ibmveth: Convert driver specific assert to BUG_ON
[linux-2.6.git] / drivers / net / ibmveth.c
1 /*
2  * IBM eServer i/pSeries Virtual Ethernet Device Driver
3  * Copyright (C) 2003 IBM Corp.
4  *  Originally written by Dave Larson (larson1@us.ibm.com)
5  *  Maintained by Santiago Leon (santil@us.ibm.com)
6  *
7  *  This program is free software; you can redistribute it and/or modify
8  *  it under the terms of the GNU General Public License as published by
9  *  the Free Software Foundation; either version 2 of the License, or
10  *  (at your option) any later version.
11  *
12  *  This program is distributed in the hope that it will be useful,
13  *  but WITHOUT ANY WARRANTY; without even the implied warranty of
14  *  MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
15  *  GNU General Public License for more details.
16  *
17  *  You should have received a copy of the GNU General Public License
18  *  along with this program; if not, write to the Free Software
19  *  Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA  02111-1307
20  *                                                                   USA
21  *
22  * This module contains the implementation of a virtual ethernet device
23  * for use with IBM i/pSeries LPAR Linux.  It utilizes the logical LAN
24  * option of the RS/6000 Platform Architechture to interface with virtual
25  * ethernet NICs that are presented to the partition by the hypervisor.
26  */
27
28 #include <linux/module.h>
29 #include <linux/moduleparam.h>
30 #include <linux/types.h>
31 #include <linux/errno.h>
32 #include <linux/ioport.h>
33 #include <linux/dma-mapping.h>
34 #include <linux/kernel.h>
35 #include <linux/netdevice.h>
36 #include <linux/etherdevice.h>
37 #include <linux/skbuff.h>
38 #include <linux/init.h>
39 #include <linux/delay.h>
40 #include <linux/mm.h>
41 #include <linux/pm.h>
42 #include <linux/ethtool.h>
43 #include <linux/in.h>
44 #include <linux/ip.h>
45 #include <linux/ipv6.h>
46 #include <linux/slab.h>
47 #include <asm/hvcall.h>
48 #include <asm/atomic.h>
49 #include <asm/vio.h>
50 #include <asm/iommu.h>
51 #include <asm/uaccess.h>
52 #include <asm/firmware.h>
53
54 #include "ibmveth.h"
55
56 static irqreturn_t ibmveth_interrupt(int irq, void *dev_instance);
57 static void ibmveth_rxq_harvest_buffer(struct ibmveth_adapter *adapter);
58 static unsigned long ibmveth_get_desired_dma(struct vio_dev *vdev);
59
60 static struct kobj_type ktype_veth_pool;
61
62
63 static const char ibmveth_driver_name[] = "ibmveth";
64 static const char ibmveth_driver_string[] = "IBM i/pSeries Virtual Ethernet "
65                                             "Driver";
66 #define ibmveth_driver_version "1.03"
67
68 MODULE_AUTHOR("Santiago Leon <santil@us.ibm.com>");
69 MODULE_DESCRIPTION("IBM i/pSeries Virtual Ethernet Driver");
70 MODULE_LICENSE("GPL");
71 MODULE_VERSION(ibmveth_driver_version);
72
73 static unsigned int tx_copybreak __read_mostly = 128;
74 module_param(tx_copybreak, uint, 0644);
75 MODULE_PARM_DESC(tx_copybreak,
76         "Maximum size of packet that is copied to a new buffer on transmit");
77
78 static unsigned int rx_copybreak __read_mostly = 128;
79 module_param(rx_copybreak, uint, 0644);
80 MODULE_PARM_DESC(rx_copybreak,
81         "Maximum size of packet that is copied to a new buffer on receive");
82
83 static unsigned int rx_flush __read_mostly = 0;
84 module_param(rx_flush, uint, 0644);
85 MODULE_PARM_DESC(rx_flush, "Flush receive buffers before use");
86
87 struct ibmveth_stat {
88         char name[ETH_GSTRING_LEN];
89         int offset;
90 };
91
92 #define IBMVETH_STAT_OFF(stat) offsetof(struct ibmveth_adapter, stat)
93 #define IBMVETH_GET_STAT(a, off) *((u64 *)(((unsigned long)(a)) + off))
94
95 struct ibmveth_stat ibmveth_stats[] = {
96         { "replenish_task_cycles", IBMVETH_STAT_OFF(replenish_task_cycles) },
97         { "replenish_no_mem", IBMVETH_STAT_OFF(replenish_no_mem) },
98         { "replenish_add_buff_failure",
99                         IBMVETH_STAT_OFF(replenish_add_buff_failure) },
100         { "replenish_add_buff_success",
101                         IBMVETH_STAT_OFF(replenish_add_buff_success) },
102         { "rx_invalid_buffer", IBMVETH_STAT_OFF(rx_invalid_buffer) },
103         { "rx_no_buffer", IBMVETH_STAT_OFF(rx_no_buffer) },
104         { "tx_map_failed", IBMVETH_STAT_OFF(tx_map_failed) },
105         { "tx_send_failed", IBMVETH_STAT_OFF(tx_send_failed) },
106         { "fw_enabled_ipv4_csum", IBMVETH_STAT_OFF(fw_ipv4_csum_support) },
107         { "fw_enabled_ipv6_csum", IBMVETH_STAT_OFF(fw_ipv6_csum_support) },
108 };
109
110 /* simple methods of getting data from the current rxq entry */
111 static inline u32 ibmveth_rxq_flags(struct ibmveth_adapter *adapter)
112 {
113         return adapter->rx_queue.queue_addr[adapter->rx_queue.index].flags_off;
114 }
115
116 static inline int ibmveth_rxq_toggle(struct ibmveth_adapter *adapter)
117 {
118         return (ibmveth_rxq_flags(adapter) & IBMVETH_RXQ_TOGGLE) >>
119                         IBMVETH_RXQ_TOGGLE_SHIFT;
120 }
121
122 static inline int ibmveth_rxq_pending_buffer(struct ibmveth_adapter *adapter)
123 {
124         return ibmveth_rxq_toggle(adapter) == adapter->rx_queue.toggle;
125 }
126
127 static inline int ibmveth_rxq_buffer_valid(struct ibmveth_adapter *adapter)
128 {
129         return ibmveth_rxq_flags(adapter) & IBMVETH_RXQ_VALID;
130 }
131
132 static inline int ibmveth_rxq_frame_offset(struct ibmveth_adapter *adapter)
133 {
134         return ibmveth_rxq_flags(adapter) & IBMVETH_RXQ_OFF_MASK;
135 }
136
137 static inline int ibmveth_rxq_frame_length(struct ibmveth_adapter *adapter)
138 {
139         return adapter->rx_queue.queue_addr[adapter->rx_queue.index].length;
140 }
141
142 static inline int ibmveth_rxq_csum_good(struct ibmveth_adapter *adapter)
143 {
144         return ibmveth_rxq_flags(adapter) & IBMVETH_RXQ_CSUM_GOOD;
145 }
146
147 /* setup the initial settings for a buffer pool */
148 static void ibmveth_init_buffer_pool(struct ibmveth_buff_pool *pool,
149                                      u32 pool_index, u32 pool_size,
150                                      u32 buff_size, u32 pool_active)
151 {
152         pool->size = pool_size;
153         pool->index = pool_index;
154         pool->buff_size = buff_size;
155         pool->threshold = pool_size * 7 / 8;
156         pool->active = pool_active;
157 }
158
159 /* allocate and setup an buffer pool - called during open */
160 static int ibmveth_alloc_buffer_pool(struct ibmveth_buff_pool *pool)
161 {
162         int i;
163
164         pool->free_map = kmalloc(sizeof(u16) * pool->size, GFP_KERNEL);
165
166         if (!pool->free_map)
167                 return -1;
168
169         pool->dma_addr = kmalloc(sizeof(dma_addr_t) * pool->size, GFP_KERNEL);
170         if (!pool->dma_addr) {
171                 kfree(pool->free_map);
172                 pool->free_map = NULL;
173                 return -1;
174         }
175
176         pool->skbuff = kcalloc(pool->size, sizeof(void *), GFP_KERNEL);
177
178         if (!pool->skbuff) {
179                 kfree(pool->dma_addr);
180                 pool->dma_addr = NULL;
181
182                 kfree(pool->free_map);
183                 pool->free_map = NULL;
184                 return -1;
185         }
186
187         memset(pool->dma_addr, 0, sizeof(dma_addr_t) * pool->size);
188
189         for (i = 0; i < pool->size; ++i)
190                 pool->free_map[i] = i;
191
192         atomic_set(&pool->available, 0);
193         pool->producer_index = 0;
194         pool->consumer_index = 0;
195
196         return 0;
197 }
198
199 static inline void ibmveth_flush_buffer(void *addr, unsigned long length)
200 {
201         unsigned long offset;
202
203         for (offset = 0; offset < length; offset += SMP_CACHE_BYTES)
204                 asm("dcbfl %0,%1" :: "b" (addr), "r" (offset));
205 }
206
207 /* replenish the buffers for a pool.  note that we don't need to
208  * skb_reserve these since they are used for incoming...
