net: ibmveth: force reconfiguring checksum settings on startup
[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_free_irq;
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_free_irq;
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_free_irq:
665         free_irq(netdev->irq, netdev);
666 err_out:
667         ibmveth_cleanup(adapter);
668         napi_disable(&adapter->napi);
669         return rc;
670 }
671
672 static int ibmveth_close(struct net_device *netdev)
673 {
674         struct ibmveth_adapter *adapter = netdev_priv(netdev);
675         long lpar_rc;
676
677         netdev_dbg(netdev, "close starting\n");
678
679         napi_disable(&adapter->napi);
680
681         if (!adapter->pool_config)
682                 netif_stop_queue(netdev);
683
684         h_vio_signal(adapter->vdev->unit_address, VIO_IRQ_DISABLE);
685
686         do {
687                 lpar_rc = h_free_logical_lan(adapter->vdev->unit_address);
688         } while (H_IS_LONG_BUSY(lpar_rc) || (lpar_rc == H_BUSY));
689
690         if (lpar_rc != H_SUCCESS) {
691                 netdev_err(netdev, "h_free_logical_lan failed with %lx, "
692                            "continuing with close\n", lpar_rc);
693         }
694
695         free_irq(netdev->irq, netdev);
696
697         adapter->rx_no_buffer = *(u64 *)(((char *)adapter->buffer_list_addr) +
698                                                 4096 - 8);
699
700         ibmveth_cleanup(adapter);
701
702         netdev_dbg(netdev, "close complete\n");
703
704         return 0;
705 }
706
707 static int netdev_get_settings(struct net_device *dev, struct ethtool_cmd *cmd)
708 {
709         cmd->supported = (SUPPORTED_1000baseT_Full | SUPPORTED_Autoneg |
710                                 SUPPORTED_FIBRE);
711         cmd->advertising = (ADVERTISED_1000baseT_Full | ADVERTISED_Autoneg |
712                                 ADVERTISED_FIBRE);
713         cmd->speed = SPEED_1000;
714         cmd->duplex = DUPLEX_FULL;
715         cmd->port = PORT_FIBRE;
716         cmd->phy_address = 0;
717         cmd->transceiver = XCVR_INTERNAL;
718         cmd->autoneg = AUTONEG_ENABLE;
719         cmd->maxtxpkt = 0;
720         cmd->maxrxpkt = 1;
721         return 0;
722 }
723
724 static void netdev_get_drvinfo(struct net_device *dev,
725                                struct ethtool_drvinfo *info)
726 {
727         strncpy(info->driver, ibmveth_driver_name, sizeof(info->driver) - 1);
728         strncpy(info->version, ibmveth_driver_version,
729                 sizeof(info->version) - 1);
730 }
731
732 static u32 ibmveth_fix_features(struct net_device *dev, u32 features)
733 {
734         /*
735          * Since the ibmveth firmware interface does not have the
736          * concept of separate tx/rx checksum offload enable, if rx
737          * checksum is disabled we also have to disable tx checksum
738          * offload. Once we disable rx checksum offload, we are no
739          * longer allowed to send tx buffers that are not properly
740          * checksummed.
741          */
742
743         if (!(features & NETIF_F_RXCSUM))
744                 features &= ~NETIF_F_ALL_CSUM;
745
746         return features;
747 }
748
749 static int ibmveth_set_csum_offload(struct net_device *dev, u32 data)
750 {
751         struct ibmveth_adapter *adapter = netdev_priv(dev);
752         unsigned long set_attr, clr_attr, ret_attr;
753         unsigned long set_attr6, clr_attr6;
754         long ret, ret6;
755         int rc1 = 0, rc2 = 0;
756         int restart = 0;
757
758         if (netif_running(dev)) {
759                 restart = 1;
760                 adapter->pool_config = 1;
761                 ibmveth_close(dev);
762                 adapter->pool_config = 0;
763         }
764
765         set_attr = 0;
766         clr_attr = 0;
767
768         if (data) {
769                 set_attr = IBMVETH_ILLAN_IPV4_TCP_CSUM;
770                 set_attr6 = IBMVETH_ILLAN_IPV6_TCP_CSUM;
771         } else {
772                 clr_attr = IBMVETH_ILLAN_IPV4_TCP_CSUM;
773                 clr_attr6 = IBMVETH_ILLAN_IPV6_TCP_CSUM;
774         }
775
776         ret = h_illan_attributes(adapter->vdev->unit_address, 0, 0, &ret_attr);
777
778         if (ret == H_SUCCESS && !(ret_attr & IBMVETH_ILLAN_ACTIVE_TRUNK) &&
779             !(ret_attr & IBMVETH_ILLAN_TRUNK_PRI_MASK) &&
780             (ret_attr & IBMVETH_ILLAN_PADDED_PKT_CSUM)) {
781                 ret = h_illan_attributes(adapter->vdev->unit_address, clr_attr,
782                                          set_attr, &ret_attr);
783
784                 if (ret != H_SUCCESS) {
785                         netdev_err(dev, "unable to change IPv4 checksum "
786                                         "offload settings. %d rc=%ld\n",
787                                         data, ret);
788
789                         ret = h_illan_attributes(adapter->vdev->unit_address,
790                                                  set_attr, clr_attr, &ret_attr);
791                 } else {
792                         adapter->fw_ipv4_csum_support = data;
793                 }
794
795                 ret6 = h_illan_attributes(adapter->vdev->unit_address,
796                                          clr_attr6, set_attr6, &ret_attr);
797
798                 if (ret6 != H_SUCCESS) {
799                         netdev_err(dev, "unable to change IPv6 checksum "
800                                         "offload settings. %d rc=%ld\n",
801                                         data, ret);
802
803                         ret = h_illan_attributes(adapter->vdev->unit_address,
804                                                  set_attr6, clr_attr6,
805                                                  &ret_attr);
806                 } else
807                         adapter->fw_ipv6_csum_support = data;
808
809                 if (ret != H_SUCCESS || ret6 != H_SUCCESS)
810                         adapter->rx_csum = data;
811                 else
812                         rc1 = -EIO;
813         } else {
814                 rc1 = -EIO;
815                 netdev_err(dev, "unable to change checksum offload settings."
