IB/mthca: Recover from catastrophic errors
[linux-2.6] / drivers / net / ibmveth.c
1 /**************************************************************************/
2 /*                                                                        */
3 /* IBM eServer i/pSeries Virtual Ethernet Device Driver                   */
4 /* Copyright (C) 2003 IBM Corp.                                           */
5 /*  Originally written by Dave Larson (larson1@us.ibm.com)                */
6 /*  Maintained by Santiago Leon (santil@us.ibm.com)                       */
7 /*                                                                        */
8 /*  This program is free software; you can redistribute it and/or modify  */
9 /*  it under the terms of the GNU General Public License as published by  */
10 /*  the Free Software Foundation; either version 2 of the License, or     */
11 /*  (at your option) any later version.                                   */
12 /*                                                                        */
13 /*  This program is distributed in the hope that it will be useful,       */
14 /*  but WITHOUT ANY WARRANTY; without even the implied warranty of        */
15 /*  MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the         */
16 /*  GNU General Public License for more details.                          */
17 /*                                                                        */
18 /*  You should have received a copy of the GNU General Public License     */
19 /*  along with this program; if not, write to the Free Software           */
20 /*  Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA  02111-1307  */
21 /*                                                                   USA  */
22 /*                                                                        */
23 /* This module contains the implementation of a virtual ethernet device   */
24 /* for use with IBM i/pSeries LPAR Linux.  It utilizes the logical LAN    */
25 /* option of the RS/6000 Platform Architechture to interface with virtual */
26 /* ethernet NICs that are presented to the partition by the hypervisor.   */
27 /*                                                                        */
28 /**************************************************************************/
29 /*
30   TODO:
31   - remove frag processing code - no longer needed
32   - add support for sysfs
33   - possibly remove procfs support
34 */
35
36 #include <linux/module.h>
37 #include <linux/types.h>
38 #include <linux/errno.h>
39 #include <linux/ioport.h>
40 #include <linux/dma-mapping.h>
41 #include <linux/kernel.h>
42 #include <linux/netdevice.h>
43 #include <linux/etherdevice.h>
44 #include <linux/skbuff.h>
45 #include <linux/init.h>
46 #include <linux/delay.h>
47 #include <linux/mm.h>
48 #include <linux/ethtool.h>
49 #include <linux/proc_fs.h>
50 #include <asm/semaphore.h>
51 #include <asm/hvcall.h>
52 #include <asm/atomic.h>
53 #include <asm/iommu.h>
54 #include <asm/vio.h>
55 #include <asm/uaccess.h>
56 #include <linux/seq_file.h>
57
58 #include "ibmveth.h"
59
60 #undef DEBUG
61
62 #define ibmveth_printk(fmt, args...) \
63   printk(KERN_DEBUG "%s: " fmt, __FILE__, ## args)
64
65 #define ibmveth_error_printk(fmt, args...) \
66   printk(KERN_ERR "(%s:%3.3d ua:%x) ERROR: " fmt, __FILE__, __LINE__ , adapter->vdev->unit_address, ## args)
67
68 #ifdef DEBUG
69 #define ibmveth_debug_printk_no_adapter(fmt, args...) \
70   printk(KERN_DEBUG "(%s:%3.3d): " fmt, __FILE__, __LINE__ , ## args)
71 #define ibmveth_debug_printk(fmt, args...) \
72   printk(KERN_DEBUG "(%s:%3.3d ua:%x): " fmt, __FILE__, __LINE__ , adapter->vdev->unit_address, ## args)
73 #define ibmveth_assert(expr) \
74   if(!(expr)) {                                   \
75     printk(KERN_DEBUG "assertion failed (%s:%3.3d ua:%x): %s\n", __FILE__, __LINE__, adapter->vdev->unit_address, #expr); \
76     BUG(); \
77   }
78 #else
79 #define ibmveth_debug_printk_no_adapter(fmt, args...)
80 #define ibmveth_debug_printk(fmt, args...)
81 #define ibmveth_assert(expr)
82 #endif
83
84 static int ibmveth_open(struct net_device *dev);
85 static int ibmveth_close(struct net_device *dev);
86 static int ibmveth_ioctl(struct net_device *dev, struct ifreq *ifr, int cmd);
87 static int ibmveth_poll(struct net_device *dev, int *budget);
88 static int ibmveth_start_xmit(struct sk_buff *skb, struct net_device *dev);
89 static struct net_device_stats *ibmveth_get_stats(struct net_device *dev);
90 static void ibmveth_set_multicast_list(struct net_device *dev);
91 static int ibmveth_change_mtu(struct net_device *dev, int new_mtu);
92 static void ibmveth_proc_register_driver(void);
93 static void ibmveth_proc_unregister_driver(void);
94 static void ibmveth_proc_register_adapter(struct ibmveth_adapter *adapter);
95 static void ibmveth_proc_unregister_adapter(struct ibmveth_adapter *adapter);
96 static irqreturn_t ibmveth_interrupt(int irq, void *dev_instance, struct pt_regs *regs);
97 static inline void ibmveth_rxq_harvest_buffer(struct ibmveth_adapter *adapter);
98 static struct kobj_type ktype_veth_pool;
99
100 #ifdef CONFIG_PROC_FS
101 #define IBMVETH_PROC_DIR "net/ibmveth"
102 static struct proc_dir_entry *ibmveth_proc_dir;
103 #endif
104
105 static const char ibmveth_driver_name[] = "ibmveth";
106 static const char ibmveth_driver_string[] = "IBM i/pSeries Virtual Ethernet Driver";
107 #define ibmveth_driver_version "1.03"
108
109 MODULE_AUTHOR("Santiago Leon <santil@us.ibm.com>");
110 MODULE_DESCRIPTION("IBM i/pSeries Virtual Ethernet Driver");
111 MODULE_LICENSE("GPL");
112 MODULE_VERSION(ibmveth_driver_version);
113
114 /* simple methods of getting data from the current rxq entry */
115 static inline int ibmveth_rxq_pending_buffer(struct ibmveth_adapter *adapter)
116 {
117         return (adapter->rx_queue.queue_addr[adapter->rx_queue.index].toggle == adapter->rx_queue.toggle);
118 }
119
120 static inline int ibmveth_rxq_buffer_valid(struct ibmveth_adapter *adapter)
121 {
122         return (adapter->rx_queue.queue_addr[adapter->rx_queue.index].valid);
123 }
124
125 static inline int ibmveth_rxq_frame_offset(struct ibmveth_adapter *adapter)
126 {
127         return (adapter->rx_queue.queue_addr[adapter->rx_queue.index].offset);
128 }
129
130 static inline int ibmveth_rxq_frame_length(struct ibmveth_adapter *adapter)
131 {
132         return (adapter->rx_queue.queue_addr[adapter->rx_queue.index].length);
133 }
134
135 /* setup the initial settings for a buffer pool */
136 static void ibmveth_init_buffer_pool(struct ibmveth_buff_pool *pool, u32 pool_index, u32 pool_size, u32 buff_size, u32 pool_active)
137 {
138         pool->size = pool_size;
139         pool->index = pool_index;
140         pool->buff_size = buff_size;
141         pool->threshold = pool_size / 2;
142         pool->active = pool_active;
143 }
144
145 /* allocate and setup an buffer pool - called during open */
146 static int ibmveth_alloc_buffer_pool(struct ibmveth_buff_pool *pool)
147 {
148         int i;
149
150         pool->free_map = kmalloc(sizeof(u16) * pool->size, GFP_KERNEL);
151
152         if(!pool->free_map) {
153                 return -1;
154         }
155
156         pool->dma_addr = kmalloc(sizeof(dma_addr_t) * pool->size, GFP_KERNEL);
157         if(!pool->dma_addr) {
158                 kfree(pool->free_map);
159                 pool->free_map = NULL;
160                 return -1;
161         }
162
163         pool->skbuff = kmalloc(sizeof(void*) * pool->size, GFP_KERNEL);
164
165         if(!pool->skbuff) {
166                 kfree(pool->dma_addr);
167                 pool->dma_addr = NULL;
168
169                 kfree(pool->free_map);
170                 pool->free_map = NULL;
171                 return -1;
172         }
173
174         memset(pool->skbuff, 0, sizeof(void*) * pool->size);
175         memset(pool->dma_addr, 0, sizeof(dma_addr_t) * pool->size);
176
177         for(i = 0; i < pool->size; ++i) {
178                 pool->free_map[i] = i;
179         }
180
181         atomic_set(&pool->available, 0);
182         pool->producer_index = 0;
183         pool->consumer_index = 0;
184
185         return 0;
186 }
187
188 /* replenish the buffers for a pool.  note that we don't need to
189  * skb_reserve these since they are used for incoming...