209  */
210 static void ibmveth_replenish_buffer_pool(struct ibmveth_adapter *adapter,
211                                           struct ibmveth_buff_pool *pool)
212 {
213         u32 i;
214         u32 count = pool->size - atomic_read(&pool->available);
215         u32 buffers_added = 0;
216         struct sk_buff *skb;
217         unsigned int free_index, index;
218         u64 correlator;
219         unsigned long lpar_rc;
220         dma_addr_t dma_addr;
221
222         mb();
223
224         for (i = 0; i < count; ++i) {
225                 union ibmveth_buf_desc desc;
226
227                 skb = netdev_alloc_skb(adapter->netdev, pool->buff_size);
228
229                 if (!skb) {
230                         netdev_dbg(adapter->netdev,
231                                    "replenish: unable to allocate skb\n");
232                         adapter->replenish_no_mem++;
233                         break;
234                 }
235
236                 free_index = pool->consumer_index;
237                 pool->consumer_index++;
238                 if (pool->consumer_index >= pool->size)
239                         pool->consumer_index = 0;
240                 index = pool->free_map[free_index];
241
242                 BUG_ON(index == IBM_VETH_INVALID_MAP);
243                 BUG_ON(pool->skbuff[index] != NULL);
244
245                 dma_addr = dma_map_single(&adapter->vdev->dev, skb->data,
246                                 pool->buff_size, DMA_FROM_DEVICE);
247
248                 if (dma_mapping_error(&adapter->vdev->dev, dma_addr))
249                         goto failure;
250
251                 pool->free_map[free_index] = IBM_VETH_INVALID_MAP;
252                 pool->dma_addr[index] = dma_addr;
253                 pool->skbuff[index] = skb;
254
255                 correlator = ((u64)pool->index << 32) | index;
256                 *(u64 *)skb->data = correlator;
257
258                 desc.fields.flags_len = IBMVETH_BUF_VALID | pool->buff_size;
259                 desc.fields.address = dma_addr;
260
261                 if (rx_flush) {
262                         unsigned int len = min(pool->buff_size,
263                                                 adapter->netdev->mtu +
264                                                 IBMVETH_BUFF_OH);
265                         ibmveth_flush_buffer(skb->data, len);
266                 }
267                 lpar_rc = h_add_logical_lan_buffer(adapter->vdev->unit_address,
268                                                    desc.desc);
269
270                 if (lpar_rc != H_SUCCESS) {
271                         goto failure;
272                 } else {
273                         buffers_added++;
274                         adapter->replenish_add_buff_success++;
275                 }
276         }
277
278         mb();
279         atomic_add(buffers_added, &(pool->available));
280         return;
281
282 failure:
283         pool->free_map[free_index] = index;
284         pool->skbuff[index] = NULL;
285         if (pool->consumer_index == 0)
286                 pool->consumer_index = pool->size - 1;
287         else
288                 pool->consumer_index--;
289         if (!dma_mapping_error(&adapter->vdev->dev, dma_addr))
290                 dma_unmap_single(&adapter->vdev->dev,
291                                  pool->dma_addr[index], pool->buff_size,
292                                  DMA_FROM_DEVICE);
293         dev_kfree_skb_any(skb);
294         adapter->replenish_add_buff_failure++;
295
296         mb();
297         atomic_add(buffers_added, &(pool->available));
298 }
299
300 /* replenish routine */
301 static void ibmveth_replenish_task(struct ibmveth_adapter *adapter)
302 {
303         int i;
304
305         adapter->replenish_task_cycles++;
306
307         for (i = (IBMVETH_NUM_BUFF_POOLS - 1); i >= 0; i--) {
308                 struct ibmveth_buff_pool *pool = &adapter->rx_buff_pool[i];
309
310                 if (pool->active &&
311                     (atomic_read(&pool->available) < pool->threshold))
312                         ibmveth_replenish_buffer_pool(adapter, pool);
313         }
314
315         adapter->rx_no_buffer = *(u64 *)(((char*)adapter->buffer_list_addr) +
316                                                 4096 - 8);
317 }
318
319 /* empty and free ana buffer pool - also used to do cleanup in error paths */
320 static void ibmveth_free_buffer_pool(struct ibmveth_adapter *adapter,
321                                      struct ibmveth_buff_pool *pool)
322 {
323         int i;
324
325         kfree(pool->free_map);
326         pool->free_map = NULL;
327
328         if (pool->skbuff && pool->dma_addr) {
329                 for (i = 0; i < pool->size; ++i) {
330                         struct sk_buff *skb = pool->skbuff[i];
331                         if (skb) {
332                                 dma_unmap_single(&adapter->vdev->dev,
333                                                  pool->dma_addr[i],
334                                                  pool->buff_size,
335                                                  DMA_FROM_DEVICE);
336                                 dev_kfree_skb_any(skb);
337                                 pool->skbuff[i] = NULL;
338                         }
339                 }
340         }
341
342         if (pool->dma_addr) {
343                 kfree(pool->dma_addr);
344                 pool->dma_addr = NULL;
345         }
346
347         if (pool->skbuff) {
348                 kfree(pool->skbuff);
349                 pool->skbuff = NULL;
350         }
351 }
352
353 /* remove a buffer from a pool */
354 static void ibmveth_remove_buffer_from_pool(struct ibmveth_adapter *adapter,
355                                             u64 correlator)
356 {
357         unsigned int pool  = correlator >> 32;
358         unsigned int index = correlator & 0xffffffffUL;
359         unsigned int free_index;
360         struct sk_buff *skb;
361
362         BUG_ON(pool >= IBMVETH_NUM_BUFF_POOLS);
363         BUG_ON(index >= adapter->rx_buff_pool[pool].size);
364
365         skb = adapter->rx_buff_pool[pool].skbuff[index];
366
367         BUG_ON(skb == NULL);
368
369         adapter->rx_buff_pool[pool].skbuff[index] = NULL;
370
371         dma_unmap_single(&adapter->vdev->dev,
372                          adapter->rx_buff_pool[pool].dma_addr[index],
373                          adapter->rx_buff_pool[pool].buff_size,
374                          DMA_FROM_DEVICE);
375
376         free_index = adapter->rx_buff_pool[pool].producer_index;
377         adapter->rx_buff_pool[pool].producer_index++;
378         if (adapter->rx_buff_pool[pool].producer_index >=
379             adapter->rx_buff_pool[pool].size)
380                 adapter->rx_buff_pool[pool].producer_index = 0;
381         adapter->rx_buff_pool[pool].free_map[free_index] = index;
382
383         mb();
384
385         atomic_dec(&(adapter->rx_buff_pool[pool].available));
386 }
387
388 /* get the current buffer on the rx queue */
389 static inline struct sk_buff *ibmveth_rxq_get_buffer(struct ibmveth_adapter *adapter)
390 {
391         u64 correlator = adapter->rx_queue.queue_addr[adapter->rx_queue.index].correlator;
392         unsigned int pool = correlator >> 32;
393         unsigned int index = correlator & 0xffffffffUL;
394
395         BUG_ON(pool >= IBMVETH_NUM_BUFF_POOLS);
396         BUG_ON(index >= adapter->rx_buff_pool[pool].size);
397
398         return adapter->rx_buff_pool[pool].skbuff[index];
399 }
400
401 /* recycle the current buffer on the rx queue */
402 static void ibmveth_rxq_recycle_buffer(struct ibmveth_adapter *adapter)
403 {
404         u32 q_index = adapter->rx_queue.index;
405         u64 correlator = adapter->rx_queue.queue_addr[q_index].correlator;
406         unsigned int pool = correlator >> 32;
407         unsigned int index = correlator & 0xffffffffUL;
408         union ibmveth_buf_desc desc;
409         unsigned long lpar_rc;
410
411         BUG_ON(pool >= IBMVETH_NUM_BUFF_POOLS);
412         BUG_ON(index >= adapter->rx_buff_pool[pool].size);
413
414         if (!adapter->rx_buff_pool[pool].active) {
415                 ibmveth_rxq_harvest_buffer(adapter);
416                 ibmveth_free_buffer_pool(adapter, &adapter->rx_buff_pool[pool]);
417                 return;
418         }
419
420         desc.fields.flags_len = IBMVETH_BUF_VALID |
421                 adapter->rx_buff_pool[pool].buff_size;
422         desc.fields.address = adapter->rx_buff_pool[pool].dma_addr[index];
423
424         lpar_rc = h_add_logical_lan_buffer(adapter->vdev->unit_address, desc.desc);
425
426         if (lpar_rc != H_SUCCESS) {
427                 netdev_dbg(adapter->netdev, "h_add_logical_lan_buffer failed "
428                            "during recycle rc=%ld", lpar_rc);
429                 ibmveth_remove_buffer_from_pool(adapter, adapter->rx_queue.queue_addr[adapter->rx_queue.index].correlator);