816                                      " %d rc=%ld ret_attr=%lx\n", data, ret,
817                                      ret_attr);
818         }
819
820         if (restart)
821                 rc2 = ibmveth_open(dev);
822
823         return rc1 ? rc1 : rc2;
824 }
825
826 static int ibmveth_set_features(struct net_device *dev, u32 features)
827 {
828         struct ibmveth_adapter *adapter = netdev_priv(dev);
829         int rx_csum = !!(features & NETIF_F_RXCSUM);
830         int rc;
831
832         if (rx_csum == adapter->rx_csum)
833                 return 0;
834
835         rc = ibmveth_set_csum_offload(dev, rx_csum);
836         if (rc && !adapter->rx_csum)
837                 dev->features = features & ~(NETIF_F_ALL_CSUM | NETIF_F_RXCSUM);
838
839         return rc;
840 }
841
842 static void ibmveth_get_strings(struct net_device *dev, u32 stringset, u8 *data)
843 {
844         int i;
845
846         if (stringset != ETH_SS_STATS)
847                 return;
848
849         for (i = 0; i < ARRAY_SIZE(ibmveth_stats); i++, data += ETH_GSTRING_LEN)
850                 memcpy(data, ibmveth_stats[i].name, ETH_GSTRING_LEN);
851 }
852
853 static int ibmveth_get_sset_count(struct net_device *dev, int sset)
854 {
855         switch (sset) {
856         case ETH_SS_STATS:
857                 return ARRAY_SIZE(ibmveth_stats);
858         default:
859                 return -EOPNOTSUPP;
860         }
861 }
862
863 static void ibmveth_get_ethtool_stats(struct net_device *dev,
864                                       struct ethtool_stats *stats, u64 *data)
865 {
866         int i;
867         struct ibmveth_adapter *adapter = netdev_priv(dev);
868
869         for (i = 0; i < ARRAY_SIZE(ibmveth_stats); i++)
870                 data[i] = IBMVETH_GET_STAT(adapter, ibmveth_stats[i].offset);
871 }
872
873 static const struct ethtool_ops netdev_ethtool_ops = {
874         .get_drvinfo            = netdev_get_drvinfo,
875         .get_settings           = netdev_get_settings,
876         .get_link               = ethtool_op_get_link,
877         .get_strings            = ibmveth_get_strings,
878         .get_sset_count         = ibmveth_get_sset_count,
879         .get_ethtool_stats      = ibmveth_get_ethtool_stats,
880 };
881
882 static int ibmveth_ioctl(struct net_device *dev, struct ifreq *ifr, int cmd)
883 {
884         return -EOPNOTSUPP;
885 }
886
887 #define page_offset(v) ((unsigned long)(v) & ((1 << 12) - 1))
888
889 static int ibmveth_send(struct ibmveth_adapter *adapter,
890                         union ibmveth_buf_desc *descs)
891 {
892         unsigned long correlator;
893         unsigned int retry_count;
894         unsigned long ret;
895
896         /*
897          * The retry count sets a maximum for the number of broadcast and
898          * multicast destinations within the system.
