapter) & IBMVETH_RXQ_VALID);
140 }
141
142 static inline int ibmveth_rxq_frame_offset(struct ibmveth_adapter *adapter)
143 {
144         return (ibmveth_rxq_flags(adapter) & IBMVETH_RXQ_OFF_MASK);
145 }
146
147 static inline int ibmveth_rxq_frame_length(struct ibmveth_adapter *adapter)
148 {
149         return (adapter->rx_queue.queue_addr[adapter->rx_queue.index].length);
150 }
151
152 static inline int ibmveth_rxq_csum_good(struct ibmveth_adapter *adapter)
153 {
154         return (ibmveth_rxq_flags(adapter) & IBMVETH_RXQ_CSUM_GOOD);
155 }
156
157 /* setup the initial settings for a buffer pool */
158 static void ibmveth_init_buffer_pool(struct ibmveth_buff_pool *pool, u32 pool_index, u32 pool_size, u32 buff_size, u32 pool_active)
159 {
160         pool->size = pool_size;
161         pool->index = pool_index;
162         pool->buff_size = buff_size;
163         pool->threshold = pool_size * 7 / 8;
164         pool->active = pool_active;
165 }
166
167 /* allocate and setup an buffer pool - called during open */
168 static int ibmveth_alloc_buffer_pool(struct ibmveth_buff_pool *pool)
169 {
170         int i;
171
172         pool->free_map = kmalloc(sizeof(u16) * pool->size, GFP_KERNEL);
173
174         if(!pool->free_map) {
175                 return -1;
176         }
177
178         pool->dma_addr = kmalloc(sizeof(dma_addr_t) * pool->size, GFP_KERNEL);
179         if(!pool->dma_addr) {
180                 kfree(pool->free_map);
181                 pool->free_map = NULL;
182                 return -1;
183         }
184
185         pool->skbuff = kcalloc(pool->size, sizeof(void *), GFP_KERNEL);
186
187         if(!pool->skbuff) {
188                 kfree(pool->dma_addr);
189                 pool->dma_addr = NULL;
190
191                 kfree(pool->free_map);
192                 pool->free_map = NULL;
193                 return -1;
194         }
195
196         memset(pool->dma_addr, 0, sizeof(dma_addr_t) * pool->size);
197
198         for(i = 0; i < pool->size; ++i) {
199                 pool->free_map[i] = i;
200         }
201
202         atomic_set(&pool->available, 0);
203         pool->producer_index = 0;
204         pool->consumer_index = 0;
205
206         return 0;
207 }
208
209 static inline void ibmveth_flush_buffer(void *addr, unsigned long length)
210 {
211         unsigned long offset;
212
213         for (offset = 0; offset < length; offset += SMP_CACHE_BYTES)
214                 asm("dcbfl %0,%1" :: "b" (addr), "r" (offset));
215 }
216
217 /* replenish the buffers for a pool.  note that we don't need to
218  * skb_reserve these since they are used for incoming...
219  */
220 static void ibmveth_replenish_buffer_pool(struct ibmveth_adapter *adapter, struct ibmveth_buff_pool *pool)
221 {
222         u32 i;
223         u32 count = pool->size - atomic_read(&pool->available);
224         u32 buffers_added = 0;
225         struct sk_buff *skb;
226         unsigned int free_index, index;
227         u64 correlator;
228         unsigned long lpar_rc;
229         dma_addr_t dma_addr;
230
231         mb();
232
233         for(i = 0; i < count; ++i) {
234                 union ibmveth_buf_desc desc;
235
236                 skb = netdev_alloc_skb(adapter->netdev, pool->buff_size);
237
238                 if(!skb) {
239                         netdev_dbg(adapter->netdev,
240                                    "replenish: unable to allocate skb\n");
241                         adapter->replenish_no_mem++;
242                         break;
243                 }
244
245                 free_index = pool->consumer_index;
246                 pool->consumer_index++;
247                 if (pool->consumer_index >= pool->size)
248                         pool->consumer_index = 0;
249                 index = pool->free_map[free_index];
250
251                 ibmveth_assert(index != IBM_VETH_INVALID_MAP);
252                 ibmveth_assert(pool->skbuff[index] == NULL);
253
254                 dma_addr = dma_map_single(&adapter->vdev->dev, skb->data,
255                                 pool->buff_size, DMA_FROM_DEVICE);
256
257                 if (dma_mapping_error(&adapter->vdev->dev, dma_addr))
258                         goto failure;
259
260                 pool->free_map[free_index] = IBM_VETH_INVALID_MAP;
261                 pool->dma_addr[index] = dma_addr;
262                 pool->skbuff[index] = skb;
263
264                 correlator = ((u64)pool->index << 32) | index;
265                 *(u64*)skb->data = correlator;
266
267                 desc.fields.flags_len = IBMVETH_BUF_VALID | pool->buff_size;
268                 desc.fields.address = dma_addr;
269
270                 if (rx_flush) {
271                         unsigned int len = min(pool->buff_size,
272                                                 adapter->netdev->mtu +
273                                                 IBMVETH_BUFF_OH);
274                         ibmveth_flush_buffer(skb->data, len);
275                 }
276                 lpar_rc = h_add_logical_lan_buffer(adapter->vdev->unit_address, desc.desc);
277
278                 if (lpar_rc != H_SUCCESS)
279                         goto failure;
280                 else {
281                         buffers_added++;
282                         adapter->replenish_add_buff_success++;
283                 }
284         }
285
286         mb();
287         atomic_add(buffers_added, &(pool->available));
288         return;
289
290 failure:
291         pool->free_map[free_index] = index;
292         pool->skbuff[index] = NULL;
293         if (pool->consumer_index == 0)
294                 pool->consumer_index = pool->size - 1;
295         else
296                 pool->consumer_index--;
297         if (!dma_mapping_error(&adapter->vdev->dev, dma_addr))
298                 dma_unmap_single(&adapter->vdev->dev,
299                                  pool->dma_addr[index], pool->buff_size,
