atomic: use <linux/atomic.h>
[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/interrupt.h>
38 #include <linux/mm.h>
39 #include <linux/pm.h>
40 #include <linux/ethtool.h>
41 #include <linux/in.h>
42 #include <linux/ip.h>
43 #include <linux/ipv6.h>
44 #include <linux/slab.h>
45 #include <asm/hvcall.h>
46 #include <linux/atomic.h>
47 #include <asm/vio.h>
48 #include <asm/iommu.h>
49 #include <asm/firmware.h>
50
51 #include "ibmveth.h"
52
53 static irqreturn_t ibmveth_interrupt(int irq, void *dev_instance);
54 static void ibmveth_rxq_harvest_buffer(struct ibmveth_adapter *adapter);
55 static unsigned long ibmveth_get_desired_dma(struct vio_dev *vdev);
56
57 static struct kobj_type ktype_veth_pool;
58
59
60 static const char ibmveth_driver_name[] = "ibmveth";
61 static const char ibmveth_driver_string[] = "IBM Power Virtual Ethernet Driver";
62 #define ibmveth_driver_version "1.04"
63
64 MODULE_AUTHOR("Santiago Leon <santil@linux.vnet.ibm.com>");
65 MODULE_DESCRIPTION("IBM Power Virtual Ethernet Driver");
66 MODULE_LICENSE("GPL");
67 MODULE_VERSION(ibmveth_driver_version);
68
69 static unsigned int tx_copybreak __read_mostly = 128;
70 module_param(tx_copybreak, uint, 0644);
71 MODULE_PARM_DESC(tx_copybreak,
72         "Maximum size of packet that is copied to a new buffer on transmit");
73
74 static unsigned int rx_copybreak __read_mostly = 128;
75 module_param(rx_copybreak, uint, 0644);
76 MODULE_PARM_DESC(rx_copybreak,
77         "Maximum size of packet that is copied to a new buffer on receive");
78
79 static unsigned int rx_flush __read_mostly = 0;
80 module_param(rx_flush, uint, 0644);
81 MODULE_PARM_DESC(rx_flush, "Flush receive buffers before use");
82
83 struct ibmveth_stat {
84         char name[ETH_GSTRING_LEN];
85         int offset;
86 };
87
88 #define IBMVETH_STAT_OFF(stat) offsetof(struct ibmveth_adapter, stat)
89 #define IBMVETH_GET_STAT(a, off) *((u64 *)(((unsigned long)(a)) + off))
90
91 struct ibmveth_stat ibmveth_stats[] = {
92         { "replenish_task_cycles", IBMVETH_STAT_OFF(replenish_task_cycles) },
93         { "replenish_no_mem", IBMVETH_STAT_OFF(replenish_no_mem) },
94         { "replenish_add_buff_failure",
95                         IBMVETH_STAT_OFF(replenish_add_buff_failure) },
96         { "replenish_add_buff_success",
97                         IBMVETH_STAT_OFF(replenish_add_buff_success) },
98         { "rx_invalid_buffer", IBMVETH_STAT_OFF(rx_invalid_buffer) },
99         { "rx_no_buffer", IBMVETH_STAT_OFF(rx_no_buffer) },
100         { "tx_map_failed", IBMVETH_STAT_OFF(tx_map_failed) },
101         { "tx_send_failed", IBMVETH_STAT_OFF(tx_send_failed) },
102         { "fw_enabled_ipv4_csum", IBMVETH_STAT_OFF(fw_ipv4_csum_support) },
103         { "fw_enabled_ipv6_csum", IBMVETH_STAT_OFF(fw_ipv6_csum_support) },
104 };
105
106 /* simple methods of getting data from the current rxq entry */
107 static inline u32 ibmveth_rxq_flags(struct ibmveth_adapter *adapter)
108 {
109         return adapter->rx_queue.queue_addr[adapter->rx_queue.index].flags_off;
110 }
111
112 static inline int ibmveth_rxq_toggle(struct ibmveth_adapter *adapter)
113 {
114         return (ibmveth_rxq_flags(adapter) & IBMVETH_RXQ_TOGGLE) >>
115                         IBMVETH_RXQ_TOGGLE_SHIFT;
116 }
117
118 static inline int ibmveth_rxq_pending_buffer(struct ibmveth_adapter *adapter)
119 {
120         return ibmveth_rxq_toggle(adapter) == adapter->rx_queue.toggle;
121 }
122
123 static inline int ibmveth_rxq_buffer_valid(struct ibmveth_adapter *adapter)
124 {
125         return ibmveth_rxq_flags(adapter) & IBMVETH_RXQ_VALID;
126 }
127
128 static inline int ibmveth_rxq_frame_offset(struct ibmveth_adapter *adapter)
129 {
130         return ibmveth_rxq_flags(adapter) & IBMVETH_RXQ_OFF_MASK;
131 }
132
133 static inline int ibmveth_rxq_frame_length(struct ibmveth_adapter *adapter)
134 {
135         return adapter->rx_queue.queue_addr[adapter->rx_queue.index].length;
136 }
137
138 static inline int ibmveth_rxq_csum_good(struct ibmveth_adapter *adapter)
139 {
140         return ibmveth_rxq_flags(adapter) & IBMVETH_RXQ_CSUM_GOOD;
141 }
142
143 /* setup the initial settings for a buffer pool */
144 static void ibmveth_init_buffer_pool(struct ibmveth_buff_pool *pool,
145                                      u32 pool_index, u32 pool_size,
146                                      u32 buff_size, u32 pool_active)
147 {
148         pool->size = pool_size;
149         pool->index = pool_index;
150         pool->buff_size = buff_size;
151         pool->threshold = pool_size * 7 / 8;
152         pool->active = pool_active;
153 }
154
155 /* allocate and setup an buffer pool - called during open */
156 static int ibmveth_alloc_buffer_pool(struct ibmveth_buff_pool *pool)
157 {
158         int i;
159
160         pool->free_map = kmalloc(sizeof(u16) * pool->size, GFP_KERNEL);
161
162         if (!pool->free_map)
163                 return -1;
164
165         pool->dma_addr = kmalloc(sizeof(dma_addr_t) * pool->size, GFP_KERNEL);
166         if (!pool->dma_addr) {
167                 kfree(pool->free_map);
168                 pool->free_map = NULL;
169                 return -1;
170         }
171
172         pool->skbuff = kcalloc(pool->size, sizeof(void *), GFP_KERNEL);
173
174         if (!pool->skbuff) {
175                 kfree(pool->dma_addr);
176                 pool->dma_addr = NULL;
177
178                 kfree(pool->free_map);
179                 pool->free_map = NULL;
180                 return -1;
181         }
182
183         memset(pool->dma_addr, 0, sizeof(dma_addr_t) * pool->size);
184
185         for (i = 0; i < pool->size; ++i)
186                 pool->free_map[i] = i;
187
188         atomic_set(&pool->available, 0);
189         pool->producer_index = 0;
190         pool->consumer_index = 0;
191
192         return 0;
193 }
194
195 static inline void ibmveth_flush_buffer(void *addr, unsigned long length)
196 {
197         unsigned long offset;
198
199         for (offset = 0; offset < length; offset += SMP_CACHE_BYTES)
200                 asm("dcbfl %0,%1" :: "b" (addr), "r" (offset));
201 }
202
203 /* replenish the buffers for a pool.  note that we don't need to
204  * skb_reserve these since they are used for incoming...
