ibmveth: Cleanup error handling inside ibmveth_open
[linux-2.6.git] / drivers / net / ibmveth.c
1 /*
2  * IBM Power Virtual Ethernet Device Driver
3  *
4  * This program is free software; you can redistribute it and/or modify
5  * it under the terms of the GNU General Public License as published by
6  * the Free Software Foundation; either version 2 of the License, or
7  * (at your option) any later version.
8  *
9  * This program is distributed in the hope that it will be useful,
10  * but WITHOUT ANY WARRANTY; without even the implied warranty of
11  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
12  * GNU General Public License for more details.
13  *
14  * You should have received a copy of the GNU General Public License
15  * along with this program; if not, write to the Free Software
16  * Foundation, Inc., 59 Temple Place - Suite 330, Boston, MA 02111-1307, USA.
17  *
18  * Copyright (C) IBM Corporation, 2003, 2010
19  *
20  * Authors: Dave Larson <larson1@us.ibm.com>
21  *          Santiago Leon <santil@linux.vnet.ibm.com>
22  *          Brian King <brking@linux.vnet.ibm.com>
23  *          Robert Jennings <rcj@linux.vnet.ibm.com>
24  *          Anton Blanchard <anton@au.ibm.com>
25  */
26
27 #include <linux/module.h>
28 #include <linux/moduleparam.h>
29 #include <linux/types.h>
30 #include <linux/errno.h>
31 #include <linux/dma-mapping.h>
32 #include <linux/kernel.h>
33 #include <linux/netdevice.h>
34 #include <linux/etherdevice.h>
35 #include <linux/skbuff.h>
36 #include <linux/init.h>
37 #include <linux/mm.h>
38 #include <linux/pm.h>
39 #include <linux/ethtool.h>
40 #include <linux/in.h>
41 #include <linux/ip.h>
42 #include <linux/ipv6.h>
43 #include <linux/slab.h>
44 #include <asm/hvcall.h>
45 #include <asm/atomic.h>
46 #include <asm/vio.h>
47 #include <asm/iommu.h>
48 #include <asm/firmware.h>
49
50 #include "ibmveth.h"
51
52 static irqreturn_t ibmveth_interrupt(int irq, void *dev_instance);
53 static void ibmveth_rxq_harvest_buffer(struct ibmveth_adapter *adapter);
54 static unsigned long ibmveth_get_desired_dma(struct vio_dev *vdev);
55
56 static struct kobj_type ktype_veth_pool;
57
58
59 static const char ibmveth_driver_name[] = "ibmveth";
60 static const char ibmveth_driver_string[] = "IBM Power Virtual Ethernet Driver";
61 #define ibmveth_driver_version "1.04"
62
63 MODULE_AUTHOR("Santiago Leon <santil@linux.vnet.ibm.com>");
64 MODULE_DESCRIPTION("IBM Power Virtual Ethernet Driver");
65 MODULE_LICENSE("GPL");
66 MODULE_VERSION(ibmveth_driver_version);
67
68 static unsigned int tx_copybreak __read_mostly = 128;
69 module_param(tx_copybreak, uint, 0644);
70 MODULE_PARM_DESC(tx_copybreak,
71         "Maximum size of packet that is copied to a new buffer on transmit");
72
73 static unsigned int rx_copybreak __read_mostly = 128;
74 module_param(rx_copybreak, uint, 0644);
75 MODULE_PARM_DESC(rx_copybreak,
76         "Maximum size of packet that is copied to a new buffer on receive");
77
78 static unsigned int rx_flush __read_mostly = 0;
79 module_param(rx_flush, uint, 0644);
80 MODULE_PARM_DESC(rx_flush, "Flush receive buffers before use");
81
82 struct ibmveth_stat {
83         char name[ETH_GSTRING_LEN];
84         int offset;
85 };
86
87 #define IBMVETH_STAT_OFF(stat) offsetof(struct ibmveth_adapter, stat)
88 #define IBMVETH_GET_STAT(a, off) *((u64 *)(((unsigned long)(a)) + off))
89
90 struct ibmveth_stat ibmveth_stats[] = {
91         { "replenish_task_cycles", IBMVETH_STAT_OFF(replenish_task_cycles) },
92         { "replenish_no_mem", IBMVETH_STAT_OFF(replenish_no_mem) },
93         { "replenish_add_buff_failure",
94                         IBMVETH_STAT_OFF(replenish_add_buff_failure) },
95         { "replenish_add_buff_success",
96                         IBMVETH_STAT_OFF(replenish_add_buff_success) },
97         { "rx_invalid_buffer", IBMVETH_STAT_OFF(rx_invalid_buffer) },
98         { "rx_no_buffer", IBMVETH_STAT_OFF(rx_no_buffer) },
99         { "tx_map_failed", IBMVETH_STAT_OFF(tx_map_failed) },
100         { "tx_send_failed", IBMVETH_STAT_OFF(tx_send_failed) },
101         { "fw_enabled_ipv4_csum", IBMVETH_STAT_OFF(fw_ipv4_csum_support) },
102         { "fw_enabled_ipv6_csum", IBMVETH_STAT_OFF(fw_ipv6_csum_support) },
103 };
104
105 /* simple methods of getting data from the current rxq entry */
106 static inline u32 ibmveth_rxq_flags(struct ibmveth_adapter *adapter)
107 {
108         return adapter->rx_queue.queue_addr[adapter->rx_queue.index].flags_off;
109 }
110
111 static inline int ibmveth_rxq_toggle(struct ibmveth_adapter *adapter)
112 {
113         return (ibmveth_rxq_flags(adapter) & IBMVETH_RXQ_TOGGLE) >>
114                         IBMVETH_RXQ_TOGGLE_SHIFT;
115 }
116
117 static inline int ibmveth_rxq_pending_buffer(struct ibmveth_adapter *adapter)
118 {
119         return ibmveth_rxq_toggle(adapter) == adapter->rx_queue.toggle;
120 }
121
122 static inline int ibmveth_rxq_buffer_valid(struct ibmveth_adapter *adapter)
123 {
124         return ibmveth_rxq_flags(adapter) & IBMVETH_RXQ_VALID;
125 }
126
127 static inline int ibmveth_rxq_frame_offset(struct ibmveth_adapter *adapter)
128 {
129         return ibmveth_rxq_flags(adapter) & IBMVETH_RXQ_OFF_MASK;
130 }
131
132 static inline int ibmveth_rxq_frame_length(struct ibmveth_adapter *adapter)
133 {
134         return adapter->rx_queue.queue_addr[adapter->rx_queue.index].length;
135 }
136
137 static inline int ibmveth_rxq_csum_good(struct ibmveth_adapter *adapter)
138 {
139         return ibmveth_rxq_flags(adapter) & IBMVETH_RXQ_CSUM_GOOD;
140 }
141
142 /* setup the initial settings for a buffer pool */
143 static void ibmveth_init_buffer_pool(struct ibmveth_buff_pool *pool,
144                                      u32 pool_index, u32 pool_size,
145                                      u32 buff_size, u32 pool_active)
146 {
147         pool->size = pool_size;
148         pool->index = pool_index;
149         pool->buff_size = buff_size;
150         pool->threshold = pool_size * 7 / 8;
151         pool->active = pool_active;
152 }
153
154 /* allocate and setup an buffer pool - called during open */
155 static int ibmveth_alloc_buffer_pool(struct ibmveth_buff_pool *pool)
156 {
157         int i;
158
159         pool->free_map = kmalloc(sizeof(u16) * pool->size, GFP_KERNEL);
160
161         if (!pool->free_map)
162                 return -1;
163
164         pool->dma_addr = kmalloc(sizeof(dma_addr_t) * pool->size, GFP_KERNEL);
165         if (!pool->dma_addr) {
166                 kfree(pool->free_map);
167                 pool->free_map = NULL;
168                 return -1;
169         }
170
171         pool->skbuff = kcalloc(pool->size, sizeof(void *), GFP_KERNEL);
172
173         if (!pool->skbuff) {
174                 kfree(pool->dma_addr);
175                 pool->dma_addr = NULL;
176
177                 kfree(pool->free_map);
178                 pool->free_map = NULL;
179                 return -1;
180         }
181
182         memset(pool->dma_addr, 0, sizeof(dma_addr_t) * pool->size);
183
184         for (i = 0; i < pool->size; ++i)
185                 pool->free_map[i] = i;
186
187         atomic_set(&pool->available, 0);
188         pool->producer_index = 0;
189         pool->consumer_index = 0;
190
191         return 0;
192 }
193
194 static inline void ibmveth_flush_buffer(void *addr, unsigned long length)
195 {
196         unsigned long offset;
197
198         for (offset = 0; offset < length; offset += SMP_CACHE_BYTES)
199                 asm("dcbfl %0,%1" :: "b" (addr), "r" (offset));
200 }
201
202 /* replenish the buffers for a pool.  note that we don't need to
203  * skb_reserve these since they are used for incoming...
