Loading...
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
53static irqreturn_t ibmveth_interrupt(int irq, void *dev_instance);
54static void ibmveth_rxq_harvest_buffer(struct ibmveth_adapter *adapter);
55static unsigned long ibmveth_get_desired_dma(struct vio_dev *vdev);
56
57static struct kobj_type ktype_veth_pool;
58
59
60static const char ibmveth_driver_name[] = "ibmveth";
61static const char ibmveth_driver_string[] = "IBM Power Virtual Ethernet Driver";
62#define ibmveth_driver_version "1.04"
63
64MODULE_AUTHOR("Santiago Leon <santil@linux.vnet.ibm.com>");
65MODULE_DESCRIPTION("IBM Power Virtual Ethernet Driver");
66MODULE_LICENSE("GPL");
67MODULE_VERSION(ibmveth_driver_version);
68
69static unsigned int tx_copybreak __read_mostly = 128;
70module_param(tx_copybreak, uint, 0644);
71MODULE_PARM_DESC(tx_copybreak,
72 "Maximum size of packet that is copied to a new buffer on transmit");
73
74static unsigned int rx_copybreak __read_mostly = 128;
75module_param(rx_copybreak, uint, 0644);
76MODULE_PARM_DESC(rx_copybreak,
77 "Maximum size of packet that is copied to a new buffer on receive");
78
79static unsigned int rx_flush __read_mostly = 0;
80module_param(rx_flush, uint, 0644);
81MODULE_PARM_DESC(rx_flush, "Flush receive buffers before use");
82
83struct 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
91struct 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 */
107static 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
112static 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
118static inline int ibmveth_rxq_pending_buffer(struct ibmveth_adapter *adapter)
119{
120 return ibmveth_rxq_toggle(adapter) == adapter->rx_queue.toggle;
121}
122
123static inline int ibmveth_rxq_buffer_valid(struct ibmveth_adapter *adapter)
124{
125 return ibmveth_rxq_flags(adapter) & IBMVETH_RXQ_VALID;
126}
127
128static inline int ibmveth_rxq_frame_offset(struct ibmveth_adapter *adapter)
129{
130 return ibmveth_rxq_flags(adapter) & IBMVETH_RXQ_OFF_MASK;
131}
132
133static 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
138static 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 */
144static 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 */
156static 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
195static 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 */
206static 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
278failure:
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 */
297static 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 */
316static 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 */
350static 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 */
385static 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 */
398static int 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 int ret = 1;
407
408 BUG_ON(pool >= IBMVETH_NUM_BUFF_POOLS);
409 BUG_ON(index >= adapter->rx_buff_pool[pool].size);
410
411 if (!adapter->rx_buff_pool[pool].active) {
412 ibmveth_rxq_harvest_buffer(adapter);
413 ibmveth_free_buffer_pool(adapter, &adapter->rx_buff_pool[pool]);
414 goto out;
415 }
416
417 desc.fields.flags_len = IBMVETH_BUF_VALID |
418 adapter->rx_buff_pool[pool].buff_size;
419 desc.fields.address = adapter->rx_buff_pool[pool].dma_addr[index];
420
421 lpar_rc = h_add_logical_lan_buffer(adapter->vdev->unit_address, desc.desc);
422
423 if (lpar_rc != H_SUCCESS) {
424 netdev_dbg(adapter->netdev, "h_add_logical_lan_buffer failed "
425 "during recycle rc=%ld", lpar_rc);
426 ibmveth_remove_buffer_from_pool(adapter, adapter->rx_queue.queue_addr[adapter->rx_queue.index].correlator);
427 ret = 0;
428 }
429
430 if (++adapter->rx_queue.index == adapter->rx_queue.num_slots) {
431 adapter->rx_queue.index = 0;
432 adapter->rx_queue.toggle = !adapter->rx_queue.toggle;
433 }
434
435out:
436 return ret;
437}
438
439static void ibmveth_rxq_harvest_buffer(struct ibmveth_adapter *adapter)
440{
441 ibmveth_remove_buffer_from_pool(adapter, adapter->rx_queue.queue_addr[adapter->rx_queue.index].correlator);
442
443 if (++adapter->rx_queue.index == adapter->rx_queue.num_slots) {
444 adapter->rx_queue.index = 0;
445 adapter->rx_queue.toggle = !adapter->rx_queue.toggle;
446 }
447}
448
449static void ibmveth_cleanup(struct ibmveth_adapter *adapter)
450{
451 int i;
452 struct device *dev = &adapter->vdev->dev;
453