899          */
900         retry_count = 1024;
901         correlator = 0;
902         do {
903                 ret = h_send_logical_lan(adapter->vdev->unit_address,
904                                              descs[0].desc, descs[1].desc,
905                                              descs[2].desc, descs[3].desc,
906                                              descs[4].desc, descs[5].desc,
907                                              correlator, &correlator);
908         } while ((ret == H_BUSY) && (retry_count--));
909
910         if (ret != H_SUCCESS && ret != H_DROPPED) {
911                 netdev_err(adapter->netdev, "tx: h_send_logical_lan failed "
912                            "with rc=%ld\n", ret);
913                 return 1;
914         }
915
916         return 0;
917 }
918
919 static netdev_tx_t ibmveth_start_xmit(struct sk_buff *skb,
920                                       struct net_device *netdev)
921 {
922         struct ibmveth_adapter *adapter = netdev_priv(netdev);
923         unsigned int desc_flags;
924         union ibmveth_buf_desc descs[6];
925         int last, i;
926         int force_bounce = 0;
927
928         /*
929          * veth handles a maximum of 6 segments including the header, so
930          * we have to linearize the skb if there are more than this.
931          */
932         if (skb_shinfo(skb)->nr_frags > 5 && __skb_linearize(skb)) {
933                 netdev->stats.tx_dropped++;
934                 goto out;
935         }
936
937         /* veth can't checksum offload UDP */
938         if (skb->ip_summed == CHECKSUM_PARTIAL &&
939             ((skb->protocol == htons(ETH_P_IP) &&
940               ip_hdr(skb)->protocol != IPPROTO_TCP) ||
941              (skb->protocol == htons(ETH_P_IPV6) &&
942               ipv6_hdr(skb)->nexthdr != IPPROTO_TCP)) &&
943             skb_checksum_help(skb)) {
944
945                 netdev_err(netdev, "tx: failed to checksum packet\n");
946                 netdev->stats.tx_dropped++;
947                 goto out;
948         }
949
950         desc_flags = IBMVETH_BUF_VALID;
951
952         if (skb->ip_summed == CHECKSUM_PARTIAL) {
953                 unsigned char *buf = skb_transport_header(skb) +
954                                                 skb->csum_offset;
955
956                 desc_flags |= (IBMVETH_BUF_NO_CSUM | IBMVETH_BUF_CSUM_GOOD);
957
958                 /* Need to zero out the checksum */
959                 buf[0] = 0;
960                 buf[1] = 0;
961         }
962
963 retry_bounce:
964         memset(descs, 0, sizeof(descs));
965
966         /*
967          * If a linear packet is below the rx threshold then
968          * copy it into the static bounce buffer. This avoids the
969          * cost of a TCE insert and remove.
970          */
971         if (force_bounce || (!skb_is_nonlinear(skb) &&
972                                 (skb->len < tx_copybreak))) {
973                 skb_copy_from_linear_data(skb, adapter->bounce_buffer,
974                                           skb->len);
975
976                 descs[0].fields.flags_len = desc_flags | skb->len;
977                 descs[0].fields.address = adapter->bounce_buffer_dma;
978
979                 if (ibmveth_send(adapter, descs)) {
980                         adapter->tx_send_failed++;
981                         netdev->stats.tx_dropped++;
982                 } else {
983                         netdev->stats.tx_packets++;
984                         netdev->stats.tx_bytes += skb->len;
985                 }
986
987                 goto out;
988         }
989
990         /* Map the header */
991         descs[0].fields.address = dma_map_single(&adapter->vdev->dev, skb->data,
992                                                  skb_headlen(skb),
993                                                  DMA_TO_DEVICE);
994         if (dma_mapping_error(&adapter->vdev->dev, descs[0].fields.address))
995                 goto map_failed;
996
997         descs[0].fields.flags_len = desc_flags | skb_headlen(skb);
998
999         /* Map the frags */
1000         for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
1001                 unsigned long dma_addr;
1002                 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
1003
1004                 dma_addr = dma_map_page(&adapter->vdev->dev, frag->page,
1005                                         frag->page_offset, frag->size,
1006                                         DMA_TO_DEVICE);
1007
1008                 if (dma_mapping_error(&adapter->vdev->dev, dma_addr))
1009                         goto map_failed_frags;
1010
1011                 descs[i+1].fields.flags_len = desc_flags | frag->size;
1012                 descs[i+1].fields.address = dma_addr;
1013         }
1014
1015         if (ibmveth_send(adapter, descs)) {
1016                 adapter->tx_send_failed++;
1017                 netdev->stats.tx_dropped++;
1018         } else {
1019                 netdev->stats.tx_packets++;
1020                 netdev->stats.tx_bytes += skb->len;
1021         }
1022
1023         for (i = 0; i < skb_shinfo(skb)->nr_frags + 1; i++)
1024                 dma_unmap_page(&adapter->vdev->dev, descs[i].fields.address,
1025                                descs[i].fields.flags_len & IBMVETH_BUF_LEN_MASK,