300                                  DMA_FROM_DEVICE);
301         dev_kfree_skb_any(skb);
302         adapter->replenish_add_buff_failure++;
303
304         mb();
305         atomic_add(buffers_added, &(pool->available));
306 }
307
308 /* replenish routine */
309 static void ibmveth_replenish_task(struct ibmveth_adapter *adapter)
310 {
311         int i;
312
313         adapter->replenish_task_cycles++;
314
315         for (i = (IbmVethNumBufferPools - 1); i >= 0; i--) {
316                 struct ibmveth_buff_pool *pool = &adapter->rx_buff_pool[i];
317
318                 if (pool->active &&
319                     (atomic_read(&pool->available) < pool->threshold))
320                         ibmveth_replenish_buffer_pool(adapter, pool);
321         }
322
323         adapter->rx_no_buffer = *(u64*)(((char*)adapter->buffer_list_addr) + 4096 - 8);
324 }
325
326 /* empty and free ana buffer pool - also used to do cleanup in error paths */
327 static void ibmveth_free_buffer_pool(struct ibmveth_adapter *adapter, 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 */
360 static void ibmveth_remove_buffer_from_pool(struct ibmveth_adapter *adapter, u64 correlator)
361 {
362         unsigned int pool  = correlator >> 32;
363         unsigned int index = correlator & 0xffffffffUL;
364         unsigned int free_index;
365         struct sk_buff *skb;
366
367         ibmveth_assert(pool < IbmVethNumBufferPools);
368         ibmveth_assert(index < adapter->rx_buff_pool[pool].size);
369
370         skb = adapter->rx_buff_pool[pool].skbuff[index];
371
372         ibmveth_assert(skb != NULL);
373
374         adapter->rx_buff_pool[pool].skbuff[index] = NULL;
375
376         dma_unmap_single(&adapter->vdev->dev,
377                          adapter->rx_buff_pool[pool].dma_addr[index],
378                          adapter->rx_buff_pool[pool].buff_size,
379                          DMA_FROM_DEVICE);
380
381         free_index = adapter->rx_buff_pool[pool].producer_index;
382         adapter->rx_buff_pool[pool].producer_index++;
383         if (adapter->rx_buff_pool[pool].producer_index >=
384             adapter->rx_buff_pool[pool].size)
385                 adapter->rx_buff_pool[pool].producer_index = 0;
386         adapter->rx_buff_pool[pool].free_map[free_index] = index;
387
388         mb();
389
390         atomic_dec(&(adapter->rx_buff_pool[pool].available));
391 }
392
393 /* get the current buffer on the rx queue */
394 static inline struct sk_buff *ibmveth_rxq_get_buffer(struct ibmveth_adapter *adapter)
395 {
396         u64 correlator = adapter->rx_queue.queue_addr[adapter->rx_queue.index].correlator;
397         unsigned int pool = correlator >> 32;
398         unsigned int index = correlator & 0xffffffffUL;
399
400         ibmveth_assert(pool < IbmVethNumBufferPools);
401         ibmveth_assert(index < adapter->rx_buff_pool[pool].size);
402
403         return adapter->rx_buff_pool[pool].skbuff[index];
404 }
405
406 /* recycle the current buffer on the rx queue */
407 static void ibmveth_rxq_recycle_buffer(struct ibmveth_adapter *adapter)
408 {
409         u32 q_index = adapter->rx_queue.index;
410         u64 correlator = adapter->rx_queue.queue_addr[q_index].correlator;
411         unsigned int pool = correlator >> 32;
412         unsigned int index = correlator & 0xffffffffUL;
413         union ibmveth_buf_desc desc;
414         unsigned long lpar_rc;
415
416         ibmveth_assert(pool < IbmVethNumBufferPools);
417         ibmveth_assert(index < adapter->rx_buff_pool[pool].size);
418
419         if(!adapter->rx_buff_pool[pool].active) {
420                 ibmveth_rxq_harvest_buffer(adapter);
421                 ibmveth_free_buffer_pool(adapter, &adapter->rx_buff_pool[pool]);
422                 return;
423         }
424
425         desc.fields.flags_len = IBMVETH_BUF_VALID |
426                 adapter->rx_buff_pool[pool].buff_size;
427         desc.fields.address = adapter->rx_buff_pool[pool].dma_addr[index];
428
429         lpar_rc = h_add_logical_lan_buffer(adapter->vdev->unit_address, desc.desc);
430
431         if(lpar_rc != H_SUCCESS) {
432                 netdev_dbg(adapter->netdev, "h_add_logical_lan_buffer failed "
433                            "during recycle rc=%ld", lpar_rc);
434                 ibmveth_remove_buffer_from_pool(adapter, adapter->rx_queue.queue_addr[adapter->rx_queue.index].correlator);
435         }
436
437         if(++adapter->rx_queue.index == adapter->rx_queue.num_slots) {
438                 adapter->rx_queue.index = 0;
439                 adapter->rx_queue.toggle = !adapter->rx_queue.toggle;
440         }
441 }
442
443 static void ibmveth_rxq_harvest_buffer(struct ibmveth_adapter *adapter)
444 {
445         ibmveth_remove_buffer_from_pool(adapter, adapter->rx_queue.queue_addr[adapter->rx_queue.index].correlator);
446
447         if(++adapter->rx_queue.index == adapter->rx_queue.num_slots) {
448                 adapter->rx_queue.index = 0;
449                 adapter->rx_queue.toggle = !adapter->rx_queue.toggle;
450         }
451 }
452
453 static void ibmveth_cleanup(struct ibmveth_adapter *adapter)
454 {
455         int i;
456         struct device *dev = &adapter->vdev->dev;
457
458         if(adapter->buffer_list_addr != NULL) {
459                 if (!dma_mapping_error(dev, adapter->buffer_list_dma)) {
460                         dma_unmap_single(dev, adapter->buffer_list_dma, 4096,
461                                         DMA_BIDIRECTIONAL);
462                         adapter->buffer_list_dma = DMA_ERROR_CODE;
463                 }
464                 free_page((unsigned long)adapter->buffer_list_addr);
465                 adapter->buffer_list_addr = NULL;
466         }
467
468         if(adapter->filter_list_addr != NULL) {
469                 if (!dma_mapping_error(dev, adapter->filter_list_dma)) {
470                         dma_unmap_single(dev, adapter->filter_list_dma, 4096,