454 if (adapter->buffer_list_addr != NULL) {
455 if (!dma_mapping_error(dev, adapter->buffer_list_dma)) {
456 dma_unmap_single(dev, adapter->buffer_list_dma, 4096,
457 DMA_BIDIRECTIONAL);
458 adapter->buffer_list_dma = DMA_ERROR_CODE;
459 }
460 free_page((unsigned long)adapter->buffer_list_addr);
461 adapter->buffer_list_addr = NULL;
462 }
463
464 if (adapter->filter_list_addr != NULL) {
465 if (!dma_mapping_error(dev, adapter->filter_list_dma)) {
466 dma_unmap_single(dev, adapter->filter_list_dma, 4096,
467 DMA_BIDIRECTIONAL);
468 adapter->filter_list_dma = DMA_ERROR_CODE;
469 }
470 free_page((unsigned long)adapter->filter_list_addr);
471 adapter->filter_list_addr = NULL;
472 }
473
474 if (adapter->rx_queue.queue_addr != NULL) {
475 dma_free_coherent(dev, adapter->rx_queue.queue_len,
476 adapter->rx_queue.queue_addr,
477 adapter->rx_queue.queue_dma);
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
499static 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 */
509retry:
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
526static 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 dev = &adapter->vdev->dev;
555
556 adapter->rx_queue.queue_len = sizeof(struct ibmveth_rx_q_entry) *
557 rxq_entries;
558 adapter->rx_queue.queue_addr =
559 dma_alloc_coherent(dev, adapter->rx_queue.queue_len,
560 &adapter->rx_queue.queue_dma, GFP_KERNEL);
561
562 if (!adapter->rx_queue.queue_addr) {
563 netdev_err(netdev, "unable to allocate rx queue pages\n");
564 rc = -ENOMEM;
565 goto err_out;
566 }
567
568 adapter->buffer_list_dma = dma_map_single(dev,
569 adapter->buffer_list_addr, 4096, DMA_BIDIRECTIONAL);
570 adapter->filter_list_dma = dma_map_single(dev,
571 adapter->filter_list_addr, 4096, DMA_BIDIRECTIONAL);
572
573 if ((dma_mapping_error(dev, adapter->buffer_list_dma)) ||
574 (dma_mapping_error(dev, adapter->filter_list_dma))) {
575 netdev_err(netdev, "unable to map filter or buffer list "
576 "pages\n");
577 rc = -ENOMEM;
578 goto err_out;
579 }
580
581 adapter->rx_queue.index = 0;
582 adapter->rx_queue.num_slots = rxq_entries;
583 adapter->rx_queue.toggle = 1;
584
585 memcpy(&mac_address, netdev->dev_addr, netdev->addr_len);
586 mac_address = mac_address >> 16;
587
588 rxq_desc.fields.flags_len = IBMVETH_BUF_VALID |
589 adapter->rx_queue.queue_len;
590 rxq_desc.fields.address = adapter->rx_queue.queue_dma;
591
592 netdev_dbg(netdev, "buffer list @ 0x%p\n", adapter->buffer_list_addr);
593 netdev_dbg(netdev, "filter list @ 0x%p\n", adapter->filter_list_addr);
594 netdev_dbg(netdev, "receive q @ 0x%p\n", adapter->rx_queue.queue_addr);
595
596 h_vio_signal(adapter->vdev->unit_address, VIO_IRQ_DISABLE);
597
598 lpar_rc = ibmveth_register_logical_lan(adapter, rxq_desc, mac_address);
599
600 if (lpar_rc != H_SUCCESS) {
601 netdev_err(netdev, "h_register_logical_lan failed with %ld\n",
602 lpar_rc);
603 netdev_err(netdev, "buffer TCE:0x%llx filter TCE:0x%llx rxq "
604 "desc:0x%llx MAC:0x%llx\n",
605 adapter->buffer_list_dma,
606 adapter->filter_list_dma,
607 rxq_desc.desc,
608 mac_address);
609 rc = -ENONET;
610 goto err_out;
611 }
612
613 for (i = 0; i < IBMVETH_NUM_BUFF_POOLS; i++) {
614 if (!adapter->rx_buff_pool[i].active)
615 continue;
616 if (ibmveth_alloc_buffer_pool(&adapter->rx_buff_pool[i])) {
617 netdev_err(netdev, "unable to alloc pool\n");
618 adapter->rx_buff_pool[i].active = 0;
619 rc = -ENOMEM;
620 goto err_out;
621 }
622 }
623
624 netdev_dbg(netdev, "registering irq 0x%x\n", netdev->irq);
625 rc = request_irq(netdev->irq, ibmveth_interrupt, 0, netdev->name,
626 netdev);
627 if (rc != 0) {
628 netdev_err(netdev, "unable to request irq 0x%x, rc %d\n",
629 netdev->irq, rc);
630 do {
631 lpar_rc = h_free_logical_lan(adapter->vdev->unit_address);
632 } while (H_IS_LONG_BUSY(lpar_rc) || (lpar_rc == H_BUSY));
633
634 goto err_out;
635 }
636
637 adapter->bounce_buffer =
638 kmalloc(netdev->mtu + IBMVETH_BUFF_OH, GFP_KERNEL);
639 if (!adapter->bounce_buffer) {
640 netdev_err(netdev, "unable to allocate bounce buffer\n");
641 rc = -ENOMEM;
642 goto err_out_free_irq;
643 }
644 adapter->bounce_buffer_dma =
645 dma_map_single(&adapter->vdev->dev, adapter->bounce_buffer,
646 netdev->mtu + IBMVETH_BUFF_OH, DMA_BIDIRECTIONAL);
647 if (dma_mapping_error(dev, adapter->bounce_buffer_dma)) {
648 netdev_err(netdev, "unable to map bounce buffer\n");
649 rc = -ENOMEM;