190  */
191 static void ibmveth_replenish_buffer_pool(struct ibmveth_adapter *adapter, struct ibmveth_buff_pool *pool)
192 {
193         u32 i;
194         u32 count = pool->size - atomic_read(&pool->available);
195         u32 buffers_added = 0;
196
197         mb();
198
199         for(i = 0; i < count; ++i) {
200                 struct sk_buff *skb;
201                 unsigned int free_index, index;
202                 u64 correlator;
203                 union ibmveth_buf_desc desc;
204                 unsigned long lpar_rc;
205                 dma_addr_t dma_addr;
206
207                 skb = alloc_skb(pool->buff_size, GFP_ATOMIC);
208
209                 if(!skb) {
210                         ibmveth_debug_printk("replenish: unable to allocate skb\n");
211                         adapter->replenish_no_mem++;
212                         break;
213                 }
214
215                 free_index = pool->consumer_index++ % pool->size;
216                 index = pool->free_map[free_index];
217
218                 ibmveth_assert(index != IBM_VETH_INVALID_MAP);
219                 ibmveth_assert(pool->skbuff[index] == NULL);
220
221                 dma_addr = dma_map_single(&adapter->vdev->dev, skb->data,
222                                 pool->buff_size, DMA_FROM_DEVICE);
223
224                 pool->free_map[free_index] = IBM_VETH_INVALID_MAP;
225                 pool->dma_addr[index] = dma_addr;
226                 pool->skbuff[index] = skb;
227
228                 correlator = ((u64)pool->index << 32) | index;
229                 *(u64*)skb->data = correlator;
230
231                 desc.desc = 0;
232                 desc.fields.valid = 1;
233                 desc.fields.length = pool->buff_size;
234                 desc.fields.address = dma_addr;
235
236                 lpar_rc = h_add_logical_lan_buffer(adapter->vdev->unit_address, desc.desc);
237
238                 if(lpar_rc != H_SUCCESS) {
239                         pool->free_map[free_index] = index;
240                         pool->skbuff[index] = NULL;
241                         pool->consumer_index--;
242                         dma_unmap_single(&adapter->vdev->dev,
243                                         pool->dma_addr[index], pool->buff_size,
244                                         DMA_FROM_DEVICE);
245                         dev_kfree_skb_any(skb);
246                         adapter->replenish_add_buff_failure++;
247                         break;
248                 } else {
249                         buffers_added++;
250                         adapter->replenish_add_buff_success++;
251                 }
252         }
253
254         mb();
255         atomic_add(buffers_added, &(pool->available));
256 }
257
258 /* replenish routine */
259 static void ibmveth_replenish_task(struct ibmveth_adapter *adapter)
260 {
261         int i;
262
263         adapter->replenish_task_cycles++;
264
265         for(i = 0; i < IbmVethNumBufferPools; i++)
266                 if(adapter->rx_buff_pool[i].active)
267                         ibmveth_replenish_buffer_pool(adapter,
268                                                      &adapter->rx_buff_pool[i]);
269
270         adapter->rx_no_buffer = *(u64*)(((char*)adapter->buffer_list_addr) + 4096 - 8);
271 }
272
273 /* empty and free ana buffer pool - also used to do cleanup in error paths */
274 static void ibmveth_free_buffer_pool(struct ibmveth_adapter *adapter, struct ibmveth_buff_pool *pool)
275 {
276         int i;
277
278         kfree(pool->free_map);
279         pool->free_map = NULL;
280
281         if(pool->skbuff && pool->dma_addr) {
282                 for(i = 0; i < pool->size; ++i) {
283                         struct sk_buff *skb = pool->skbuff[i];
284                         if(skb) {
285                                 dma_unmap_single(&adapter->vdev->dev,
286                                                  pool->dma_addr[i],
287                                                  pool->buff_size,
288                                                  DMA_FROM_DEVICE);
289                                 dev_kfree_skb_any(skb);
290                                 pool->skbuff[i] = NULL;
291                         }
292                 }
293         }
294
295         if(pool->dma_addr) {
296                 kfree(pool->dma_addr);
297                 pool->dma_addr = NULL;
298         }
299
300         if(pool->skbuff) {
301                 kfree(pool->skbuff);
302                 pool->skbuff = NULL;
303         }
304 }
305
306 /* remove a buffer from a pool */
307 static void ibmveth_remove_buffer_from_pool(struct ibmveth_adapter *adapter, u64 correlator)
308 {
309         unsigned int pool  = correlator >> 32;
310         unsigned int index = correlator & 0xffffffffUL;
311         unsigned int free_index;
312         struct sk_buff *skb;
313
314         ibmveth_assert(pool < IbmVethNumBufferPools);
315         ibmveth_assert(index < adapter->rx_buff_pool[pool].size);
316
317         skb = adapter->rx_buff_pool[pool].skbuff[index];
318
319         ibmveth_assert(skb != NULL);
320
321         adapter->rx_buff_pool[pool].skbuff[index] = NULL;
322
323         dma_unmap_single(&adapter->vdev->dev,
324                          adapter->rx_buff_pool[pool].dma_addr[index],
325                          adapter->rx_buff_pool[pool].buff_size,
326                          DMA_FROM_DEVICE);
327
328         free_index = adapter->rx_buff_pool[pool].producer_index++ % adapter->rx_buff_pool[pool].size;
329         adapter->rx_buff_pool[pool].free_map[free_index] = index;
330
331         mb();
332
333         atomic_dec(&(adapter->rx_buff_pool[pool].available));
334 }
335
336 /* get the current buffer on the rx queue */
337 static inline struct sk_buff *ibmveth_rxq_get_buffer(struct ibmveth_adapter *adapter)
338 {
339         u64 correlator = adapter->rx_queue.queue_addr[adapter->rx_queue.index].correlator;
340         unsigned int pool = correlator >> 32;
341         unsigned int index = correlator & 0xffffffffUL;
342
343         ibmveth_assert(pool < IbmVethNumBufferPools);
344         ibmveth_assert(index < adapter->rx_buff_pool[pool].size);
345
346         return adapter->rx_buff_pool[pool].skbuff[index];
347 }
348
349 /* recycle the current buffer on the rx queue */
350 static void ibmveth_rxq_recycle_buffer(struct ibmveth_adapter *adapter)
351 {
352         u32 q_index = adapter->rx_queue.index;
353         u64 correlator = adapter->rx_queue.queue_addr[q_index].correlator;
354         unsigned int pool = correlator >> 32;