204  */
205 static void ibmveth_replenish_buffer_pool(struct ibmveth_adapter *adapter,
206                                           struct ibmveth_buff_pool *pool)
207 {
208         u32 i;
209         u32 count = pool->size - atomic_read(&pool->available);
210         u32 buffers_added = 0;
211         struct sk_buff *skb;
212         unsigned int free_index, index;
213         u64 correlator;
214         unsigned long lpar_rc;
215         dma_addr_t dma_addr;
216
217         mb();
218
219         for (i = 0; i < count; ++i) {
220                 union ibmveth_buf_desc desc;
221
222                 skb = netdev_alloc_skb(adapter->netdev, pool->buff_size);
223
224                 if (!skb) {
225                         netdev_dbg(adapter->netdev,
226                                    "replenish: unable to allocate skb\n");
227                         adapter->replenish_no_mem++;
228                         break;
229                 }
230
231                 free_index = pool->consumer_index;
232                 pool->consumer_index++;
233                 if (pool->consumer_index >= pool->size)
234                         pool->consumer_index = 0;
235                 index = pool->free_map[free_index];
236
237                 BUG_ON(index == IBM_VETH_INVALID_MAP);
238                 BUG_ON(pool->skbuff[index] != NULL);
239
240                 dma_addr = dma_map_single(&adapter->vdev->dev, skb->data,
241                                 pool->buff_size, DMA_FROM_DEVICE);
242
243                 if (dma_mapping_error(&adapter->vdev->dev, dma_addr))
244                         goto failure;
245
246                 pool->free_map[free_index] = IBM_VETH_INVALID_MAP;
247                 pool->dma_addr[index] = dma_addr;
248                 pool->skbuff[index] = skb;
249
250                 correlator = ((u64)pool->index << 32) | index;
251                 *(u64 *)skb->data = correlator;
252
253                 desc.fields.flags_len = IBMVETH_BUF_VALID | pool->buff_size;
254                 desc.fields.address = dma_addr;
255
256                 if (rx_flush) {
257                         unsigned int len = min(pool->buff_size,
258                                                 adapter->netdev->mtu +
259                                                 IBMVETH_BUFF_OH);
260                         ibmveth_flush_buffer(skb->data, len);
261                 }
262                 lpar_rc = h_add_logical_lan_buffer(adapter->vdev->unit_address,
263                                                    desc.desc);
264
265                 if (lpar_rc != H_SUCCESS) {
266                         goto failure;
267                 } else {
268                         buffers_added++;
269                         adapter->replenish_add_buff_success++;
270                 }
271         }
272
273         mb();
274         atomic_add(buffers_added, &(pool->available));
275         return;
276
277 failure:
278         pool->free_map[free_index] = index;
279         pool->skbuff[index] = NULL;
280         if (pool->consumer_index == 0)
281                 pool->consumer_index = pool->size - 1;
282         else
283                 pool->consumer_index--;
284         if (!dma_mapping_error(&adapter->vdev->dev, dma_addr))
285                 dma_unmap_single(&adapter->vdev->dev,
286                                  pool->dma_addr[index], pool->buff_size,
287                                  DMA_FROM_DEVICE);
288         dev_kfree_skb_any(skb);
289         adapter->replenish_add_buff_failure++;
290
291         mb();
292         atomic_add(buffers_added, &(pool->available));
293 }
294
295 /* replenish routine */
296 static void ibmveth_replenish_task(struct ibmveth_adapter *adapter)
297 {
298         int i;
299
300         adapter->replenish_task_cycles++;
301
302         for (i = (IBMVETH_NUM_BUFF_POOLS - 1); i >= 0; i--) {
303                 struct ibmveth_buff_pool *pool = &adapter->rx_buff_pool[i];
304
305                 if (pool->active &&
306                     (atomic_read(&pool->available) < pool->threshold))
307                         ibmveth_replenish_buffer_pool(adapter, pool);
308         }
309
310         adapter->rx_no_buffer = *(u64 *)(((char*)adapter->buffer_list_addr) +
311                                                 4096 - 8);
312 }
313
314 /* empty and free ana buffer pool - also used to do cleanup in error paths */
315 static void ibmveth_free_buffer_pool(struct ibmveth_adapter *adapter,
316                                      struct ibmveth_buff_pool *pool)
317 {
318         int i;
319
320         kfree(pool->free_map);
321         pool->free_map = NULL;
322
323         if (pool->skbuff && pool->dma_addr) {
324                 for (i = 0; i < pool->size; ++i) {
325                         struct sk_buff *skb = pool->skbuff[i];
326                         if (skb) {
327                                 dma_unmap_single(&adapter->vdev->dev,
328                                                  pool->dma_addr[i],
329                                                  pool->buff_size,
330                                                  DMA_FROM_DEVICE);
331                                 dev_kfree_skb_any(skb);
332                                 pool->skbuff[i] = NULL;
333                         }
334                 }
335         }
336
337         if (pool->dma_addr) {
338                 kfree(pool->dma_addr);
339                 pool->dma_addr = NULL;
340         }
341
342         if (pool->skbuff) {
343                 kfree(pool->skbuff);
344                 pool->skbuff = NULL;
345         }
346 }
347
348 /* remove a buffer from a pool */
349 static void ibmveth_remove_buffer_from_pool(struct ibmveth_adapter *adapter,
350                                             u64 correlator)
351 {
352         unsigned int pool  = correlator >> 32;
353         unsigned int index = correlator & 0xffffffffUL;
354         unsigned int free_index;
355         struct sk_buff *skb;
356
357         BUG_ON(pool >= IBMVETH_NUM_BUFF_POOLS);
358         BUG_ON(index >= adapter->rx_buff_pool[pool].size);
359
360         skb = adapter->rx_buff_pool[pool].skbuff[index];
361
362         BUG_ON(skb == NULL);
363
364         adapter->rx_buff_pool[pool].skbuff[index] = NULL;
365
366         dma_unmap_single(&adapter->vdev->dev,
367                          adapter->rx_buff_pool[pool].dma_addr[index],
368                          adapter->rx_buff_pool[pool].buff_size,
369                          DMA_FROM_DEVICE);
370
371         free_index = adapter->rx_buff_pool[pool].producer_index;
372         adapter->rx_buff_pool[pool].producer_index++;
373         if (adapter->rx_buff_pool[pool].producer_index >=
374             adapter->rx_buff_pool[pool].size)
375                 adapter->rx_buff_pool[pool].producer_index = 0;
376         adapter->rx_buff_pool[pool].free_map[free_index] = index;
377
378         mb();
379
380         atomic_dec(&(adapter->rx_buff_pool[pool].available));
381 }
382
383 /* get the current buffer on the rx queue */
384 static inline struct sk_buff *ibmveth_rxq_get_buffer(struct ibmveth_adapter *adapter)
385 {
386         u64 correlator = adapter->rx_queue.queue_addr[adapter->rx_queue.index].correlator;
387         unsigned int pool = correlator >> 32;
388         unsigned int index = correlator & 0xffffffffUL;
389
390         BUG_ON(pool >= IBMVETH_NUM_BUFF_POOLS);
391         BUG_ON(index >= adapter->rx_buff_pool[pool].size);
392
393         return adapter->rx_buff_pool[pool].skbuff[index];
394 }
395
396 /* recycle the current buffer on the rx queue */
397 static void ibmveth_rxq_recycle_buffer(struct ibmveth_adapter *adapter)
398 {
399         u32 q_index = adapter->rx_queue.index;