471                                         DMA_BIDIRECTIONAL);
472                         adapter->filter_list_dma = DMA_ERROR_CODE;
473                 }
474                 free_page((unsigned long)adapter->filter_list_addr);
475                 adapter->filter_list_addr = NULL;
476         }
477
478         if(adapter->rx_queue.queue_addr != NULL) {
479                 if (!dma_mapping_error(dev, adapter->rx_queue.queue_dma)) {
480                         dma_unmap_single(dev,
481                                         adapter->rx_queue.queue_dma,
482                                         adapter->rx_queue.queue_len,
483                                         DMA_BIDIRECTIONAL);
484                         adapter->rx_queue.queue_dma = DMA_ERROR_CODE;
485                 }
486                 kfree(adapter->rx_queue.queue_addr);
487                 adapter->rx_queue.queue_addr = NULL;
488         }
489
490         for(i = 0; i<IbmVethNumBufferPools; i++)
491                 if (adapter->rx_buff_pool[i].active)
492                         ibmveth_free_buffer_pool(adapter,
493                                                  &adapter->rx_buff_pool[i]);
494
495         if (adapter->bounce_buffer != NULL) {
496                 if (!dma_mapping_error(dev, adapter->bounce_buffer_dma)) {
497                         dma_unmap_single(&adapter->vdev->dev,
498                                         adapter->bounce_buffer_dma,
499                                         adapter->netdev->mtu + IBMVETH_BUFF_OH,
500                                         DMA_BIDIRECTIONAL);
501                         adapter->bounce_buffer_dma = DMA_ERROR_CODE;
502                 }
503                 kfree(adapter->bounce_buffer);
504                 adapter->bounce_buffer = NULL;
505         }
506 }
507
508 static int ibmveth_register_logical_lan(struct ibmveth_adapter *adapter,
509         union ibmveth_buf_desc rxq_desc, u64 mac_address)
510 {
511         int rc, try_again = 1;
512
513         /* After a kexec the adapter will still be open, so our attempt to
514         * open it will fail. So if we get a failure we free the adapter and
515         * try again, but only once. */
516 retry:
517         rc = h_register_logical_lan(adapter->vdev->unit_address,
518                                     adapter->buffer_list_dma, rxq_desc.desc,
519                                     adapter->filter_list_dma, mac_address);
520
521         if (rc != H_SUCCESS && try_again) {
522                 do {
523                         rc = h_free_logical_lan(adapter->vdev->unit_address);
524                 } while (H_IS_LONG_BUSY(rc) || (rc == H_BUSY));
525
526                 try_again = 0;
527                 goto retry;
528         }
529
530         return rc;
531 }
532
533 static int ibmveth_open(struct net_device *netdev)
534 {
535         struct ibmveth_adapter *adapter = netdev_priv(netdev);
536         u64 mac_address = 0;
537         int rxq_entries = 1;
538         unsigned long lpar_rc;
539         int rc;
540         union ibmveth_buf_desc rxq_desc;
541         int i;
542         struct device *dev;
543
544         netdev_dbg(netdev, "open starting\n");
545
546         napi_enable(&adapter->napi);
547
548         for(i = 0; i<IbmVethNumBufferPools; i++)
549                 rxq_entries += adapter->rx_buff_pool[i].size;
550
551         adapter->buffer_list_addr = (void*) get_zeroed_page(GFP_KERNEL);
552         adapter->filter_list_addr = (void*) get_zeroed_page(GFP_KERNEL);
553
554         if(!adapter->buffer_list_addr || !adapter->filter_list_addr) {
555                 netdev_err(netdev, "unable to allocate filter or buffer list "
556                            "pages\n");
557                 ibmveth_cleanup(adapter);
558                 napi_disable(&adapter->napi);
559                 return -ENOMEM;
560         }
561
562         adapter->rx_queue.queue_len = sizeof(struct ibmveth_rx_q_entry) * rxq_entries;
563         adapter->rx_queue.queue_addr = kmalloc(adapter->rx_queue.queue_len, GFP_KERNEL);
564
565         if(!adapter->rx_queue.queue_addr) {
566                 netdev_err(netdev, "unable to allocate rx queue pages\n");
567                 ibmveth_cleanup(adapter);
568                 napi_disable(&adapter->napi);
569                 return -ENOMEM;
570         }
571
572         dev = &adapter->vdev->dev;
573
574         adapter->buffer_list_dma = dma_map_single(dev,
575                         adapter->buffer_list_addr, 4096, DMA_BIDIRECTIONAL);
576         adapter->filter_list_dma = dma_map_single(dev,
577                         adapter->filter_list_addr, 4096, DMA_BIDIRECTIONAL);
578         adapter->rx_queue.queue_dma = dma_map_single(dev,
579                         adapter->rx_queue.queue_addr,
580                         adapter->rx_queue.queue_len, DMA_BIDIRECTIONAL);
581
582         if ((dma_mapping_error(dev, adapter->buffer_list_dma)) ||
583             (dma_mapping_error(dev, adapter->filter_list_dma)) ||
584             (dma_mapping_error(dev, adapter->rx_queue.queue_dma))) {
585                 netdev_err(netdev, "unable to map filter or buffer list "
586                            "pages\n");
587                 ibmveth_cleanup(adapter);
588                 napi_disable(&adapter->napi);
589                 return -ENOMEM;
590         }
591
592         adapter->rx_queue.index = 0;
593         adapter->rx_queue.num_slots = rxq_entries;
594         adapter->rx_queue.toggle = 1;
595
596         memcpy(&mac_address, netdev->dev_addr, netdev->addr_len);
597         mac_address = mac_address >> 16;
598
599         rxq_desc.fields.flags_len = IBMVETH_BUF_VALID | adapter->rx_queue.queue_len;
600         rxq_desc.fields.address = adapter->rx_queue.queue_dma;
601
602         netdev_dbg(netdev, "buffer list @ 0x%p\n", adapter->buffer_list_addr);
603         netdev_dbg(netdev, "filter list @ 0x%p\n", adapter->filter_list_addr);