205  */
206 static void ibmveth_replenish_buffer_pool(struct ibmveth_adapter *adapter,
207                                           struct ibmveth_buff_pool *pool)
208 {
209         u32 i;
210         u32 count = pool->size - atomic_read(&pool->available);
211         u32 buffers_added = 0;
212         struct sk_buff *skb;
213         unsigned int free_index, index;
214         u64 correlator;
215         unsigned long lpar_rc;
216         dma_addr_t dma_addr;
217
218         mb();
219
220         for (i = 0; i < count; ++i) {
221                 union ibmveth_buf_desc desc;
222
223                 skb = netdev_alloc_skb(adapter->netdev, pool->buff_size);
224
225                 if (!skb) {
226                         netdev_dbg(adapter->netdev,
227                                    "replenish: unable to allocate skb\n");
228                         adapter->replenish_no_mem++;
229                         break;
230                 }
231
232                 free_index = pool->consumer_index;
233                 pool->consumer_index++;
234                 if (pool->consumer_index >= pool->size)
235                         pool->consumer_index = 0;
236                 index = pool->free_map[free_index];
237
238                 BUG_ON(index == IBM_VETH_INVALID_MAP);
239                 BUG_ON(pool->skbuff[index] != NULL);
240
241                 dma_addr = dma_map_single(&adapter->vdev->dev, skb->data,
242                                 pool->buff_size, DMA_FROM_DEVICE);
243
244                 if (dma_mapping_error(&adapter->vdev->dev, dma_addr))
245                         goto failure;
246
247                 pool->free_map[free_index] = IBM_VETH_INVALID_MAP;
248                 pool->dma_addr[index] = dma_addr;
249                 pool->skbuff[index] = skb;
250
251                 correlator = ((u64)pool->index << 32) | index;
252                 *(u64 *)skb->data = correlator;
253
254                 desc.fields.flags_len = IBMVETH_BUF_VALID | pool->buff_size;
255                 desc.fields.address = dma_addr;
256
257                 if (rx_flush) {
258                         unsigned int len = min(pool->buff_size,
259                                                 adapter->netdev->mtu +
260                                                 IBMVETH_BUFF_OH);
261                         ibmveth_flush_buffer(skb->data, len);
262                 }
263                 lpar_rc = h_add_logical_lan_buffer(adapter->vdev->unit_address,
264                                                    desc.desc);
265
266                 if (lpar_rc != H_SUCCESS) {
267                         goto failure;
268                 } else {
269                         buffers_added++;
270                         adapter->replenish_add_buff_success++;
271                 }
272         }
273
274         mb();
275         atomic_add(buffers_added, &(pool->available));
276         return;
277
278 failure:
279         pool->free_map[free_index] = index;
280         pool->skbuff[index] = NULL;
281         if (pool->consumer_index == 0)
282                 pool->consumer_index = pool->size - 1;
283         else
284                 pool->consumer_index--;
285         if (!dma_mapping_error(&adapter->vdev->dev, dma_addr))
286                 dma_unmap_single(&adapter->vdev->dev,
287                                  pool->dma_addr[index], pool->buff_size,
288                                  DMA_FROM_DEVICE);
289         dev_kfree_skb_any(skb);
290         adapter->replenish_add_buff_failure++;
291
292         mb();
293         atomic_add(buffers_added, &(pool->available));
294 }
295
296 /* replenish routine */
297 static void ibmveth_replenish_task(struct ibmveth_adapter *adapter)
298 {
299         int i;
300
301         adapter->replenish_task_cycles++;
302
303         for (i = (IBMVETH_NUM_BUFF_POOLS - 1); i >= 0; i--) {
304                 struct ibmveth_buff_pool *pool = &adapter->rx_buff_pool[i];
305
306                 if (pool->active &&
307                     (atomic_read(&pool->available) < pool->threshold))
308                         ibmveth_replenish_buffer_pool(adapter, pool);
309         }
310
311         adapter->rx_no_buffer = *(u64 *)(((char*)adapter->buffer_list_addr) +
312                                                 4096 - 8);
313 }
314
315 /* empty and free ana buffer pool - also used to do cleanup in error paths */
316 static void ibmveth_free_buffer_pool(struct ibmveth_adapter *adapter,
317                                      struct ibmveth_buff_pool *pool)
318 {
319         int i;
320
321         kfree(pool->free_map);
322         pool->free_map = NULL;
323
324         if (pool->skbuff && pool->dma_addr) {
325                 for (i = 0; i < pool->size; ++i) {
326                         struct sk_buff *skb = pool->skbuff[i];
327                         if (skb) {
328                                 dma_unmap_single(&adapter->vdev->dev,
329                                                  pool->dma_addr[i],
330                                                  pool->buff_size,
331                                                  DMA_FROM_DEVICE);
332                                 dev_kfree_skb_any(skb);
333                                 pool->skbuff[i] = NULL;
334                         }
335                 }
336         }
337
338         if (pool->dma_addr) {
339                 kfree(pool->dma_addr);
340                 pool->dma_addr = NULL;
341         }
342
343         if (pool->skbuff) {
344                 kfree(pool->skbuff);
345                 pool->skbuff = NULL;
346         }
347 }
348
349 /* remove a buffer from a pool */
350 static void ibmveth_remove_buffer_from_pool(struct ibmveth_adapter *adapter,
351                                             u64 correlator)
352 {
353         unsigned int pool  = correlator >> 32;
354         unsigned int index = correlator & 0xffffffffUL;
355         unsigned int free_index;
356         struct sk_buff *skb;
357
358         BUG_ON(pool >= IBMVETH_NUM_BUFF_POOLS);
359         BUG_ON(index >= adapter->rx_buff_pool[pool].size);
360
361         skb = adapter->rx_buff_pool[pool].skbuff[index];
362
363         BUG_ON(skb == NULL);
364
365         adapter->rx_buff_pool[pool].skbuff[index] = NULL;
366
367         dma_unmap_single(&adapter->vdev->dev,
368                          adapter->rx_buff_pool[pool].dma_addr[index],
369                          adapter->rx_buff_pool[pool].buff_size,
370                          DMA_FROM_DEVICE);
371
372         free_index = adapter->rx_buff_pool[pool].producer_index;
373         adapter->rx_buff_pool[pool].producer_index++;
374         if (adapter->rx_buff_pool[pool].producer_index >=
375             adapter->rx_buff_pool[pool].size)
376                 adapter->rx_buff_pool[pool].producer_index = 0;
377         adapter->rx_buff_pool[pool].free_map[free_index] = index;
378
379         mb();
380
381         atomic_dec(&(adapter->rx_buff_pool[pool].available));
382 }
383
384 /* get the current buffer on the rx queue */
385 static inline struct sk_buff *ibmveth_rxq_get_buffer(struct ibmveth_adapter *adapter)
386 {
387         u64 correlator = adapter->rx_queue.queue_addr[adapter->rx_queue.index].correlator;
388         unsigned int pool = correlator >> 32;
389         unsigned int index = correlator & 0xffffffffUL;
390
391         BUG_ON(pool >= IBMVETH_NUM_BUFF_POOLS);
392         BUG_ON(index >= adapter->rx_buff_pool[pool].size);
393
394         return adapter->rx_buff_pool[pool].skbuff[index];
395 }
396
397 /* recycle the current buffer on the rx queue */
398 static void ibmveth_rxq_recycle_buffer(struct ibmveth_adapter *adapter)
399 {
400         u32 q_index = adapter->rx_queue.index;
401         u64 correlator = adapter->rx_queue.queue_addr[q_index].correlator;
402         unsigned int pool = correlator >> 32;
403         unsigned int index = correlator & 0xffffffffUL;
404         union ibmveth_buf_desc desc;
405         unsigned long lpar_rc;
406
407         BUG_ON(pool >= IBMVETH_NUM_BUFF_POOLS);
408         BUG_ON(index >= adapter->rx_buff_pool[pool].size);
409
410         if (!adapter->rx_buff_pool[pool].active) {
411                 ibmveth_rxq_harvest_buffer(adapter);
412                 ibmveth_free_buffer_pool(adapter, &adapter->rx_buff_pool[pool]);
413                 return;
414         }
415
416         desc.fields.flags_len = IBMVETH_BUF_VALID |
417                 adapter->rx_buff_pool[pool].buff_size;