400         u64 correlator = adapter->rx_queue.queue_addr[q_index].correlator;
401         unsigned int pool = correlator >> 32;
402         unsigned int index = correlator & 0xffffffffUL;
403         union ibmveth_buf_desc desc;
404         unsigned long lpar_rc;
405
406         BUG_ON(pool >= IBMVETH_NUM_BUFF_POOLS);
407         BUG_ON(index >= adapter->rx_buff_pool[pool].size);
408
409         if (!adapter->rx_buff_pool[pool].active) {
410                 ibmveth_rxq_harvest_buffer(adapter);
411                 ibmveth_free_buffer_pool(adapter, &adapter->rx_buff_pool[pool]);
412                 return;
413         }
414
415         desc.fields.flags_len = IBMVETH_BUF_VALID |
416                 adapter->rx_buff_pool[pool].buff_size;
417         desc.fields.address = adapter->rx_buff_pool[pool].dma_addr[index];
418
419         lpar_rc = h_add_logical_lan_buffer(adapter->vdev->unit_address, desc.desc);
420
421         if (lpar_rc != H_SUCCESS) {
422                 netdev_dbg(adapter->netdev, "h_add_logical_lan_buffer failed "
423                            "during recycle rc=%ld", lpar_rc);
424                 ibmveth_remove_buffer_from_pool(adapter, adapter->rx_queue.queue_addr[adapter->rx_queue.index].correlator);
425         }
426
427         if (++adapter->rx_queue.index == adapter->rx_queue.num_slots) {
428                 adapter->rx_queue.index = 0;
429                 adapter->rx_queue.toggle = !adapter->rx_queue.toggle;
430         }
431 }
432
433 static void ibmveth_rxq_harvest_buffer(struct ibmveth_adapter *adapter)
434 {
435         ibmveth_remove_buffer_from_pool(adapter, adapter->rx_queue.queue_addr[adapter->rx_queue.index].correlator);
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_cleanup(struct ibmveth_adapter *adapter)
444 {
445         int i;
446         struct device *dev = &adapter->vdev->dev;
447
448         if (adapter->buffer_list_addr != NULL) {
449                 if (!dma_mapping_error(dev, adapter->buffer_list_dma)) {
450                         dma_unmap_single(dev, adapter->buffer_list_dma, 4096,
451                                         DMA_BIDIRECTIONAL);
452                         adapter->buffer_list_dma = DMA_ERROR_CODE;
453                 }
454                 free_page((unsigned long)adapter->buffer_list_addr);
455                 adapter->buffer_list_addr = NULL;
456         }
457
458         if (adapter->filter_list_addr != NULL) {
459                 if (!dma_mapping_error(dev, adapter->filter_list_dma)) {
460                         dma_unmap_single(dev, adapter->filter_list_dma, 4096,
461                                         DMA_BIDIRECTIONAL);
462                         adapter->filter_list_dma = DMA_ERROR_CODE;
463                 }
464                 free_page((unsigned long)adapter->filter_list_addr);
465                 adapter->filter_list_addr = NULL;
466         }
467
468         if (adapter->rx_queue.queue_addr != NULL) {
469                 if (!dma_mapping_error(dev, adapter->rx_queue.queue_dma)) {
470                         dma_unmap_single(dev,
471                                         adapter->rx_queue.queue_dma,
472                                         adapter->rx_queue.queue_len,
473                                         DMA_BIDIRECTIONAL);
474                         adapter->rx_queue.queue_dma = DMA_ERROR_CODE;
475                 }
476                 kfree(adapter->rx_queue.queue_addr);
477                 adapter->rx_queue.queue_addr = NULL;
478         }
479
480         for (i = 0; i < IBMVETH_NUM_BUFF_POOLS; i++)
481                 if (adapter->rx_buff_pool[i].active)
482                         ibmveth_free_buffer_pool(adapter,
483                                                  &adapter->rx_buff_pool[i]);
484
485         if (adapter->bounce_buffer != NULL) {
486                 if (!dma_mapping_error(dev, adapter->bounce_buffer_dma)) {
487                         dma_unmap_single(&adapter->vdev->dev,
488                                         adapter->bounce_buffer_dma,
489                                         adapter->netdev->mtu + IBMVETH_BUFF_OH,
490                                         DMA_BIDIRECTIONAL);
491                         adapter->bounce_buffer_dma = DMA_ERROR_CODE;
492                 }
493                 kfree(adapter->bounce_buffer);
494                 adapter->bounce_buffer = NULL;
495         }
496 }
497
498 static int ibmveth_register_logical_lan(struct ibmveth_adapter *adapter,
499         union ibmveth_buf_desc rxq_desc, u64 mac_address)
500 {
501         int rc, try_again = 1;
502
503         /*
504          * After a kexec the adapter will still be open, so our attempt to
505          * open it will fail. So if we get a failure we free the adapter and
506          * try again, but only once.
507          */
508 retry:
509         rc = h_register_logical_lan(adapter->vdev->unit_address,
510                                     adapter->buffer_list_dma, rxq_desc.desc,
511                                     adapter->filter_list_dma, mac_address);
512
513         if (rc != H_SUCCESS && try_again) {
514                 do {
515                         rc = h_free_logical_lan(adapter->vdev->unit_address);
516                 } while (H_IS_LONG_BUSY(rc) || (rc == H_BUSY));
517
518                 try_again = 0;
519                 goto retry;
520         }
521
522         return rc;
523 }
524
525 static int ibmveth_open(struct net_device *netdev)
526 {
527         struct ibmveth_adapter *adapter = netdev_priv(netdev);
528         u64 mac_address = 0;
529         int rxq_entries = 1;
530         unsigned long lpar_rc;
531         int rc;
532         union ibmveth_buf_desc rxq_desc;
533         int i;
534         struct device *dev;
535
536         netdev_dbg(netdev, "open starting\n");
537
538         napi_enable(&adapter->napi);
539
540         for(i = 0; i < IBMVETH_NUM_BUFF_POOLS; i++)
541                 rxq_entries += adapter->rx_buff_pool[i].size;
542
543         adapter->buffer_list_addr = (void*) get_zeroed_page(GFP_KERNEL);
544         adapter->filter_list_addr = (void*) get_zeroed_page(GFP_KERNEL);
545
546         if (!adapter->buffer_list_addr || !adapter->filter_list_addr) {
547                 netdev_err(netdev, "unable to allocate filter or buffer list "
548                            "pages\n");
549                 rc = -ENOMEM;
550                 goto err_out;
551         }
552
553         adapter->rx_queue.queue_len = sizeof(struct ibmveth_rx_q_entry) *
554                                                 rxq_entries;
555         adapter->rx_queue.queue_addr = kmalloc(adapter->rx_queue.queue_len,
556                                                 GFP_KERNEL);
557
558         if (!adapter->rx_queue.queue_addr) {
559                 netdev_err(netdev, "unable to allocate rx queue pages\n");
560                 rc = -ENOMEM;
561                 goto err_out;
562         }
563
564         dev = &adapter->vdev->dev;
565
566         adapter->buffer_list_dma = dma_map_single(dev,
567                         adapter->buffer_list_addr, 4096, DMA_BIDIRECTIONAL);
568         adapter->filter_list_dma = dma_map_single(dev,
569                         adapter->filter_list_addr, 4096, DMA_BIDIRECTIONAL);
570         adapter->rx_queue.queue_dma = dma_map_single(dev,
571                         adapter->rx_queue.queue_addr,
572                         adapter->rx_queue.queue_len, DMA_BIDIRECTIONAL);
573
574         if ((dma_mapping_error(dev, adapter->buffer_list_dma)) ||
575             (dma_mapping_error(dev, adapter->filter_list_dma)) ||
576             (dma_mapping_error(dev, adapter->rx_queue.queue_dma))) {