604         netdev_dbg(netdev, "receive q   @ 0x%p\n", adapter->rx_queue.queue_addr);
605
606         h_vio_signal(adapter->vdev->unit_address, VIO_IRQ_DISABLE);
607
608         lpar_rc = ibmveth_register_logical_lan(adapter, rxq_desc, mac_address);
609
610         if(lpar_rc != H_SUCCESS) {
611                 netdev_err(netdev, "h_register_logical_lan failed with %ld\n",
612                            lpar_rc);
613                 netdev_err(netdev, "buffer TCE:0x%llx filter TCE:0x%llx rxq "
614                            "desc:0x%llx MAC:0x%llx\n",
615                                      adapter->buffer_list_dma,
616                                      adapter->filter_list_dma,
617                                      rxq_desc.desc,
618                                      mac_address);
619                 ibmveth_cleanup(adapter);
620                 napi_disable(&adapter->napi);
621                 return -ENONET;
622         }
623
624         for(i = 0; i<IbmVethNumBufferPools; i++) {
625                 if(!adapter->rx_buff_pool[i].active)
626                         continue;
627                 if (ibmveth_alloc_buffer_pool(&adapter->rx_buff_pool[i])) {
628                         netdev_err(netdev, "unable to alloc pool\n");
629                         adapter->rx_buff_pool[i].active = 0;
630                         ibmveth_cleanup(adapter);
631                         napi_disable(&adapter->napi);
632                         return -ENOMEM ;
633                 }
634         }
635
636         netdev_dbg(netdev, "registering irq 0x%x\n", netdev->irq);
637         if((rc = request_irq(netdev->irq, ibmveth_interrupt, 0, netdev->name, netdev)) != 0) {
638                 netdev_err(netdev, "unable to request irq 0x%x, rc %d\n",
639                            netdev->irq, rc);
640                 do {
641                         rc = h_free_logical_lan(adapter->vdev->unit_address);
642                 } while (H_IS_LONG_BUSY(rc) || (rc == H_BUSY));
643
644                 ibmveth_cleanup(adapter);
645                 napi_disable(&adapter->napi);
646                 return rc;
647         }
648
649         adapter->bounce_buffer =
650             kmalloc(netdev->mtu + IBMVETH_BUFF_OH, GFP_KERNEL);
651         if (!adapter->bounce_buffer) {
652                 netdev_err(netdev, "unable to allocate bounce buffer\n");
653                 ibmveth_cleanup(adapter);
654                 napi_disable(&adapter->napi);
655                 return -ENOMEM;
656         }
657         adapter->bounce_buffer_dma =
658             dma_map_single(&adapter->vdev->dev, adapter->bounce_buffer,
659                            netdev->mtu + IBMVETH_BUFF_OH, DMA_BIDIRECTIONAL);
660         if (dma_mapping_error(dev, adapter->bounce_buffer_dma)) {
661                 netdev_err(netdev, "unable to map bounce buffer\n");
662                 ibmveth_cleanup(adapter);
663                 napi_disable(&adapter->napi);
664                 return -ENOMEM;
665         }
666
667         netdev_dbg(netdev, "initial replenish cycle\n");
668         ibmveth_interrupt(netdev->irq, netdev);
669
670         netif_start_queue(netdev);
671
672         netdev_dbg(netdev, "open complete\n");
673
674         return 0;
675 }
676
677 static int ibmveth_close(struct net_device *netdev)
678 {
679         struct ibmveth_adapter *adapter = netdev_priv(netdev);
680         long lpar_rc;
681
682         netdev_dbg(netdev, "close starting\n");
683
684         napi_disable(&adapter->napi);
685
686         if (!adapter->pool_config)
687                 netif_stop_queue(netdev);
688
689         h_vio_signal(adapter->vdev->unit_address, VIO_IRQ_DISABLE);
690
691         do {
692                 lpar_rc = h_free_logical_lan(adapter->vdev->unit_address);
693         } while (H_IS_LONG_BUSY(lpar_rc) || (lpar_rc == H_BUSY));
694
695         if(lpar_rc != H_SUCCESS)
696         {
697                 netdev_err(netdev, "h_free_logical_lan failed with %lx, "
698                            "continuing with close\n", lpar_rc);
699         }
700
701         free_irq(netdev->irq, netdev);
702
703         adapter->rx_no_buffer = *(u64*)(((char*)adapter->buffer_list_addr) + 4096 - 8);
704
705         ibmveth_cleanup(adapter);
706
707         netdev_dbg(netdev, "close complete\n");
708
709         return 0;
710 }
711
712 static int netdev_get_settings(struct net_device *dev, struct ethtool_cmd *cmd) {
713         cmd->supported = (SUPPORTED_1000baseT_Full | SUPPORTED_Autoneg | SUPPORTED_FIBRE);
714         cmd->advertising = (ADVERTISED_1000baseT_Full | ADVERTISED_Autoneg | ADVERTISED_FIBRE);
715         cmd->speed = SPEED_1000;
716         cmd->duplex = DUPLEX_FULL;
717         cmd->port = PORT_FIBRE;
718         cmd->phy_address = 0;
719         cmd->transceiver = XCVR_INTERNAL;
720         cmd->autoneg = AUTONEG_ENABLE;
721         cmd->maxtxpkt = 0;
722         cmd->maxrxpkt = 1;
723         return 0;
724 }
725
726 static void netdev_get_drvinfo (struct net_device *dev, struct ethtool_drvinfo *info) {
727         strncpy(info->driver, ibmveth_driver_name, sizeof(info->driver) - 1);
728         strncpy(info->version, ibmveth_driver_version, sizeof(info->version) - 1);
729 }
730
731 static u32 netdev_get_link(struct net_device *dev) {
732         return 1;
733 }
734
735 static void ibmveth_set_rx_csum_flags(struct net_device *dev, u32 data)
736 {
737         struct ibmveth_adapter *adapter = netdev_priv(dev);
738
739         if (data)
740                 adapter->rx_csum = 1;
741         else {
742                 /*
743                  * Since the ibmveth firmware interface does not have the concept of
744                  * separate tx/rx checksum offload enable, if rx checksum is disabled
745                  * we also have to disable tx checksum offload. Once we disable rx
746                  * checksum offload, we are no longer allowed to send tx buffers that
747                  * are not properly checksummed.