1026                                DMA_TO_DEVICE);
1027
1028 out:
1029         dev_kfree_skb(skb);
1030         return NETDEV_TX_OK;
1031
1032 map_failed_frags:
1033         last = i+1;
1034         for (i = 0; i < last; i++)
1035                 dma_unmap_page(&adapter->vdev->dev, descs[i].fields.address,
1036                                descs[i].fields.flags_len & IBMVETH_BUF_LEN_MASK,
1037                                DMA_TO_DEVICE);
1038
1039 map_failed:
1040         if (!firmware_has_feature(FW_FEATURE_CMO))
1041                 netdev_err(netdev, "tx: unable to map xmit buffer\n");
1042         adapter->tx_map_failed++;
1043         skb_linearize(skb);
1044         force_bounce = 1;
1045         goto retry_bounce;
1046 }
1047
1048 static int ibmveth_poll(struct napi_struct *napi, int budget)
1049 {
1050         struct ibmveth_adapter *adapter =
1051                         container_of(napi, struct ibmveth_adapter, napi);
1052         struct net_device *netdev = adapter->netdev;
1053         int frames_processed = 0;
1054         unsigned long lpar_rc;
1055
1056 restart_poll:
1057         do {
1058                 if (!ibmveth_rxq_pending_buffer(adapter))
1059                         break;
1060
1061                 smp_rmb();
1062                 if (!ibmveth_rxq_buffer_valid(adapter)) {
1063                         wmb(); /* suggested by larson1 */
1064                         adapter->rx_invalid_buffer++;
1065                         netdev_dbg(netdev, "recycling invalid buffer\n");
1066                         ibmveth_rxq_recycle_buffer(adapter);
1067                 } else {
1068                         struct sk_buff *skb, *new_skb;
1069                         int length = ibmveth_rxq_frame_length(adapter);
1070                         int offset = ibmveth_rxq_frame_offset(adapter);
1071                         int csum_good = ibmveth_rxq_csum_good(adapter);
1072
1073                         skb = ibmveth_rxq_get_buffer(adapter);
1074
1075                         new_skb = NULL;
1076                         if (length < rx_copybreak)
1077                                 new_skb = netdev_alloc_skb(netdev, length);
1078
1079                         if (new_skb) {
1080                                 skb_copy_to_linear_data(new_skb,
1081                                                         skb->data + offset,
1082                                                         length);
1083                                 if (rx_flush)
1084                                         ibmveth_flush_buffer(skb->data,
1085                                                 length + offset);
1086                                 skb = new_skb;
1087                                 ibmveth_rxq_recycle_buffer(adapter);
1088                         } else {
1089                                 ibmveth_rxq_harvest_buffer(adapter);
1090                                 skb_reserve(skb, offset);
1091                         }
1092
1093                         skb_put(skb, length);
1094                         skb->protocol = eth_type_trans(skb, netdev);
1095
1096                         if (csum_good)
1097                                 skb->ip_summed = CHECKSUM_UNNECESSARY;
1098
1099                         netif_receive_skb(skb); /* send it up */
1100
1101                         netdev->stats.rx_packets++;
1102                         netdev->stats.rx_bytes += length;
1103                         frames_processed++;
1104                 }
1105         } while (frames_processed < budget);
1106
1107         ibmveth_replenish_task(adapter);
1108
1109         if (frames_processed < budget) {
1110                 /* We think we are done - reenable interrupts,
1111                  * then check once more to make sure we are done.
1112                  */
1113                 lpar_rc = h_vio_signal(adapter->vdev->unit_address,
1114                                        VIO_IRQ_ENABLE);
1115
1116                 BUG_ON(lpar_rc != H_SUCCESS);
1117
1118                 napi_complete(napi);
1119
1120                 if (ibmveth_rxq_pending_buffer(adapter) &&
1121                     napi_reschedule(napi)) {
1122                         lpar_rc = h_vio_signal(adapter->vdev->unit_address,
1123                                                VIO_IRQ_DISABLE);
1124                         goto restart_poll;
1125                 }
1126         }
1127
1128         return frames_processed;
1129 }
1130
1131 static irqreturn_t ibmveth_interrupt(int irq, void *dev_instance)
1132 {
1133         struct net_device *netdev = dev_instance;
1134         struct ibmveth_adapter *adapter = netdev_priv(netdev);
1135         unsigned long lpar_rc;
1136
1137         if (napi_schedule_prep(&adapter->napi)) {
1138                 lpar_rc = h_vio_signal(adapter->vdev->unit_address,
1139                                        VIO_IRQ_DISABLE);
1140                 BUG_ON(lpar_rc != H_SUCCESS);
1141                 __napi_schedule(&adapter->napi);
1142         }
1143         return IRQ_HANDLED;
1144 }
1145
1146 static void ibmveth_set_multicast_list(struct net_device *netdev)
1147 {
1148         struct ibmveth_adapter *adapter = netdev_priv(netdev);
1149         unsigned long lpar_rc;
1150
1151         if ((netdev->flags & IFF_PROMISC) ||