577                 netdev_err(netdev, "unable to map filter or buffer list "
578                            "pages\n");
579                 rc = -ENOMEM;
580                 goto err_out;
581         }
582
583         adapter->rx_queue.index = 0;
584         adapter->rx_queue.num_slots = rxq_entries;
585         adapter->rx_queue.toggle = 1;
586
587         memcpy(&mac_address, netdev->dev_addr, netdev->addr_len);
588         mac_address = mac_address >> 16;
589
590         rxq_desc.fields.flags_len = IBMVETH_BUF_VALID |
591                                         adapter->rx_queue.queue_len;
592         rxq_desc.fields.address = adapter->rx_queue.queue_dma;
593
594         netdev_dbg(netdev, "buffer list @ 0x%p\n", adapter->buffer_list_addr);
595         netdev_dbg(netdev, "filter list @ 0x%p\n", adapter->filter_list_addr);
596         netdev_dbg(netdev, "receive q   @ 0x%p\n", adapter->rx_queue.queue_addr);
597
598         h_vio_signal(adapter->vdev->unit_address, VIO_IRQ_DISABLE);
599
600         lpar_rc = ibmveth_register_logical_lan(adapter, rxq_desc, mac_address);
601
602         if (lpar_rc != H_SUCCESS) {
603                 netdev_err(netdev, "h_register_logical_lan failed with %ld\n",
604                            lpar_rc);
605                 netdev_err(netdev, "buffer TCE:0x%llx filter TCE:0x%llx rxq "
606                            "desc:0x%llx MAC:0x%llx\n",
607                                      adapter->buffer_list_dma,
608                                      adapter->filter_list_dma,
609                                      rxq_desc.desc,
610                                      mac_address);
611                 rc = -ENONET;
612                 goto err_out;
613         }
614
615         for (i = 0; i < IBMVETH_NUM_BUFF_POOLS; i++) {
616                 if (!adapter->rx_buff_pool[i].active)
617                         continue;
618                 if (ibmveth_alloc_buffer_pool(&adapter->rx_buff_pool[i])) {
619                         netdev_err(netdev, "unable to alloc pool\n");
620                         adapter->rx_buff_pool[i].active = 0;
621                         rc = -ENOMEM;
622                         goto err_out;
623                 }
624         }
625
626         netdev_dbg(netdev, "registering irq 0x%x\n", netdev->irq);
627         rc = request_irq(netdev->irq, ibmveth_interrupt, 0, netdev->name,
628                          netdev);
629         if (rc != 0) {
630                 netdev_err(netdev, "unable to request irq 0x%x, rc %d\n",
631                            netdev->irq, rc);
632                 do {
633                         rc = h_free_logical_lan(adapter->vdev->unit_address);
634                 } while (H_IS_LONG_BUSY(rc) || (rc == H_BUSY));
635
636                 goto err_out;
637         }
638
639         adapter->bounce_buffer =
640             kmalloc(netdev->mtu + IBMVETH_BUFF_OH, GFP_KERNEL);
641         if (!adapter->bounce_buffer) {
642                 netdev_err(netdev, "unable to allocate bounce buffer\n");
643                 rc = -ENOMEM;
644                 goto err_out;
645         }
646         adapter->bounce_buffer_dma =
647             dma_map_single(&adapter->vdev->dev, adapter->bounce_buffer,
648                            netdev->mtu + IBMVETH_BUFF_OH, DMA_BIDIRECTIONAL);
649         if (dma_mapping_error(dev, adapter->bounce_buffer_dma)) {
650                 netdev_err(netdev, "unable to map bounce buffer\n");
651                 rc = -ENOMEM;
652                 goto err_out;
653         }
654
655         netdev_dbg(netdev, "initial replenish cycle\n");
656         ibmveth_interrupt(netdev->irq, netdev);
657
658         netif_start_queue(netdev);
659
660         netdev_dbg(netdev, "open complete\n");
661
662         return 0;
663
664 err_out:
665         ibmveth_cleanup(adapter);
666         napi_disable(&adapter->napi);
667         return rc;
668 }
669
670 static int ibmveth_close(struct net_device *netdev)
671 {
672         struct ibmveth_adapter *adapter = netdev_priv(netdev);
673         long lpar_rc;
674
675         netdev_dbg(netdev, "close starting\n");
676
677         napi_disable(&adapter->napi);
678
679         if (!adapter->pool_config)
680                 netif_stop_queue(netdev);
681
682         h_vio_signal(adapter->vdev->unit_address, VIO_IRQ_DISABLE);
683
684         do {
685                 lpar_rc = h_free_logical_lan(adapter->vdev->unit_address);
686         } while (H_IS_LONG_BUSY(lpar_rc) || (lpar_rc == H_BUSY));
687
688         if (lpar_rc != H_SUCCESS) {
689                 netdev_err(netdev, "h_free_logical_lan failed with %lx, "
690                            "continuing with close\n", lpar_rc);
691         }
692
693         free_irq(netdev->irq, netdev);
694
695         adapter->rx_no_buffer = *(u64 *)(((char *)adapter->buffer_list_addr) +
696                                                 4096 - 8);
697
698         ibmveth_cleanup(adapter);
699
700         netdev_dbg(netdev, "close complete\n");
701
702         return 0;
703 }
704
705 static int netdev_get_settings(struct net_device *dev, struct ethtool_cmd *cmd)
706 {
707         cmd->supported = (SUPPORTED_1000baseT_Full | SUPPORTED_Autoneg |
708                                 SUPPORTED_FIBRE);
709         cmd->advertising = (ADVERTISED_1000baseT_Full | ADVERTISED_Autoneg |
710                                 ADVERTISED_FIBRE);
711         cmd->speed = SPEED_1000;
712         cmd->duplex = DUPLEX_FULL;
713         cmd->port = PORT_FIBRE;
714         cmd->phy_address = 0;
715         cmd->transceiver = XCVR_INTERNAL;
716         cmd->autoneg = AUTONEG_ENABLE;
717         cmd->maxtxpkt = 0;
718         cmd->maxrxpkt = 1;
719         return 0;
720 }
721
722 static void netdev_get_drvinfo(struct net_device *dev,
723                                struct ethtool_drvinfo *info)
724 {
725         strncpy(info->driver, ibmveth_driver_name, sizeof(info->driver) - 1);
726         strncpy(info->version, ibmveth_driver_version,
727                 sizeof(info->version) - 1);
728 }
729
730 static u32 netdev_get_link(struct net_device *dev)
731 {
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
744                  * concept of separate tx/rx checksum offload enable, if rx
745                  * checksum is disabled we also have to disable tx checksum
746                  * offload. Once we disable rx checksum offload, we are no
747                  * longer allowed to send tx buffers that are not properly
748                  * 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                         netdev_err(dev, "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
819                 ret6 = h_illan_attributes(adapter->vdev->unit_address,
820                                          clr_attr6, set_attr6, &ret_attr);
821
822                 if (ret6 != H_SUCCESS) {
823                         netdev_err(dev, "unable to change IPv6 checksum "
824                                         "offload settings. %d rc=%ld\n",
825                                         data, ret);
826
827                         ret = h_illan_attributes(adapter->vdev->unit_address,
828                                                  set_attr6, clr_attr6,
829                                                  &ret_attr);
830                 } else
831                         adapter->fw_ipv6_csum_support = data;
832
833                 if (ret == H_SUCCESS || ret6 == H_SUCCESS)
834                         done(dev, data);
835                 else
836                         rc1 = -EIO;
837         } else {
838                 rc1 = -EIO;
839                 netdev_err(dev, "unable to change checksum offload settings."