748                  */
749                 adapter->rx_csum = 0;
750                 dev->features &= ~NETIF_F_IP_CSUM;
751                 dev->features &= ~NETIF_F_IPV6_CSUM;
752         }
753 }
754
755 static void ibmveth_set_tx_csum_flags(struct net_device *dev, u32 data)
756 {
757         struct ibmveth_adapter *adapter = netdev_priv(dev);
758
759         if (data) {
760                 if (adapter->fw_ipv4_csum_support)
761                         dev->features |= NETIF_F_IP_CSUM;
762                 if (adapter->fw_ipv6_csum_support)
763                         dev->features |= NETIF_F_IPV6_CSUM;
764                 adapter->rx_csum = 1;
765         } else {
766                 dev->features &= ~NETIF_F_IP_CSUM;
767                 dev->features &= ~NETIF_F_IPV6_CSUM;
768         }
769 }
770
771 static int ibmveth_set_csum_offload(struct net_device *dev, u32 data,
772                                     void (*done) (struct net_device *, u32))
773 {
774         struct ibmveth_adapter *adapter = netdev_priv(dev);
775         unsigned long set_attr, clr_attr, ret_attr;
776         unsigned long set_attr6, clr_attr6;
777         long ret, ret6;
778         int rc1 = 0, rc2 = 0;
779         int restart = 0;
780
781         if (netif_running(dev)) {
782                 restart = 1;
783                 adapter->pool_config = 1;
784                 ibmveth_close(dev);
785                 adapter->pool_config = 0;
786         }
787
788         set_attr = 0;
789         clr_attr = 0;
790
791         if (data) {
792                 set_attr = IBMVETH_ILLAN_IPV4_TCP_CSUM;
793                 set_attr6 = IBMVETH_ILLAN_IPV6_TCP_CSUM;
794         } else {
795                 clr_attr = IBMVETH_ILLAN_IPV4_TCP_CSUM;
796                 clr_attr6 = IBMVETH_ILLAN_IPV6_TCP_CSUM;
797         }
798
799         ret = h_illan_attributes(adapter->vdev->unit_address, 0, 0, &ret_attr);
800
801         if (ret == H_SUCCESS && !(ret_attr & IBMVETH_ILLAN_ACTIVE_TRUNK) &&
802             !(ret_attr & IBMVETH_ILLAN_TRUNK_PRI_MASK) &&
803             (ret_attr & IBMVETH_ILLAN_PADDED_PKT_CSUM)) {
804                 ret = h_illan_attributes(adapter->vdev->unit_address, clr_attr,
805                                          set_attr, &ret_attr);
806
807                 if (ret != H_SUCCESS) {
808                         netdev_err(dev, "unable to change IPv4 checksum "
809                                         "offload settings. %d rc=%ld\n",
810                                         data, ret);
811
812                         ret = h_illan_attributes(adapter->vdev->unit_address,
813                                                  set_attr, clr_attr, &ret_attr);
814                 } else
815                         adapter->fw_ipv4_csum_support = data;
816
817                 ret6 = h_illan_attributes(adapter->vdev->unit_address,
818                                          clr_attr6, set_attr6, &ret_attr);
819
820                 if (ret6 != H_SUCCESS) {
821                         netdev_err(dev, "unable to change IPv6 checksum "
822                                         "offload settings. %d rc=%ld\n",
823                                         data, ret);
824
825                         ret = h_illan_attributes(adapter->vdev->unit_address,
826                                                  set_attr6, clr_attr6,
827                                                  &ret_attr);
828                 } else
829                         adapter->fw_ipv6_csum_support = data;
830
831                 if (ret == H_SUCCESS || ret6 == H_SUCCESS)
832                         done(dev, data);
833                 else
834                         rc1 = -EIO;
835         } else {
836                 rc1 = -EIO;
837                 netdev_err(dev, "unable to change checksum offload settings."