355         unsigned int index = correlator & 0xffffffffUL;
356         union ibmveth_buf_desc desc;
357         unsigned long lpar_rc;
358
359         ibmveth_assert(pool < IbmVethNumBufferPools);
360         ibmveth_assert(index < adapter->rx_buff_pool[pool].size);
361
362         if(!adapter->rx_buff_pool[pool].active) {
363                 ibmveth_rxq_harvest_buffer(adapter);
364                 ibmveth_free_buffer_pool(adapter, &adapter->rx_buff_pool[pool]);
365                 return;
366         }
367
368         desc.desc = 0;
369         desc.fields.valid = 1;
370         desc.fields.length = adapter->rx_buff_pool[pool].buff_size;
371         desc.fields.address = adapter->rx_buff_pool[pool].dma_addr[index];
372
373         lpar_rc = h_add_logical_lan_buffer(adapter->vdev->unit_address, desc.desc);
374
375         if(lpar_rc != H_SUCCESS) {
376                 ibmveth_debug_printk("h_add_logical_lan_buffer failed during recycle rc=%ld", lpar_rc);
377                 ibmveth_remove_buffer_from_pool(adapter, adapter->rx_queue.queue_addr[adapter->rx_queue.index].correlator);
378         }
379
380         if(++adapter->rx_queue.index == adapter->rx_queue.num_slots) {
381                 adapter->rx_queue.index = 0;
382                 adapter->rx_queue.toggle = !adapter->rx_queue.toggle;
383         }
384 }
385
386 static inline void ibmveth_rxq_harvest_buffer(struct ibmveth_adapter *adapter)
387 {
388         ibmveth_remove_buffer_from_pool(adapter, adapter->rx_queue.queue_addr[adapter->rx_queue.index].correlator);
389
390         if(++adapter->rx_queue.index == adapter->rx_queue.num_slots) {
391                 adapter->rx_queue.index = 0;
392                 adapter->rx_queue.toggle = !adapter->rx_queue.toggle;
393         }
394 }
395
396 static void ibmveth_cleanup(struct ibmveth_adapter *adapter)
397 {
398         int i;
399
400         if(adapter->buffer_list_addr != NULL) {
401                 if(!dma_mapping_error(adapter->buffer_list_dma)) {
402                         dma_unmap_single(&adapter->vdev->dev,
403                                         adapter->buffer_list_dma, 4096,
404                                         DMA_BIDIRECTIONAL);
405                         adapter->buffer_list_dma = DMA_ERROR_CODE;
406                 }
407                 free_page((unsigned long)adapter->buffer_list_addr);
408                 adapter->buffer_list_addr = NULL;
409         }
410
411         if(adapter->filter_list_addr != NULL) {
412                 if(!dma_mapping_error(adapter->filter_list_dma)) {
413                         dma_unmap_single(&adapter->vdev->dev,
414                                         adapter->filter_list_dma, 4096,
415                                         DMA_BIDIRECTIONAL);
416                         adapter->filter_list_dma = DMA_ERROR_CODE;
417                 }
418                 free_page((unsigned long)adapter->filter_list_addr);
419                 adapter->filter_list_addr = NULL;
420         }
421
422         if(adapter->rx_queue.queue_addr != NULL) {
423                 if(!dma_mapping_error(adapter->rx_queue.queue_dma)) {
424                         dma_unmap_single(&adapter->vdev->dev,
425                                         adapter->rx_queue.queue_dma,
426                                         adapter->rx_queue.queue_len,
427                                         DMA_BIDIRECTIONAL);
428                         adapter->rx_queue.queue_dma = DMA_ERROR_CODE;
429                 }
430                 kfree(adapter->rx_queue.queue_addr);
431                 adapter->rx_queue.queue_addr = NULL;
432         }
433
434         for(i = 0; i<IbmVethNumBufferPools; i++)
435                 if (adapter->rx_buff_pool[i].active)
436                         ibmveth_free_buffer_pool(adapter,
437                                                  &adapter->rx_buff_pool[i]);
438 }
439
440 static int ibmveth_open(struct net_device *netdev)
441 {
442         struct ibmveth_adapter *adapter = netdev->priv;
443         u64 mac_address = 0;
444         int rxq_entries = 1;
445         unsigned long lpar_rc;
446         int rc;
447         union ibmveth_buf_desc rxq_desc;
448         int i;
449
450         ibmveth_debug_printk("open starting\n");
451
452         for(i = 0; i<IbmVethNumBufferPools; i++)
453                 rxq_entries += adapter->rx_buff_pool[i].size;
454
455         adapter->buffer_list_addr = (void*) get_zeroed_page(GFP_KERNEL);
456         adapter->filter_list_addr = (void*) get_zeroed_page(GFP_KERNEL);
457
458         if(!adapter->buffer_list_addr || !adapter->filter_list_addr) {
459                 ibmveth_error_printk("unable to allocate filter or buffer list pages\n");
460                 ibmveth_cleanup(adapter);
461                 return -ENOMEM;
462         }
463
464         adapter->rx_queue.queue_len = sizeof(struct ibmveth_rx_q_entry) * rxq_entries;
465         adapter->rx_queue.queue_addr = kmalloc(adapter->rx_queue.queue_len, GFP_KERNEL);
466
467         if(!adapter->rx_queue.queue_addr) {
468                 ibmveth_error_printk("unable to allocate rx queue pages\n");
469                 ibmveth_cleanup(adapter);
470                 return -ENOMEM;
471         }
472
473         adapter->buffer_list_dma = dma_map_single(&adapter->vdev->dev,
474                         adapter->buffer_list_addr, 4096, DMA_BIDIRECTIONAL);
475         adapter->filter_list_dma = dma_map_single(&adapter->vdev->dev,
476                         adapter->filter_list_addr, 4096, DMA_BIDIRECTIONAL);
477         adapter->rx_queue.queue_dma = dma_map_single(&adapter->vdev->dev,
478                         adapter->rx_queue.queue_addr,
479                         adapter->rx_queue.queue_len, DMA_BIDIRECTIONAL);
480
481         if((dma_mapping_error(adapter->buffer_list_dma) ) ||
482            (dma_mapping_error(adapter->filter_list_dma)) ||
483            (dma_mapping_error(adapter->rx_queue.queue_dma))) {
484                 ibmveth_error_printk("unable to map filter or buffer list pages\n");
485                 ibmveth_cleanup(adapter);
486                 return -ENOMEM;
487         }
488
489         adapter->rx_queue.index = 0;
490         adapter->rx_queue.num_slots = rxq_entries;
491         adapter->rx_queue.toggle = 1;
492
493         memcpy(&mac_address, netdev->dev_addr, netdev->addr_len);
494         mac_address = mac_address >> 16;
495
496         rxq_desc.desc = 0;
497         rxq_desc.fields.valid = 1;
498         rxq_desc.fields.length = adapter->rx_queue.queue_len;
499         rxq_desc.fields.address = adapter->rx_queue.queue_dma;