840                                      " %d rc=%ld ret_attr=%lx\n", data, ret,
841                                      ret_attr);
842         }
843
844         if (restart)
845                 rc2 = ibmveth_open(dev);
846
847         return rc1 ? rc1 : rc2;
848 }
849
850 static int ibmveth_set_rx_csum(struct net_device *dev, u32 data)
851 {
852         struct ibmveth_adapter *adapter = netdev_priv(dev);
853
854         if ((data && adapter->rx_csum) || (!data && !adapter->rx_csum))
855                 return 0;
856
857         return ibmveth_set_csum_offload(dev, data, ibmveth_set_rx_csum_flags);
858 }
859
860 static int ibmveth_set_tx_csum(struct net_device *dev, u32 data)
861 {
862         struct ibmveth_adapter *adapter = netdev_priv(dev);
863         int rc = 0;
864
865         if (data && (dev->features & (NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM)))
866                 return 0;
867         if (!data && !(dev->features & (NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM)))
868                 return 0;
869
870         if (data && !adapter->rx_csum)
871                 rc = ibmveth_set_csum_offload(dev, data,
872                                               ibmveth_set_tx_csum_flags);
873         else
874                 ibmveth_set_tx_csum_flags(dev, data);
875
876         return rc;
877 }
878
879 static u32 ibmveth_get_rx_csum(struct net_device *dev)
880 {
881         struct ibmveth_adapter *adapter = netdev_priv(dev);
882         return adapter->rx_csum;
883 }
884
885 static void ibmveth_get_strings(struct net_device *dev, u32 stringset, u8 *data)
886 {
887         int i;
888
889         if (stringset != ETH_SS_STATS)
890                 return;
891
892         for (i = 0; i < ARRAY_SIZE(ibmveth_stats); i++, data += ETH_GSTRING_LEN)
893                 memcpy(data, ibmveth_stats[i].name, ETH_GSTRING_LEN);
894 }
895
896 static int ibmveth_get_sset_count(struct net_device *dev, int sset)
897 {
898         switch (sset) {
899         case ETH_SS_STATS:
900                 return ARRAY_SIZE(ibmveth_stats);
901         default:
902                 return -EOPNOTSUPP;
903         }
904 }
905
906 static void ibmveth_get_ethtool_stats(struct net_device *dev,
907                                       struct ethtool_stats *stats, u64 *data)
908 {
909         int i;
910         struct ibmveth_adapter *adapter = netdev_priv(dev);
911
912         for (i = 0; i < ARRAY_SIZE(ibmveth_stats); i++)
913                 data[i] = IBMVETH_GET_STAT(adapter, ibmveth_stats[i].offset);
914 }
915
916 static const struct ethtool_ops netdev_ethtool_ops = {
917         .get_drvinfo            = netdev_get_drvinfo,
918         .get_settings           = netdev_get_settings,
919         .get_link               = netdev_get_link,
920         .set_tx_csum            = ibmveth_set_tx_csum,
921         .get_rx_csum            = ibmveth_get_rx_csum,
922         .set_rx_csum            = ibmveth_set_rx_csum,
923         .get_strings            = ibmveth_get_strings,
924         .get_sset_count         = ibmveth_get_sset_count,
925         .get_ethtool_stats      = ibmveth_get_ethtool_stats,
926         .set_sg                 = ethtool_op_set_sg,
927 };
928
929 static int ibmveth_ioctl(struct net_device *dev, struct ifreq *ifr, int cmd)
930 {
931         return -EOPNOTSUPP;
932 }
933
934 #define page_offset(v) ((unsigned long)(v) & ((1 << 12) - 1))
935
936 static int ibmveth_send(struct ibmveth_adapter *adapter,
937                         union ibmveth_buf_desc *descs)
938 {
939         unsigned long correlator;
940         unsigned int retry_count;
941         unsigned long ret;
942
943         /*
944          * The retry count sets a maximum for the number of broadcast and
945          * multicast destinations within the system.
946          */
947         retry_count = 1024;
948         correlator = 0;
949         do {
950                 ret = h_send_logical_lan(adapter->vdev->unit_address,
951                                              descs[0].desc, descs[1].desc,
952                                              descs[2].desc, descs[3].desc,
953                                              descs[4].desc, descs[5].desc,
954                                              correlator, &correlator);
955         } while ((ret == H_BUSY) && (retry_count--));
956
957         if (ret != H_SUCCESS && ret != H_DROPPED) {
958                 netdev_err(adapter->netdev, "tx: h_send_logical_lan failed "
959                            "with rc=%ld\n", ret);
960                 return 1;
961         }
962
963         return 0;
964 }
965
966 static netdev_tx_t ibmveth_start_xmit(struct sk_buff *skb,
967                                       struct net_device *netdev)
968 {
969         struct ibmveth_adapter *adapter = netdev_priv(netdev);
970         unsigned int desc_flags;
971         union ibmveth_buf_desc descs[6];
972         int last, i;
973         int force_bounce = 0;
974
975         /*
976          * veth handles a maximum of 6 segments including the header, so
977          * we have to linearize the skb if there are more than this.
978          */
979         if (skb_shinfo(skb)->nr_frags > 5 && __skb_linearize(skb)) {
980                 netdev->stats.tx_dropped++;
981                 goto out;
982         }
983
984         /* veth can't checksum offload UDP */
985         if (skb->ip_summed == CHECKSUM_PARTIAL &&
986             ((skb->protocol == htons(ETH_P_IP) &&
987               ip_hdr(skb)->protocol != IPPROTO_TCP) ||
988              (skb->protocol == htons(ETH_P_IPV6) &&
989               ipv6_hdr(skb)->nexthdr != IPPROTO_TCP)) &&
990             skb_checksum_help(skb)) {
991
992                 netdev_err(netdev, "tx: failed to checksum packet\n");
993                 netdev->stats.tx_dropped++;
994                 goto out;
995         }
996
997         desc_flags = IBMVETH_BUF_VALID;
998
999         if (skb->ip_summed == CHECKSUM_PARTIAL) {
1000                 unsigned char *buf = skb_transport_header(skb) +
1001                                                 skb->csum_offset;
1002
1003                 desc_flags |= (IBMVETH_BUF_NO_CSUM | IBMVETH_BUF_CSUM_GOOD);
1004
1005                 /* Need to zero out the checksum */
1006                 buf[0] = 0;
1007                 buf[1] = 0;
1008         }
1009
1010 retry_bounce:
1011         memset(descs, 0, sizeof(descs));
1012
1013         /*
1014          * If a linear packet is below the rx threshold then
1015          * copy it into the static bounce buffer. This avoids the
1016          * cost of a TCE insert and remove.