430         }
431
432         if (++adapter->rx_queue.index == adapter->rx_queue.num_slots) {
433                 adapter->rx_queue.index = 0;
434                 adapter->rx_queue.toggle = !adapter->rx_queue.toggle;
435         }
436 }
437
438 static void ibmveth_rxq_harvest_buffer(struct ibmveth_adapter *adapter)
439 {
440         ibmveth_remove_buffer_from_pool(adapter, adapter->rx_queue.queue_addr[adapter->rx_queue.index].correlator);
441
442         if (++adapter->rx_queue.index == adapter->rx_queue.num_slots) {
443                 adapter->rx_queue.index = 0;
444                 adapter->rx_queue.toggle = !adapter->rx_queue.toggle;
445         }
446 }
447
448 static void ibmveth_cleanup(struct ibmveth_adapter *adapter)
449 {
450         int i;
451         struct device *dev = &adapter->vdev->dev;
452
453         if (adapter->buffer_list_addr != NULL) {
454                 if (!dma_mapping_error(dev, adapter->buffer_list_dma)) {
455                         dma_unmap_single(dev, adapter->buffer_list_dma, 4096,
456                                         DMA_BIDIRECTIONAL);
457                         adapter->buffer_list_dma = DMA_ERROR_CODE;
458                 }
459                 free_page((unsigned long)adapter->buffer_list_addr);
460                 adapter->buffer_list_addr = NULL;
461         }
462
463         if (adapter->filter_list_addr != NULL) {
464                 if (!dma_mapping_error(dev, adapter->filter_list_dma)) {
465                         dma_unmap_single(dev, adapter->filter_list_dma, 4096,
466                                         DMA_BIDIRECTIONAL);
467                         adapter->filter_list_dma = DMA_ERROR_CODE;
468                 }
469                 free_page((unsigned long)adapter->filter_list_addr);
470                 adapter->filter_list_addr = NULL;
471         }
472
473         if (adapter->rx_queue.queue_addr != NULL) {
474                 if (!dma_mapping_error(dev, adapter->rx_queue.queue_dma)) {
475                         dma_unmap_single(dev,
476                                         adapter->rx_queue.queue_dma,
477                                         adapter->rx_queue.queue_len,
478                                         DMA_BIDIRECTIONAL);
479                         adapter->rx_queue.queue_dma = DMA_ERROR_CODE;
480                 }
481                 kfree(adapter->rx_queue.queue_addr);
482                 adapter->rx_queue.queue_addr = NULL;
483         }
484
485         for (i = 0; i < IBMVETH_NUM_BUFF_POOLS; i++)
486                 if (adapter->rx_buff_pool[i].active)
487                         ibmveth_free_buffer_pool(adapter,
488                                                  &adapter->rx_buff_pool[i]);
489
490         if (adapter->bounce_buffer != NULL) {
491                 if (!dma_mapping_error(dev, adapter->bounce_buffer_dma)) {
492                         dma_unmap_single(&adapter->vdev->dev,
493                                         adapter->bounce_buffer_dma,
494                                         adapter->netdev->mtu + IBMVETH_BUFF_OH,
495                                         DMA_BIDIRECTIONAL);
496                         adapter->bounce_buffer_dma = DMA_ERROR_CODE;
497                 }
498                 kfree(adapter->bounce_buffer);
499                 adapter->bounce_buffer = NULL;
500         }
501 }
502
503 static int ibmveth_register_logical_lan(struct ibmveth_adapter *adapter,
504         union ibmveth_buf_desc rxq_desc, u64 mac_address)
505 {
506         int rc, try_again = 1;
507
508         /*
509          * After a kexec the adapter will still be open, so our attempt to
510          * open it will fail. So if we get a failure we free the adapter and
511          * try again, but only once.
512          */
513 retry:
514         rc = h_register_logical_lan(adapter->vdev->unit_address,
515                                     adapter->buffer_list_dma, rxq_desc.desc,
516                                     adapter->filter_list_dma, mac_address);
517
518         if (rc != H_SUCCESS && try_again) {
519                 do {
520                         rc = h_free_logical_lan(adapter->vdev->unit_address);
521                 } while (H_IS_LONG_BUSY(rc) || (rc == H_BUSY));
522
523                 try_again = 0;
524                 goto retry;
525         }
526
527         return rc;
528 }
529
530 static int ibmveth_open(struct net_device *netdev)
531 {
532         struct ibmveth_adapter *adapter = netdev_priv(netdev);
533         u64 mac_address = 0;
534         int rxq_entries = 1;
535         unsigned long lpar_rc;
536         int rc;
537         union ibmveth_buf_desc rxq_desc;
538         int i;
539         struct device *dev;
540
541         netdev_dbg(netdev, "open starting\n");
542
543         napi_enable(&adapter->napi);
544
545         for(i = 0; i < IBMVETH_NUM_BUFF_POOLS; i++)
546                 rxq_entries += adapter->rx_buff_pool[i].size;
547
548         adapter->buffer_list_addr = (void*) get_zeroed_page(GFP_KERNEL);
549         adapter->filter_list_addr = (void*) get_zeroed_page(GFP_KERNEL);
550
551         if (!adapter->buffer_list_addr || !adapter->filter_list_addr) {
552                 netdev_err(netdev, "unable to allocate filter or buffer list "
553                            "pages\n");
554                 ibmveth_cleanup(adapter);
555                 napi_disable(&adapter->napi);
556                 return -ENOMEM;
557         }
558
559         adapter->rx_queue.queue_len = sizeof(struct ibmveth_rx_q_entry) *
560                                                 rxq_entries;
561         adapter->rx_queue.queue_addr = kmalloc(adapter->rx_queue.queue_len,
562                                                 GFP_KERNEL);
563
564         if (!adapter->rx_queue.queue_addr) {
565                 netdev_err(netdev, "unable to allocate rx queue pages\n");
566                 ibmveth_cleanup(adapter);
567                 napi_disable(&adapter->napi);
568                 return -ENOMEM;
569         }
570
571         dev = &adapter->vdev->dev;
572
573         adapter->buffer_list_dma = dma_map_single(dev,
574                         adapter->buffer_list_addr, 4096, DMA_BIDIRECTIONAL);
575         adapter->filter_list_dma = dma_map_single(dev,
576                         adapter->filter_list_addr, 4096, DMA_BIDIRECTIONAL);
577         adapter->rx_queue.queue_dma = dma_map_single(dev,
578                         adapter->rx_queue.queue_addr,
579                         adapter->rx_queue.queue_len, DMA_BIDIRECTIONAL);
580
581         if ((dma_mapping_error(dev, adapter->buffer_list_dma)) ||
582             (dma_mapping_error(dev, adapter->filter_list_dma)) ||
583             (dma_mapping_error(dev, adapter->rx_queue.queue_dma))) {
584                 netdev_err(netdev, "unable to map filter or buffer list "
585                            "pages\n");
586                 ibmveth_cleanup(adapter);
587                 napi_disable(&adapter->napi);
588                 return -ENOMEM;
589         }
590
591         adapter->rx_queue.index = 0;
592         adapter->rx_queue.num_slots = rxq_entries;
593         adapter->rx_queue.toggle = 1;
594
595         memcpy(&mac_address, netdev->dev_addr, netdev->addr_len);
596         mac_address = mac_address >> 16;
597
598         rxq_desc.fields.flags_len = IBMVETH_BUF_VALID |
599                                         adapter->rx_queue.queue_len;
600         rxq_desc.fields.address = adapter->rx_queue.queue_dma;
601
602         netdev_dbg(netdev, "buffer list @ 0x%p\n", adapter->buffer_list_addr);
603         netdev_dbg(netdev, "filter list @ 0x%p\n", adapter->filter_list_addr);
604         netdev_dbg(netdev, "receive q   @ 0x%p\n", adapter->rx_queue.queue_addr);
605
606         h_vio_signal(adapter->vdev->unit_address, VIO_IRQ_DISABLE);
607
608         lpar_rc = ibmveth_register_logical_lan(adapter, rxq_desc, mac_address);
609
610         if (lpar_rc != H_SUCCESS) {
611                 netdev_err(netdev, "h_register_logical_lan failed with %ld\n",
612                            lpar_rc);
613                 netdev_err(netdev, "buffer TCE:0x%llx filter TCE:0x%llx rxq "
614                            "desc:0x%llx MAC:0x%llx\n",
615                                      adapter->buffer_list_dma,
616                                      adapter->filter_list_dma,
617                                      rxq_desc.desc,
618                                      mac_address);
619                 ibmveth_cleanup(adapter);
620                 napi_disable(&adapter->napi);
621                 return -ENONET;
622         }
623
624         for (i = 0; i < IBMVETH_NUM_BUFF_POOLS; i++) {
625                 if (!adapter->rx_buff_pool[i].active)
626                         continue;
627                 if (ibmveth_alloc_buffer_pool(&adapter->rx_buff_pool[i])) {