418         desc.fields.address = adapter->rx_buff_pool[pool].dma_addr[index];
419
420         lpar_rc = h_add_logical_lan_buffer(adapter->vdev->unit_address, desc.desc);
421
422         if (lpar_rc != H_SUCCESS) {
423                 netdev_dbg(adapter->netdev, "h_add_logical_lan_buffer failed "
424                            "during recycle rc=%ld", lpar_rc);
425                 ibmveth_remove_buffer_from_pool(adapter, adapter->rx_queue.queue_addr[adapter->rx_queue.index].correlator);
426         }
427
428         if (++adapter->rx_queue.index == adapter->rx_queue.num_slots) {
429                 adapter->rx_queue.index = 0;
430                 adapter->rx_queue.toggle = !adapter->rx_queue.toggle;
431         }
432 }
433
434 static void ibmveth_rxq_harvest_buffer(struct ibmveth_adapter *adapter)
435 {
436         ibmveth_remove_buffer_from_pool(adapter, adapter->rx_queue.queue_addr[adapter->rx_queue.index].correlator);
437
438         if (++adapter->rx_queue.index == adapter->rx_queue.num_slots) {
439                 adapter->rx_queue.index = 0;
440                 adapter->rx_queue.toggle = !adapter->rx_queue.toggle;
441         }
442 }
443
444 static void ibmveth_cleanup(struct ibmveth_adapter *adapter)
445 {
446         int i;
447         struct device *dev = &adapter->vdev->dev;
448
449         if (adapter->buffer_list_addr != NULL) {
450                 if (!dma_mapping_error(dev, adapter->buffer_list_dma)) {
451                         dma_unmap_single(dev, adapter->buffer_list_dma, 4096,
452                                         DMA_BIDIRECTIONAL);
453                         adapter->buffer_list_dma = DMA_ERROR_CODE;
454                 }
455                 free_page((unsigned long)adapter->buffer_list_addr);
456                 adapter->buffer_list_addr = NULL;
457         }
458
459         if (adapter->filter_list_addr != NULL) {
460                 if (!dma_mapping_error(dev, adapter->filter_list_dma)) {
461                         dma_unmap_single(dev, adapter->filter_list_dma, 4096,
462                                         DMA_BIDIRECTIONAL);
463                         adapter->filter_list_dma = DMA_ERROR_CODE;
464                 }
465                 free_page((unsigned long)adapter->filter_list_addr);
466                 adapter->filter_list_addr = NULL;
467         }
468
469         if (adapter->rx_queue.queue_addr != NULL) {
470                 if (!dma_mapping_error(dev, adapter->rx_queue.queue_dma)) {
471                         dma_unmap_single(dev,
472                                         adapter->rx_queue.queue_dma,
473                                         adapter->rx_queue.queue_len,
474                                         DMA_BIDIRECTIONAL);
475                         adapter->rx_queue.queue_dma = DMA_ERROR_CODE;
476                 }
477                 kfree(adapter->rx_queue.queue_addr);
478                 adapter->rx_queue.queue_addr = NULL;
479         }
480
481         for (i = 0; i < IBMVETH_NUM_BUFF_POOLS; i++)
482                 if (adapter->rx_buff_pool[i].active)
483                         ibmveth_free_buffer_pool(adapter,
484                                                  &adapter->rx_buff_pool[i]);
485
486         if (adapter->bounce_buffer != NULL) {
487                 if (!dma_mapping_error(dev, adapter->bounce_buffer_dma)) {
488                         dma_unmap_single(&adapter->vdev->dev,
489                                         adapter->bounce_buffer_dma,
490                                         adapter->netdev->mtu + IBMVETH_BUFF_OH,
491                                         DMA_BIDIRECTIONAL);
492                         adapter->bounce_buffer_dma = DMA_ERROR_CODE;
493                 }
494                 kfree(adapter->bounce_buffer);
495                 adapter->bounce_buffer = NULL;
496         }
497 }
498
499 static int ibmveth_register_logical_lan(struct ibmveth_adapter *adapter,
500         union ibmveth_buf_desc rxq_desc, u64 mac_address)
501 {
502         int rc, try_again = 1;
503
504         /*
505          * After a kexec the adapter will still be open, so our attempt to
506          * open it will fail. So if we get a failure we free the adapter and
507          * try again, but only once.
508          */
509 retry:
510         rc = h_register_logical_lan(adapter->vdev->unit_address,
511                                     adapter->buffer_list_dma, rxq_desc.desc,
512                                     adapter->filter_list_dma, mac_address);
513
514         if (rc != H_SUCCESS && try_again) {
515                 do {
516                         rc = h_free_logical_lan(adapter->vdev->unit_address);
517                 } while (H_IS_LONG_BUSY(rc) || (rc == H_BUSY));
518
519                 try_again = 0;
520                 goto retry;
521         }
522
523         return rc;
524 }
525
526 static int ibmveth_open(struct net_device *netdev)
527 {
528         struct ibmveth_adapter *adapter = netdev_priv(netdev);
529         u64 mac_address = 0;
530         int rxq_entries = 1;
531         unsigned long lpar_rc;
532         int rc;
533         union ibmveth_buf_desc rxq_desc;
534         int i;
535         struct device *dev;
536
537         netdev_dbg(netdev, "open starting\n");
538
539         napi_enable(&adapter->napi);
540
541         for(i = 0; i < IBMVETH_NUM_BUFF_POOLS; i++)
542                 rxq_entries += adapter->rx_buff_pool[i].size;
543
544         adapter->buffer_list_addr = (void*) get_zeroed_page(GFP_KERNEL);
545         adapter->filter_list_addr = (void*) get_zeroed_page(GFP_KERNEL);
546
547         if (!adapter->buffer_list_addr || !adapter->filter_list_addr) {
548                 netdev_err(netdev, "unable to allocate filter or buffer list "
549                            "pages\n");
550                 rc = -ENOMEM;
551                 goto err_out;
552         }
553
554         adapter->rx_queue.queue_len = sizeof(struct ibmveth_rx_q_entry) *
555                                                 rxq_entries;
556         adapter->rx_queue.queue_addr = kmalloc(adapter->rx_queue.queue_len,
557                                                 GFP_KERNEL);
558
559         if (!adapter->rx_queue.queue_addr) {
560                 netdev_err(netdev, "unable to allocate rx queue pages\n");
561                 rc = -ENOMEM;
562                 goto err_out;
563         }
564
565         dev = &adapter->vdev->dev;
566
567         adapter->buffer_list_dma = dma_map_single(dev,
568                         adapter->buffer_list_addr, 4096, DMA_BIDIRECTIONAL);
569         adapter->filter_list_dma = dma_map_single(dev,
570                         adapter->filter_list_addr, 4096, DMA_BIDIRECTIONAL);
571         adapter->rx_queue.queue_dma = dma_map_single(dev,
572                         adapter->rx_queue.queue_addr,
573                         adapter->rx_queue.queue_len, DMA_BIDIRECTIONAL);
574
575         if ((dma_mapping_error(dev, adapter->buffer_list_dma)) ||
576             (dma_mapping_error(dev, adapter->filter_list_dma)) ||
577             (dma_mapping_error(dev, adapter->rx_queue.queue_dma))) {
578                 netdev_err(netdev, "unable to map filter or buffer list "
579                            "pages\n");
580                 rc = -ENOMEM;
581                 goto err_out;
582         }
583
584         adapter->rx_queue.index = 0;
585         adapter->rx_queue.num_slots = rxq_entries;
586         adapter->rx_queue.toggle = 1;
587
588         memcpy(&mac_address, netdev->dev_addr, netdev->addr_len);
589         mac_address = mac_address >> 16;
590
591         rxq_desc.fields.flags_len = IBMVETH_BUF_VALID |
592                                         adapter->rx_queue.queue_len;
593         rxq_desc.fields.address = adapter->rx_queue.queue_dma;
594
595         netdev_dbg(netdev, "buffer list @ 0x%p\n", adapter->buffer_list_addr);
596         netdev_dbg(netdev, "filter list @ 0x%p\n", adapter->filter_list_addr);
597         netdev_dbg(netdev, "receive q   @ 0x%p\n", adapter->rx_queue.queue_addr);
598
599         h_vio_signal(adapter->vdev->unit_address, VIO_IRQ_DISABLE);
600
601         lpar_rc = ibmveth_register_logical_lan(adapter, rxq_desc, mac_address);
602
603         if (lpar_rc != H_SUCCESS) {
604                 netdev_err(netdev, "h_register_logical_lan failed with %ld\n",
605                            lpar_rc);
606                 netdev_err(netdev, "buffer TCE:0x%llx filter TCE:0x%llx rxq "