1017          */
1018         if (force_bounce || (!skb_is_nonlinear(skb) &&
1019                                 (skb->len < tx_copybreak))) {
1020                 skb_copy_from_linear_data(skb, adapter->bounce_buffer,
1021                                           skb->len);
1022
1023                 descs[0].fields.flags_len = desc_flags | skb->len;
1024                 descs[0].fields.address = adapter->bounce_buffer_dma;
1025
1026                 if (ibmveth_send(adapter, descs)) {
1027                         adapter->tx_send_failed++;
1028                         netdev->stats.tx_dropped++;
1029                 } else {
1030                         netdev->stats.tx_packets++;
1031                         netdev->stats.tx_bytes += skb->len;
1032                 }
1033
1034                 goto out;
1035         }
1036
1037         /* Map the header */
1038         descs[0].fields.address = dma_map_single(&adapter->vdev->dev, skb->data,
1039                                                  skb_headlen(skb),
1040                                                  DMA_TO_DEVICE);
1041         if (dma_mapping_error(&adapter->vdev->dev, descs[0].fields.address))
1042                 goto map_failed;
1043
1044         descs[0].fields.flags_len = desc_flags | skb_headlen(skb);
1045
1046         /* Map the frags */
1047         for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
1048                 unsigned long dma_addr;
1049                 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
1050
1051                 dma_addr = dma_map_page(&adapter->vdev->dev, frag->page,
1052                                         frag->page_offset, frag->size,
1053                                         DMA_TO_DEVICE);
1054
1055                 if (dma_mapping_error(&adapter->vdev->dev, dma_addr))
1056                         goto map_failed_frags;
1057
1058                 descs[i+1].fields.flags_len = desc_flags | frag->size;
1059                 descs[i+1].fields.address = dma_addr;
1060         }
1061
1062         if (ibmveth_send(adapter, descs)) {
1063                 adapter->tx_send_failed++;
1064                 netdev->stats.tx_dropped++;
1065         } else {
1066                 netdev->stats.tx_packets++;
1067                 netdev->stats.tx_bytes += skb->len;
1068         }
1069
1070         for (i = 0; i < skb_shinfo(skb)->nr_frags + 1; i++)
1071                 dma_unmap_page(&adapter->vdev->dev, descs[i].fields.address,
1072                                descs[i].fields.flags_len & IBMVETH_BUF_LEN_MASK,
1073                                DMA_TO_DEVICE);
1074
1075 out:
1076         dev_kfree_skb(skb);
1077         return NETDEV_TX_OK;
1078
1079 map_failed_frags:
1080         last = i+1;
1081         for (i = 0; i < last; i++)
1082                 dma_unmap_page(&adapter->vdev->dev, descs[i].fields.address,
1083                                descs[i].fields.flags_len & IBMVETH_BUF_LEN_MASK,
1084                                DMA_TO_DEVICE);
1085
1086 map_failed:
1087         if (!firmware_has_feature(FW_FEATURE_CMO))
1088                 netdev_err(netdev, "tx: unable to map xmit buffer\n");
1089         adapter->tx_map_failed++;
1090         skb_linearize(skb);
1091         force_bounce = 1;
1092         goto retry_bounce;
1093 }
1094
1095 static int ibmveth_poll(struct napi_struct *napi, int budget)
1096 {
1097         struct ibmveth_adapter *adapter =
1098                         container_of(napi, struct ibmveth_adapter, napi);
1099         struct net_device *netdev = adapter->netdev;
1100         int frames_processed = 0;
1101         unsigned long lpar_rc;
1102
1103 restart_poll:
1104         do {
1105                 if (!ibmveth_rxq_pending_buffer(adapter))
1106                         break;
1107
1108                 smp_rmb();
1109                 if (!ibmveth_rxq_buffer_valid(adapter)) {
1110                         wmb(); /* suggested by larson1 */
1111                         adapter->rx_invalid_buffer++;
1112                         netdev_dbg(netdev, "recycling invalid buffer\n");
1113                         ibmveth_rxq_recycle_buffer(adapter);
1114                 } else {
1115                         struct sk_buff *skb, *new_skb;
1116                         int length = ibmveth_rxq_frame_length(adapter);
1117                         int offset = ibmveth_rxq_frame_offset(adapter);
1118                         int csum_good = ibmveth_rxq_csum_good(adapter);
1119
1120                         skb = ibmveth_rxq_get_buffer(adapter);
1121
1122                         new_skb = NULL;
1123                         if (length < rx_copybreak)
1124                                 new_skb = netdev_alloc_skb(netdev, length);
1125
1126                         if (new_skb) {
1127                                 skb_copy_to_linear_data(new_skb,
1128                                                         skb->data + offset,
1129                                                         length);
1130                                 if (rx_flush)
1131                                         ibmveth_flush_buffer(skb->data,
1132                                                 length + offset);
1133                                 skb = new_skb;
1134                                 ibmveth_rxq_recycle_buffer(adapter);
1135                         } else {
1136                                 ibmveth_rxq_harvest_buffer(adapter);
1137                                 skb_reserve(skb, offset);
1138                         }
1139
1140                         skb_put(skb, length);
1141                         skb->protocol = eth_type_trans(skb, netdev);
1142
1143                         if (csum_good)
1144                                 skb->ip_summed = CHECKSUM_UNNECESSARY;
1145
1146                         netif_receive_skb(skb); /* send it up */
1147
1148                         netdev->stats.rx_packets++;
1149                         netdev->stats.rx_bytes += length;
1150                         frames_processed++;
1151                 }
1152         } while (frames_processed < budget);
1153
1154         ibmveth_replenish_task(adapter);
1155
1156         if (frames_processed < budget) {
1157                 /* We think we are done - reenable interrupts,
1158                  * then check once more to make sure we are done.
1159                  */
1160                 lpar_rc = h_vio_signal(adapter->vdev->unit_address,
1161                                        VIO_IRQ_ENABLE);
1162
1163                 BUG_ON(lpar_rc != H_SUCCESS);
1164
1165                 napi_complete(napi);
1166
1167                 if (ibmveth_rxq_pending_buffer(adapter) &&
1168                     napi_reschedule(napi)) {
1169                         lpar_rc = h_vio_signal(adapter->vdev->unit_address,
1170                                                VIO_IRQ_DISABLE);
1171                         goto restart_poll;
1172                 }
1173         }
1174
1175         return frames_processed;
1176 }
1177
1178 static irqreturn_t ibmveth_interrupt(int irq, void *dev_instance)
1179 {
1180         struct net_device *netdev = dev_instance;
1181         struct ibmveth_adapter *adapter = netdev_priv(netdev);
1182         unsigned long lpar_rc;
1183
1184         if (napi_schedule_prep(&adapter->napi)) {
1185                 lpar_rc = h_vio_signal(adapter->vdev->unit_address,
1186                                        VIO_IRQ_DISABLE);
1187                 BUG_ON(lpar_rc != H_SUCCESS);
1188                 __napi_schedule(&adapter->napi);
1189         }
1190         return IRQ_HANDLED;
1191 }
1192
1193 static void ibmveth_set_multicast_list(struct net_device *netdev)
1194 {
1195         struct ibmveth_adapter *adapter = netdev_priv(netdev);
1196         unsigned long lpar_rc;
1197
1198         if ((netdev->flags & IFF_PROMISC) ||
1199             (netdev_mc_count(netdev) > adapter->mcastFilterSize)) {
1200                 lpar_rc = h_multicast_ctrl(adapter->vdev->unit_address,
1201                                            IbmVethMcastEnableRecv |
1202                                            IbmVethMcastDisableFiltering,
1203                                            0);
1204                 if (lpar_rc != H_SUCCESS) {
1205                         netdev_err(netdev, "h_multicast_ctrl rc=%ld when "
1206                                    "entering promisc mode\n", lpar_rc);
1207                 }
1208         } else {
1209                 struct netdev_hw_addr *ha;
1210                 /* clear the filter table & disable filtering */
1211                 lpar_rc = h_multicast_ctrl(adapter->vdev->unit_address,
1212                                            IbmVethMcastEnableRecv |
1213                                            IbmVethMcastDisableFiltering |
1214                                            IbmVethMcastClearFilterTable,
1215                                            0);
1216                 if (lpar_rc != H_SUCCESS) {
1217                         netdev_err(netdev, "h_multicast_ctrl rc=%ld when "
1218                                    "attempting to clear filter table\n",
1219                                    lpar_rc);
1220                 }
1221                 /* add the addresses to the filter table */
1222                 netdev_for_each_mc_addr(ha, netdev) {
1223                         /* add the multicast address to the filter table */
1224                         unsigned long mcast_addr = 0;
1225                         memcpy(((char *)&mcast_addr)+2, ha->addr, 6);
1226                         lpar_rc = h_multicast_ctrl(adapter->vdev->unit_address,
1227                                                    IbmVethMcastAddFilter,
1228                                                    mcast_addr);
1229                         if (lpar_rc != H_SUCCESS) {
1230                                 netdev_err(netdev, "h_multicast_ctrl rc=%ld "
1231                                            "when adding an entry to the filter "
1232                                            "table\n", lpar_rc);
1233                         }
1234                 }
1235
1236                 /* re-enable filtering */
1237                 lpar_rc = h_multicast_ctrl(adapter->vdev->unit_address,
1238                                            IbmVethMcastEnableFiltering,
1239                                            0);
1240                 if (lpar_rc != H_SUCCESS) {
1241                         netdev_err(netdev, "h_multicast_ctrl rc=%ld when "
1242                                    "enabling filtering\n", lpar_rc);
1243                 }
1244         }
1245 }
1246
1247 static int ibmveth_change_mtu(struct net_device *dev, int new_mtu)
1248 {
1249         struct ibmveth_adapter *adapter = netdev_priv(dev);
1250         struct vio_dev *viodev = adapter->vdev;
1251         int new_mtu_oh = new_mtu + IBMVETH_BUFF_OH;
1252         int i, rc;
1253         int need_restart = 0;
1254
1255         if (new_mtu < IBMVETH_MIN_MTU)
1256                 return -EINVAL;
1257
1258         for (i = 0; i < IBMVETH_NUM_BUFF_POOLS; i++)
1259                 if (new_mtu_oh < adapter->rx_buff_pool[i].buff_size)
1260                         break;
1261
1262         if (i == IBMVETH_NUM_BUFF_POOLS)
1263                 return -EINVAL;
1264
1265         /* Deactivate all the buffer pools so that the next loop can activate
1266            only the buffer pools necessary to hold the new MTU */
1267         if (netif_running(adapter->netdev)) {
1268                 need_restart = 1;
1269                 adapter->pool_config = 1;
1270                 ibmveth_close(adapter->netdev);
1271                 adapter->pool_config = 0;
1272         }
1273
1274         /* Look for an active buffer pool that can hold the new MTU */
1275         for (i = 0; i < IBMVETH_NUM_BUFF_POOLS; i++) {
1276                 adapter->rx_buff_pool[i].active = 1;
1277
1278                 if (new_mtu_oh < adapter->rx_buff_pool[i].buff_size) {
1279                         dev->mtu = new_mtu;
1280                         vio_cmo_set_dev_desired(viodev,
1281                                                 ibmveth_get_desired_dma
1282                                                 (viodev));
1283                         if (need_restart) {
1284                                 return ibmveth_open(adapter->netdev);
1285                         }
1286                         return 0;
1287                 }
1288         }
1289
1290         if (need_restart && (rc = ibmveth_open(adapter->netdev)))
1291                 return rc;
1292
1293         return -EINVAL;
1294 }
1295
1296 #ifdef CONFIG_NET_POLL_CONTROLLER
1297 static void ibmveth_poll_controller(struct net_device *dev)
1298 {
1299         ibmveth_replenish_task(netdev_priv(dev));
1300         ibmveth_interrupt(dev->irq, dev);
1301 }
1302 #endif
1303
1304 /**
1305  * ibmveth_get_desired_dma - Calculate IO memory desired by the driver
1306  *
1307  * @vdev: struct vio_dev for the device whose desired IO mem is to be returned
1308  *
1309  * Return value:
1310  *      Number of bytes of IO data the driver will need to perform well.