650 goto err_out_free_irq;
651 }
652
653 netdev_dbg(netdev, "initial replenish cycle\n");
654 ibmveth_interrupt(netdev->irq, netdev);
655
656 netif_start_queue(netdev);
657
658 netdev_dbg(netdev, "open complete\n");
659
660 return 0;
661
662err_out_free_irq:
663 free_irq(netdev->irq, netdev);
664err_out:
665 ibmveth_cleanup(adapter);
666 napi_disable(&adapter->napi);
667 return rc;
668}
669
670static int ibmveth_close(struct net_device *netdev)
671{
672 struct ibmveth_adapter *adapter = netdev_priv(netdev);
673 long lpar_rc;
674
675 netdev_dbg(netdev, "close starting\n");
676
677 napi_disable(&adapter->napi);
678
679 if (!adapter->pool_config)
680 netif_stop_queue(netdev);
681
682 h_vio_signal(adapter->vdev->unit_address, VIO_IRQ_DISABLE);
683
684 do {
685 lpar_rc = h_free_logical_lan(adapter->vdev->unit_address);
686 } while (H_IS_LONG_BUSY(lpar_rc) || (lpar_rc == H_BUSY));
687
688 if (lpar_rc != H_SUCCESS) {
689 netdev_err(netdev, "h_free_logical_lan failed with %lx, "
690 "continuing with close\n", lpar_rc);
691 }
692
693 free_irq(netdev->irq, netdev);
694
695 adapter->rx_no_buffer = *(u64 *)(((char *)adapter->buffer_list_addr) +
696 4096 - 8);
697
698 ibmveth_cleanup(adapter);
699
700 netdev_dbg(netdev, "close complete\n");
701
702 return 0;
703}
704
705static int netdev_get_settings(struct net_device *dev, struct ethtool_cmd *cmd)
706{
707 cmd->supported = (SUPPORTED_1000baseT_Full | SUPPORTED_Autoneg |
708 SUPPORTED_FIBRE);
709 cmd->advertising = (ADVERTISED_1000baseT_Full | ADVERTISED_Autoneg |
710 ADVERTISED_FIBRE);
711 ethtool_cmd_speed_set(cmd, SPEED_1000);
712 cmd->duplex = DUPLEX_FULL;
713 cmd->port = PORT_FIBRE;
714 cmd->phy_address = 0;
715 cmd->transceiver = XCVR_INTERNAL;
716 cmd->autoneg = AUTONEG_ENABLE;
717 cmd->maxtxpkt = 0;
718 cmd->maxrxpkt = 1;
719 return 0;
720}
721
722static void netdev_get_drvinfo(struct net_device *dev,
723 struct ethtool_drvinfo *info)
724{
725 strncpy(info->driver, ibmveth_driver_name, sizeof(info->driver) - 1);
726 strncpy(info->version, ibmveth_driver_version,
727 sizeof(info->version) - 1);
728}
729
730static netdev_features_t ibmveth_fix_features(struct net_device *dev,
731 netdev_features_t features)
732{
733 /*
734 * Since the ibmveth firmware interface does not have the
735 * concept of separate tx/rx checksum offload enable, if rx
736 * checksum is disabled we also have to disable tx checksum
737 * offload. Once we disable rx checksum offload, we are no
738 * longer allowed to send tx buffers that are not properly
739 * checksummed.
740 */
741
742 if (!(features & NETIF_F_RXCSUM))
743 features &= ~NETIF_F_ALL_CSUM;
744
745 return features;
746}
747
748static int ibmveth_set_csum_offload(struct net_device *dev, u32 data)
749{
750 struct ibmveth_adapter *adapter = netdev_priv(dev);
751 unsigned long set_attr, clr_attr, ret_attr;
752 unsigned long set_attr6, clr_attr6;
753 long ret, ret4, ret6;
754 int rc1 = 0, rc2 = 0;
755 int restart = 0;
756
757 if (netif_running(dev)) {
758 restart = 1;
759 adapter->pool_config = 1;
760 ibmveth_close(dev);
761 adapter->pool_config = 0;
762 }
763
764 set_attr = 0;
765 clr_attr = 0;
766 set_attr6 = 0;
767 clr_attr6 = 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 ret4 = h_illan_attributes(adapter->vdev->unit_address, clr_attr,
783 set_attr, &ret_attr);
784
785 if (ret4 != H_SUCCESS) {
786 netdev_err(dev, "unable to change IPv4 checksum "
787 "offload settings. %d rc=%ld\n",
788 data, ret4);
789
790 h_illan_attributes(adapter->vdev->unit_address,
791 set_attr, clr_attr, &ret_attr);
792
793 if (data == 1)
794 dev->features &= ~NETIF_F_IP_CSUM;
795
796 } else {
797 adapter->fw_ipv4_csum_support = data;
798 }
799
800 ret6 = h_illan_attributes(adapter->vdev->unit_address,
801 clr_attr6, set_attr6, &ret_attr);
802
803 if (ret6 != H_SUCCESS) {
804 netdev_err(dev, "unable to change IPv6 checksum "
805 "offload settings. %d rc=%ld\n",
806 data, ret6);
807
808 h_illan_attributes(adapter->vdev->unit_address,
809 set_attr6, clr_attr6, &ret_attr);
810
811 if (data == 1)
812 dev->features &= ~NETIF_F_IPV6_CSUM;
813
814 } else
815 adapter->fw_ipv6_csum_support = data;
816
817 if (ret4 == H_SUCCESS || ret6 == H_SUCCESS)
818 adapter->rx_csum = data;
819 else
820 rc1 = -EIO;
821 } else {
822 rc1 = -EIO;
823 netdev_err(dev, "unable to change checksum offload settings."