500
501         ibmveth_debug_printk("buffer list @ 0x%p\n", adapter->buffer_list_addr);
502         ibmveth_debug_printk("filter list @ 0x%p\n", adapter->filter_list_addr);
503         ibmveth_debug_printk("receive q   @ 0x%p\n", adapter->rx_queue.queue_addr);
504
505
506         lpar_rc = h_register_logical_lan(adapter->vdev->unit_address,
507                                          adapter->buffer_list_dma,
508                                          rxq_desc.desc,
509                                          adapter->filter_list_dma,
510                                          mac_address);
511
512         if(lpar_rc != H_SUCCESS) {
513                 ibmveth_error_printk("h_register_logical_lan failed with %ld\n", lpar_rc);
514                 ibmveth_error_printk("buffer TCE:0x%lx filter TCE:0x%lx rxq desc:0x%lx MAC:0x%lx\n",
515                                      adapter->buffer_list_dma,
516                                      adapter->filter_list_dma,
517                                      rxq_desc.desc,
518                                      mac_address);
519                 ibmveth_cleanup(adapter);
520                 return -ENONET;
521         }
522
523         for(i = 0; i<IbmVethNumBufferPools; i++) {
524                 if(!adapter->rx_buff_pool[i].active)
525                         continue;
526                 if (ibmveth_alloc_buffer_pool(&adapter->rx_buff_pool[i])) {
527                         ibmveth_error_printk("unable to alloc pool\n");
528                         adapter->rx_buff_pool[i].active = 0;
529                         ibmveth_cleanup(adapter);
530                         return -ENOMEM ;
531                 }
532         }
533
534         ibmveth_debug_printk("registering irq 0x%x\n", netdev->irq);
535         if((rc = request_irq(netdev->irq, &ibmveth_interrupt, 0, netdev->name, netdev)) != 0) {
536                 ibmveth_error_printk("unable to request irq 0x%x, rc %d\n", netdev->irq, rc);
537                 do {
538                         rc = h_free_logical_lan(adapter->vdev->unit_address);
539                 } while (H_IS_LONG_BUSY(rc) || (rc == H_BUSY));
540
541                 ibmveth_cleanup(adapter);
542                 return rc;
543         }
544
545         ibmveth_debug_printk("initial replenish cycle\n");
546         ibmveth_interrupt(netdev->irq, netdev, NULL);
547
548         netif_start_queue(netdev);
549
550         ibmveth_debug_printk("open complete\n");
551
552         return 0;
553 }
554
555 static int ibmveth_close(struct net_device *netdev)
556 {
557         struct ibmveth_adapter *adapter = netdev->priv;
558         long lpar_rc;
559
560         ibmveth_debug_printk("close starting\n");
561
562         if (!adapter->pool_config)
563                 netif_stop_queue(netdev);
564
565         free_irq(netdev->irq, netdev);
566
567         do {
568                 lpar_rc = h_free_logical_lan(adapter->vdev->unit_address);
569         } while (H_IS_LONG_BUSY(lpar_rc) || (lpar_rc == H_BUSY));
570
571         if(lpar_rc != H_SUCCESS)
572         {
573                 ibmveth_error_printk("h_free_logical_lan failed with %lx, continuing with close\n",
574                                      lpar_rc);
575         }
576
577         adapter->rx_no_buffer = *(u64*)(((char*)adapter->buffer_list_addr) + 4096 - 8);
578
579         ibmveth_cleanup(adapter);
580
581         ibmveth_debug_printk("close complete\n");
582
583         return 0;
584 }
585
586 static int netdev_get_settings(struct net_device *dev, struct ethtool_cmd *cmd) {
587         cmd->supported = (SUPPORTED_1000baseT_Full | SUPPORTED_Autoneg | SUPPORTED_FIBRE);
588         cmd->advertising = (ADVERTISED_1000baseT_Full | ADVERTISED_Autoneg | ADVERTISED_FIBRE);
589         cmd->speed = SPEED_1000;
590         cmd->duplex = DUPLEX_FULL;
591         cmd->port = PORT_FIBRE;
592         cmd->phy_address = 0;
593         cmd->transceiver = XCVR_INTERNAL;
594         cmd->autoneg = AUTONEG_ENABLE;
595         cmd->maxtxpkt = 0;
596         cmd->maxrxpkt = 1;
597         return 0;
598 }
599
600 static void netdev_get_drvinfo (struct net_device *dev, struct ethtool_drvinfo *info) {
601         strncpy(info->driver, ibmveth_driver_name, sizeof(info->driver) - 1);
602         strncpy(info->version, ibmveth_driver_version, sizeof(info->version) - 1);
603 }
604
605 static u32 netdev_get_link(struct net_device *dev) {
606         return 1;
607 }
608
609 static struct ethtool_ops netdev_ethtool_ops = {
610         .get_drvinfo            = netdev_get_drvinfo,
611         .get_settings           = netdev_get_settings,
612         .get_link               = netdev_get_link,
613         .get_sg                 = ethtool_op_get_sg,
614         .get_tx_csum            = ethtool_op_get_tx_csum,
615 };
616
617 static int ibmveth_ioctl(struct net_device *dev, struct ifreq *ifr, int cmd)
618 {
619         return -EOPNOTSUPP;
620 }
621
622 #define page_offset(v) ((unsigned long)(v) & ((1 << 12) - 1))
623
624 static int ibmveth_start_xmit(struct sk_buff *skb, struct net_device *netdev)
625 {
626         struct ibmveth_adapter *adapter = netdev->priv;
627         union ibmveth_buf_desc desc[IbmVethMaxSendFrags];
628         unsigned long lpar_rc;
629         int nfrags = 0, curfrag;
630         unsigned long correlator;
631         unsigned long flags;
632         unsigned int retry_count;
633         unsigned int tx_dropped = 0;
634         unsigned int tx_bytes = 0;
635         unsigned int tx_packets = 0;
636         unsigned int tx_send_failed = 0;
637         unsigned int tx_map_failed = 0;
638
639
640         if ((skb_shinfo(skb)->nr_frags + 1) > IbmVethMaxSendFrags) {
641                 tx_dropped++;
642                 goto out;
643         }
644
645         memset(&desc, 0, sizeof(desc));
646
647         /* nfrags = number of frags after the initial fragment */
648         nfrags = skb_shinfo(skb)->nr_frags;
649
650         if(nfrags)
651                 adapter->tx_multidesc_send++;
652
653         /* map the initial fragment */
654         desc[0].fields.length  = nfrags ? skb->len - skb->data_len : skb->len;
655         desc[0].fields.address = dma_map_single(&adapter->vdev->dev, skb->data,
656                                         desc[0].fields.length, DMA_TO_DEVICE);
657         desc[0].fields.valid   = 1;
658
659         if(dma_mapping_error(desc[0].fields.address)) {
660                 ibmveth_error_printk("tx: unable to map initial fragment\n");
661                 tx_map_failed++;
662                 tx_dropped++;