1311  */
1312 static unsigned long ibmveth_get_desired_dma(struct vio_dev *vdev)
1313 {
1314         struct net_device *netdev = dev_get_drvdata(&vdev->dev);
1315         struct ibmveth_adapter *adapter;
1316         unsigned long ret;
1317         int i;
1318         int rxqentries = 1;
1319
1320         /* netdev inits at probe time along with the structures we need below*/
1321         if (netdev == NULL)
1322                 return IOMMU_PAGE_ALIGN(IBMVETH_IO_ENTITLEMENT_DEFAULT);
1323
1324         adapter = netdev_priv(netdev);
1325
1326         ret = IBMVETH_BUFF_LIST_SIZE + IBMVETH_FILT_LIST_SIZE;
1327         ret += IOMMU_PAGE_ALIGN(netdev->mtu);
1328
1329         for (i = 0; i < IBMVETH_NUM_BUFF_POOLS; i++) {
1330                 /* add the size of the active receive buffers */
1331                 if (adapter->rx_buff_pool[i].active)
1332                         ret +=
1333                             adapter->rx_buff_pool[i].size *
1334                             IOMMU_PAGE_ALIGN(adapter->rx_buff_pool[i].
1335                                     buff_size);
1336                 rxqentries += adapter->rx_buff_pool[i].size;
1337         }
1338         /* add the size of the receive queue entries */
1339         ret += IOMMU_PAGE_ALIGN(rxqentries * sizeof(struct ibmveth_rx_q_entry));
1340
1341         return ret;
1342 }
1343
1344 static const struct net_device_ops ibmveth_netdev_ops = {
1345         .ndo_open               = ibmveth_open,
1346         .ndo_stop               = ibmveth_close,
1347         .ndo_start_xmit         = ibmveth_start_xmit,
1348         .ndo_set_multicast_list = ibmveth_set_multicast_list,
1349         .ndo_do_ioctl           = ibmveth_ioctl,
1350         .ndo_change_mtu         = ibmveth_change_mtu,
1351         .ndo_validate_addr      = eth_validate_addr,
1352         .ndo_set_mac_address    = eth_mac_addr,
1353 #ifdef CONFIG_NET_POLL_CONTROLLER
1354         .ndo_poll_controller    = ibmveth_poll_controller,
1355 #endif
1356 };
1357
1358 static int __devinit ibmveth_probe(struct vio_dev *dev,
1359                                    const struct vio_device_id *id)
1360 {
1361         int rc, i;
1362         struct net_device *netdev;
1363         struct ibmveth_adapter *adapter;
1364         unsigned char *mac_addr_p;
1365         unsigned int *mcastFilterSize_p;
1366
1367         dev_dbg(&dev->dev, "entering ibmveth_probe for UA 0x%x\n",
1368                 dev->unit_address);
1369
1370         mac_addr_p = (unsigned char *)vio_get_attribute(dev, VETH_MAC_ADDR,
1371                                                         NULL);
1372         if (!mac_addr_p) {
1373                 dev_err(&dev->dev, "Can't find VETH_MAC_ADDR attribute\n");
1374                 return -EINVAL;
1375         }
1376
1377         mcastFilterSize_p = (unsigned int *)vio_get_attribute(dev,
1378                                                 VETH_MCAST_FILTER_SIZE, NULL);
1379         if (!mcastFilterSize_p) {
1380                 dev_err(&dev->dev, "Can't find VETH_MCAST_FILTER_SIZE "
1381                         "attribute\n");
1382                 return -EINVAL;
1383         }
1384
1385         netdev = alloc_etherdev(sizeof(struct ibmveth_adapter));
1386
1387         if (!netdev)
1388                 return -ENOMEM;
1389
1390         adapter = netdev_priv(netdev);
1391         dev_set_drvdata(&dev->dev, netdev);
1392
1393         adapter->vdev = dev;
1394         adapter->netdev = netdev;
1395         adapter->mcastFilterSize = *mcastFilterSize_p;
1396         adapter->pool_config = 0;
1397
1398         netif_napi_add(netdev, &adapter->napi, ibmveth_poll, 16);
1399
1400         /*
1401          * Some older boxes running PHYP non-natively have an OF that returns
1402          * a 8-byte local-mac-address field (and the first 2 bytes have to be
1403          * ignored) while newer boxes' OF return a 6-byte field. Note that
1404          * IEEE 1275 specifies that local-mac-address must be a 6-byte field.