824 " %d rc=%ld ret_attr=%lx\n", data, ret,
825 ret_attr);
826 }
827
828 if (restart)
829 rc2 = ibmveth_open(dev);
830
831 return rc1 ? rc1 : rc2;
832}
833
834static int ibmveth_set_features(struct net_device *dev,
835 netdev_features_t features)
836{
837 struct ibmveth_adapter *adapter = netdev_priv(dev);
838 int rx_csum = !!(features & NETIF_F_RXCSUM);
839 int rc;
840
841 if (rx_csum == adapter->rx_csum)
842 return 0;
843
844 rc = ibmveth_set_csum_offload(dev, rx_csum);
845 if (rc && !adapter->rx_csum)
846 dev->features = features & ~(NETIF_F_ALL_CSUM | NETIF_F_RXCSUM);
847
848 return rc;
849}
850
851static void ibmveth_get_strings(struct net_device *dev, u32 stringset, u8 *data)
852{
853 int i;
854
855 if (stringset != ETH_SS_STATS)
856 return;
857
858 for (i = 0; i < ARRAY_SIZE(ibmveth_stats); i++, data += ETH_GSTRING_LEN)
859 memcpy(data, ibmveth_stats[i].name, ETH_GSTRING_LEN);
860}
861
862static int ibmveth_get_sset_count(struct net_device *dev, int sset)
863{
864 switch (sset) {
865 case ETH_SS_STATS:
866 return ARRAY_SIZE(ibmveth_stats);
867 default:
868 return -EOPNOTSUPP;
869 }
870}
871
872static void ibmveth_get_ethtool_stats(struct net_device *dev,
873 struct ethtool_stats *stats, u64 *data)
874{
875 int i;
876 struct ibmveth_adapter *adapter = netdev_priv(dev);
877
878 for (i = 0; i < ARRAY_SIZE(ibmveth_stats); i++)
879 data[i] = IBMVETH_GET_STAT(adapter, ibmveth_stats[i].offset);
880}
881
882static const struct ethtool_ops netdev_ethtool_ops = {
883 .get_drvinfo = netdev_get_drvinfo,
884 .get_settings = netdev_get_settings,
885 .get_link = ethtool_op_get_link,
886 .get_strings = ibmveth_get_strings,
887 .get_sset_count = ibmveth_get_sset_count,
888 .get_ethtool_stats = ibmveth_get_ethtool_stats,
889};
890
891static int ibmveth_ioctl(struct net_device *dev, struct ifreq *ifr, int cmd)
892{
893 return -EOPNOTSUPP;
894}
895
896#define page_offset(v) ((unsigned long)(v) & ((1 << 12) - 1))
897
898static int ibmveth_send(struct ibmveth_adapter *adapter,
899 union ibmveth_buf_desc *descs)
900{
901 unsigned long correlator;
902 unsigned int retry_count;
903 unsigned long ret;
904
905 /*
906 * The retry count sets a maximum for the number of broadcast and
907 * multicast destinations within the system.
908 */
909 retry_count = 1024;
910 correlator = 0;
911 do {
912 ret = h_send_logical_lan(adapter->vdev->unit_address,
913 descs[0].desc, descs[1].desc,
914 descs[2].desc, descs[3].desc,
915 descs[4].desc, descs[5].desc,
916 correlator, &correlator);
917 } while ((ret == H_BUSY) && (retry_count--));
918
919 if (ret != H_SUCCESS && ret != H_DROPPED) {
920 netdev_err(adapter->netdev, "tx: h_send_logical_lan failed "
921 "with rc=%ld\n", ret);
922 return 1;
923 }
924
925 return 0;
926}
927
928static netdev_tx_t ibmveth_start_xmit(struct sk_buff *skb,
929 struct net_device *netdev)
930{
931 struct ibmveth_adapter *adapter = netdev_priv(netdev);
932 unsigned int desc_flags;
933 union ibmveth_buf_desc descs[6];
934 int last, i;
935 int force_bounce = 0;
936 dma_addr_t dma_addr;
937
938 /*
939 * veth handles a maximum of 6 segments including the header, so
940 * we have to linearize the skb if there are more than this.
941 */
942 if (skb_shinfo(skb)->nr_frags > 5 && __skb_linearize(skb)) {
943 netdev->stats.tx_dropped++;
944 goto out;
945 }
946
947 /* veth can't checksum offload UDP */
948 if (skb->ip_summed == CHECKSUM_PARTIAL &&
949 ((skb->protocol == htons(ETH_P_IP) &&
950 ip_hdr(skb)->protocol != IPPROTO_TCP) ||
951 (skb->protocol == htons(ETH_P_IPV6) &&
952 ipv6_hdr(skb)->nexthdr != IPPROTO_TCP)) &&
953 skb_checksum_help(skb)) {
954
955 netdev_err(netdev, "tx: failed to checksum packet\n");
956 netdev->stats.tx_dropped++;
957 goto out;
958 }
959
960 desc_flags = IBMVETH_BUF_VALID;
961
962 if (skb->ip_summed == CHECKSUM_PARTIAL) {
963 unsigned char *buf = skb_transport_header(skb) +
964 skb->csum_offset;
965
966 desc_flags |= (IBMVETH_BUF_NO_CSUM | IBMVETH_BUF_CSUM_GOOD);
967
968 /* Need to zero out the checksum */
969 buf[0] = 0;
970 buf[1] = 0;
971 }
972
973retry_bounce:
974 memset(descs, 0, sizeof(descs));
975
976 /*
977 * If a linear packet is below the rx threshold then
978 * copy it into the static bounce buffer. This avoids the
979 * cost of a TCE insert and remove.