607                            "desc:0x%llx MAC:0x%llx\n",
608                                      adapter->buffer_list_dma,
609                                      adapter->filter_list_dma,
610                                      rxq_desc.desc,
611                                      mac_address);
612                 rc = -ENONET;
613                 goto err_out;
614         }
615
616         for (i = 0; i < IBMVETH_NUM_BUFF_POOLS; i++) {
617                 if (!adapter->rx_buff_pool[i].active)
618                         continue;
619                 if (ibmveth_alloc_buffer_pool(&adapter->rx_buff_pool[i])) {
620                         netdev_err(netdev, "unable to alloc pool\n");
621                         adapter->rx_buff_pool[i].active = 0;
622                         rc = -ENOMEM;
623                         goto err_out;
624                 }
625         }
626
627         netdev_dbg(netdev, "registering irq 0x%x\n", netdev->irq);
628         rc = request_irq(netdev->irq, ibmveth_interrupt, 0, netdev->name,
629                          netdev);
630         if (rc != 0) {
631                 netdev_err(netdev, "unable to request irq 0x%x, rc %d\n",
632                            netdev->irq, rc);
633                 do {
634                         rc = h_free_logical_lan(adapter->vdev->unit_address);
635                 } while (H_IS_LONG_BUSY(rc) || (rc == H_BUSY));
636
637                 goto err_out;
638         }
639
640         adapter->bounce_buffer =
641             kmalloc(netdev->mtu + IBMVETH_BUFF_OH, GFP_KERNEL);
642         if (!adapter->bounce_buffer) {
643                 netdev_err(netdev, "unable to allocate bounce buffer\n");
644                 rc = -ENOMEM;
645                 goto err_out_free_irq;
646         }
647         adapter->bounce_buffer_dma =
648             dma_map_single(&adapter->vdev->dev, adapter->bounce_buffer,
649                            netdev->mtu + IBMVETH_BUFF_OH, DMA_BIDIRECTIONAL);
650         if (dma_mapping_error(dev, adapter->bounce_buffer_dma)) {
651                 netdev_err(netdev, "unable to map bounce buffer\n");
652                 rc = -ENOMEM;
653                 goto err_out_free_irq;
654         }
655
656         netdev_dbg(netdev, "initial replenish cycle\n");
657         ibmveth_interrupt(netdev->irq, netdev);
658
659         netif_start_queue(netdev);
660
661         netdev_dbg(netdev, "open complete\n");
662
663         return 0;
664
665 err_out_free_irq:
666         free_irq(netdev->irq, netdev);
667 err_out:
668         ibmveth_cleanup(adapter);
669         napi_disable(&adapter->napi);
670         return rc;
671 }
672
673 static int ibmveth_close(struct net_device *netdev)
674 {
675         struct ibmveth_adapter *adapter = netdev_priv(netdev);
676         long lpar_rc;
677
678         netdev_dbg(netdev, "close starting\n");
679
680         napi_disable(&adapter->napi);
681
682         if (!adapter->pool_config)
683                 netif_stop_queue(netdev);
684
685         h_vio_signal(adapter->vdev->unit_address, VIO_IRQ_DISABLE);
686
687         do {
688                 lpar_rc = h_free_logical_lan(adapter->vdev->unit_address);
689         } while (H_IS_LONG_BUSY(lpar_rc) || (lpar_rc == H_BUSY));
690
691         if (lpar_rc != H_SUCCESS) {
692                 netdev_err(netdev, "h_free_logical_lan failed with %lx, "
693                            "continuing with close\n", lpar_rc);
694         }
695
696         free_irq(netdev->irq, netdev);
697
698         adapter->rx_no_buffer = *(u64 *)(((char *)adapter->buffer_list_addr) +
699                                                 4096 - 8);
700
701         ibmveth_cleanup(adapter);
702
703         netdev_dbg(netdev, "close complete\n");
704
705         return 0;
706 }
707
708 static int netdev_get_settings(struct net_device *dev, struct ethtool_cmd *cmd)
709 {
710         cmd->supported = (SUPPORTED_1000baseT_Full | SUPPORTED_Autoneg |
711                                 SUPPORTED_FIBRE);
712         cmd->advertising = (ADVERTISED_1000baseT_Full | ADVERTISED_Autoneg |
713                                 ADVERTISED_FIBRE);
714         ethtool_cmd_speed_set(cmd, SPEED_1000);
715         cmd->duplex = DUPLEX_FULL;
716         cmd->port = PORT_FIBRE;
717         cmd->phy_address = 0;
718         cmd->transceiver = XCVR_INTERNAL;
719         cmd->autoneg = AUTONEG_ENABLE;
720         cmd->maxtxpkt = 0;
721         cmd->maxrxpkt = 1;
722         return 0;
723 }
724
725 static void netdev_get_drvinfo(struct net_device *dev,
726                                struct ethtool_drvinfo *info)
727 {
728         strncpy(info->driver, ibmveth_driver_name, sizeof(info->driver) - 1);
729         strncpy(info->version, ibmveth_driver_version,
730                 sizeof(info->version) - 1);
731 }
732
733 static u32 ibmveth_fix_features(struct net_device *dev, u32 features)
734 {
735         /*
736          * Since the ibmveth firmware interface does not have the
737          * concept of separate tx/rx checksum offload enable, if rx
738          * checksum is disabled we also have to disable tx checksum
739          * offload. Once we disable rx checksum offload, we are no
740          * longer allowed to send tx buffers that are not properly
741          * checksummed.
742          */
743
744         if (!(features & NETIF_F_RXCSUM))
745                 features &= ~NETIF_F_ALL_CSUM;
746
747         return features;
748 }
749
750 static int ibmveth_set_csum_offload(struct net_device *dev, u32 data)
751 {
752         struct ibmveth_adapter *adapter = netdev_priv(dev);
753         unsigned long set_attr, clr_attr, ret_attr;
754         unsigned long set_attr6, clr_attr6;
755         long ret, ret6;
756         int rc1 = 0, rc2 = 0;
757         int restart = 0;
758
759         if (netif_running(dev)) {
760                 restart = 1;
761                 adapter->pool_config = 1;
762                 ibmveth_close(dev);
763                 adapter->pool_config = 0;
764         }
765
766         set_attr = 0;
767         clr_attr = 0;
768
769         if (data) {
770                 set_attr = IBMVETH_ILLAN_IPV4_TCP_CSUM;
771                 set_attr6 = IBMVETH_ILLAN_IPV6_TCP_CSUM;
772         } else {
773                 clr_attr = IBMVETH_ILLAN_IPV4_TCP_CSUM;
774                 clr_attr6 = IBMVETH_ILLAN_IPV6_TCP_CSUM;
775         }
776
777         ret = h_illan_attributes(adapter->vdev->unit_address, 0, 0, &ret_attr);
778
779         if (ret == H_SUCCESS && !(ret_attr & IBMVETH_ILLAN_ACTIVE_TRUNK) &&
780             !(ret_attr & IBMVETH_ILLAN_TRUNK_PRI_MASK) &&
781             (ret_attr & IBMVETH_ILLAN_PADDED_PKT_CSUM)) {
782                 ret = h_illan_attributes(adapter->vdev->unit_address, clr_attr,
783                                          set_attr, &ret_attr);
784
785                 if (ret != H_SUCCESS) {
786                         netdev_err(dev, "unable to change IPv4 checksum "
787                                         "offload settings. %d rc=%ld\n",
788                                         data, ret);
789
790                         ret = h_illan_attributes(adapter->vdev->unit_address,
791                                                  set_attr, clr_attr, &ret_attr);
792                 } else {
793                         adapter->fw_ipv4_csum_support = data;
794                 }
795
796                 ret6 = h_illan_attributes(adapter->vdev->unit_address,
797                                          clr_attr6, set_attr6, &ret_attr);
798
799                 if (ret6 != H_SUCCESS) {
800                         netdev_err(dev, "unable to change IPv6 checksum "
801                                         "offload settings. %d rc=%ld\n",
802                                         data, ret);
803
804                         ret = h_illan_attributes(adapter->vdev->unit_address,
805                                                  set_attr6, clr_attr6,
806                                                  &ret_attr);
807                 } else
808                         adapter->fw_ipv6_csum_support = data;
809
810                 if (ret != H_SUCCESS || ret6 != H_SUCCESS)
811                         adapter->rx_csum = data;
812                 else
813                         rc1 = -EIO;
814         } else {
815                 rc1 = -EIO;
816                 netdev_err(dev, "unable to change checksum offload settings."