628                         netdev_err(netdev, "unable to alloc pool\n");
629                         adapter->rx_buff_pool[i].active = 0;
630                         ibmveth_cleanup(adapter);
631                         napi_disable(&adapter->napi);
632                         return -ENOMEM ;
633                 }
634         }
635
636         netdev_dbg(netdev, "registering irq 0x%x\n", netdev->irq);
637         rc = request_irq(netdev->irq, ibmveth_interrupt, 0, netdev->name,
638                          netdev);
639         if (rc != 0) {
640                 netdev_err(netdev, "unable to request irq 0x%x, rc %d\n",
641                            netdev->irq, rc);
642                 do {
643                         rc = h_free_logical_lan(adapter->vdev->unit_address);
644                 } while (H_IS_LONG_BUSY(rc) || (rc == H_BUSY));
645
646                 ibmveth_cleanup(adapter);
647                 napi_disable(&adapter->napi);
648                 return rc;
649         }
650
651         adapter->bounce_buffer =
652             kmalloc(netdev->mtu + IBMVETH_BUFF_OH, GFP_KERNEL);
653         if (!adapter->bounce_buffer) {
654                 netdev_err(netdev, "unable to allocate bounce buffer\n");
655                 ibmveth_cleanup(adapter);
656                 napi_disable(&adapter->napi);
657                 return -ENOMEM;
658         }
659         adapter->bounce_buffer_dma =
660             dma_map_single(&adapter->vdev->dev, adapter->bounce_buffer,
661                            netdev->mtu + IBMVETH_BUFF_OH, DMA_BIDIRECTIONAL);
662         if (dma_mapping_error(dev, adapter->bounce_buffer_dma)) {
663                 netdev_err(netdev, "unable to map bounce buffer\n");
664                 ibmveth_cleanup(adapter);
665                 napi_disable(&adapter->napi);
666                 return -ENOMEM;
667         }
668
669         netdev_dbg(netdev, "initial replenish cycle\n");
670         ibmveth_interrupt(netdev->irq, netdev);
671
672         netif_start_queue(netdev);
673
674         netdev_dbg(netdev, "open complete\n");
675
676         return 0;
677 }
678
679 static int ibmveth_close(struct net_device *netdev)
680 {
681         struct ibmveth_adapter *adapter = netdev_priv(netdev);
682         long lpar_rc;
683
684         netdev_dbg(netdev, "close starting\n");
685
686         napi_disable(&adapter->napi);
687
688         if (!adapter->pool_config)
689                 netif_stop_queue(netdev);
690
691         h_vio_signal(adapter->vdev->unit_address, VIO_IRQ_DISABLE);
692
693         do {
694                 lpar_rc = h_free_logical_lan(adapter->vdev->unit_address);
695         } while (H_IS_LONG_BUSY(lpar_rc) || (lpar_rc == H_BUSY));
696
697         if (lpar_rc != H_SUCCESS) {
698                 netdev_err(netdev, "h_free_logical_lan failed with %lx, "
699                            "continuing with close\n", lpar_rc);
700         }
701
702         free_irq(netdev->irq, netdev);
703
704         adapter->rx_no_buffer = *(u64 *)(((char *)adapter->buffer_list_addr) +
705                                                 4096 - 8);
706
707         ibmveth_cleanup(adapter);
708
709         netdev_dbg(netdev, "close complete\n");
710
711         return 0;
712 }
713
714 static int netdev_get_settings(struct net_device *dev, struct ethtool_cmd *cmd)
715 {
716         cmd->supported = (SUPPORTED_1000baseT_Full | SUPPORTED_Autoneg |
717                                 SUPPORTED_FIBRE);
718         cmd->advertising = (ADVERTISED_1000baseT_Full | ADVERTISED_Autoneg |
719                                 ADVERTISED_FIBRE);
720         cmd->speed = SPEED_1000;
721         cmd->duplex = DUPLEX_FULL;
722         cmd->port = PORT_FIBRE;
723         cmd->phy_address = 0;
724         cmd->transceiver = XCVR_INTERNAL;
725         cmd->autoneg = AUTONEG_ENABLE;
726         cmd->maxtxpkt = 0;
727         cmd->maxrxpkt = 1;
728         return 0;
729 }
730
731 static void netdev_get_drvinfo(struct net_device *dev,
732                                struct ethtool_drvinfo *info)
733 {
734         strncpy(info->driver, ibmveth_driver_name, sizeof(info->driver) - 1);
735         strncpy(info->version, ibmveth_driver_version,
736                 sizeof(info->version) - 1);
737 }
738
739 static u32 netdev_get_link(struct net_device *dev)
740 {
741         return 1;
742 }
743
744 static void ibmveth_set_rx_csum_flags(struct net_device *dev, u32 data)
745 {
746         struct ibmveth_adapter *adapter = netdev_priv(dev);
747
748         if (data) {
749                 adapter->rx_csum = 1;
750         } else {
751                 /*
752                  * Since the ibmveth firmware interface does not have the
753                  * concept of separate tx/rx checksum offload enable, if rx
754                  * checksum is disabled we also have to disable tx checksum
755                  * offload. Once we disable rx checksum offload, we are no
756                  * longer allowed to send tx buffers that are not properly
757                  * checksummed.
758                  */
759                 adapter->rx_csum = 0;
760                 dev->features &= ~NETIF_F_IP_CSUM;
761                 dev->features &= ~NETIF_F_IPV6_CSUM;
762         }
763 }
764
765 static void ibmveth_set_tx_csum_flags(struct net_device *dev, u32 data)
766 {
767         struct ibmveth_adapter *adapter = netdev_priv(dev);
768
769         if (data) {
770                 if (adapter->fw_ipv4_csum_support)
771                         dev->features |= NETIF_F_IP_CSUM;
772                 if (adapter->fw_ipv6_csum_support)
773                         dev->features |= NETIF_F_IPV6_CSUM;
774                 adapter->rx_csum = 1;
775         } else {
776                 dev->features &= ~NETIF_F_IP_CSUM;
777                 dev->features &= ~NETIF_F_IPV6_CSUM;
778         }
779 }
780
781 static int ibmveth_set_csum_offload(struct net_device *dev, u32 data,
782                                     void (*done) (struct net_device *, u32))
783 {
784         struct ibmveth_adapter *adapter = netdev_priv(dev);
785         unsigned long set_attr, clr_attr, ret_attr;
786         unsigned long set_attr6, clr_attr6;
787         long ret, ret6;
788         int rc1 = 0, rc2 = 0;
789         int restart = 0;
790
791         if (netif_running(dev)) {
792                 restart = 1;
793                 adapter->pool_config = 1;
794                 ibmveth_close(dev);
795                 adapter->pool_config = 0;
796         }
797
798         set_attr = 0;
799         clr_attr = 0;
800
801         if (data) {
802                 set_attr = IBMVETH_ILLAN_IPV4_TCP_CSUM;
803                 set_attr6 = IBMVETH_ILLAN_IPV6_TCP_CSUM;
804         } else {
805                 clr_attr = IBMVETH_ILLAN_IPV4_TCP_CSUM;
806                 clr_attr6 = IBMVETH_ILLAN_IPV6_TCP_CSUM;
807         }
808
809         ret = h_illan_attributes(adapter->vdev->unit_address, 0, 0, &ret_attr);
810
811         if (ret == H_SUCCESS && !(ret_attr & IBMVETH_ILLAN_ACTIVE_TRUNK) &&
812             !(ret_attr & IBMVETH_ILLAN_TRUNK_PRI_MASK) &&
813             (ret_attr & IBMVETH_ILLAN_PADDED_PKT_CSUM)) {
814                 ret = h_illan_attributes(adapter->vdev->unit_address, clr_attr,
815                                          set_attr, &ret_attr);
816
817                 if (ret != H_SUCCESS) {
818                         netdev_err(dev, "unable to change IPv4 checksum "
819                                         "offload settings. %d rc=%ld\n",
820                                         data, ret);
821
822                         ret = h_illan_attributes(adapter->vdev->unit_address,
823                                                  set_attr, clr_attr, &ret_attr);
824                 } else {
825                         adapter->fw_ipv4_csum_support = data;
826                 }
827
828                 ret6 = h_illan_attributes(adapter->vdev->unit_address,
829                                          clr_attr6, set_attr6, &ret_attr);
830
831                 if (ret6 != H_SUCCESS) {
832                         netdev_err(dev, "unable to change IPv6 checksum "
833                                         "offload settings. %d rc=%ld\n",
834                                         data, ret);
835
836                         ret = h_illan_attributes(adapter->vdev->unit_address,
837                                                  set_attr6, clr_attr6,
838                                                  &ret_attr);
839                 } else
840                         adapter->fw_ipv6_csum_support = data;
841
842                 if (ret == H_SUCCESS || ret6 == H_SUCCESS)
843                         done(dev, data);
844                 else
845                         rc1 = -EIO;
846         } else {
847                 rc1 = -EIO;
848                 netdev_err(dev, "unable to change checksum offload settings."