980 */
981 if (force_bounce || (!skb_is_nonlinear(skb) &&
982 (skb->len < tx_copybreak))) {
983 skb_copy_from_linear_data(skb, adapter->bounce_buffer,
984 skb->len);
985
986 descs[0].fields.flags_len = desc_flags | skb->len;
987 descs[0].fields.address = adapter->bounce_buffer_dma;
988
989 if (ibmveth_send(adapter, descs)) {
990 adapter->tx_send_failed++;
991 netdev->stats.tx_dropped++;
992 } else {
993 netdev->stats.tx_packets++;
994 netdev->stats.tx_bytes += skb->len;
995 }
996
997 goto out;
998 }
999
1000 /* Map the header */
1001 dma_addr = dma_map_single(&adapter->vdev->dev, skb->data,
1002 skb_headlen(skb), DMA_TO_DEVICE);
1003 if (dma_mapping_error(&adapter->vdev->dev, dma_addr))
1004 goto map_failed;
1005
1006 descs[0].fields.flags_len = desc_flags | skb_headlen(skb);
1007 descs[0].fields.address = dma_addr;
1008
1009 /* Map the frags */
1010 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
1011 const skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
1012
1013 dma_addr = skb_frag_dma_map(&adapter->vdev->dev, frag, 0,
1014 skb_frag_size(frag), DMA_TO_DEVICE);
1015
1016 if (dma_mapping_error(&adapter->vdev->dev, dma_addr))
1017 goto map_failed_frags;
1018
1019 descs[i+1].fields.flags_len = desc_flags | skb_frag_size(frag);
1020 descs[i+1].fields.address = dma_addr;
1021 }
1022
1023 if (ibmveth_send(adapter, descs)) {
1024 adapter->tx_send_failed++;
1025 netdev->stats.tx_dropped++;
1026 } else {
1027 netdev->stats.tx_packets++;
1028 netdev->stats.tx_bytes += skb->len;
1029 }
1030
1031 dma_unmap_single(&adapter->vdev->dev,
1032 descs[0].fields.address,
1033 descs[0].fields.flags_len & IBMVETH_BUF_LEN_MASK,
1034 DMA_TO_DEVICE);
1035
1036 for (i = 1; i < skb_shinfo(skb)->nr_frags + 1; i++)
1037 dma_unmap_page(&adapter->vdev->dev, descs[i].fields.address,
1038 descs[i].fields.flags_len & IBMVETH_BUF_LEN_MASK,
1039 DMA_TO_DEVICE);
1040
1041out:
1042 dev_kfree_skb(skb);
1043 return NETDEV_TX_OK;
1044
1045map_failed_frags:
1046 last = i+1;
1047 for (i = 0; i < last; i++)
1048 dma_unmap_page(&adapter->vdev->dev, descs[i].fields.address,
1049 descs[i].fields.flags_len & IBMVETH_BUF_LEN_MASK,
1050 DMA_TO_DEVICE);
1051
1052map_failed:
1053 if (!firmware_has_feature(FW_FEATURE_CMO))
1054 netdev_err(netdev, "tx: unable to map xmit buffer\n");
1055 adapter->tx_map_failed++;
1056 skb_linearize(skb);
1057 force_bounce = 1;
1058 goto retry_bounce;
1059}
1060
1061static int ibmveth_poll(struct napi_struct *napi, int budget)
1062{
1063 struct ibmveth_adapter *adapter =
1064 container_of(napi, struct ibmveth_adapter, napi);
1065 struct net_device *netdev = adapter->netdev;
1066 int frames_processed = 0;
1067 unsigned long lpar_rc;
1068
1069restart_poll:
1070 do {
1071 if (!ibmveth_rxq_pending_buffer(adapter))
1072 break;
1073
1074 smp_rmb();
1075 if (!ibmveth_rxq_buffer_valid(adapter)) {
1076 wmb(); /* suggested by larson1 */
1077 adapter->rx_invalid_buffer++;
1078 netdev_dbg(netdev, "recycling invalid buffer\n");
1079 ibmveth_rxq_recycle_buffer(adapter);
1080 } else {
1081 struct sk_buff *skb, *new_skb;
1082 int length = ibmveth_rxq_frame_length(adapter);
1083 int offset = ibmveth_rxq_frame_offset(adapter);
1084 int csum_good = ibmveth_rxq_csum_good(adapter);
1085
1086 skb = ibmveth_rxq_get_buffer(adapter);
1087
1088 new_skb = NULL;
1089 if (length < rx_copybreak)
1090 new_skb = netdev_alloc_skb(netdev, length);
1091
1092 if (new_skb) {
1093 skb_copy_to_linear_data(new_skb,
1094 skb->data + offset,
1095 length);
1096 if (rx_flush)
1097 ibmveth_flush_buffer(skb->data,
1098 length + offset);
1099 if (!ibmveth_rxq_recycle_buffer(adapter))
1100 kfree_skb(skb);
1101 skb = new_skb;
1102 } else {
1103 ibmveth_rxq_harvest_buffer(adapter);
1104 skb_reserve(skb, offset);
1105 }
1106
1107 skb_put(skb, length);
1108 skb->protocol = eth_type_trans(skb, netdev);
1109
1110 if (csum_good)
1111 skb->ip_summed = CHECKSUM_UNNECESSARY;
1112
1113 netif_receive_skb(skb); /* send it up */
1114
1115 netdev->stats.rx_packets++;
1116 netdev->stats.rx_bytes += length;
1117 frames_processed++;
1118 }
1119 } while (frames_processed < budget);
1120
1121 ibmveth_replenish_task(adapter);
1122
1123 if (frames_processed < budget) {
1124 /* We think we are done - reenable interrupts,
1125 * then check once more to make sure we are done.