817                                      " %d rc=%ld ret_attr=%lx\n", data, ret,
818                                      ret_attr);
819         }
820
821         if (restart)
822                 rc2 = ibmveth_open(dev);
823
824         return rc1 ? rc1 : rc2;
825 }
826
827 static int ibmveth_set_features(struct net_device *dev, u32 features)
828 {
829         struct ibmveth_adapter *adapter = netdev_priv(dev);
830         int rx_csum = !!(features & NETIF_F_RXCSUM);
831         int rc;
832
833         if (rx_csum == adapter->rx_csum)
834                 return 0;
835
836         rc = ibmveth_set_csum_offload(dev, rx_csum);
837         if (rc && !adapter->rx_csum)
838                 dev->features = features & ~(NETIF_F_ALL_CSUM | NETIF_F_RXCSUM);
839
840         return rc;
841 }
842
843 static void ibmveth_get_strings(struct net_device *dev, u32 stringset, u8 *data)
844 {
845         int i;
846
847         if (stringset != ETH_SS_STATS)
848                 return;
849
850         for (i = 0; i < ARRAY_SIZE(ibmveth_stats); i++, data += ETH_GSTRING_LEN)
851                 memcpy(data, ibmveth_stats[i].name, ETH_GSTRING_LEN);
852 }
853
854 static int ibmveth_get_sset_count(struct net_device *dev, int sset)
855 {
856         switch (sset) {
857         case ETH_SS_STATS:
858                 return ARRAY_SIZE(ibmveth_stats);
859         default:
860                 return -EOPNOTSUPP;
861         }
862 }
863
864 static void ibmveth_get_ethtool_stats(struct net_device *dev,
865                                       struct ethtool_stats *stats, u64 *data)
866 {
867         int i;
868         struct ibmveth_adapter *adapter = netdev_priv(dev);
869
870         for (i = 0; i < ARRAY_SIZE(ibmveth_stats); i++)
871                 data[i] = IBMVETH_GET_STAT(adapter, ibmveth_stats[i].offset);
872 }
873
874 static const struct ethtool_ops netdev_ethtool_ops = {
875         .get_drvinfo            = netdev_get_drvinfo,
876         .get_settings           = netdev_get_settings,
877         .get_link               = ethtool_op_get_link,
878         .get_strings            = ibmveth_get_strings,
879         .get_sset_count         = ibmveth_get_sset_count,
880         .get_ethtool_stats      = ibmveth_get_ethtool_stats,
881 };
882
883 static int ibmveth_ioctl(struct net_device *dev, struct ifreq *ifr, int cmd)
884 {
885         return -EOPNOTSUPP;
886 }
887
888 #define page_offset(v) ((unsigned long)(v) & ((1 << 12) - 1))
889
890 static int ibmveth_send(struct ibmveth_adapter *adapter,
891                         union ibmveth_buf_desc *descs)
892 {
893         unsigned long correlator;
894         unsigned int retry_count;
895         unsigned long ret;
896
897         /*
898          * The retry count sets a maximum for the number of broadcast and
899          * multicast destinations within the system.
900          */
901         retry_count = 1024;
902         correlator = 0;
903         do {
904                 ret = h_send_logical_lan(adapter->vdev->unit_address,
905                                              descs[0].desc, descs[1].desc,
906                                              descs[2].desc, descs[3].desc,
907                                              descs[4].desc, descs[5].desc,
908                                              correlator, &correlator);
909         } while ((ret == H_BUSY) && (retry_count--));
910
911         if (ret != H_SUCCESS && ret != H_DROPPED) {
912                 netdev_err(adapter->netdev, "tx: h_send_logical_lan failed "
913                            "with rc=%ld\n", ret);
914                 return 1;
915         }
916
917         return 0;
918 }
919
920 static netdev_tx_t ibmveth_start_xmit(struct sk_buff *skb,
921                                       struct net_device *netdev)
922 {
923         struct ibmveth_adapter *adapter = netdev_priv(netdev);
924         unsigned int desc_flags;
925         union ibmveth_buf_desc descs[6];
926         int last, i;
927         int force_bounce = 0;
928
929         /*
930          * veth handles a maximum of 6 segments including the header, so
931          * we have to linearize the skb if there are more than this.
932          */
933         if (skb_shinfo(skb)->nr_frags > 5 && __skb_linearize(skb)) {
934                 netdev->stats.tx_dropped++;
935                 goto out;
936         }
937
938         /* veth can't checksum offload UDP */
939         if (skb->ip_summed == CHECKSUM_PARTIAL &&
940             ((skb->protocol == htons(ETH_P_IP) &&
941               ip_hdr(skb)->protocol != IPPROTO_TCP) ||
942              (skb->protocol == htons(ETH_P_IPV6) &&
943               ipv6_hdr(skb)->nexthdr != IPPROTO_TCP)) &&
944             skb_checksum_help(skb)) {
945
946                 netdev_err(netdev, "tx: failed to checksum packet\n");
947                 netdev->stats.tx_dropped++;
948                 goto out;
949         }
950
951         desc_flags = IBMVETH_BUF_VALID;
952
953         if (skb->ip_summed == CHECKSUM_PARTIAL) {
954                 unsigned char *buf = skb_transport_header(skb) +
955                                                 skb->csum_offset;
956
957                 desc_flags |= (IBMVETH_BUF_NO_CSUM | IBMVETH_BUF_CSUM_GOOD);
958
959                 /* Need to zero out the checksum */
960                 buf[0] = 0;
961                 buf[1] = 0;
962         }
963
964 retry_bounce:
965         memset(descs, 0, sizeof(descs));
966
967         /*
968          * If a linear packet is below the rx threshold then
969          * copy it into the static bounce buffer. This avoids the
970          * cost of a TCE insert and remove.