849                                      " %d rc=%ld ret_attr=%lx\n", data, ret,
850                                      ret_attr);
851         }
852
853         if (restart)
854                 rc2 = ibmveth_open(dev);
855
856         return rc1 ? rc1 : rc2;
857 }
858
859 static int ibmveth_set_rx_csum(struct net_device *dev, u32 data)
860 {
861         struct ibmveth_adapter *adapter = netdev_priv(dev);
862
863         if ((data && adapter->rx_csum) || (!data && !adapter->rx_csum))
864                 return 0;
865
866         return ibmveth_set_csum_offload(dev, data, ibmveth_set_rx_csum_flags);
867 }
868
869 static int ibmveth_set_tx_csum(struct net_device *dev, u32 data)
870 {
871         struct ibmveth_adapter *adapter = netdev_priv(dev);
872         int rc = 0;
873
874         if (data && (dev->features & (NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM)))
875                 return 0;
876         if (!data && !(dev->features & (NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM)))
877                 return 0;
878
879         if (data && !adapter->rx_csum)
880                 rc = ibmveth_set_csum_offload(dev, data,
881                                               ibmveth_set_tx_csum_flags);
882         else
883                 ibmveth_set_tx_csum_flags(dev, data);
884
885         return rc;
886 }
887
888 static u32 ibmveth_get_rx_csum(struct net_device *dev)
889 {
890         struct ibmveth_adapter *adapter = netdev_priv(dev);
891         return adapter->rx_csum;
892 }
893
894 static void ibmveth_get_strings(struct net_device *dev, u32 stringset, u8 *data)
895 {
896         int i;
897
898         if (stringset != ETH_SS_STATS)
899                 return;
900
901         for (i = 0; i < ARRAY_SIZE(ibmveth_stats); i++, data += ETH_GSTRING_LEN)
902                 memcpy(data, ibmveth_stats[i].name, ETH_GSTRING_LEN);
903 }
904
905 static int ibmveth_get_sset_count(struct net_device *dev, int sset)
906 {
907         switch (sset) {
908         case ETH_SS_STATS:
909                 return ARRAY_SIZE(ibmveth_stats);
910         default:
911                 return -EOPNOTSUPP;
912         }
913 }
914
915 static void ibmveth_get_ethtool_stats(struct net_device *dev,
916                                       struct ethtool_stats *stats, u64 *data)
917 {
918         int i;
919         struct ibmveth_adapter *adapter = netdev_priv(dev);
920
921         for (i = 0; i < ARRAY_SIZE(ibmveth_stats); i++)
922                 data[i] = IBMVETH_GET_STAT(adapter, ibmveth_stats[i].offset);
923 }
924
925 static const struct ethtool_ops netdev_ethtool_ops = {
926         .get_drvinfo            = netdev_get_drvinfo,
927         .get_settings           = netdev_get_settings,
928         .get_link               = netdev_get_link,
929         .set_tx_csum            = ibmveth_set_tx_csum,
930         .get_rx_csum            = ibmveth_get_rx_csum,
931         .set_rx_csum            = ibmveth_set_rx_csum,
932         .get_strings            = ibmveth_get_strings,
933         .get_sset_count         = ibmveth_get_sset_count,
934         .get_ethtool_stats      = ibmveth_get_ethtool_stats,
935         .set_sg                 = ethtool_op_set_sg,
936 };
937
938 static int ibmveth_ioctl(struct net_device *dev, struct ifreq *ifr, int cmd)
939 {
940         return -EOPNOTSUPP;
941 }
942
943 #define page_offset(v) ((unsigned long)(v) & ((1 << 12) - 1))
944
945 static int ibmveth_send(struct ibmveth_adapter *adapter,
946                         union ibmveth_buf_desc *descs)
947 {
948         unsigned long correlator;
949         unsigned int retry_count;
950         unsigned long ret;
951
952         /*
953          * The retry count sets a maximum for the number of broadcast and
954          * multicast destinations within the system.
955          */
956         retry_count = 1024;
957         correlator = 0;
958         do {
959                 ret = h_send_logical_lan(adapter->vdev->unit_address,
960                                              descs[0].desc, descs[1].desc,
961                                              descs[2].desc, descs[3].desc,
962                                              descs[4].desc, descs[5].desc,
963                                              correlator, &correlator);
964         } while ((ret == H_BUSY) && (retry_count--));
965
966         if (ret != H_SUCCESS && ret != H_DROPPED) {
967                 netdev_err(adapter->netdev, "tx: h_send_logical_lan failed "
968                            "with rc=%ld\n", ret);
969                 return 1;
970         }
971
972         return 0;
973 }
974
975 static netdev_tx_t ibmveth_start_xmit(struct sk_buff *skb,
976                                       struct net_device *netdev)
977 {
978         struct ibmveth_adapter *adapter = netdev_priv(netdev);
979         unsigned int desc_flags;
980         union ibmveth_buf_desc descs[6];
981         int last, i;
982         int force_bounce = 0;
983
984         /*
985          * veth handles a maximum of 6 segments including the header, so
986          * we have to linearize the skb if there are more than this.
987          */
988         if (skb_shinfo(skb)->nr_frags > 5 && __skb_linearize(skb)) {
989                 netdev->stats.tx_dropped++;
990                 goto out;
991         }
992
993         /* veth can't checksum offload UDP */
994         if (skb->ip_summed == CHECKSUM_PARTIAL &&
995             ((skb->protocol == htons(ETH_P_IP) &&
996               ip_hdr(skb)->protocol != IPPROTO_TCP) ||
997              (skb->protocol == htons(ETH_P_IPV6) &&
998               ipv6_hdr(skb)->nexthdr != IPPROTO_TCP)) &&
999             skb_checksum_help(skb)) {
1000
1001                 netdev_err(netdev, "tx: failed to checksum packet\n");
1002                 netdev->stats.tx_dropped++;
1003                 goto out;
1004         }
1005
1006         desc_flags = IBMVETH_BUF_VALID;
1007
1008         if (skb->ip_summed == CHECKSUM_PARTIAL) {
1009                 unsigned char *buf = skb_transport_header(skb) +
1010                                                 skb->csum_offset;
1011
1012                 desc_flags |= (IBMVETH_BUF_NO_CSUM | IBMVETH_BUF_CSUM_GOOD);
1013
1014                 /* Need to zero out the checksum */
1015                 buf[0] = 0;
1016                 buf[1] = 0;
1017         }
1018
1019 retry_bounce:
1020         memset(descs, 0, sizeof(descs));
1021
1022         /*
1023          * If a linear packet is below the rx threshold then
1024          * copy it into the static bounce buffer. This avoids the
1025          * cost of a TCE insert and remove.