1126 */
1127 lpar_rc = h_vio_signal(adapter->vdev->unit_address,
1128 VIO_IRQ_ENABLE);
1129
1130 BUG_ON(lpar_rc != H_SUCCESS);
1131
1132 napi_complete(napi);
1133
1134 if (ibmveth_rxq_pending_buffer(adapter) &&
1135 napi_reschedule(napi)) {
1136 lpar_rc = h_vio_signal(adapter->vdev->unit_address,
1137 VIO_IRQ_DISABLE);
1138 goto restart_poll;
1139 }
1140 }
1141
1142 return frames_processed;
1143}
1144
1145static irqreturn_t ibmveth_interrupt(int irq, void *dev_instance)
1146{
1147 struct net_device *netdev = dev_instance;
1148 struct ibmveth_adapter *adapter = netdev_priv(netdev);
1149 unsigned long lpar_rc;
1150
1151 if (napi_schedule_prep(&adapter->napi)) {
1152 lpar_rc = h_vio_signal(adapter->vdev->unit_address,
1153 VIO_IRQ_DISABLE);
1154 BUG_ON(lpar_rc != H_SUCCESS);
1155 __napi_schedule(&adapter->napi);
1156 }
1157 return IRQ_HANDLED;
1158}
1159
1160static void ibmveth_set_multicast_list(struct net_device *netdev)
1161{
1162 struct ibmveth_adapter *adapter = netdev_priv(netdev);
1163 unsigned long lpar_rc;
1164
1165 if ((netdev->flags & IFF_PROMISC) ||
1166 (netdev_mc_count(netdev) > adapter->mcastFilterSize)) {
1167 lpar_rc = h_multicast_ctrl(adapter->vdev->unit_address,
1168 IbmVethMcastEnableRecv |
1169 IbmVethMcastDisableFiltering,
1170 0);
1171 if (lpar_rc != H_SUCCESS) {
1172 netdev_err(netdev, "h_multicast_ctrl rc=%ld when "
1173 "entering promisc mode\n", lpar_rc);
1174 }
1175 } else {
1176 struct netdev_hw_addr *ha;
1177 /* clear the filter table & disable filtering */
1178 lpar_rc = h_multicast_ctrl(adapter->vdev->unit_address,
1179 IbmVethMcastEnableRecv |
1180 IbmVethMcastDisableFiltering |
1181 IbmVethMcastClearFilterTable,
1182 0);
1183 if (lpar_rc != H_SUCCESS) {
1184 netdev_err(netdev, "h_multicast_ctrl rc=%ld when "
1185 "attempting to clear filter table\n",
1186 lpar_rc);
1187 }
1188 /* add the addresses to the filter table */
1189 netdev_for_each_mc_addr(ha, netdev) {
1190 /* add the multicast address to the filter table */
1191 unsigned long mcast_addr = 0;
1192 memcpy(((char *)&mcast_addr)+2, ha->addr, 6);
1193 lpar_rc = h_multicast_ctrl(adapter->vdev->unit_address,
1194 IbmVethMcastAddFilter,
1195 mcast_addr);
1196 if (lpar_rc != H_SUCCESS) {
1197 netdev_err(netdev, "h_multicast_ctrl rc=%ld "
1198 "when adding an entry to the filter "
1199 "table\n", lpar_rc);
1200 }
1201 }
1202
1203 /* re-enable filtering */
1204 lpar_rc = h_multicast_ctrl(adapter->vdev->unit_address,
1205 IbmVethMcastEnableFiltering,
1206 0);
1207 if (lpar_rc != H_SUCCESS) {
1208 netdev_err(netdev, "h_multicast_ctrl rc=%ld when "
1209 "enabling filtering\n", lpar_rc);
1210 }
1211 }
1212}
1213
1214static int ibmveth_change_mtu(struct net_device *dev, int new_mtu)
1215{
1216 struct ibmveth_adapter *adapter = netdev_priv(dev);
1217 struct vio_dev *viodev = adapter->vdev;
1218 int new_mtu_oh = new_mtu + IBMVETH_BUFF_OH;
1219 int i, rc;
1220 int need_restart = 0;
1221
1222 if (new_mtu < IBMVETH_MIN_MTU)
1223 return -EINVAL;
1224
1225 for (i = 0; i < IBMVETH_NUM_BUFF_POOLS; i++)
1226 if (new_mtu_oh < adapter->rx_buff_pool[i].buff_size)
1227 break;
1228
1229 if (i == IBMVETH_NUM_BUFF_POOLS)
1230 return -EINVAL;
1231
1232 /* Deactivate all the buffer pools so that the next loop can activate
1233 only the buffer pools necessary to hold the new MTU */
1234 if (netif_running(adapter->netdev)) {
1235 need_restart = 1;
1236 adapter->pool_config = 1;
1237 ibmveth_close(adapter->netdev);
1238 adapter->pool_config = 0;
1239 }
1240
1241 /* Look for an active buffer pool that can hold the new MTU */
1242 for (i = 0; i < IBMVETH_NUM_BUFF_POOLS; i++) {
1243 adapter->rx_buff_pool[i].active = 1;
1244
1245 if (new_mtu_oh < adapter->rx_buff_pool[i].buff_size) {
1246 dev->mtu = new_mtu;
1247 vio_cmo_set_dev_desired(viodev,
1248 ibmveth_get_desired_dma
1249 (viodev));
1250 if (need_restart) {
1251 return ibmveth_open(adapter->netdev);
1252 }
1253 return 0;
1254 }
1255 }
1256
1257 if (need_restart && (rc = ibmveth_open(adapter->netdev)))
1258 return rc;
1259
1260 return -EINVAL;
1261}
1262
1263#ifdef CONFIG_NET_POLL_CONTROLLER
1264static void ibmveth_poll_controller(struct net_device *dev)
1265{
1266 ibmveth_replenish_task(netdev_priv(dev));
1267 ibmveth_interrupt(dev->irq, dev);
1268}
1269#endif
1270
1271/**
1272 * ibmveth_get_desired_dma - Calculate IO memory desired by the driver
1273 *
1274 * @vdev: struct vio_dev for the device whose desired IO mem is to be returned
1275 *
1276 * Return value:
1277 * Number of bytes of IO data the driver will need to perform well.