971          */
972         if (force_bounce || (!skb_is_nonlinear(skb) &&
973                                 (skb->len < tx_copybreak))) {
974                 skb_copy_from_linear_data(skb, adapter->bounce_buffer,
975                                           skb->len);
976
977                 descs[0].fields.flags_len = desc_flags | skb->len;
978                 descs[0].fields.address = adapter->bounce_buffer_dma;
979
980                 if (ibmveth_send(adapter, descs)) {
981                         adapter->tx_send_failed++;
982                         netdev->stats.tx_dropped++;
983                 } else {
984                         netdev->stats.tx_packets++;
985                         netdev->stats.tx_bytes += skb->len;
986                 }
987
988                 goto out;
989         }
990
991         /* Map the header */
992         descs[0].fields.address = dma_map_single(&adapter->vdev->dev, skb->data,
993                                                  skb_headlen(skb),
994                                                  DMA_TO_DEVICE);
995         if (dma_mapping_error(&adapter->vdev->dev, descs[0].fields.address))
996                 goto map_failed;
997
998         descs[0].fields.flags_len = desc_flags | skb_headlen(skb);
999
1000         /* Map the frags */
1001         for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
1002                 unsigned long dma_addr;
1003                 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
1004
1005                 dma_addr = dma_map_page(&adapter->vdev->dev, frag->page,
1006                                         frag->page_offset, frag->size,
1007                                         DMA_TO_DEVICE);
1008
1009                 if (dma_mapping_error(&adapter->vdev->dev, dma_addr))
1010                         goto map_failed_frags;
1011
1012                 descs[i+1].fields.flags_len = desc_flags | frag->size;
1013                 descs[i+1].fields.address = dma_addr;
1014         }
1015
1016         if (ibmveth_send(adapter, descs)) {
1017                 adapter->tx_send_failed++;
1018                 netdev->stats.tx_dropped++;
1019         } else {
1020                 netdev->stats.tx_packets++;
1021                 netdev->stats.tx_bytes += skb->len;
1022         }
1023
1024         for (i = 0; i < skb_shinfo(skb)->nr_frags + 1; i++)
1025                 dma_unmap_page(&adapter->vdev->dev, descs[i].fields.address,
1026                                descs[i].fields.flags_len & IBMVETH_BUF_LEN_MASK,
1027                                DMA_TO_DEVICE);
1028
1029 out:
1030         dev_kfree_skb(skb);
1031         return NETDEV_TX_OK;
1032
1033 map_failed_frags:
1034         last = i+1;
1035         for (i = 0; i < last; i++)
1036                 dma_unmap_page(&adapter->vdev->dev, descs[i].fields.address,
1037                                descs[i].fields.flags_len & IBMVETH_BUF_LEN_MASK,
1038                                DMA_TO_DEVICE);
1039
1040 map_failed:
1041         if (!firmware_has_feature(FW_FEATURE_CMO))
1042                 netdev_err(netdev, "tx: unable to map xmit buffer\n");
1043         adapter->tx_map_failed++;
1044         skb_linearize(skb);
1045         force_bounce = 1;
1046         goto retry_bounce;
1047 }
1048
1049 static int ibmveth_poll(struct napi_struct *napi, int budget)
1050 {
1051         struct ibmveth_adapter *adapter =
1052                         container_of(napi, struct ibmveth_adapter, napi);
1053         struct net_device *netdev = adapter->netdev;
1054         int frames_processed = 0;
1055         unsigned long lpar_rc;
1056
1057 restart_poll:
1058         do {
1059                 if (!ibmveth_rxq_pending_buffer(adapter))
1060                         break;
1061
1062                 smp_rmb();
1063                 if (!ibmveth_rxq_buffer_valid(adapter)) {
1064                         wmb(); /* suggested by larson1 */
1065                         adapter->rx_invalid_buffer++;
1066                         netdev_dbg(netdev, "recycling invalid buffer\n");
1067                         ibmveth_rxq_recycle_buffer(adapter);
1068                 } else {
1069                         struct sk_buff *skb, *new_skb;
1070                         int length = ibmveth_rxq_frame_length(adapter);
1071                         int offset = ibmveth_rxq_frame_offset(adapter);
1072                         int csum_good = ibmveth_rxq_csum_good(adapter);
1073
1074                         skb = ibmveth_rxq_get_buffer(adapter);
1075
1076                         new_skb = NULL;
1077                         if (length < rx_copybreak)
1078                                 new_skb = netdev_alloc_skb(netdev, length);
1079
1080                         if (new_skb) {
1081                                 skb_copy_to_linear_data(new_skb,
1082                                                         skb->data + offset,
1083                                                         length);
1084                                 if (rx_flush)
1085                                         ibmveth_flush_buffer(skb->data,
1086                                                 length + offset);
1087                                 skb = new_skb;
1088                                 ibmveth_rxq_recycle_buffer(adapter);
1089                         } else {
1090                                 ibmveth_rxq_harvest_buffer(adapter);
1091                                 skb_reserve(skb, offset);
1092                         }
1093
1094                         skb_put(skb, length);
1095                         skb->protocol = eth_type_trans(skb, netdev);
1096
1097                         if (csum_good)
1098                                 skb->ip_summed = CHECKSUM_UNNECESSARY;
1099
1100                         netif_receive_skb(skb); /* send it up */
1101
1102                         netdev->stats.rx_packets++;
1103                         netdev->stats.rx_bytes += length;
1104                         frames_processed++;
1105                 }
1106         } while (frames_processed < budget);
1107
1108         ibmveth_replenish_task(adapter);
1109
1110         if (frames_processed < budget) {
1111                 /* We think we are done - reenable interrupts,
1112                  * then check once more to make sure we are done.
1113                  */
1114                 lpar_rc = h_vio_signal(adapter->vdev->unit_address,
1115                                        VIO_IRQ_ENABLE);
1116
1117                 BUG_ON(lpar_rc != H_SUCCESS);
1118
1119                 napi_complete(napi);
1120
1121                 if (ibmveth_rxq_pending_buffer(adapter) &&
1122                     napi_reschedule(napi)) {
1123                         lpar_rc = h_vio_signal(adapter->vdev->unit_address,
1124                                                VIO_IRQ_DISABLE);
1125                         goto restart_poll;
1126                 }
1127         }
1128
1129         return frames_processed;
1130 }
1131
1132 static irqreturn_t ibmveth_interrupt(int irq, void *dev_instance)
1133 {
1134         struct net_device *netdev = dev_instance;
1135         struct ibmveth_adapter *adapter = netdev_priv(netdev);
1136         unsigned long lpar_rc;
1137
1138         if (napi_schedule_prep(&adapter->napi)) {
1139                 lpar_rc = h_vio_signal(adapter->vdev->unit_address,
1140                                        VIO_IRQ_DISABLE);
1141                 BUG_ON(lpar_rc != H_SUCCESS);
1142                 __napi_schedule(&adapter->napi);
1143         }
1144         return IRQ_HANDLED;
1145 }
1146
1147 static void ibmveth_set_multicast_list(struct net_device *netdev)
1148 {
1149         struct ibmveth_adapter *adapter = netdev_priv(netdev);
1150         unsigned long lpar_rc;
1151
1152         if ((netdev->flags & IFF_PROMISC) ||
1153             (netdev_mc_count(netdev) > adapter->mcastFilterSize)) {
1154                 lpar_rc = h_multicast_ctrl(adapter->vdev->unit_address,
1155                                            IbmVethMcastEnableRecv |
1156                                            IbmVethMcastDisableFiltering,
1157                                            0);
1158                 if (lpar_rc != H_SUCCESS) {
1159                         netdev_err(netdev, "h_multicast_ctrl rc=%ld when "
1160                                    "entering promisc mode\n", lpar_rc);
1161                 }
1162         } else {
1163                 struct netdev_hw_addr *ha;
1164                 /* clear the filter table & disable filtering */
1165                 lpar_rc = h_multicast_ctrl(adapter->vdev->unit_address,
1166                                            IbmVethMcastEnableRecv |
1167                                            IbmVethMcastDisableFiltering |
1168                                            IbmVethMcastClearFilterTable,
1169                                            0);
1170                 if (lpar_rc != H_SUCCESS) {
1171                         netdev_err(netdev, "h_multicast_ctrl rc=%ld when "
1172                                    "attempting to clear filter table\n",
1173                                    lpar_rc);
1174                 }
1175                 /* add the addresses to the filter table */
1176                 netdev_for_each_mc_addr(ha, netdev) {
1177                         /* add the multicast address to the filter table */
1178                         unsigned long mcast_addr = 0;
1179                         memcpy(((char *)&mcast_addr)+2, ha->addr, 6);
1180                         lpar_rc = h_multicast_ctrl(adapter->vdev->unit_address,
1181                                                    IbmVethMcastAddFilter,
1182                                                    mcast_addr);
1183                         if (lpar_rc != H_SUCCESS) {
1184                                 netdev_err(netdev, "h_multicast_ctrl rc=%ld "
1185                                            "when adding an entry to the filter "
1186                                            "table\n", lpar_rc);
1187                         }
1188                 }
1189
1190                 /* re-enable filtering */
1191                 lpar_rc = h_multicast_ctrl(adapter->vdev->unit_address,
1192                                            IbmVethMcastEnableFiltering,
1193                                            0);
1194                 if (lpar_rc != H_SUCCESS) {
1195                         netdev_err(netdev, "h_multicast_ctrl rc=%ld when "
1196                                    "enabling filtering\n", lpar_rc);
1197                 }
1198         }
1199 }
1200
1201 static int ibmveth_change_mtu(struct net_device *dev, int new_mtu)
1202 {
1203         struct ibmveth_adapter *adapter = netdev_priv(dev);
1204         struct vio_dev *viodev = adapter->vdev;
1205         int new_mtu_oh = new_mtu + IBMVETH_BUFF_OH;
1206         int i, rc;
1207         int need_restart = 0;
1208
1209         if (new_mtu < IBMVETH_MIN_MTU)
1210                 return -EINVAL;
1211
1212         for (i = 0; i < IBMVETH_NUM_BUFF_POOLS; i++)
1213                 if (new_mtu_oh < adapter->rx_buff_pool[i].buff_size)
1214                         break;
1215
1216         if (i == IBMVETH_NUM_BUFF_POOLS)
1217                 return -EINVAL;
1218
1219         /* Deactivate all the buffer pools so that the next loop can activate
1220            only the buffer pools necessary to hold the new MTU */
1221         if (netif_running(adapter->netdev)) {
1222                 need_restart = 1;
1223                 adapter->pool_config = 1;
1224                 ibmveth_close(adapter->netdev);
1225                 adapter->pool_config = 0;
1226         }
1227
1228         /* Look for an active buffer pool that can hold the new MTU */
1229         for (i = 0; i < IBMVETH_NUM_BUFF_POOLS; i++) {
1230                 adapter->rx_buff_pool[i].active = 1;
1231
1232                 if (new_mtu_oh < adapter->rx_buff_pool[i].buff_size) {
1233                         dev->mtu = new_mtu;
1234                         vio_cmo_set_dev_desired(viodev,
1235                                                 ibmveth_get_desired_dma
1236                                                 (viodev));
1237                         if (need_restart) {
1238                                 return ibmveth_open(adapter->netdev);
1239                         }
1240                         return 0;
1241                 }
1242         }
1243
1244         if (need_restart && (rc = ibmveth_open(adapter->netdev)))
1245                 return rc;
1246
1247         return -EINVAL;
1248 }
1249
1250 #ifdef CONFIG_NET_POLL_CONTROLLER
1251 static void ibmveth_poll_controller(struct net_device *dev)
1252 {
1253         ibmveth_replenish_task(netdev_priv(dev));
1254         ibmveth_interrupt(dev->irq, dev);
1255 }
1256 #endif
1257
1258 /**
1259  * ibmveth_get_desired_dma - Calculate IO memory desired by the driver
1260  *
1261  * @vdev: struct vio_dev for the device whose desired IO mem is to be returned
1262  *
1263  * Return value:
1264  *      Number of bytes of IO data the driver will need to perform well.