838                                      " %d rc=%ld ret_attr=%lx\n", data, ret,
839                                      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                 netdev_err(adapter->netdev, "tx: h_send_logical_lan failed "
956                            "with 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                 netdev_err(netdev, "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                 netdev_err(netdev, "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                         netdev_err(netdev, "h_multicast_ctrl rc=%ld when "
1202                                    "entering promisc mode\n", lpar_rc);
1203                 }
1204         } else {
1205                 struct netdev_hw_addr *ha;
1206                 /* clear the filter table & disable filtering */
1207                 lpar_rc = h_multicast_ctrl(adapter->vdev->unit_address,
1208                                            IbmVethMcastEnableRecv |
1209                                            IbmVethMcastDisableFiltering |
1210                                            IbmVethMcastClearFilterTable,
1211                                            0);
1212                 if(lpar_rc != H_SUCCESS) {
1213                         netdev_err(netdev, "h_multicast_ctrl rc=%ld when "
1214                                    "attempting to clear filter table\n",
1215                                    lpar_rc);
1216                 }
1217                 /* add the addresses to the filter table */
1218                 netdev_for_each_mc_addr(ha, netdev) {
1219                         // add the multicast address to the filter table
1220                         unsigned long mcast_addr = 0;
1221                         memcpy(((char *)&mcast_addr)+2, ha->addr, 6);
1222                         lpar_rc = h_multicast_ctrl(adapter->vdev->unit_address,
1223                                                    IbmVethMcastAddFilter,
1224                                                    mcast_addr);
1225                         if(lpar_rc != H_SUCCESS) {
1226                                 netdev_err(netdev, "h_multicast_ctrl rc=%ld "
1227                                            "when adding an entry to the filter "
1228                                            "table\n", lpar_rc);
1229                         }
1230                 }
1231
1232                 /* re-enable filtering */
1233                 lpar_rc = h_multicast_ctrl(adapter->vdev->unit_address,
1234                                            IbmVethMcastEnableFiltering,
1235                                            0);
1236                 if(lpar_rc != H_SUCCESS) {
1237                         netdev_err(netdev, "h_multicast_ctrl rc=%ld when "
1238                                    "enabling filtering\n", lpar_rc);
1239                 }
1240         }
1241 }
1242
1243 static int ibmveth_change_mtu(struct net_device *dev, int new_mtu)
1244 {
1245         struct ibmveth_adapter *adapter = netdev_priv(dev);
1246         struct vio_dev *viodev = adapter->vdev;
1247         int new_mtu_oh = new_mtu + IBMVETH_BUFF_OH;
1248         int i, rc;
1249         int need_restart = 0;
1250
1251         if (new_mtu < IBMVETH_MAX_MTU)
1252                 return -EINVAL;
1253
1254         for (i = 0; i < IbmVethNumBufferPools; i++)
1255                 if (new_mtu_oh < adapter->rx_buff_pool[i].buff_size)
1256                         break;
1257
1258         if (i == IbmVethNumBufferPools)
1259                 return -EINVAL;
1260
1261         /* Deactivate all the buffer pools so that the next loop can activate
1262            only the buffer pools necessary to hold the new MTU */
1263         if (netif_running(adapter->netdev)) {
1264                 need_restart = 1;
1265                 adapter->pool_config = 1;
1266                 ibmveth_close(adapter->netdev);
1267                 adapter->pool_config = 0;
1268         }
1269
1270         /* Look for an active buffer pool that can hold the new MTU */
1271         for(i = 0; i<IbmVethNumBufferPools; i++) {
1272                 adapter->rx_buff_pool[i].active = 1;
1273
1274                 if (new_mtu_oh < adapter->rx_buff_pool[i].buff_size) {
1275                         dev->mtu = new_mtu;
1276                         vio_cmo_set_dev_desired(viodev,
1277                                                 ibmveth_get_desired_dma
1278                                                 (viodev));
1279                         if (need_restart) {
1280                                 return ibmveth_open(adapter->netdev);
1281                         }
1282                         return 0;
1283                 }
1284         }
1285
1286         if (need_restart && (rc = ibmveth_open(adapter->netdev)))
1287                 return rc;
1288
1289         return -EINVAL;
1290 }
1291
1292 #ifdef CONFIG_NET_POLL_CONTROLLER
1293 static void ibmveth_poll_controller(struct net_device *dev)
1294 {
1295         ibmveth_replenish_task(netdev_priv(dev));
1296         ibmveth_interrupt(dev->irq, dev);
1297 }
1298 #endif
1299
1300 /**
1301  * ibmveth_get_desired_dma - Calculate IO memory desired by the driver
1302  *
1303  * @vdev: struct vio_dev for the device whose desired IO mem is to be returned
1304  *
1305  * Return value:
1306  *      Number of bytes of IO data the driver will need to perform well.
1307  */
1308 static unsigned long ibmveth_get_desired_dma(struct vio_dev *vdev)
1309 {
1310         struct net_device *netdev = dev_get_drvdata(&vdev->dev);
1311         struct ibmveth_adapter *adapter;
1312         unsigned long ret;
1313         int i;
1314         int rxqentries = 1;
1315
1316         /* netdev inits at probe time along with the structures we need below*/
1317         if (netdev == NULL)
1318                 return IOMMU_PAGE_ALIGN(IBMVETH_IO_ENTITLEMENT_DEFAULT);
1319
1320         adapter = netdev_priv(netdev);
1321
1322         ret = IBMVETH_BUFF_LIST_SIZE + IBMVETH_FILT_LIST_SIZE;
1323         ret += IOMMU_PAGE_ALIGN(netdev->mtu);
1324
1325         for (i = 0; i < IbmVethNumBufferPools; i++) {
1326                 /* add the size of the active receive buffers */
1327                 if (adapter->rx_buff_pool[i].active)