1152             (netdev_mc_count(netdev) > adapter->mcastFilterSize)) {
1153                 lpar_rc = h_multicast_ctrl(adapter->vdev->unit_address,
1154                                            IbmVethMcastEnableRecv |
1155                                            IbmVethMcastDisableFiltering,
1156                                            0);
1157                 if (lpar_rc != H_SUCCESS) {
1158                         netdev_err(netdev, "h_multicast_ctrl rc=%ld when "
1159                                    "entering promisc mode\n", lpar_rc);
1160                 }
1161         } else {
1162                 struct netdev_hw_addr *ha;
1163                 /* clear the filter table & disable filtering */
1164                 lpar_rc = h_multicast_ctrl(adapter->vdev->unit_address,
1165                                            IbmVethMcastEnableRecv |
1166                                            IbmVethMcastDisableFiltering |
1167                                            IbmVethMcastClearFilterTable,
1168                                            0);
1169                 if (lpar_rc != H_SUCCESS) {
1170                         netdev_err(netdev, "h_multicast_ctrl rc=%ld when "
1171                                    "attempting to clear filter table\n",
1172                                    lpar_rc);
1173                 }
1174                 /* add the addresses to the filter table */
1175                 netdev_for_each_mc_addr(ha, netdev) {
1176                         /* add the multicast address to the filter table */
1177                         unsigned long mcast_addr = 0;
1178                         memcpy(((char *)&mcast_addr)+2, ha->addr, 6);
1179                         lpar_rc = h_multicast_ctrl(adapter->vdev->unit_address,
1180                                                    IbmVethMcastAddFilter,
1181                                                    mcast_addr);
1182                         if (lpar_rc != H_SUCCESS) {
1183                                 netdev_err(netdev, "h_multicast_ctrl rc=%ld "
1184                                            "when adding an entry to the filter "
1185                                            "table\n", lpar_rc);
1186                         }
1187                 }
1188
1189                 /* re-enable filtering */
1190                 lpar_rc = h_multicast_ctrl(adapter->vdev->unit_address,
1191                                            IbmVethMcastEnableFiltering,
1192                                            0);
1193                 if (lpar_rc != H_SUCCESS) {
1194                         netdev_err(netdev, "h_multicast_ctrl rc=%ld when "
1195                                    "enabling filtering\n", lpar_rc);
1196                 }
1197         }
1198 }
1199
1200 static int ibmveth_change_mtu(struct net_device *dev, int new_mtu)
1201 {
1202         struct ibmveth_adapter *adapter = netdev_priv(dev);
1203         struct vio_dev *viodev = adapter->vdev;
1204         int new_mtu_oh = new_mtu + IBMVETH_BUFF_OH;
1205         int i, rc;
1206         int need_restart = 0;
1207
1208         if (new_mtu < IBMVETH_MIN_MTU)
1209                 return -EINVAL;
1210
1211         for (i = 0; i < IBMVETH_NUM_BUFF_POOLS; i++)
1212                 if (new_mtu_oh < adapter->rx_buff_pool[i].buff_size)
1213                         break;
1214
1215         if (i == IBMVETH_NUM_BUFF_POOLS)
1216                 return -EINVAL;
1217
1218         /* Deactivate all the buffer pools so that the next loop can activate
1219            only the buffer pools necessary to hold the new MTU */
1220         if (netif_running(adapter->netdev)) {
1221                 need_restart = 1;
1222                 adapter->pool_config = 1;
1223                 ibmveth_close(adapter->netdev);
1224                 adapter->pool_config = 0;
1225         }
1226
1227         /* Look for an active buffer pool that can hold the new MTU */
1228         for (i = 0; i < IBMVETH_NUM_BUFF_POOLS; i++) {
1229                 adapter->rx_buff_pool[i].active = 1;
1230
1231                 if (new_mtu_oh < adapter->rx_buff_pool[i].buff_size) {
1232                         dev->mtu = new_mtu;
1233                         vio_cmo_set_dev_desired(viodev,
1234                                                 ibmveth_get_desired_dma
1235                                                 (viodev));
1236                         if (need_restart) {
1237                                 return ibmveth_open(adapter->netdev);
1238                         }
1239                         return 0;
1240                 }
1241         }
1242
1243         if (need_restart && (rc = ibmveth_open(adapter->netdev)))
1244                 return rc;
1245
1246         return -EINVAL;
1247 }
1248
1249 #ifdef CONFIG_NET_POLL_CONTROLLER
1250 static void ibmveth_poll_controller(struct net_device *dev)
1251 {
1252         ibmveth_replenish_task(netdev_priv(dev));
1253         ibmveth_interrupt(dev->irq, dev);
1254 }
1255 #endif
1256
1257 /**
1258  * ibmveth_get_desired_dma - Calculate IO memory desired by the driver
1259  *
1260  * @vdev: struct vio_dev for the device whose desired IO mem is to be returned
1261  *
1262  * Return value:
1263  *      Number of bytes of IO data the driver will need to perform well.