663                 goto out;
664         }
665
666         curfrag = nfrags;
667
668         /* map fragments past the initial portion if there are any */
669         while(curfrag--) {
670                 skb_frag_t *frag = &skb_shinfo(skb)->frags[curfrag];
671                 desc[curfrag+1].fields.address
672                         = dma_map_single(&adapter->vdev->dev,
673                                 page_address(frag->page) + frag->page_offset,
674                                 frag->size, DMA_TO_DEVICE);
675                 desc[curfrag+1].fields.length = frag->size;
676                 desc[curfrag+1].fields.valid  = 1;
677
678                 if(dma_mapping_error(desc[curfrag+1].fields.address)) {
679                         ibmveth_error_printk("tx: unable to map fragment %d\n", curfrag);
680                         tx_map_failed++;
681                         tx_dropped++;
682                         /* Free all the mappings we just created */
683                         while(curfrag < nfrags) {
684                                 dma_unmap_single(&adapter->vdev->dev,
685                                                  desc[curfrag+1].fields.address,
686                                                  desc[curfrag+1].fields.length,
687                                                  DMA_TO_DEVICE);
688                                 curfrag++;
689                         }
690                         goto out;
691                 }
692         }
693
694         /* send the frame. Arbitrarily set retrycount to 1024 */
695         correlator = 0;
696         retry_count = 1024;
697         do {
698                 lpar_rc = h_send_logical_lan(adapter->vdev->unit_address,
699                                              desc[0].desc,
700                                              desc[1].desc,
701                                              desc[2].desc,
702                                              desc[3].desc,
703                                              desc[4].desc,
704                                              desc[5].desc,
705                                              correlator,
706                                              &correlator);
707         } while ((lpar_rc == H_BUSY) && (retry_count--));
708
709         if(lpar_rc != H_SUCCESS && lpar_rc != H_DROPPED) {
710                 int i;
711                 ibmveth_error_printk("tx: h_send_logical_lan failed with rc=%ld\n", lpar_rc);
712                 for(i = 0; i < 6; i++) {
713                         ibmveth_error_printk("tx: desc[%i] valid=%d, len=%d, address=0x%d\n", i,
714                                              desc[i].fields.valid, desc[i].fields.length, desc[i].fields.address);
715                 }
716                 tx_send_failed++;
717                 tx_dropped++;
718         } else {
719                 tx_packets++;
720                 tx_bytes += skb->len;
721                 netdev->trans_start = jiffies;
722         }
723
724         do {
725                 dma_unmap_single(&adapter->vdev->dev,
726                                 desc[nfrags].fields.address,
727                                 desc[nfrags].fields.length, DMA_TO_DEVICE);
728         } while(--nfrags >= 0);
729
730 out:    spin_lock_irqsave(&adapter->stats_lock, flags);
731         adapter->stats.tx_dropped += tx_dropped;
732         adapter->stats.tx_bytes += tx_bytes;
733         adapter->stats.tx_packets += tx_packets;
734         adapter->tx_send_failed += tx_send_failed;
735         adapter->tx_map_failed += tx_map_failed;
736         spin_unlock_irqrestore(&adapter->stats_lock, flags);
737
738         dev_kfree_skb(skb);
739         return 0;
740 }
741
742 static int ibmveth_poll(struct net_device *netdev, int *budget)
743 {
744         struct ibmveth_adapter *adapter = netdev->priv;
745         int max_frames_to_process = netdev->quota;
746         int frames_processed = 0;
747         int more_work = 1;
748         unsigned long lpar_rc;
749
750  restart_poll:
751         do {
752                 struct net_device *netdev = adapter->netdev;
753
754                 if(ibmveth_rxq_pending_buffer(adapter)) {
755                         struct sk_buff *skb;
756
757                         rmb();
758
759                         if(!ibmveth_rxq_buffer_valid(adapter)) {
760                                 wmb(); /* suggested by larson1 */
761                                 adapter->rx_invalid_buffer++;
762                                 ibmveth_debug_printk("recycling invalid buffer\n");
763                                 ibmveth_rxq_recycle_buffer(adapter);
764                         } else {
765                                 int length = ibmveth_rxq_frame_length(adapter);
766                                 int offset = ibmveth_rxq_frame_offset(adapter);
767                                 skb = ibmveth_rxq_get_buffer(adapter);
768
769                                 ibmveth_rxq_harvest_buffer(adapter);
770
771                                 skb_reserve(skb, offset);
772                                 skb_put(skb, length);
773                                 skb->dev = netdev;
774                                 skb->protocol = eth_type_trans(skb, netdev);
775
776                                 netif_receive_skb(skb); /* send it up */
777
778                                 adapter->stats.rx_packets++;
779                                 adapter->stats.rx_bytes += length;
780                                 frames_processed++;
781                                 netdev->last_rx = jiffies;
782                         }
783                 } else {
784                         more_work = 0;
785                 }
786         } while(more_work && (frames_processed < max_frames_to_process));
787
788         ibmveth_replenish_task(adapter);
789
790         if(more_work) {
791                 /* more work to do - return that we are not done yet */
792                 netdev->quota -= frames_processed;
793                 *budget -= frames_processed;
794                 return 1;
795         }
796
797         /* we think we are done - reenable interrupts, then check once more to make sure we are done */
798         lpar_rc = h_vio_signal(adapter->vdev->unit_address, VIO_IRQ_ENABLE);
799
800         ibmveth_assert(lpar_rc == H_SUCCESS);
801
802         netif_rx_complete(netdev);
803
804         if(ibmveth_rxq_pending_buffer(adapter) && netif_rx_reschedule(netdev, frames_processed))
805         {
806                 lpar_rc = h_vio_signal(adapter->vdev->unit_address, VIO_IRQ_DISABLE);