1328                         ret +=
1329                             adapter->rx_buff_pool[i].size *
1330                             IOMMU_PAGE_ALIGN(adapter->rx_buff_pool[i].
1331                                     buff_size);
1332                 rxqentries += adapter->rx_buff_pool[i].size;
1333         }
1334         /* add the size of the receive queue entries */
1335         ret += IOMMU_PAGE_ALIGN(rxqentries * sizeof(struct ibmveth_rx_q_entry));
1336
1337         return ret;
1338 }
1339
1340 static const struct net_device_ops ibmveth_netdev_ops = {
1341         .ndo_open               = ibmveth_open,
1342         .ndo_stop               = ibmveth_close,
1343         .ndo_start_xmit         = ibmveth_start_xmit,
1344         .ndo_set_multicast_list = ibmveth_set_multicast_list,
1345         .ndo_do_ioctl           = ibmveth_ioctl,
1346         .ndo_change_mtu         = ibmveth_change_mtu,
1347         .ndo_validate_addr      = eth_validate_addr,
1348         .ndo_set_mac_address    = eth_mac_addr,
1349 #ifdef CONFIG_NET_POLL_CONTROLLER
1350         .ndo_poll_controller    = ibmveth_poll_controller,
1351 #endif
1352 };
1353
1354 static int __devinit ibmveth_probe(struct vio_dev *dev, const struct vio_device_id *id)
1355 {
1356         int rc, i;
1357         struct net_device *netdev;
1358         struct ibmveth_adapter *adapter;
1359
1360         unsigned char *mac_addr_p;
1361         unsigned int *mcastFilterSize_p;
1362
1363
1364         dev_dbg(&dev->dev, "entering ibmveth_probe for UA 0x%x\n",
1365                 dev->unit_address);
1366
1367         mac_addr_p = (unsigned char *) vio_get_attribute(dev,
1368                                                 VETH_MAC_ADDR, NULL);
1369         if(!mac_addr_p) {
1370                 dev_err(&dev->dev, "Can't find VETH_MAC_ADDR attribute\n");
1371                 return 0;
1372         }
1373
1374         mcastFilterSize_p = (unsigned int *) vio_get_attribute(dev,
1375                                                 VETH_MCAST_FILTER_SIZE, NULL);
1376         if(!mcastFilterSize_p) {
1377                 dev_err(&dev->dev, "Can't find VETH_MCAST_FILTER_SIZE "
1378                         "attribute\n");
1379                 return 0;
1380         }
1381
1382         netdev = alloc_etherdev(sizeof(struct ibmveth_adapter));
1383
1384         if(!netdev)
1385                 return -ENOMEM;
1386
1387         adapter = netdev_priv(netdev);
1388         dev_set_drvdata(&dev->dev, netdev);
1389
1390         adapter->vdev = dev;
1391         adapter->netdev = netdev;
1392         adapter->mcastFilterSize= *mcastFilterSize_p;
1393         adapter->pool_config = 0;
1394
1395         netif_napi_add(netdev, &adapter->napi, ibmveth_poll, 16);
1396
1397         /*      Some older boxes running PHYP non-natively have an OF that
1398                 returns a 8-byte local-mac-address field (and the first
1399                 2 bytes have to be ignored) while newer boxes' OF return
1400                 a 6-byte field. Note that IEEE 1275 specifies that
1401                 local-mac-address must be a 6-byte field.
1402                 The RPA doc specifies that the first byte must be 10b, so
1403                 we'll just look for it to solve this 8 vs. 6 byte field issue */
1404
1405         if ((*mac_addr_p & 0x3) != 0x02)
1406                 mac_addr_p += 2;
1407
1408         adapter->mac_addr = 0;
1409         memcpy(&adapter->mac_addr, mac_addr_p, 6);
1410
1411         netdev->irq = dev->irq;
1412         netdev->netdev_ops = &ibmveth_netdev_ops;
1413         netdev->ethtool_ops = &netdev_ethtool_ops;
1414         SET_NETDEV_DEV(netdev, &dev->dev);
1415         netdev->features |= NETIF_F_SG;
1416
1417         memcpy(netdev->dev_addr, &adapter->mac_addr, netdev->addr_len);
1418
1419         for(i = 0; i<IbmVethNumBufferPools; i++) {
1420                 struct kobject *kobj = &adapter->rx_buff_pool[i].kobj;
1421                 int error;
1422
1423                 ibmveth_init_buffer_pool(&adapter->rx_buff_pool[i], i,
1424                                          pool_count[i], pool_size[i],
1425                                          pool_active[i]);
1426                 error = kobject_init_and_add(kobj, &ktype_veth_pool,
1427                                              &dev->dev.kobj, "pool%d", i);
1428                 if (!error)
1429                         kobject_uevent(kobj, KOBJ_ADD);
1430         }
1431
1432         netdev_dbg(netdev, "adapter @ 0x%p\n", adapter);
1433
1434         adapter->buffer_list_dma = DMA_ERROR_CODE;
1435         adapter->filter_list_dma = DMA_ERROR_CODE;
1436         adapter->rx_queue.queue_dma = DMA_ERROR_CODE;
1437
1438         netdev_dbg(netdev, "registering netdev...\n");
1439
1440         ibmveth_set_csum_offload(netdev, 1, ibmveth_set_tx_csum_flags);
1441
1442         rc = register_netdev(netdev);
1443
1444         if(rc) {
1445                 netdev_dbg(netdev, "failed to register netdev rc=%d\n", rc);
1446                 free_netdev(netdev);
1447                 return rc;
1448         }
1449
1450         netdev_dbg(netdev, "registered\n");
1451
1452         return 0;
1453 }
1454
1455 static int __devexit ibmveth_remove(struct vio_dev *dev)
1456 {
1457         struct net_device *netdev = dev_get_drvdata(&dev->dev);
1458         struct ibmveth_adapter *adapter = netdev_priv(netdev);
1459         int i;
1460
1461         for(i = 0; i<IbmVethNumBufferPools; i++)
1462                 kobject_put(&adapter->rx_buff_pool[i].kobj);
1463
1464         unregister_netdev(netdev);
1465
1466         free_netdev(netdev);
1467         dev_set_drvdata(&dev->dev, NULL);
1468
1469         return 0;
1470 }
1471
1472 static struct attribute veth_active_attr;
1473 static struct attribute veth_num_attr;
1474 static struct attribute veth_size_attr;
1475
1476 static ssize_t veth_pool_show(struct kobject * kobj,
1477                               struct attribute * attr, char * buf)
1478 {
1479         struct ibmveth_buff_pool *pool = container_of(kobj,
1480                                                       struct ibmveth_buff_pool,
1481                                                       kobj);
1482
1483         if (attr == &veth_active_attr)
1484                 return sprintf(buf, "%d\n", pool->active);
1485         else if (attr == &veth_num_attr)
1486                 return sprintf(buf, "%d\n", pool->size);
1487         else if (attr == &veth_size_attr)
1488                 return sprintf(buf, "%d\n", pool->buff_size);
1489         return 0;
1490 }
1491
1492 static ssize_t veth_pool_store(struct kobject * kobj, struct attribute * attr,
1493 const char * buf, size_t count)
1494 {
1495         struct ibmveth_buff_pool *pool = container_of(kobj,
1496                                                       struct ibmveth_buff_pool,