1026          */
1027         if (force_bounce || (!skb_is_nonlinear(skb) &&
1028                                 (skb->len < tx_copybreak))) {
1029                 skb_copy_from_linear_data(skb, adapter->bounce_buffer,
1030                                           skb->len);
1031
1032                 descs[0].fields.flags_len = desc_flags | skb->len;
1033                 descs[0].fields.address = adapter->bounce_buffer_dma;
1034
1035                 if (ibmveth_send(adapter, descs)) {
1036                         adapter->tx_send_failed++;
1037                         netdev->stats.tx_dropped++;
1038                 } else {
1039                         netdev->stats.tx_packets++;
1040                         netdev->stats.tx_bytes += skb->len;
1041                 }
1042
1043                 goto out;
1044         }
1045
1046         /* Map the header */
1047         descs[0].fields.address = dma_map_single(&adapter->vdev->dev, skb->data,
1048                                                  skb_headlen(skb),
1049                                                  DMA_TO_DEVICE);
1050         if (dma_mapping_error(&adapter->vdev->dev, descs[0].fields.address))
1051                 goto map_failed;
1052
1053         descs[0].fields.flags_len = desc_flags | skb_headlen(skb);
1054
1055         /* Map the frags */
1056         for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
1057                 unsigned long dma_addr;
1058                 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
1059
1060                 dma_addr = dma_map_page(&adapter->vdev->dev, frag->page,
1061                                         frag->page_offset, frag->size,
1062                                         DMA_TO_DEVICE);
1063
1064                 if (dma_mapping_error(&adapter->vdev->dev, dma_addr))
1065                         goto map_failed_frags;
1066
1067                 descs[i+1].fields.flags_len = desc_flags | frag->size;
1068                 descs[i+1].fields.address = dma_addr;
1069         }
1070
1071         if (ibmveth_send(adapter, descs)) {
1072                 adapter->tx_send_failed++;
1073                 netdev->stats.tx_dropped++;
1074         } else {
1075                 netdev->stats.tx_packets++;
1076                 netdev->stats.tx_bytes += skb->len;
1077         }
1078
1079         for (i = 0; i < skb_shinfo(skb)->nr_frags + 1; i++)
1080                 dma_unmap_page(&adapter->vdev->dev, descs[i].fields.address,
1081                                descs[i].fields.flags_len & IBMVETH_BUF_LEN_MASK,
1082                                DMA_TO_DEVICE);
1083
1084 out:
1085         dev_kfree_skb(skb);
1086         return NETDEV_TX_OK;
1087
1088 map_failed_frags:
1089         last = i+1;
1090         for (i = 0; i < last; i++)
1091                 dma_unmap_page(&adapter->vdev->dev, descs[i].fields.address,
1092                                descs[i].fields.flags_len & IBMVETH_BUF_LEN_MASK,
1093                                DMA_TO_DEVICE);
1094
1095 map_failed:
1096         if (!firmware_has_feature(FW_FEATURE_CMO))
1097                 netdev_err(netdev, "tx: unable to map xmit buffer\n");
1098         adapter->tx_map_failed++;
1099         skb_linearize(skb);
1100         force_bounce = 1;
1101         goto retry_bounce;
1102 }
1103
1104 static int ibmveth_poll(struct napi_struct *napi, int budget)
1105 {
1106         struct ibmveth_adapter *adapter =
1107                         container_of(napi, struct ibmveth_adapter, napi);
1108         struct net_device *netdev = adapter->netdev;
1109         int frames_processed = 0;
1110         unsigned long lpar_rc;
1111
1112 restart_poll:
1113         do {
1114                 if (!ibmveth_rxq_pending_buffer(adapter))
1115                         break;
1116
1117                 smp_rmb();
1118                 if (!ibmveth_rxq_buffer_valid(adapter)) {
1119                         wmb(); /* suggested by larson1 */
1120                         adapter->rx_invalid_buffer++;
1121                         netdev_dbg(netdev, "recycling invalid buffer\n");
1122                         ibmveth_rxq_recycle_buffer(adapter);
1123                 } else {
1124                         struct sk_buff *skb, *new_skb;
1125                         int length = ibmveth_rxq_frame_length(adapter);
1126                         int offset = ibmveth_rxq_frame_offset(adapter);
1127                         int csum_good = ibmveth_rxq_csum_good(adapter);
1128
1129                         skb = ibmveth_rxq_get_buffer(adapter);
1130
1131                         new_skb = NULL;
1132                         if (length < rx_copybreak)
1133                                 new_skb = netdev_alloc_skb(netdev, length);
1134
1135                         if (new_skb) {
1136                                 skb_copy_to_linear_data(new_skb,
1137                                                         skb->data + offset,
1138                                                         length);
1139                                 if (rx_flush)
1140                                         ibmveth_flush_buffer(skb->data,
1141                                                 length + offset);
1142                                 skb = new_skb;
1143                                 ibmveth_rxq_recycle_buffer(adapter);
1144                         } else {
1145                                 ibmveth_rxq_harvest_buffer(adapter);
1146                                 skb_reserve(skb, offset);
1147                         }
1148
1149                         skb_put(skb, length);
1150                         skb->protocol = eth_type_trans(skb, netdev);
1151
1152                         if (csum_good)
1153                                 skb->ip_summed = CHECKSUM_UNNECESSARY;
1154
1155                         netif_receive_skb(skb); /* send it up */
1156
1157                         netdev->stats.rx_packets++;
1158                         netdev->stats.rx_bytes += length;
1159                         frames_processed++;
1160                 }
1161         } while (frames_processed < budget);
1162
1163         ibmveth_replenish_task(adapter);
1164
1165         if (frames_processed < budget) {
1166                 /* We think we are done - reenable interrupts,
1167                  * then check once more to make sure we are done.
1168                  */
1169                 lpar_rc = h_vio_signal(adapter->vdev->unit_address,
1170                                        VIO_IRQ_ENABLE);
1171
1172                 BUG_ON(lpar_rc != H_SUCCESS);
1173
1174                 napi_complete(napi);
1175
1176                 if (ibmveth_rxq_pending_buffer(adapter) &&
1177                     napi_reschedule(napi)) {
1178                         lpar_rc = h_vio_signal(adapter->vdev->unit_address,
1179                                                VIO_IRQ_DISABLE);
1180                         goto restart_poll;
1181                 }
1182         }
1183
1184         return frames_processed;
1185 }
1186
1187 static irqreturn_t ibmveth_interrupt(int irq, void *dev_instance)
1188 {
1189         struct net_device *netdev = dev_instance;
1190         struct ibmveth_adapter *adapter = netdev_priv(netdev);
1191         unsigned long lpar_rc;
1192
1193         if (napi_schedule_prep(&adapter->napi)) {
1194                 lpar_rc = h_vio_signal(adapter->vdev->unit_address,
1195                                        VIO_IRQ_DISABLE);
1196                 BUG_ON(lpar_rc != H_SUCCESS);
1197                 __napi_schedule(&adapter->napi);
1198         }
1199         return IRQ_HANDLED;
1200 }
1201
1202 static void ibmveth_set_multicast_list(struct net_device *netdev)
1203 {
1204         struct ibmveth_adapter *adapter = netdev_priv(netdev);
1205         unsigned long lpar_rc;
1206
1207         if ((netdev->flags & IFF_PROMISC) ||
1208             (netdev_mc_count(netdev) > adapter->mcastFilterSize)) {
1209                 lpar_rc = h_multicast_ctrl(adapter->vdev->unit_address,
1210                                            IbmVethMcastEnableRecv |
1211                                            IbmVethMcastDisableFiltering,
1212                                            0);
1213                 if (lpar_rc != H_SUCCESS) {
1214                         netdev_err(netdev, "h_multicast_ctrl rc=%ld when "
1215                                    "entering promisc mode\n", lpar_rc);
1216                 }
1217         } else {
1218                 struct netdev_hw_addr *ha;
1219                 /* clear the filter table & disable filtering */