1278 */
1279static unsigned long ibmveth_get_desired_dma(struct vio_dev *vdev)
1280{
1281 struct net_device *netdev = dev_get_drvdata(&vdev->dev);
1282 struct ibmveth_adapter *adapter;
1283 unsigned long ret;
1284 int i;
1285 int rxqentries = 1;
1286
1287 /* netdev inits at probe time along with the structures we need below*/
1288 if (netdev == NULL)
1289 return IOMMU_PAGE_ALIGN(IBMVETH_IO_ENTITLEMENT_DEFAULT);
1290
1291 adapter = netdev_priv(netdev);
1292
1293 ret = IBMVETH_BUFF_LIST_SIZE + IBMVETH_FILT_LIST_SIZE;
1294 ret += IOMMU_PAGE_ALIGN(netdev->mtu);
1295
1296 for (i = 0; i < IBMVETH_NUM_BUFF_POOLS; i++) {
1297 /* add the size of the active receive buffers */
1298 if (adapter->rx_buff_pool[i].active)
1299 ret +=
1300 adapter->rx_buff_pool[i].size *
1301 IOMMU_PAGE_ALIGN(adapter->rx_buff_pool[i].
1302 buff_size);
1303 rxqentries += adapter->rx_buff_pool[i].size;
1304 }
1305 /* add the size of the receive queue entries */
1306 ret += IOMMU_PAGE_ALIGN(rxqentries * sizeof(struct ibmveth_rx_q_entry));
1307
1308 return ret;
1309}
1310
1311static const struct net_device_ops ibmveth_netdev_ops = {
1312 .ndo_open = ibmveth_open,
1313 .ndo_stop = ibmveth_close,
1314 .ndo_start_xmit = ibmveth_start_xmit,
1315 .ndo_set_rx_mode = ibmveth_set_multicast_list,
1316 .ndo_do_ioctl = ibmveth_ioctl,
1317 .ndo_change_mtu = ibmveth_change_mtu,
1318 .ndo_fix_features = ibmveth_fix_features,
1319 .ndo_set_features = ibmveth_set_features,
1320 .ndo_validate_addr = eth_validate_addr,
1321 .ndo_set_mac_address = eth_mac_addr,
1322#ifdef CONFIG_NET_POLL_CONTROLLER
1323 .ndo_poll_controller = ibmveth_poll_controller,
1324#endif
1325};
1326
1327static int __devinit ibmveth_probe(struct vio_dev *dev,
1328 const struct vio_device_id *id)
1329{
1330 int rc, i;
1331 struct net_device *netdev;
1332 struct ibmveth_adapter *adapter;
1333 unsigned char *mac_addr_p;
1334 unsigned int *mcastFilterSize_p;
1335
1336 dev_dbg(&dev->dev, "entering ibmveth_probe for UA 0x%x\n",
1337 dev->unit_address);
1338
1339 mac_addr_p = (unsigned char *)vio_get_attribute(dev, VETH_MAC_ADDR,
1340 NULL);
1341 if (!mac_addr_p) {
1342 dev_err(&dev->dev, "Can't find VETH_MAC_ADDR attribute\n");
1343 return -EINVAL;
1344 }
1345
1346 mcastFilterSize_p = (unsigned int *)vio_get_attribute(dev,
1347 VETH_MCAST_FILTER_SIZE, NULL);
1348 if (!mcastFilterSize_p) {
1349 dev_err(&dev->dev, "Can't find VETH_MCAST_FILTER_SIZE "
1350 "attribute\n");
1351 return -EINVAL;
1352 }
1353
1354 netdev = alloc_etherdev(sizeof(struct ibmveth_adapter));
1355
1356 if (!netdev)
1357 return -ENOMEM;
1358
1359 adapter = netdev_priv(netdev);
1360 dev_set_drvdata(&dev->dev, netdev);
1361
1362 adapter->vdev = dev;
1363 adapter->netdev = netdev;
1364 adapter->mcastFilterSize = *mcastFilterSize_p;
1365 adapter->pool_config = 0;
1366
1367 netif_napi_add(netdev, &adapter->napi, ibmveth_poll, 16);
1368
1369 /*
1370 * Some older boxes running PHYP non-natively have an OF that returns
1371 * a 8-byte local-mac-address field (and the first 2 bytes have to be
1372 * ignored) while newer boxes' OF return a 6-byte field. Note that
1373 * IEEE 1275 specifies that local-mac-address must be a 6-byte field.