1265  */
1266 static unsigned long ibmveth_get_desired_dma(struct vio_dev *vdev)
1267 {
1268         struct net_device *netdev = dev_get_drvdata(&vdev->dev);
1269         struct ibmveth_adapter *adapter;
1270         unsigned long ret;
1271         int i;
1272         int rxqentries = 1;
1273
1274         /* netdev inits at probe time along with the structures we need below*/
1275         if (netdev == NULL)
1276                 return IOMMU_PAGE_ALIGN(IBMVETH_IO_ENTITLEMENT_DEFAULT);
1277
1278         adapter = netdev_priv(netdev);
1279
1280         ret = IBMVETH_BUFF_LIST_SIZE + IBMVETH_FILT_LIST_SIZE;
1281         ret += IOMMU_PAGE_ALIGN(netdev->mtu);
1282
1283         for (i = 0; i < IBMVETH_NUM_BUFF_POOLS; i++) {
1284                 /* add the size of the active receive buffers */
1285                 if (adapter->rx_buff_pool[i].active)
1286                         ret +=
1287                             adapter->rx_buff_pool[i].size *
1288                             IOMMU_PAGE_ALIGN(adapter->rx_buff_pool[i].
1289                                     buff_size);
1290                 rxqentries += adapter->rx_buff_pool[i].size;
1291         }
1292         /* add the size of the receive queue entries */
1293         ret += IOMMU_PAGE_ALIGN(rxqentries * sizeof(struct ibmveth_rx_q_entry));
1294
1295         return ret;
1296 }
1297
1298 static const struct net_device_ops ibmveth_netdev_ops = {
1299         .ndo_open               = ibmveth_open,
1300         .ndo_stop               = ibmveth_close,
1301         .ndo_start_xmit         = ibmveth_start_xmit,
1302         .ndo_set_multicast_list = ibmveth_set_multicast_list,
1303         .ndo_do_ioctl           = ibmveth_ioctl,
1304         .ndo_change_mtu         = ibmveth_change_mtu,
1305         .ndo_fix_features       = ibmveth_fix_features,
1306         .ndo_set_features       = ibmveth_set_features,
1307         .ndo_validate_addr      = eth_validate_addr,
1308         .ndo_set_mac_address    = eth_mac_addr,
1309 #ifdef CONFIG_NET_POLL_CONTROLLER
1310         .ndo_poll_controller    = ibmveth_poll_controller,
1311 #endif
1312 };
1313
1314 static int __devinit ibmveth_probe(struct vio_dev *dev,
1315                                    const struct vio_device_id *id)
1316 {
1317         int rc, i;
1318         struct net_device *netdev;
1319         struct ibmveth_adapter *adapter;
1320         unsigned char *mac_addr_p;
1321         unsigned int *mcastFilterSize_p;
1322
1323         dev_dbg(&dev->dev, "entering ibmveth_probe for UA 0x%x\n",
1324                 dev->unit_address);
1325
1326         mac_addr_p = (unsigned char *)vio_get_attribute(dev, VETH_MAC_ADDR,
1327                                                         NULL);
1328         if (!mac_addr_p) {
1329                 dev_err(&dev->dev, "Can't find VETH_MAC_ADDR attribute\n");
1330                 return -EINVAL;
1331         }
1332
1333         mcastFilterSize_p = (unsigned int *)vio_get_attribute(dev,
1334                                                 VETH_MCAST_FILTER_SIZE, NULL);
1335         if (!mcastFilterSize_p) {
1336                 dev_err(&dev->dev, "Can't find VETH_MCAST_FILTER_SIZE "
1337                         "attribute\n");
1338                 return -EINVAL;
1339         }
1340
1341         netdev = alloc_etherdev(sizeof(struct ibmveth_adapter));
1342
1343         if (!netdev)
1344                 return -ENOMEM;
1345
1346         adapter = netdev_priv(netdev);
1347         dev_set_drvdata(&dev->dev, netdev);
1348
1349         adapter->vdev = dev;
1350         adapter->netdev = netdev;
1351         adapter->mcastFilterSize = *mcastFilterSize_p;
1352         adapter->pool_config = 0;
1353
1354         netif_napi_add(netdev, &adapter->napi, ibmveth_poll, 16);
1355
1356         /*
1357          * Some older boxes running PHYP non-natively have an OF that returns
1358          * a 8-byte local-mac-address field (and the first 2 bytes have to be
1359          * ignored) while newer boxes' OF return a 6-byte field. Note that
1360          * IEEE 1275 specifies that local-mac-address must be a 6-byte field.