1220                 lpar_rc = h_multicast_ctrl(adapter->vdev->unit_address,
1221                                            IbmVethMcastEnableRecv |
1222                                            IbmVethMcastDisableFiltering |
1223                                            IbmVethMcastClearFilterTable,
1224                                            0);
1225                 if (lpar_rc != H_SUCCESS) {
1226                         netdev_err(netdev, "h_multicast_ctrl rc=%ld when "
1227                                    "attempting to clear filter table\n",
1228                                    lpar_rc);
1229                 }
1230                 /* add the addresses to the filter table */
1231                 netdev_for_each_mc_addr(ha, netdev) {
1232                         /* add the multicast address to the filter table */
1233                         unsigned long mcast_addr = 0;
1234                         memcpy(((char *)&mcast_addr)+2, ha->addr, 6);
1235                         lpar_rc = h_multicast_ctrl(adapter->vdev->unit_address,
1236                                                    IbmVethMcastAddFilter,
1237                                                    mcast_addr);
1238                         if (lpar_rc != H_SUCCESS) {
1239                                 netdev_err(netdev, "h_multicast_ctrl rc=%ld "
1240                                            "when adding an entry to the filter "
1241                                            "table\n", lpar_rc);
1242                         }
1243                 }
1244
1245                 /* re-enable filtering */
1246                 lpar_rc = h_multicast_ctrl(adapter->vdev->unit_address,
1247                                            IbmVethMcastEnableFiltering,
1248                                            0);
1249                 if (lpar_rc != H_SUCCESS) {
1250                         netdev_err(netdev, "h_multicast_ctrl rc=%ld when "
1251                                    "enabling filtering\n", lpar_rc);
1252                 }
1253         }
1254 }
1255
1256 static int ibmveth_change_mtu(struct net_device *dev, int new_mtu)
1257 {
1258         struct ibmveth_adapter *adapter = netdev_priv(dev);
1259         struct vio_dev *viodev = adapter->vdev;
1260         int new_mtu_oh = new_mtu + IBMVETH_BUFF_OH;
1261         int i, rc;
1262         int need_restart = 0;
1263
1264         if (new_mtu < IBMVETH_MIN_MTU)
1265                 return -EINVAL;
1266
1267         for (i = 0; i < IBMVETH_NUM_BUFF_POOLS; i++)
1268                 if (new_mtu_oh < adapter->rx_buff_pool[i].buff_size)
1269                         break;
1270
1271         if (i == IBMVETH_NUM_BUFF_POOLS)
1272                 return -EINVAL;
1273
1274         /* Deactivate all the buffer pools so that the next loop can activate
1275            only the buffer pools necessary to hold the new MTU */
1276         if (netif_running(adapter->netdev)) {
1277                 need_restart = 1;
1278                 adapter->pool_config = 1;
1279                 ibmveth_close(adapter->netdev);
1280                 adapter->pool_config = 0;
1281         }
1282
1283         /* Look for an active buffer pool that can hold the new MTU */
1284         for (i = 0; i < IBMVETH_NUM_BUFF_POOLS; i++) {
1285                 adapter->rx_buff_pool[i].active = 1;
1286
1287                 if (new_mtu_oh < adapter->rx_buff_pool[i].buff_size) {
1288                         dev->mtu = new_mtu;
1289                         vio_cmo_set_dev_desired(viodev,
1290                                                 ibmveth_get_desired_dma
1291                                                 (viodev));
1292                         if (need_restart) {
1293                                 return ibmveth_open(adapter->netdev);
1294                         }
1295                         return 0;
1296                 }
1297         }
1298
1299         if (need_restart && (rc = ibmveth_open(adapter->netdev)))
1300                 return rc;
1301
1302         return -EINVAL;
1303 }
1304
1305 #ifdef CONFIG_NET_POLL_CONTROLLER
1306 static void ibmveth_poll_controller(struct net_device *dev)
1307 {
1308         ibmveth_replenish_task(netdev_priv(dev));
1309         ibmveth_interrupt(dev->irq, dev);
1310 }
1311 #endif
1312
1313 /**
1314  * ibmveth_get_desired_dma - Calculate IO memory desired by the driver
1315  *
1316  * @vdev: struct vio_dev for the device whose desired IO mem is to be returned
1317  *
1318  * Return value:
1319  *      Number of bytes of IO data the driver will need to perform well.
1320  */
1321 static unsigned long ibmveth_get_desired_dma(struct vio_dev *vdev)
1322 {
1323         struct net_device *netdev = dev_get_drvdata(&vdev->dev);
1324         struct ibmveth_adapter *adapter;
1325         unsigned long ret;
1326         int i;
1327         int rxqentries = 1;
1328
1329         /* netdev inits at probe time along with the structures we need below*/
1330         if (netdev == NULL)
1331                 return IOMMU_PAGE_ALIGN(IBMVETH_IO_ENTITLEMENT_DEFAULT);
1332
1333         adapter = netdev_priv(netdev);
1334
1335         ret = IBMVETH_BUFF_LIST_SIZE + IBMVETH_FILT_LIST_SIZE;
1336         ret += IOMMU_PAGE_ALIGN(netdev->mtu);
1337
1338         for (i = 0; i < IBMVETH_NUM_BUFF_POOLS; i++) {
1339                 /* add the size of the active receive buffers */
1340                 if (adapter->rx_buff_pool[i].active)
1341                         ret +=
1342                             adapter->rx_buff_pool[i].size *
1343                             IOMMU_PAGE_ALIGN(adapter->rx_buff_pool[i].
1344                                     buff_size);
1345                 rxqentries += adapter->rx_buff_pool[i].size;
1346         }
1347         /* add the size of the receive queue entries */
1348         ret += IOMMU_PAGE_ALIGN(rxqentries * sizeof(struct ibmveth_rx_q_entry));
1349
1350         return ret;
1351 }
1352
1353 static const struct net_device_ops ibmveth_netdev_ops = {
1354         .ndo_open               = ibmveth_open,
1355         .ndo_stop               = ibmveth_close,
1356         .ndo_start_xmit         = ibmveth_start_xmit,
1357         .ndo_set_multicast_list = ibmveth_set_multicast_list,
1358         .ndo_do_ioctl           = ibmveth_ioctl,
1359         .ndo_change_mtu         = ibmveth_change_mtu,
1360         .ndo_validate_addr      = eth_validate_addr,
1361         .ndo_set_mac_address    = eth_mac_addr,
1362 #ifdef CONFIG_NET_POLL_CONTROLLER
1363         .ndo_poll_controller    = ibmveth_poll_controller,
1364 #endif
1365 };
1366
1367 static int __devinit ibmveth_probe(struct vio_dev *dev,
1368                                    const struct vio_device_id *id)
1369 {
1370         int rc, i;
1371         struct net_device *netdev;
1372         struct ibmveth_adapter *adapter;
1373         unsigned char *mac_addr_p;
1374         unsigned int *mcastFilterSize_p;
1375
1376         dev_dbg(&dev->dev, "entering ibmveth_probe for UA 0x%x\n",
1377                 dev->unit_address);
1378
1379         mac_addr_p = (unsigned char *)vio_get_attribute(dev, VETH_MAC_ADDR,
1380                                                         NULL);
1381         if (!mac_addr_p) {
1382                 dev_err(&dev->dev, "Can't find VETH_MAC_ADDR attribute\n");
1383                 return -EINVAL;
1384         }
1385
1386         mcastFilterSize_p = (unsigned int *)vio_get_attribute(dev,
1387                                                 VETH_MCAST_FILTER_SIZE, NULL);
1388         if (!mcastFilterSize_p) {
1389                 dev_err(&dev->dev, "Can't find VETH_MCAST_FILTER_SIZE "
1390                         "attribute\n");
1391                 return -EINVAL;
1392         }
1393
1394         netdev = alloc_etherdev(sizeof(struct ibmveth_adapter));
1395
1396         if (!netdev)
1397                 return -ENOMEM;
1398
1399         adapter = netdev_priv(netdev);
1400         dev_set_drvdata(&dev->dev, netdev);
1401
1402         adapter->vdev = dev;
1403         adapter->netdev = netdev;
1404         adapter->mcastFilterSize = *mcastFilterSize_p;
1405         adapter->pool_config = 0;
1406
1407         netif_napi_add(netdev, &adapter->napi, ibmveth_poll, 16);
1408
1409         /*
1410          * Some older boxes running PHYP non-natively have an OF that returns
1411          * a 8-byte local-mac-address field (and the first 2 bytes have to be
1412          * ignored) while newer boxes' OF return a 6-byte field. Note that
1413          * IEEE 1275 specifies that local-mac-address must be a 6-byte field.