807                 ibmveth_assert(lpar_rc == H_SUCCESS);
808                 more_work = 1;
809                 goto restart_poll;
810         }
811
812         netdev->quota -= frames_processed;
813         *budget -= frames_processed;
814
815         /* we really are done */
816         return 0;
817 }
818
819 static irqreturn_t ibmveth_interrupt(int irq, void *dev_instance, struct pt_regs *regs)
820 {
821         struct net_device *netdev = dev_instance;
822         struct ibmveth_adapter *adapter = netdev->priv;
823         unsigned long lpar_rc;
824
825         if(netif_rx_schedule_prep(netdev)) {
826                 lpar_rc = h_vio_signal(adapter->vdev->unit_address, VIO_IRQ_DISABLE);
827                 ibmveth_assert(lpar_rc == H_SUCCESS);
828                 __netif_rx_schedule(netdev);
829         }
830         return IRQ_HANDLED;
831 }
832
833 static struct net_device_stats *ibmveth_get_stats(struct net_device *dev)
834 {
835         struct ibmveth_adapter *adapter = dev->priv;
836         return &adapter->stats;
837 }
838
839 static void ibmveth_set_multicast_list(struct net_device *netdev)
840 {
841         struct ibmveth_adapter *adapter = netdev->priv;
842         unsigned long lpar_rc;
843
844         if((netdev->flags & IFF_PROMISC) || (netdev->mc_count > adapter->mcastFilterSize)) {
845                 lpar_rc = h_multicast_ctrl(adapter->vdev->unit_address,
846                                            IbmVethMcastEnableRecv |
847                                            IbmVethMcastDisableFiltering,
848                                            0);
849                 if(lpar_rc != H_SUCCESS) {
850                         ibmveth_error_printk("h_multicast_ctrl rc=%ld when entering promisc mode\n", lpar_rc);
851                 }
852         } else {
853                 struct dev_mc_list *mclist = netdev->mc_list;
854                 int i;
855                 /* clear the filter table & disable filtering */
856                 lpar_rc = h_multicast_ctrl(adapter->vdev->unit_address,
857                                            IbmVethMcastEnableRecv |
858                                            IbmVethMcastDisableFiltering |
859                                            IbmVethMcastClearFilterTable,
860                                            0);
861                 if(lpar_rc != H_SUCCESS) {
862                         ibmveth_error_printk("h_multicast_ctrl rc=%ld when attempting to clear filter table\n", lpar_rc);
863                 }
864                 /* add the addresses to the filter table */
865                 for(i = 0; i < netdev->mc_count; ++i, mclist = mclist->next) {
866                         // add the multicast address to the filter table
867                         unsigned long mcast_addr = 0;
868                         memcpy(((char *)&mcast_addr)+2, mclist->dmi_addr, 6);
869                         lpar_rc = h_multicast_ctrl(adapter->vdev->unit_address,
870                                                    IbmVethMcastAddFilter,
871                                                    mcast_addr);
872                         if(lpar_rc != H_SUCCESS) {
873                                 ibmveth_error_printk("h_multicast_ctrl rc=%ld when adding an entry to the filter table\n", lpar_rc);
874                         }
875                 }
876
877                 /* re-enable filtering */
878                 lpar_rc = h_multicast_ctrl(adapter->vdev->unit_address,
879                                            IbmVethMcastEnableFiltering,
880                                            0);
881                 if(lpar_rc != H_SUCCESS) {
882                         ibmveth_error_printk("h_multicast_ctrl rc=%ld when enabling filtering\n", lpar_rc);
883                 }
884         }
885 }
886
887 static int ibmveth_change_mtu(struct net_device *dev, int new_mtu)
888 {
889         struct ibmveth_adapter *adapter = dev->priv;
890         int new_mtu_oh = new_mtu + IBMVETH_BUFF_OH;
891         int i;
892
893         if (new_mtu < IBMVETH_MAX_MTU)
894                 return -EINVAL;
895
896         /* Look for an active buffer pool that can hold the new MTU */
897         for(i = 0; i<IbmVethNumBufferPools; i++) {
898                 if (!adapter->rx_buff_pool[i].active)
899                         continue;
900                 if (new_mtu_oh < adapter->rx_buff_pool[i].buff_size) {
901                         dev->mtu = new_mtu;
902                         return 0;
903                 }
904         }
905         return -EINVAL;
906 }
907
908 static int __devinit ibmveth_probe(struct vio_dev *dev, const struct vio_device_id *id)
909 {
910         int rc, i;
911         struct net_device *netdev;
912         struct ibmveth_adapter *adapter = NULL;
913
914         unsigned char *mac_addr_p;
915         unsigned int *mcastFilterSize_p;
916
917
918         ibmveth_debug_printk_no_adapter("entering ibmveth_probe for UA 0x%x\n",
919                                         dev->unit_address);
920
921         mac_addr_p = (unsigned char *) vio_get_attribute(dev, VETH_MAC_ADDR, 0);
922         if(!mac_addr_p) {
923                 printk(KERN_ERR "(%s:%3.3d) ERROR: Can't find VETH_MAC_ADDR "
924                                 "attribute\n", __FILE__, __LINE__);
925                 return 0;
926         }
927
928         mcastFilterSize_p= (unsigned int *) vio_get_attribute(dev, VETH_MCAST_FILTER_SIZE, 0);
929         if(!mcastFilterSize_p) {
930                 printk(KERN_ERR "(%s:%3.3d) ERROR: Can't find "
931                                 "VETH_MCAST_FILTER_SIZE attribute\n",
932                                 __FILE__, __LINE__);
933                 return 0;
934         }
935
936         netdev = alloc_etherdev(sizeof(struct ibmveth_adapter));
937
938         if(!netdev)
939                 return -ENOMEM;
940
941         SET_MODULE_OWNER(netdev);
942
943         adapter = netdev->priv;
944         memset(adapter, 0, sizeof(adapter));
945         dev->dev.driver_data = netdev;
946
947         adapter->vdev = dev;
948         adapter->netdev = netdev;
949         adapter->mcastFilterSize= *mcastFilterSize_p;
950         adapter->pool_config = 0;
951
952         /*      Some older boxes running PHYP non-natively have an OF that
953                 returns a 8-byte local-mac-address field (and the first
954                 2 bytes have to be ignored) while newer boxes' OF return
955                 a 6-byte field. Note that IEEE 1275 specifies that
956                 local-mac-address must be a 6-byte field.