1361          * The RPA doc specifies that the first byte must be 10b, so we'll
1362          * just look for it to solve this 8 vs. 6 byte field issue
1363          */
1364         if ((*mac_addr_p & 0x3) != 0x02)
1365                 mac_addr_p += 2;
1366
1367         adapter->mac_addr = 0;
1368         memcpy(&adapter->mac_addr, mac_addr_p, 6);
1369
1370         netdev->irq = dev->irq;
1371         netdev->netdev_ops = &ibmveth_netdev_ops;
1372         netdev->ethtool_ops = &netdev_ethtool_ops;
1373         SET_NETDEV_DEV(netdev, &dev->dev);
1374         netdev->hw_features = NETIF_F_SG | NETIF_F_RXCSUM |
1375                 NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM;
1376         netdev->features |= netdev->hw_features;
1377
1378         memcpy(netdev->dev_addr, &adapter->mac_addr, netdev->addr_len);
1379
1380         for (i = 0; i < IBMVETH_NUM_BUFF_POOLS; i++) {
1381                 struct kobject *kobj = &adapter->rx_buff_pool[i].kobj;
1382                 int error;
1383
1384                 ibmveth_init_buffer_pool(&adapter->rx_buff_pool[i], i,
1385                                          pool_count[i], pool_size[i],
1386                                          pool_active[i]);
1387                 error = kobject_init_and_add(kobj, &ktype_veth_pool,
1388                                              &dev->dev.kobj, "pool%d", i);
1389                 if (!error)
1390                         kobject_uevent(kobj, KOBJ_ADD);
1391         }
1392
1393         netdev_dbg(netdev, "adapter @ 0x%p\n", adapter);
1394
1395         adapter->buffer_list_dma = DMA_ERROR_CODE;
1396         adapter->filter_list_dma = DMA_ERROR_CODE;
1397         adapter->rx_queue.queue_dma = DMA_ERROR_CODE;
1398
1399         netdev_dbg(netdev, "registering netdev...\n");
1400
1401         ibmveth_set_features(netdev, netdev->features);
1402
1403         rc = register_netdev(netdev);
1404
1405         if (rc) {
1406                 netdev_dbg(netdev, "failed to register netdev rc=%d\n", rc);
1407                 free_netdev(netdev);
1408                 return rc;
1409         }
1410
1411         netdev_dbg(netdev, "registered\n");
1412
1413         return 0;
1414 }
1415
1416 static int __devexit ibmveth_remove(struct vio_dev *dev)
1417 {
1418         struct net_device *netdev = dev_get_drvdata(&dev->dev);
1419         struct ibmveth_adapter *adapter = netdev_priv(netdev);
1420         int i;
1421
1422         for (i = 0; i < IBMVETH_NUM_BUFF_POOLS; i++)
1423                 kobject_put(&adapter->rx_buff_pool[i].kobj);
1424
1425         unregister_netdev(netdev);
1426
1427         free_netdev(netdev);
1428         dev_set_drvdata(&dev->dev, NULL);
1429
1430         return 0;
1431 }
1432
1433 static struct attribute veth_active_attr;
1434 static struct attribute veth_num_attr;
1435 static struct attribute veth_size_attr;
1436
1437 static ssize_t veth_pool_show(struct kobject *kobj,
1438                               struct attribute *attr, char *buf)
1439 {
1440         struct ibmveth_buff_pool *pool = container_of(kobj,
1441                                                       struct ibmveth_buff_pool,
1442                                                       kobj);
1443
1444         if (attr == &veth_active_attr)
1445                 return sprintf(buf, "%d\n", pool->active);
1446         else if (attr == &veth_num_attr)
1447                 return sprintf(buf, "%d\n", pool->size);
1448         else if (attr == &veth_size_attr)
1449                 return sprintf(buf, "%d\n", pool->buff_size);
1450         return 0;
1451 }
1452
1453 static ssize_t veth_pool_store(struct kobject *kobj, struct attribute *attr,
1454                                const char *buf, size_t count)
1455 {
1456         struct ibmveth_buff_pool *pool = container_of(kobj,
1457                                                       struct ibmveth_buff_pool,
1458                                                       kobj);
1459         struct net_device *netdev = dev_get_drvdata(
1460             container_of(kobj->parent, struct device, kobj));
1461         struct ibmveth_adapter *adapter = netdev_priv(netdev);
1462         long value = simple_strtol(buf, NULL, 10);
1463         long rc;
1464
1465         if (attr == &veth_active_attr) {
1466                 if (value && !pool->active) {
1467                         if (netif_running(netdev)) {
1468                                 if (ibmveth_alloc_buffer_pool(pool)) {
1469                                         netdev_err(netdev,
1470                                                    "unable to alloc pool\n");
1471                                         return -ENOMEM;
1472                                 }
1473                                 pool->active = 1;
1474                                 adapter->pool_config = 1;
1475                                 ibmveth_close(netdev);
1476                                 adapter->pool_config = 0;
1477                                 if ((rc = ibmveth_open(netdev)))
1478                                         return rc;
1479                         } else {
1480                                 pool->active = 1;
1481                         }
1482                 } else if (!value && pool->active) {
1483                         int mtu = netdev->mtu + IBMVETH_BUFF_OH;
1484                         int i;
1485                         /* Make sure there is a buffer pool with buffers that
1486                            can hold a packet of the size of the MTU */
1487                         for (i = 0; i < IBMVETH_NUM_BUFF_POOLS; i++) {
1488                                 if (pool == &adapter->rx_buff_pool[i])
1489                                         continue;
1490                                 if (!adapter->rx_buff_pool[i].active)
1491                                         continue;
1492                                 if (mtu <= adapter->rx_buff_pool[i].buff_size)
1493                                         break;
1494                         }
1495
1496                         if (i == IBMVETH_NUM_BUFF_POOLS) {
1497                                 netdev_err(netdev, "no active pool >= MTU\n");
1498                                 return -EPERM;
1499                         }
1500
1501                         if (netif_running(netdev)) {
1502                                 adapter->pool_config = 1;
1503                                 ibmveth_close(netdev);
1504                                 pool->active = 0;
1505                                 adapter->pool_config = 0;
1506                                 if ((rc = ibmveth_open(netdev)))
1507                                         return rc;
1508                         }
1509                         pool->active = 0;
1510                 }
1511         } else if (attr == &veth_num_attr) {
1512                 if (value <= 0 || value > IBMVETH_MAX_POOL_COUNT) {
1513                         return -EINVAL;
1514                 } else {
1515                         if (netif_running(netdev)) {
1516                                 adapter->pool_config = 1;
1517                                 ibmveth_close(netdev);
1518                                 adapter->pool_config = 0;
1519                                 pool->size = value;
1520                                 if ((rc = ibmveth_open(netdev)))
1521                                         return rc;
1522                         } else {
1523                                 pool->size = value;
1524                         }
1525                 }
1526         } else if (attr == &veth_size_attr) {
1527                 if (value <= IBMVETH_BUFF_OH || value > IBMVETH_MAX_BUF_SIZE) {
1528                         return -EINVAL;
1529                 } else {
1530                         if (netif_running(netdev)) {
1531                                 adapter->pool_config = 1;
1532                                 ibmveth_close(netdev);
1533                                 adapter->pool_config = 0;
1534                                 pool->buff_size = value;
1535                                 if ((rc = ibmveth_open(netdev)))
1536                                         return rc;
1537                         } else {
1538                                 pool->buff_size = value;
1539                         }
1540                 }
1541         }
1542
1543         /* kick the interrupt handler to allocate/deallocate pools */
1544         ibmveth_interrupt(netdev->irq, netdev);
1545         return count;
1546 }
1547
1548
1549 #define ATTR(_name, _mode)                              \
1550         struct attribute veth_##_name##_attr = {        \
1551         .name = __stringify(_name), .mode = _mode,      \
1552         };
1553
1554 static ATTR(active, 0644);
1555 static ATTR(num, 0644);
1556 static ATTR(size, 0644);
1557
1558 static struct attribute *veth_pool_attrs[] = {
1559         &veth_active_attr,
1560         &veth_num_attr,
1561         &veth_size_attr,
1562         NULL,
1563 };
1564
1565 static const struct sysfs_ops veth_pool_ops = {
1566         .show   = veth_pool_show,
1567         .store  = veth_pool_store,
1568 };
1569
1570 static struct kobj_type ktype_veth_pool = {
1571         .release        = NULL,
1572         .sysfs_ops      = &veth_pool_ops,
1573         .default_attrs  = veth_pool_attrs,
1574 };
1575
1576 static int ibmveth_resume(struct device *dev)
1577 {
1578         struct net_device *netdev = dev_get_drvdata(dev);
1579         ibmveth_interrupt(netdev->irq, netdev);
1580         return 0;
1581 }
1582
1583 static struct vio_device_id ibmveth_device_table[] __devinitdata = {
1584         { "network", "IBM,l-lan"},
1585         { "", "" }
1586 };
1587 MODULE_DEVICE_TABLE(vio, ibmveth_device_table);
1588
1589 static struct dev_pm_ops ibmveth_pm_ops = {
1590         .resume = ibmveth_resume
1591 };
1592
1593 static struct vio_driver ibmveth_driver = {
1594         .id_table       = ibmveth_device_table,
1595         .probe          = ibmveth_probe,
1596         .remove         = ibmveth_remove,
1597         .get_desired_dma = ibmveth_get_desired_dma,
1598         .driver         = {
1599                 .name   = ibmveth_driver_name,
1600                 .owner  = THIS_MODULE,
1601                 .pm = &ibmveth_pm_ops,
1602         }
1603 };
1604
1605 static int __init ibmveth_module_init(void)
1606 {
1607         printk(KERN_DEBUG "%s: %s %s\n", ibmveth_driver_name,
1608                ibmveth_driver_string, ibmveth_driver_version);
1609
1610         return vio_register_driver(&ibmveth_driver);
1611 }
1612
1613 static void __exit ibmveth_module_exit(void)
1614 {
1615         vio_unregister_driver(&ibmveth_driver);
1616 }
1617
1618 module_init(ibmveth_module_init);
1619 module_exit(ibmveth_module_exit);