1264  */
1265 static unsigned long ibmveth_get_desired_dma(struct vio_dev *vdev)
1266 {
1267         struct net_device *netdev = dev_get_drvdata(&vdev->dev);
1268         struct ibmveth_adapter *adapter;
1269         unsigned long ret;
1270         int i;
1271         int rxqentries = 1;
1272
1273         /* netdev inits at probe time along with the structures we need below*/
1274         if (netdev == NULL)
1275                 return IOMMU_PAGE_ALIGN(IBMVETH_IO_ENTITLEMENT_DEFAULT);
1276
1277         adapter = netdev_priv(netdev);
1278
1279         ret = IBMVETH_BUFF_LIST_SIZE + IBMVETH_FILT_LIST_SIZE;
1280         ret += IOMMU_PAGE_ALIGN(netdev->mtu);
1281
1282         for (i = 0; i < IBMVETH_NUM_BUFF_POOLS; i++) {
1283                 /* add the size of the active receive buffers */
1284                 if (adapter->rx_buff_pool[i].active)
1285                         ret +=
1286                             adapter->rx_buff_pool[i].size *
1287                             IOMMU_PAGE_ALIGN(adapter->rx_buff_pool[i].
1288                                     buff_size);
1289                 rxqentries += adapter->rx_buff_pool[i].size;
1290         }
1291         /* add the size of the receive queue entries */
1292         ret += IOMMU_PAGE_ALIGN(rxqentries * sizeof(struct ibmveth_rx_q_entry));
1293
1294         return ret;
1295 }
1296
1297 static const struct net_device_ops ibmveth_netdev_ops = {
1298         .ndo_open               = ibmveth_open,
1299         .ndo_stop               = ibmveth_close,
1300         .ndo_start_xmit         = ibmveth_start_xmit,
1301         .ndo_set_multicast_list = ibmveth_set_multicast_list,
1302         .ndo_do_ioctl           = ibmveth_ioctl,
1303         .ndo_change_mtu         = ibmveth_change_mtu,
1304         .ndo_fix_features       = ibmveth_fix_features,
1305         .ndo_set_features       = ibmveth_set_features,
1306         .ndo_validate_addr      = eth_validate_addr,
1307         .ndo_set_mac_address    = eth_mac_addr,
1308 #ifdef CONFIG_NET_POLL_CONTROLLER
1309         .ndo_poll_controller    = ibmveth_poll_controller,
1310 #endif
1311 };
1312
1313 static int __devinit ibmveth_probe(struct vio_dev *dev,
1314                                    const struct vio_device_id *id)
1315 {
1316         int rc, i;
1317         struct net_device *netdev;
1318         struct ibmveth_adapter *adapter;
1319         unsigned char *mac_addr_p;
1320         unsigned int *mcastFilterSize_p;
1321
1322         dev_dbg(&dev->dev, "entering ibmveth_probe for UA 0x%x\n",
1323                 dev->unit_address);
1324
1325         mac_addr_p = (unsigned char *)vio_get_attribute(dev, VETH_MAC_ADDR,
1326                                                         NULL);
1327         if (!mac_addr_p) {
1328                 dev_err(&dev->dev, "Can't find VETH_MAC_ADDR attribute\n");
1329                 return -EINVAL;
1330         }
1331
1332         mcastFilterSize_p = (unsigned int *)vio_get_attribute(dev,
1333                                                 VETH_MCAST_FILTER_SIZE, NULL);
1334         if (!mcastFilterSize_p) {
1335                 dev_err(&dev->dev, "Can't find VETH_MCAST_FILTER_SIZE "
1336                         "attribute\n");
1337                 return -EINVAL;
1338         }
1339
1340         netdev = alloc_etherdev(sizeof(struct ibmveth_adapter));
1341
1342         if (!netdev)
1343                 return -ENOMEM;
1344
1345         adapter = netdev_priv(netdev);
1346         dev_set_drvdata(&dev->dev, netdev);
1347
1348         adapter->vdev = dev;
1349         adapter->netdev = netdev;
1350         adapter->mcastFilterSize = *mcastFilterSize_p;
1351         adapter->pool_config = 0;
1352
1353         netif_napi_add(netdev, &adapter->napi, ibmveth_poll, 16);
1354
1355         /*
1356          * Some older boxes running PHYP non-natively have an OF that returns
1357          * a 8-byte local-mac-address field (and the first 2 bytes have to be
1358          * ignored) while newer boxes' OF return a 6-byte field. Note that
1359          * IEEE 1275 specifies that local-mac-address must be a 6-byte field.