957                 The RPA doc specifies that the first byte must be 10b, so
958                 we'll just look for it to solve this 8 vs. 6 byte field issue */
959
960         if ((*mac_addr_p & 0x3) != 0x02)
961                 mac_addr_p += 2;
962
963         adapter->mac_addr = 0;
964         memcpy(&adapter->mac_addr, mac_addr_p, 6);
965
966         adapter->liobn = dev->iommu_table->it_index;
967
968         netdev->irq = dev->irq;
969         netdev->open               = ibmveth_open;
970         netdev->poll               = ibmveth_poll;
971         netdev->weight             = 16;
972         netdev->stop               = ibmveth_close;
973         netdev->hard_start_xmit    = ibmveth_start_xmit;
974         netdev->get_stats          = ibmveth_get_stats;
975         netdev->set_multicast_list = ibmveth_set_multicast_list;
976         netdev->do_ioctl           = ibmveth_ioctl;
977         netdev->ethtool_ops           = &netdev_ethtool_ops;
978         netdev->change_mtu         = ibmveth_change_mtu;
979         SET_NETDEV_DEV(netdev, &dev->dev);
980         netdev->features |= NETIF_F_LLTX;
981         spin_lock_init(&adapter->stats_lock);
982
983         memcpy(&netdev->dev_addr, &adapter->mac_addr, netdev->addr_len);
984
985         for(i = 0; i<IbmVethNumBufferPools; i++) {
986                 struct kobject *kobj = &adapter->rx_buff_pool[i].kobj;
987                 ibmveth_init_buffer_pool(&adapter->rx_buff_pool[i], i,
988                                          pool_count[i], pool_size[i],
989                                          pool_active[i]);
990                 kobj->parent = &dev->dev.kobj;
991                 sprintf(kobj->name, "pool%d", i);
992                 kobj->ktype = &ktype_veth_pool;
993                 kobject_register(kobj);
994         }
995
996         ibmveth_debug_printk("adapter @ 0x%p\n", adapter);
997
998         adapter->buffer_list_dma = DMA_ERROR_CODE;
999         adapter->filter_list_dma = DMA_ERROR_CODE;
1000         adapter->rx_queue.queue_dma = DMA_ERROR_CODE;
1001
1002         ibmveth_debug_printk("registering netdev...\n");
1003
1004         rc = register_netdev(netdev);
1005
1006         if(rc) {
1007                 ibmveth_debug_printk("failed to register netdev rc=%d\n", rc);
1008                 free_netdev(netdev);
1009                 return rc;
1010         }
1011
1012         ibmveth_debug_printk("registered\n");
1013
1014         ibmveth_proc_register_adapter(adapter);
1015
1016         return 0;
1017 }
1018
1019 static int __devexit ibmveth_remove(struct vio_dev *dev)
1020 {
1021         struct net_device *netdev = dev->dev.driver_data;
1022         struct ibmveth_adapter *adapter = netdev->priv;
1023         int i;
1024
1025         for(i = 0; i<IbmVethNumBufferPools; i++)
1026                 kobject_unregister(&adapter->rx_buff_pool[i].kobj);
1027
1028         unregister_netdev(netdev);
1029
1030         ibmveth_proc_unregister_adapter(adapter);
1031
1032         free_netdev(netdev);
1033         return 0;
1034 }
1035
1036 #ifdef CONFIG_PROC_FS
1037 static void ibmveth_proc_register_driver(void)
1038 {
1039         ibmveth_proc_dir = proc_mkdir(IBMVETH_PROC_DIR, NULL);
1040         if (ibmveth_proc_dir) {
1041                 SET_MODULE_OWNER(ibmveth_proc_dir);
1042         }
1043 }
1044
1045 static void ibmveth_proc_unregister_driver(void)
1046 {
1047         remove_proc_entry(IBMVETH_PROC_DIR, NULL);
1048 }
1049
1050 static void *ibmveth_seq_start(struct seq_file *seq, loff_t *pos)
1051 {
1052         if (*pos == 0) {
1053                 return (void *)1;
1054         } else {
1055                 return NULL;
1056         }
1057 }
1058
1059 static void *ibmveth_seq_next(struct seq_file *seq, void *v, loff_t *pos)
1060 {
1061         ++*pos;
1062         return NULL;
1063 }
1064
1065 static void ibmveth_seq_stop(struct seq_file *seq, void *v)
1066 {
1067 }
1068
1069 static int ibmveth_seq_show(struct seq_file *seq, void *v)
1070 {
1071         struct ibmveth_adapter *adapter = seq->private;
1072         char *current_mac = ((char*) &adapter->netdev->dev_addr);
1073         char *firmware_mac = ((char*) &adapter->mac_addr) ;
1074
1075         seq_printf(seq, "%s %s\n\n", ibmveth_driver_string, ibmveth_driver_version);
1076
1077         seq_printf(seq, "Unit Address:    0x%x\n", adapter->vdev->unit_address);
1078         seq_printf(seq, "LIOBN:           0x%lx\n", adapter->liobn);
1079         seq_printf(seq, "Current MAC:     %02X:%02X:%02X:%02X:%02X:%02X\n",
1080                    current_mac[0], current_mac[1], current_mac[2],
1081                    current_mac[3], current_mac[4], current_mac[5]);
1082         seq_printf(seq, "Firmware MAC:    %02X:%02X:%02X:%02X:%02X:%02X\n",
1083                    firmware_mac[0], firmware_mac[1], firmware_mac[2],
1084                    firmware_mac[3], firmware_mac[4], firmware_mac[5]);
1085
1086         seq_printf(seq, "\nAdapter Statistics:\n");
1087         seq_printf(seq, "  TX:  skbuffs linearized:          %ld\n", adapter->tx_linearized);
1088         seq_printf(seq, "       multi-descriptor sends:      %ld\n", adapter->tx_multidesc_send);
1089         seq_printf(seq, "       skb_linearize failures:      %ld\n", adapter->tx_linearize_failed);
1090         seq_printf(seq, "       vio_map_single failres:      %ld\n", adapter->tx_map_failed);
1091         seq_printf(seq, "       send failures:               %ld\n", adapter->tx_send_failed);
1092         seq_printf(seq, "  RX:  replenish task cycles:       %ld\n", adapter->replenish_task_cycles);
1093         seq_printf(seq, "       alloc_skb_failures:          %ld\n", adapter->replenish_no_mem);
1094         seq_printf(seq, "       add buffer failures:         %ld\n", adapter->replenish_add_buff_failure);
1095         seq_printf(seq, "       invalid buffers:             %ld\n", adapter->rx_invalid_buffer);
1096         seq_printf(seq, "       no buffers:                  %ld\n", adapter->rx_no_buffer);
1097
1098         return 0;
1099 }
1100 static struct seq_operations ibmveth_seq_ops = {
1101         .start = ibmveth_seq_start,
1102         .next  = ibmveth_seq_next,
1103         .stop  = ibmveth_seq_stop,
1104         .show  = ibmveth_seq_show,
1105 };
1106
1107 static int ibmveth_proc_open(struct inode *inode, struct file *file)
1108 {
1109         struct seq_file *seq;
1110         struct proc_dir_entry *proc;
1111         int rc;
1112
1113         rc = seq_open(file, &ibmveth_seq_ops);
1114         if (!rc) {
1115                 /* recover the pointer buried in proc_dir_entry data */
1116                 seq = file->private_data;
1117                 proc = PDE(inode);
1118                 seq->private = proc->data;
1119         }
1120         return rc;
1121 }
1122
1123 static struct file_operations ibmveth_proc_fops = {
1124         .owner   = THIS_MODULE,
1125         .open    = ibmveth_proc_open,
1126         .read    = seq_read,
1127         .llseek  = seq_lseek,
1128         .release = seq_release,
1129 };
1130
1131 static void ibmveth_proc_register_adapter(struct ibmveth_adapter *adapter)
1132 {
1133         struct proc_dir_entry *entry;
1134         if (ibmveth_proc_dir) {
1135                 entry = create_proc_entry(adapter->netdev->name, S_IFREG, ibmveth_proc_dir);
1136                 if (!entry) {
1137                         ibmveth_error_printk("Cannot create adapter proc entry");
1138                 } else {
1139                         entry->data = (void *) adapter;
1140                         entry->proc_fops = &ibmveth_proc_fops;