1405          * The RPA doc specifies that the first byte must be 10b, so we'll
1406          * just look for it to solve this 8 vs. 6 byte field issue
1407          */
1408         if ((*mac_addr_p & 0x3) != 0x02)
1409                 mac_addr_p += 2;
1410
1411         adapter->mac_addr = 0;
1412         memcpy(&adapter->mac_addr, mac_addr_p, 6);
1413
1414         netdev->irq = dev->irq;
1415         netdev->netdev_ops = &ibmveth_netdev_ops;
1416         netdev->ethtool_ops = &netdev_ethtool_ops;
1417         SET_NETDEV_DEV(netdev, &dev->dev);
1418         netdev->features |= NETIF_F_SG;
1419
1420         memcpy(netdev->dev_addr, &adapter->mac_addr, netdev->addr_len);
1421
1422         for (i = 0; i < IBMVETH_NUM_BUFF_POOLS; i++) {
1423                 struct kobject *kobj = &adapter->rx_buff_pool[i].kobj;
1424                 int error;
1425
1426                 ibmveth_init_buffer_pool(&adapter->rx_buff_pool[i], i,
1427                                          pool_count[i], pool_size[i],
1428                                          pool_active[i]);
1429                 error = kobject_init_and_add(kobj, &ktype_veth_pool,
1430                                              &dev->dev.kobj, "pool%d", i);
1431                 if (!error)
1432                         kobject_uevent(kobj, KOBJ_ADD);
1433         }
1434
1435         netdev_dbg(netdev, "adapter @ 0x%p\n", adapter);
1436
1437         adapter->buffer_list_dma = DMA_ERROR_CODE;
1438         adapter->filter_list_dma = DMA_ERROR_CODE;
1439         adapter->rx_queue.queue_dma = DMA_ERROR_CODE;
1440
1441         netdev_dbg(netdev, "registering netdev...\n");
1442
1443         ibmveth_set_csum_offload(netdev, 1, ibmveth_set_tx_csum_flags);
1444
1445         rc = register_netdev(netdev);
1446
1447         if (rc) {
1448                 netdev_dbg(netdev, "failed to register netdev rc=%d\n", rc);
1449                 free_netdev(netdev);
1450                 return rc;
1451         }
1452
1453         netdev_dbg(netdev, "registered\n");
1454
1455         return 0;
1456 }
1457
1458 static int __devexit ibmveth_remove(struct vio_dev *dev)
1459 {
1460         struct net_device *netdev = dev_get_drvdata(&dev->dev);
1461         struct ibmveth_adapter *adapter = netdev_priv(netdev);
1462         int i;
1463
1464         for (i = 0; i < IBMVETH_NUM_BUFF_POOLS; i++)
1465                 kobject_put(&adapter->rx_buff_pool[i].kobj);
1466
1467         unregister_netdev(netdev);
1468
1469         free_netdev(netdev);
1470         dev_set_drvdata(&dev->dev, NULL);
1471
1472         return 0;
1473 }
1474
1475 static struct attribute veth_active_attr;
1476 static struct attribute veth_num_attr;
1477 static struct attribute veth_size_attr;
1478
1479 static ssize_t veth_pool_show(struct kobject *kobj,
1480                               struct attribute *attr, char *buf)
1481 {
1482         struct ibmveth_buff_pool *pool = container_of(kobj,
1483                                                       struct ibmveth_buff_pool,
1484                                                       kobj);
1485
1486         if (attr == &veth_active_attr)
1487                 return sprintf(buf, "%d\n", pool->active);
1488         else if (attr == &veth_num_attr)
1489                 return sprintf(buf, "%d\n", pool->size);
1490         else if (attr == &veth_size_attr)
1491                 return sprintf(buf, "%d\n", pool->buff_size);
1492         return 0;
1493 }
1494
1495 static ssize_t veth_pool_store(struct kobject *kobj, struct attribute *attr,
1496                                const char *buf, size_t count)
1497 {
1498         struct ibmveth_buff_pool *pool = container_of(kobj,
1499                                                       struct ibmveth_buff_pool,
1500                                                       kobj);
1501         struct net_device *netdev = dev_get_drvdata(
1502             container_of(kobj->parent, struct device, kobj));
1503         struct ibmveth_adapter *adapter = netdev_priv(netdev);
1504         long value = simple_strtol(buf, NULL, 10);
1505         long rc;
1506
1507         if (attr == &veth_active_attr) {
1508                 if (value && !pool->active) {
1509                         if (netif_running(netdev)) {
1510                                 if (ibmveth_alloc_buffer_pool(pool)) {
1511                                         netdev_err(netdev,
1512                                                    "unable to alloc pool\n");
1513                                         return -ENOMEM;
1514                                 }
1515                                 pool->active = 1;
1516                                 adapter->pool_config = 1;
1517                                 ibmveth_close(netdev);
1518                                 adapter->pool_config = 0;
1519                                 if ((rc = ibmveth_open(netdev)))
1520                                         return rc;
1521                         } else {
1522                                 pool->active = 1;
1523                         }
1524                 } else if (!value && pool->active) {
1525                         int mtu = netdev->mtu + IBMVETH_BUFF_OH;
1526                         int i;
1527                         /* Make sure there is a buffer pool with buffers that
1528                            can hold a packet of the size of the MTU */
1529                         for (i = 0; i < IBMVETH_NUM_BUFF_POOLS; i++) {
1530                                 if (pool == &adapter->rx_buff_pool[i])
1531                                         continue;
1532                                 if (!adapter->rx_buff_pool[i].active)
1533                                         continue;
1534                                 if (mtu <= adapter->rx_buff_pool[i].buff_size)
1535                                         break;
1536                         }
1537
1538                         if (i == IBMVETH_NUM_BUFF_POOLS) {
1539                                 netdev_err(netdev, "no active pool >= MTU\n");
1540                                 return -EPERM;
1541                         }
1542
1543                         if (netif_running(netdev)) {
1544                                 adapter->pool_config = 1;
1545                                 ibmveth_close(netdev);
1546                                 pool->active = 0;
1547                                 adapter->pool_config = 0;
1548                                 if ((rc = ibmveth_open(netdev)))
1549                                         return rc;
1550                         }
1551                         pool->active = 0;
1552                 }
1553         } else if (attr == &veth_num_attr) {
1554                 if (value <= 0 || value > IBMVETH_MAX_POOL_COUNT) {
1555                         return -EINVAL;
1556                 } else {
1557                         if (netif_running(netdev)) {
1558                                 adapter->pool_config = 1;
1559                                 ibmveth_close(netdev);
1560                                 adapter->pool_config = 0;
1561                                 pool->size = value;
1562                                 if ((rc = ibmveth_open(netdev)))
1563                                         return rc;
1564                         } else {
1565                                 pool->size = value;
1566                         }
1567                 }
1568         } else if (attr == &veth_size_attr) {
1569                 if (value <= IBMVETH_BUFF_OH || value > IBMVETH_MAX_BUF_SIZE) {
1570                         return -EINVAL;
1571                 } else {
1572                         if (netif_running(netdev)) {
1573                                 adapter->pool_config = 1;
1574                                 ibmveth_close(netdev);
1575                                 adapter->pool_config = 0;
1576                                 pool->buff_size = value;
1577                                 if ((rc = ibmveth_open(netdev)))
1578                                         return rc;
1579                         } else {
1580                                 pool->buff_size = value;
1581                         }
1582                 }
1583         }
1584
1585         /* kick the interrupt handler to allocate/deallocate pools */
1586         ibmveth_interrupt(netdev->irq, netdev);
1587         return count;
1588 }
1589
1590
1591 #define ATTR(_name, _mode)                              \
1592         struct attribute veth_##_name##_attr = {        \
1593         .name = __stringify(_name), .mode = _mode,      \
1594         };
1595
1596 static ATTR(active, 0644);
1597 static ATTR(num, 0644);
1598 static ATTR(size, 0644);
1599
1600 static struct attribute *veth_pool_attrs[] = {
1601         &veth_active_attr,
1602         &veth_num_attr,
1603         &veth_size_attr,
1604         NULL,
1605 };
1606
1607 static const struct sysfs_ops veth_pool_ops = {
1608         .show   = veth_pool_show,
1609         .store  = veth_pool_store,
1610 };
1611
1612 static struct kobj_type ktype_veth_pool = {
1613         .release        = NULL,
1614         .sysfs_ops      = &veth_pool_ops,
1615         .default_attrs  = veth_pool_attrs,
1616 };
1617
1618 static int ibmveth_resume(struct device *dev)
1619 {
1620         struct net_device *netdev = dev_get_drvdata(dev);
1621         ibmveth_interrupt(netdev->irq, netdev);
1622         return 0;
1623 }
1624
1625 static struct vio_device_id ibmveth_device_table[] __devinitdata = {
1626         { "network", "IBM,l-lan"},
1627         { "", "" }
1628 };
1629 MODULE_DEVICE_TABLE(vio, ibmveth_device_table);
1630
1631 static struct dev_pm_ops ibmveth_pm_ops = {
1632         .resume = ibmveth_resume
1633 };
1634
1635 static struct vio_driver ibmveth_driver = {
1636         .id_table       = ibmveth_device_table,
1637         .probe          = ibmveth_probe,
1638         .remove         = ibmveth_remove,
1639         .get_desired_dma = ibmveth_get_desired_dma,
1640         .driver         = {
1641                 .name   = ibmveth_driver_name,
1642                 .owner  = THIS_MODULE,
1643                 .pm = &ibmveth_pm_ops,
1644         }
1645 };
1646
1647 static int __init ibmveth_module_init(void)
1648 {
1649         printk(KERN_DEBUG "%s: %s %s\n", ibmveth_driver_name,
1650                ibmveth_driver_string, ibmveth_driver_version);
1651
1652         return vio_register_driver(&ibmveth_driver);
1653 }
1654
1655 static void __exit ibmveth_module_exit(void)
1656 {
1657         vio_unregister_driver(&ibmveth_driver);
1658 }
1659
1660 module_init(ibmveth_module_init);
1661 module_exit(ibmveth_module_exit);