1360          * The RPA doc specifies that the first byte must be 10b, so we'll
1361          * just look for it to solve this 8 vs. 6 byte field issue
1362          */
1363         if ((*mac_addr_p & 0x3) != 0x02)
1364                 mac_addr_p += 2;
1365
1366         adapter->mac_addr = 0;
1367         memcpy(&adapter->mac_addr, mac_addr_p, 6);
1368
1369         netdev->irq = dev->irq;
1370         netdev->netdev_ops = &ibmveth_netdev_ops;
1371         netdev->ethtool_ops = &netdev_ethtool_ops;
1372         SET_NETDEV_DEV(netdev, &dev->dev);
1373         netdev->hw_features = NETIF_F_SG | NETIF_F_RXCSUM |
1374                 NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM;
1375         netdev->features |= netdev->hw_features;
1376
1377         memcpy(netdev->dev_addr, &adapter->mac_addr, netdev->addr_len);
1378
1379         for (i = 0; i < IBMVETH_NUM_BUFF_POOLS; i++) {
1380                 struct kobject *kobj = &adapter->rx_buff_pool[i].kobj;
1381                 int error;
1382
1383                 ibmveth_init_buffer_pool(&adapter->rx_buff_pool[i], i,
1384                                          pool_count[i], pool_size[i],
1385                                          pool_active[i]);
1386                 error = kobject_init_and_add(kobj, &ktype_veth_pool,
1387                                              &dev->dev.kobj, "pool%d", i);
1388                 if (!error)
1389                         kobject_uevent(kobj, KOBJ_ADD);
1390         }
1391
1392         netdev_dbg(netdev, "adapter @ 0x%p\n", adapter);
1393
1394         adapter->buffer_list_dma = DMA_ERROR_CODE;
1395         adapter->filter_list_dma = DMA_ERROR_CODE;
1396         adapter->rx_queue.queue_dma = DMA_ERROR_CODE;
1397
1398         netdev_dbg(netdev, "registering netdev...\n");
1399
1400         ibmveth_set_features(netdev, netdev->features);
1401
1402         rc = register_netdev(netdev);
1403
1404         if (rc) {
1405                 netdev_dbg(netdev, "failed to register netdev rc=%d\n", rc);
1406                 free_netdev(netdev);
1407                 return rc;
1408         }
1409
1410         netdev_dbg(netdev, "registered\n");
1411
1412         return 0;
1413 }
1414
1415 static int __devexit ibmveth_remove(struct vio_dev *dev)
1416 {
1417         struct net_device *netdev = dev_get_drvdata(&dev->dev);
1418         struct ibmveth_adapter *adapter = netdev_priv(netdev);
1419         int i;
1420
1421         for (i = 0; i < IBMVETH_NUM_BUFF_POOLS; i++)
1422                 kobject_put(&adapter->rx_buff_pool[i].kobj);
1423
1424         unregister_netdev(netdev);
1425
1426         free_netdev(netdev);
1427         dev_set_drvdata(&dev->dev, NULL);
1428
1429         return 0;
1430 }
1431
1432 static struct attribute veth_active_attr;
1433 static struct attribute veth_num_attr;
1434 static struct attribute veth_size_attr;
1435
1436 static ssize_t veth_pool_show(struct kobject *kobj,
1437                               struct attribute *attr, char *buf)
1438 {
1439         struct ibmveth_buff_pool *pool = container_of(kobj,
1440                                                       struct ibmveth_buff_pool,
1441                                                       kobj);
1442
1443         if (attr == &veth_active_attr)
1444                 return sprintf(buf, "%d\n", pool->active);
1445         else if (attr == &veth_num_attr)
1446                 return sprintf(buf, "%d\n", pool->size);
1447         else if (attr == &veth_size_attr)
1448                 return sprintf(buf, "%d\n", pool->buff_size);
1449         return 0;
1450 }
1451
1452 static ssize_t veth_pool_store(struct kobject *kobj, struct attribute *attr,
1453                                const char *buf, size_t count)
1454 {
1455         struct ibmveth_buff_pool *pool = container_of(kobj,
1456                                                       struct ibmveth_buff_pool,
1457                                                       kobj);
1458         struct net_device *netdev = dev_get_drvdata(
1459             container_of(kobj->parent, struct device, kobj));
1460         struct ibmveth_adapter *adapter = netdev_priv(netdev);
1461         long value = simple_strtol(buf, NULL, 10);
1462         long rc;
1463
1464         if (attr == &veth_active_attr) {
1465                 if (value && !pool->active) {
1466                         if (netif_running(netdev)) {
1467                                 if (ibmveth_alloc_buffer_pool(pool)) {
1468                                         netdev_err(netdev,
1469                                                    "unable to alloc pool\n");
1470                                         return -ENOMEM;
1471                                 }
1472                                 pool->active = 1;
1473                                 adapter->pool_config = 1;
1474                                 ibmveth_close(netdev);
1475                                 adapter->pool_config = 0;
1476                                 if ((rc = ibmveth_open(netdev)))
1477                                         return rc;
1478                         } else {
1479                                 pool->active = 1;
1480                         }
1481                 } else if (!value && pool->active) {
1482                         int mtu = netdev->mtu + IBMVETH_BUFF_OH;
1483                         int i;