1141                         SET_MODULE_OWNER(entry);
1142                 }
1143         }
1144         return;
1145 }
1146
1147 static void ibmveth_proc_unregister_adapter(struct ibmveth_adapter *adapter)
1148 {
1149         if (ibmveth_proc_dir) {
1150                 remove_proc_entry(adapter->netdev->name, ibmveth_proc_dir);
1151         }
1152 }
1153
1154 #else /* CONFIG_PROC_FS */
1155 static void ibmveth_proc_register_adapter(struct ibmveth_adapter *adapter)
1156 {
1157 }
1158
1159 static void ibmveth_proc_unregister_adapter(struct ibmveth_adapter *adapter)
1160 {
1161 }
1162 static void ibmveth_proc_register_driver(void)
1163 {
1164 }
1165
1166 static void ibmveth_proc_unregister_driver(void)
1167 {
1168 }
1169 #endif /* CONFIG_PROC_FS */
1170
1171 static struct attribute veth_active_attr;
1172 static struct attribute veth_num_attr;
1173 static struct attribute veth_size_attr;
1174
1175 static ssize_t veth_pool_show(struct kobject * kobj,
1176                               struct attribute * attr, char * buf)
1177 {
1178         struct ibmveth_buff_pool *pool = container_of(kobj,
1179                                                       struct ibmveth_buff_pool,
1180                                                       kobj);
1181
1182         if (attr == &veth_active_attr)
1183                 return sprintf(buf, "%d\n", pool->active);
1184         else if (attr == &veth_num_attr)
1185                 return sprintf(buf, "%d\n", pool->size);
1186         else if (attr == &veth_size_attr)
1187                 return sprintf(buf, "%d\n", pool->buff_size);
1188         return 0;
1189 }
1190
1191 static ssize_t veth_pool_store(struct kobject * kobj, struct attribute * attr,
1192 const char * buf, size_t count)
1193 {
1194         struct ibmveth_buff_pool *pool = container_of(kobj,
1195                                                       struct ibmveth_buff_pool,
1196                                                       kobj);
1197         struct net_device *netdev =
1198             container_of(kobj->parent, struct device, kobj)->driver_data;
1199         struct ibmveth_adapter *adapter = netdev->priv;
1200         long value = simple_strtol(buf, NULL, 10);
1201         long rc;
1202
1203         if (attr == &veth_active_attr) {
1204                 if (value && !pool->active) {
1205                         if(ibmveth_alloc_buffer_pool(pool)) {
1206                                 ibmveth_error_printk("unable to alloc pool\n");
1207                                 return -ENOMEM;
1208                         }
1209                         pool->active = 1;
1210                         adapter->pool_config = 1;
1211                         ibmveth_close(netdev);
1212                         adapter->pool_config = 0;
1213                         if ((rc = ibmveth_open(netdev)))
1214                                 return rc;
1215                 } else if (!value && pool->active) {
1216                         int mtu = netdev->mtu + IBMVETH_BUFF_OH;
1217                         int i;
1218                         /* Make sure there is a buffer pool with buffers that
1219                            can hold a packet of the size of the MTU */
1220                         for(i = 0; i<IbmVethNumBufferPools; i++) {
1221                                 if (pool == &adapter->rx_buff_pool[i])
1222                                         continue;
1223                                 if (!adapter->rx_buff_pool[i].active)
1224                                         continue;
1225                                 if (mtu < adapter->rx_buff_pool[i].buff_size) {
1226                                         pool->active = 0;
1227                                         h_free_logical_lan_buffer(adapter->
1228                                                                   vdev->
1229                                                                   unit_address,
1230                                                                   pool->
1231                                                                   buff_size);
1232                                 }
1233                         }
1234                         if (pool->active) {
1235                                 ibmveth_error_printk("no active pool >= MTU\n");
1236                                 return -EPERM;
1237                         }
1238                 }
1239         } else if (attr == &veth_num_attr) {
1240                 if (value <= 0 || value > IBMVETH_MAX_POOL_COUNT)
1241                         return -EINVAL;
1242                 else {
1243                         adapter->pool_config = 1;
1244                         ibmveth_close(netdev);
1245                         adapter->pool_config = 0;
1246                         pool->size = value;
1247                         if ((rc = ibmveth_open(netdev)))
1248                                 return rc;
1249                 }
1250         } else if (attr == &veth_size_attr) {
1251                 if (value <= IBMVETH_BUFF_OH || value > IBMVETH_MAX_BUF_SIZE)
1252                         return -EINVAL;
1253                 else {
1254                         adapter->pool_config = 1;
1255                         ibmveth_close(netdev);
1256                         adapter->pool_config = 0;
1257                         pool->buff_size = value;
1258                         if ((rc = ibmveth_open(netdev)))
1259                                 return rc;
1260                 }
1261         }
1262
1263         /* kick the interrupt handler to allocate/deallocate pools */
1264         ibmveth_interrupt(netdev->irq, netdev, NULL);
1265         return count;
1266 }
1267
1268
1269 #define ATTR(_name, _mode)      \
1270         struct attribute veth_##_name##_attr = {               \
1271         .name = __stringify(_name), .mode = _mode, .owner = THIS_MODULE \
1272         };
1273
1274 static ATTR(active, 0644);
1275 static ATTR(num, 0644);
1276 static ATTR(size, 0644);
1277
1278 static struct attribute * veth_pool_attrs[] = {
1279         &veth_active_attr,
1280         &veth_num_attr,
1281         &veth_size_attr,
1282         NULL,
1283 };
1284
1285 static struct sysfs_ops veth_pool_ops = {
1286         .show   = veth_pool_show,
1287         .store  = veth_pool_store,
1288 };
1289
1290 static struct kobj_type ktype_veth_pool = {
1291         .release        = NULL,
1292         .sysfs_ops      = &veth_pool_ops,
1293         .default_attrs  = veth_pool_attrs,
1294 };
1295
1296
1297 static struct vio_device_id ibmveth_device_table[] __devinitdata= {
1298         { "network", "IBM,l-lan"},
1299         { "", "" }
1300 };
1301 MODULE_DEVICE_TABLE(vio, ibmveth_device_table);
1302
1303 static struct vio_driver ibmveth_driver = {
1304         .id_table       = ibmveth_device_table,
1305         .probe          = ibmveth_probe,
1306         .remove         = ibmveth_remove,
1307         .driver         = {
1308                 .name   = ibmveth_driver_name,
1309                 .owner  = THIS_MODULE,
1310         }
1311 };
1312
1313 static int __init ibmveth_module_init(void)
1314 {
1315         ibmveth_printk("%s: %s %s\n", ibmveth_driver_name, ibmveth_driver_string, ibmveth_driver_version);
1316
1317         ibmveth_proc_register_driver();
1318
1319         return vio_register_driver(&ibmveth_driver);
1320 }
1321
1322 static void __exit ibmveth_module_exit(void)
1323 {
1324         vio_unregister_driver(&ibmveth_driver);
1325         ibmveth_proc_unregister_driver();
1326 }
1327
1328 module_init(ibmveth_module_init);
1329 module_exit(ibmveth_module_exit);