1414          * The RPA doc specifies that the first byte must be 10b, so we'll
1415          * just look for it to solve this 8 vs. 6 byte field issue
1416          */
1417         if ((*mac_addr_p & 0x3) != 0x02)
1418                 mac_addr_p += 2;
1419
1420         adapter->mac_addr = 0;
1421         memcpy(&adapter->mac_addr, mac_addr_p, 6);
1422
1423         netdev->irq = dev->irq;
1424         netdev->netdev_ops = &ibmveth_netdev_ops;
1425         netdev->ethtool_ops = &netdev_ethtool_ops;
1426         SET_NETDEV_DEV(netdev, &dev->dev);
1427         netdev->features |= NETIF_F_SG;
1428
1429         memcpy(netdev->dev_addr, &adapter->mac_addr, netdev->addr_len);
1430
1431         for (i = 0; i < IBMVETH_NUM_BUFF_POOLS; i++) {
1432                 struct kobject *kobj = &adapter->rx_buff_pool[i].kobj;
1433                 int error;
1434
1435                 ibmveth_init_buffer_pool(&adapter->rx_buff_pool[i], i,
1436                                          pool_count[i], pool_size[i],
1437                                          pool_active[i]);
1438                 error = kobject_init_and_add(kobj, &ktype_veth_pool,
1439                                              &dev->dev.kobj, "pool%d", i);
1440                 if (!error)
1441                         kobject_uevent(kobj, KOBJ_ADD);
1442         }
1443
1444         netdev_dbg(netdev, "adapter @ 0x%p\n", adapter);
1445
1446         adapter->buffer_list_dma = DMA_ERROR_CODE;
1447         adapter->filter_list_dma = DMA_ERROR_CODE;
1448         adapter->rx_queue.queue_dma = DMA_ERROR_CODE;
1449
1450         netdev_dbg(netdev, "registering netdev...\n");
1451
1452         ibmveth_set_csum_offload(netdev, 1, ibmveth_set_tx_csum_flags);
1453
1454         rc = register_netdev(netdev);
1455
1456         if (rc) {
1457                 netdev_dbg(netdev, "failed to register netdev rc=%d\n", rc);
1458                 free_netdev(netdev);
1459                 return rc;
1460         }
1461
1462         netdev_dbg(netdev, "registered\n");
1463
1464         return 0;
1465 }
1466
1467 static int __devexit ibmveth_remove(struct vio_dev *dev)
1468 {
1469         struct net_device *netdev = dev_get_drvdata(&dev->dev);
1470         struct ibmveth_adapter *adapter = netdev_priv(netdev);
1471         int i;
1472
1473         for (i = 0; i < IBMVETH_NUM_BUFF_POOLS; i++)
1474                 kobject_put(&adapter->rx_buff_pool[i].kobj);
1475
1476         unregister_netdev(netdev);
1477
1478         free_netdev(netdev);
1479         dev_set_drvdata(&dev->dev, NULL);
1480
1481         return 0;
1482 }
1483
1484 static struct attribute veth_active_attr;
1485 static struct attribute veth_num_attr;
1486 static struct attribute veth_size_attr;
1487
1488 static ssize_t veth_pool_show(struct kobject *kobj,
1489                               struct attribute *attr, char *buf)
1490 {
1491         struct ibmveth_buff_pool *pool = container_of(kobj,
1492                                                       struct ibmveth_buff_pool,
1493                                                       kobj);
1494
1495         if (attr == &veth_active_attr)
1496                 return sprintf(buf, "%d\n", pool->active);
1497         else if (attr == &veth_num_attr)
1498                 return sprintf(buf, "%d\n", pool->size);
1499         else if (attr == &veth_size_attr)
1500                 return sprintf(buf, "%d\n", pool->buff_size);
1501         return 0;
1502 }
1503
1504 static ssize_t veth_pool_store(struct kobject *kobj, struct attribute *attr,
1505                                const char *buf, size_t count)
1506 {
1507         struct ibmveth_buff_pool *pool = container_of(kobj,
1508                                                       struct ibmveth_buff_pool,
1509                                                       kobj);
1510         struct net_device *netdev = dev_get_drvdata(
1511             container_of(kobj->parent, struct device, kobj));
1512         struct ibmveth_adapter *adapter = netdev_priv(netdev);
1513         long value = simple_strtol(buf, NULL, 10);
1514         long rc;
1515
1516         if (attr == &veth_active_attr) {
1517                 if (value && !pool->active) {
1518                         if (netif_running(netdev)) {
1519                                 if (ibmveth_alloc_buffer_pool(pool)) {
1520                                         netdev_err(netdev,
1521                                                    "unable to alloc pool\n");
1522                                         return -ENOMEM;
1523                                 }
1524                                 pool->active = 1;
1525                                 adapter->pool_config = 1;
1526                                 ibmveth_close(netdev);
1527                                 adapter->pool_config = 0;
1528                                 if ((rc = ibmveth_open(netdev)))
1529                                         return rc;
1530                         } else {
1531                                 pool->active = 1;
1532                         }
1533                 } else if (!value && pool->active) {
1534                         int mtu = netdev->mtu + IBMVETH_BUFF_OH;
1535                         int i;
1536                         /* Make sure there is a buffer pool with buffers that
1537                            can hold a packet of the size of the MTU */
1538                         for (i = 0; i < IBMVETH_NUM_BUFF_POOLS; i++) {
1539                                 if (pool == &adapter->rx_buff_pool[i])
1540                                         continue;
1541                                 if (!adapter->rx_buff_pool[i].active)
1542                                         continue;
1543                                 if (mtu <= adapter->rx_buff_pool[i].buff_size)
1544                                         break;
1545                         }
1546
1547                         if (i == IBMVETH_NUM_BUFF_POOLS) {
1548                                 netdev_err(netdev, "no active pool >= MTU\n");
1549                                 return -EPERM;
1550                         }
1551
1552                         if (netif_running(netdev)) {
1553                                 adapter->pool_config = 1;
1554                                 ibmveth_close(netdev);
1555                                 pool->active = 0;
1556                                 adapter->pool_config = 0;
1557                                 if ((rc = ibmveth_open(netdev)))
1558                                         return rc;
1559                         }
1560                         pool->active = 0;
1561                 }
1562         } else if (attr == &veth_num_attr) {
1563                 if (value <= 0 || value > IBMVETH_MAX_POOL_COUNT) {
1564                         return -EINVAL;
1565                 } else {
1566                         if (netif_running(netdev)) {
1567                                 adapter->pool_config = 1;
1568                                 ibmveth_close(netdev);
1569                                 adapter->pool_config = 0;
1570                                 pool->size = value;
1571                                 if ((rc = ibmveth_open(netdev)))
1572                                         return rc;
1573                         } else {
1574                                 pool->size = value;
1575                         }
1576                 }
1577         } else if (attr == &veth_size_attr) {
1578                 if (value <= IBMVETH_BUFF_OH || value > IBMVETH_MAX_BUF_SIZE) {
1579                         return -EINVAL;
1580                 } else {
1581                         if (netif_running(netdev)) {
1582                                 adapter->pool_config = 1;
1583                                 ibmveth_close(netdev);
1584                                 adapter->pool_config = 0;
1585                                 pool->buff_size = value;
1586                                 if ((rc = ibmveth_open(netdev)))
1587                                         return rc;
1588                         } else {
1589                                 pool->buff_size = value;
1590                         }
1591                 }
1592         }
1593
1594         /* kick the interrupt handler to allocate/deallocate pools */
1595         ibmveth_interrupt(netdev->irq, netdev);
1596         return count;
1597 }
1598
1599
1600 #define ATTR(_name, _mode)                              \
1601         struct attribute veth_##_name##_attr = {        \
1602         .name = __stringify(_name), .mode = _mode,      \
1603         };
1604
1605 static ATTR(active, 0644);
1606 static ATTR(num, 0644);
1607 static ATTR(size, 0644);
1608
1609 static struct attribute *veth_pool_attrs[] = {
1610         &veth_active_attr,
1611         &veth_num_attr,
1612         &veth_size_attr,
1613         NULL,
1614 };
1615
1616 static const struct sysfs_ops veth_pool_ops = {
1617         .show   = veth_pool_show,
1618         .store  = veth_pool_store,
1619 };
1620
1621 static struct kobj_type ktype_veth_pool = {
1622         .release        = NULL,
1623         .sysfs_ops      = &veth_pool_ops,
1624         .default_attrs  = veth_pool_attrs,
1625 };
1626
1627 static int ibmveth_resume(struct device *dev)
1628 {
1629         struct net_device *netdev = dev_get_drvdata(dev);
1630         ibmveth_interrupt(netdev->irq, netdev);
1631         return 0;
1632 }
1633
1634 static struct vio_device_id ibmveth_device_table[] __devinitdata = {
1635         { "network", "IBM,l-lan"},
1636         { "", "" }
1637 };
1638 MODULE_DEVICE_TABLE(vio, ibmveth_device_table);
1639
1640 static struct dev_pm_ops ibmveth_pm_ops = {
1641         .resume = ibmveth_resume
1642 };
1643
1644 static struct vio_driver ibmveth_driver = {
1645         .id_table       = ibmveth_device_table,
1646         .probe          = ibmveth_probe,
1647         .remove         = ibmveth_remove,
1648         .get_desired_dma = ibmveth_get_desired_dma,
1649         .driver         = {
1650                 .name   = ibmveth_driver_name,
1651                 .owner  = THIS_MODULE,
1652                 .pm = &ibmveth_pm_ops,
1653         }
1654 };
1655
1656 static int __init ibmveth_module_init(void)
1657 {
1658         printk(KERN_DEBUG "%s: %s %s\n", ibmveth_driver_name,
1659                ibmveth_driver_string, ibmveth_driver_version);
1660
1661         return vio_register_driver(&ibmveth_driver);
1662 }
1663
1664 static void __exit ibmveth_module_exit(void)
1665 {
1666         vio_unregister_driver(&ibmveth_driver);
1667 }
1668
1669 module_init(ibmveth_module_init);
1670 module_exit(ibmveth_module_exit);