1484                         /* Make sure there is a buffer pool with buffers that
1485                            can hold a packet of the size of the MTU */
1486                         for (i = 0; i < IBMVETH_NUM_BUFF_POOLS; i++) {
1487                                 if (pool == &adapter->rx_buff_pool[i])
1488                                         continue;
1489                                 if (!adapter->rx_buff_pool[i].active)
1490                                         continue;
1491                                 if (mtu <= adapter->rx_buff_pool[i].buff_size)
1492                                         break;
1493                         }
1494
1495                         if (i == IBMVETH_NUM_BUFF_POOLS) {
1496                                 netdev_err(netdev, "no active pool >= MTU\n");
1497                                 return -EPERM;
1498                         }
1499
1500                         if (netif_running(netdev)) {
1501                                 adapter->pool_config = 1;
1502                                 ibmveth_close(netdev);
1503                                 pool->active = 0;
1504                                 adapter->pool_config = 0;
1505                                 if ((rc = ibmveth_open(netdev)))
1506                                         return rc;
1507                         }
1508                         pool->active = 0;
1509                 }
1510         } else if (attr == &veth_num_attr) {
1511                 if (value <= 0 || value > IBMVETH_MAX_POOL_COUNT) {
1512                         return -EINVAL;
1513                 } else {
1514                         if (netif_running(netdev)) {
1515                                 adapter->pool_config = 1;
1516                                 ibmveth_close(netdev);
1517                                 adapter->pool_config = 0;
1518                                 pool->size = value;
1519                                 if ((rc = ibmveth_open(netdev)))
1520                                         return rc;
1521                         } else {
1522                                 pool->size = value;
1523                         }
1524                 }
1525         } else if (attr == &veth_size_attr) {
1526                 if (value <= IBMVETH_BUFF_OH || value > IBMVETH_MAX_BUF_SIZE) {
1527                         return -EINVAL;
1528                 } else {
1529                         if (netif_running(netdev)) {
1530                                 adapter->pool_config = 1;
1531                                 ibmveth_close(netdev);
1532                                 adapter->pool_config = 0;
1533                                 pool->buff_size = value;
1534                                 if ((rc = ibmveth_open(netdev)))
1535                                         return rc;
1536                         } else {
1537                                 pool->buff_size = value;
1538                         }
1539                 }
1540         }
1541
1542         /* kick the interrupt handler to allocate/deallocate pools */
1543         ibmveth_interrupt(netdev->irq, netdev);
1544         return count;
1545 }
1546
1547
1548 #define ATTR(_name, _mode)                              \
1549         struct attribute veth_##_name##_attr = {        \
1550         .name = __stringify(_name), .mode = _mode,      \
1551         };
1552
1553 static ATTR(active, 0644);
1554 static ATTR(num, 0644);
1555 static ATTR(size, 0644);
1556
1557 static struct attribute *veth_pool_attrs[] = {
1558         &veth_active_attr,
1559         &veth_num_attr,
1560         &veth_size_attr,
1561         NULL,
1562 };
1563
1564 static const struct sysfs_ops veth_pool_ops = {
1565         .show   = veth_pool_show,
1566         .store  = veth_pool_store,
1567 };
1568
1569 static struct kobj_type ktype_veth_pool = {
1570         .release        = NULL,
1571         .sysfs_ops      = &veth_pool_ops,
1572         .default_attrs  = veth_pool_attrs,
1573 };
1574
1575 static int ibmveth_resume(struct device *dev)
1576 {
1577         struct net_device *netdev = dev_get_drvdata(dev);
1578         ibmveth_interrupt(netdev->irq, netdev);
1579         return 0;
1580 }
1581
1582 static struct vio_device_id ibmveth_device_table[] __devinitdata = {
1583         { "network", "IBM,l-lan"},
1584         { "", "" }
1585 };
1586 MODULE_DEVICE_TABLE(vio, ibmveth_device_table);
1587
1588 static struct dev_pm_ops ibmveth_pm_ops = {
1589         .resume = ibmveth_resume
1590 };
1591
1592 static struct vio_driver ibmveth_driver = {
1593         .id_table       = ibmveth_device_table,
1594         .probe          = ibmveth_probe,
1595         .remove         = ibmveth_remove,
1596         .get_desired_dma = ibmveth_get_desired_dma,
1597         .driver         = {
1598                 .name   = ibmveth_driver_name,
1599                 .owner  = THIS_MODULE,
1600                 .pm = &ibmveth_pm_ops,
1601         }
1602 };
1603
1604 static int __init ibmveth_module_init(void)
1605 {
1606         printk(KERN_DEBUG "%s: %s %s\n", ibmveth_driver_name,
1607                ibmveth_driver_string, ibmveth_driver_version);
1608
1609         return vio_register_driver(&ibmveth_driver);
1610 }
1611
1612 static void __exit ibmveth_module_exit(void)
1613 {
1614         vio_unregister_driver(&ibmveth_driver);
1615 }
1616
1617 module_init(ibmveth_module_init);
1618 module_exit(ibmveth_module_exit);