1497                                                       kobj);
1498         struct net_device *netdev = dev_get_drvdata(
1499             container_of(kobj->parent, struct device, kobj));
1500         struct ibmveth_adapter *adapter = netdev_priv(netdev);
1501         long value = simple_strtol(buf, NULL, 10);
1502         long rc;
1503
1504         if (attr == &veth_active_attr) {
1505                 if (value && !pool->active) {
1506                         if (netif_running(netdev)) {
1507                                 if(ibmveth_alloc_buffer_pool(pool)) {
1508                                         netdev_err(netdev,
1509                                                    "unable to alloc pool\n");
1510                                         return -ENOMEM;
1511                                 }
1512                                 pool->active = 1;
1513                                 adapter->pool_config = 1;
1514                                 ibmveth_close(netdev);
1515                                 adapter->pool_config = 0;
1516                                 if ((rc = ibmveth_open(netdev)))
1517                                         return rc;
1518                         } else
1519                                 pool->active = 1;
1520                 } else if (!value && pool->active) {
1521                         int mtu = netdev->mtu + IBMVETH_BUFF_OH;
1522                         int i;
1523                         /* Make sure there is a buffer pool with buffers that
1524                            can hold a packet of the size of the MTU */
1525                         for (i = 0; i < IbmVethNumBufferPools; i++) {
1526                                 if (pool == &adapter->rx_buff_pool[i])
1527                                         continue;
1528                                 if (!adapter->rx_buff_pool[i].active)
1529                                         continue;
1530                                 if (mtu <= adapter->rx_buff_pool[i].buff_size)
1531                                         break;
1532                         }
1533
1534                         if (i == IbmVethNumBufferPools) {
1535                                 netdev_err(netdev, "no active pool >= MTU\n");
1536                                 return -EPERM;
1537                         }
1538
1539                         if (netif_running(netdev)) {
1540                                 adapter->pool_config = 1;
1541                                 ibmveth_close(netdev);
1542                                 pool->active = 0;
1543                                 adapter->pool_config = 0;
1544                                 if ((rc = ibmveth_open(netdev)))
1545                                         return rc;
1546                         }
1547                         pool->active = 0;
1548                 }
1549         } else if (attr == &veth_num_attr) {
1550                 if (value <= 0 || value > IBMVETH_MAX_POOL_COUNT)
1551                         return -EINVAL;
1552                 else {
1553                         if (netif_running(netdev)) {
1554                                 adapter->pool_config = 1;
1555                                 ibmveth_close(netdev);
1556                                 adapter->pool_config = 0;
1557                                 pool->size = value;
1558                                 if ((rc = ibmveth_open(netdev)))
1559                                         return rc;
1560                         } else
1561                                 pool->size = value;
1562                 }
1563         } else if (attr == &veth_size_attr) {
1564                 if (value <= IBMVETH_BUFF_OH || value > IBMVETH_MAX_BUF_SIZE)
1565                         return -EINVAL;
1566                 else {
1567                         if (netif_running(netdev)) {
1568                                 adapter->pool_config = 1;
1569                                 ibmveth_close(netdev);
1570                                 adapter->pool_config = 0;
1571                                 pool->buff_size = value;
1572                                 if ((rc = ibmveth_open(netdev)))
1573                                         return rc;
1574                         } else
1575                                 pool->buff_size = value;
1576                 }
1577         }
1578
1579         /* kick the interrupt handler to allocate/deallocate pools */
1580         ibmveth_interrupt(netdev->irq, netdev);
1581         return count;
1582 }
1583
1584
1585 #define ATTR(_name, _mode)      \
1586         struct attribute veth_##_name##_attr = {               \
1587         .name = __stringify(_name), .mode = _mode, \
1588         };
1589
1590 static ATTR(active, 0644);
1591 static ATTR(num, 0644);
1592 static ATTR(size, 0644);
1593
1594 static struct attribute * veth_pool_attrs[] = {
1595         &veth_active_attr,
1596         &veth_num_attr,
1597         &veth_size_attr,
1598         NULL,
1599 };
1600
1601 static const struct sysfs_ops veth_pool_ops = {
1602         .show   = veth_pool_show,
1603         .store  = veth_pool_store,
1604 };
1605
1606 static struct kobj_type ktype_veth_pool = {
1607         .release        = NULL,
1608         .sysfs_ops      = &veth_pool_ops,
1609         .default_attrs  = veth_pool_attrs,
1610 };
1611
1612 static int ibmveth_resume(struct device *dev)
1613 {
1614         struct net_device *netdev = dev_get_drvdata(dev);
1615         ibmveth_interrupt(netdev->irq, netdev);
1616         return 0;
1617 }
1618
1619 static struct vio_device_id ibmveth_device_table[] __devinitdata= {
1620         { "network", "IBM,l-lan"},
1621         { "", "" }
1622 };
1623 MODULE_DEVICE_TABLE(vio, ibmveth_device_table);
1624
1625 static struct dev_pm_ops ibmveth_pm_ops = {
1626         .resume = ibmveth_resume
1627 };
1628
1629 static struct vio_driver ibmveth_driver = {
1630         .id_table       = ibmveth_device_table,
1631         .probe          = ibmveth_probe,
1632         .remove         = ibmveth_remove,
1633         .get_desired_dma = ibmveth_get_desired_dma,
1634         .driver         = {
1635                 .name   = ibmveth_driver_name,
1636                 .owner  = THIS_MODULE,
1637                 .pm = &ibmveth_pm_ops,
1638         }
1639 };
1640
1641 static int __init ibmveth_module_init(void)
1642 {
1643         printk(KERN_DEBUG "%s: %s %s\n", ibmveth_driver_name,
1644                ibmveth_driver_string, ibmveth_driver_version);
1645
1646         return vio_register_driver(&ibmveth_driver);
1647 }
1648
1649 static void __exit ibmveth_module_exit(void)
1650 {
1651         vio_unregister_driver(&ibmveth_driver);
1652 }
1653
1654 module_init(ibmveth_module_init);
1655 module_exit(ibmveth_module_exit);