1374 * The RPA doc specifies that the first byte must be 10b, so we'll
1375 * just look for it to solve this 8 vs. 6 byte field issue
1376 */
1377 if ((*mac_addr_p & 0x3) != 0x02)
1378 mac_addr_p += 2;
1379
1380 adapter->mac_addr = 0;
1381 memcpy(&adapter->mac_addr, mac_addr_p, 6);
1382
1383 netdev->irq = dev->irq;
1384 netdev->netdev_ops = &ibmveth_netdev_ops;
1385 netdev->ethtool_ops = &netdev_ethtool_ops;
1386 SET_NETDEV_DEV(netdev, &dev->dev);
1387 netdev->hw_features = NETIF_F_SG | NETIF_F_RXCSUM |
1388 NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM;
1389 netdev->features |= netdev->hw_features;
1390
1391 memcpy(netdev->dev_addr, &adapter->mac_addr, netdev->addr_len);
1392
1393 for (i = 0; i < IBMVETH_NUM_BUFF_POOLS; i++) {
1394 struct kobject *kobj = &adapter->rx_buff_pool[i].kobj;
1395 int error;
1396
1397 ibmveth_init_buffer_pool(&adapter->rx_buff_pool[i], i,
1398 pool_count[i], pool_size[i],
1399 pool_active[i]);
1400 error = kobject_init_and_add(kobj, &ktype_veth_pool,
1401 &dev->dev.kobj, "pool%d", i);
1402 if (!error)
1403 kobject_uevent(kobj, KOBJ_ADD);
1404 }
1405
1406 netdev_dbg(netdev, "adapter @ 0x%p\n", adapter);
1407
1408 adapter->buffer_list_dma = DMA_ERROR_CODE;
1409 adapter->filter_list_dma = DMA_ERROR_CODE;
1410 adapter->rx_queue.queue_dma = DMA_ERROR_CODE;
1411
1412 netdev_dbg(netdev, "registering netdev...\n");
1413
1414 ibmveth_set_features(netdev, netdev->features);
1415
1416 rc = register_netdev(netdev);
1417
1418 if (rc) {
1419 netdev_dbg(netdev, "failed to register netdev rc=%d\n", rc);
1420 free_netdev(netdev);
1421 return rc;
1422 }
1423
1424 netdev_dbg(netdev, "registered\n");
1425
1426 return 0;
1427}
1428
1429static int __devexit ibmveth_remove(struct vio_dev *dev)
1430{
1431 struct net_device *netdev = dev_get_drvdata(&dev->dev);
1432 struct ibmveth_adapter *adapter = netdev_priv(netdev);
1433 int i;
1434
1435 for (i = 0; i < IBMVETH_NUM_BUFF_POOLS; i++)
1436 kobject_put(&adapter->rx_buff_pool[i].kobj);
1437
1438 unregister_netdev(netdev);
1439
1440 free_netdev(netdev);
1441 dev_set_drvdata(&dev->dev, NULL);
1442
1443 return 0;
1444}
1445
1446static struct attribute veth_active_attr;
1447static struct attribute veth_num_attr;
1448static struct attribute veth_size_attr;
1449
1450static ssize_t veth_pool_show(struct kobject *kobj,
1451 struct attribute *attr, char *buf)
1452{
1453 struct ibmveth_buff_pool *pool = container_of(kobj,
1454 struct ibmveth_buff_pool,
1455 kobj);
1456
1457 if (attr == &veth_active_attr)
1458 return sprintf(buf, "%d\n", pool->active);
1459 else if (attr == &veth_num_attr)
1460 return sprintf(buf, "%d\n", pool->size);
1461 else if (attr == &veth_size_attr)
1462 return sprintf(buf, "%d\n", pool->buff_size);
1463 return 0;
1464}
1465
1466static ssize_t veth_pool_store(struct kobject *kobj, struct attribute *attr,
1467 const char *buf, size_t count)
1468{
1469 struct ibmveth_buff_pool *pool = container_of(kobj,
1470 struct ibmveth_buff_pool,
1471 kobj);
1472 struct net_device *netdev = dev_get_drvdata(
1473 container_of(kobj->parent, struct device, kobj));
1474 struct ibmveth_adapter *adapter = netdev_priv(netdev);
1475 long value = simple_strtol(buf, NULL, 10);
1476 long rc;
1477
1478 if (attr == &veth_active_attr) {
1479 if (value && !pool->active) {
1480 if (netif_running(netdev)) {
1481 if (ibmveth_alloc_buffer_pool(pool)) {
1482 netdev_err(netdev,
1483 "unable to alloc pool\n");
1484 return -ENOMEM;
1485 }
1486 pool->active = 1;
1487 adapter->pool_config = 1;
1488 ibmveth_close(netdev);
1489 adapter->pool_config = 0;
1490 if ((rc = ibmveth_open(netdev)))
1491 return rc;
1492 } else {
1493 pool->active = 1;
1494 }
1495 } else if (!value && pool->active) {
1496 int mtu = netdev->mtu + IBMVETH_BUFF_OH;
1497 int i;
1498 /* Make sure there is a buffer pool with buffers that
1499 can hold a packet of the size of the MTU */
1500 for (i = 0; i < IBMVETH_NUM_BUFF_POOLS; i++) {
1501 if (pool == &adapter->rx_buff_pool[i])
1502 continue;
1503 if (!adapter->rx_buff_pool[i].active)
1504 continue;
1505 if (mtu <= adapter->rx_buff_pool[i].buff_size)
1506 break;
1507 }
1508
1509 if (i == IBMVETH_NUM_BUFF_POOLS) {
1510 netdev_err(netdev, "no active pool >= MTU\n");
1511 return -EPERM;
1512 }
1513
1514 if (netif_running(netdev)) {
1515 adapter->pool_config = 1;
1516 ibmveth_close(netdev);
1517 pool->active = 0;
1518 adapter->pool_config = 0;
1519 if ((rc = ibmveth_open(netdev)))
1520 return rc;
1521 }
1522 pool->active = 0;
1523 }
1524 } else if (attr == &veth_num_attr) {
1525 if (value <= 0 || value > IBMVETH_MAX_POOL_COUNT) {
1526 return -EINVAL;
1527 } else {
1528 if (netif_running(netdev)) {
1529 adapter->pool_config = 1;
1530 ibmveth_close(netdev);
1531 adapter->pool_config = 0;
1532 pool->size = value;
1533 if ((rc = ibmveth_open(netdev)))
1534 return rc;
1535 } else {
1536 pool->size = value;
1537 }
1538 }
1539 } else if (attr == &veth_size_attr) {
1540 if (value <= IBMVETH_BUFF_OH || value > IBMVETH_MAX_BUF_SIZE) {
1541 return -EINVAL;
1542 } else {
1543 if (netif_running(netdev)) {
1544 adapter->pool_config = 1;
1545 ibmveth_close(netdev);
1546 adapter->pool_config = 0;
1547 pool->buff_size = value;
1548 if ((rc = ibmveth_open(netdev)))
1549 return rc;
1550 } else {
1551 pool->buff_size = value;
1552 }
1553 }
1554 }
1555
1556 /* kick the interrupt handler to allocate/deallocate pools */
1557 ibmveth_interrupt(netdev->irq, netdev);
1558 return count;
1559}
1560
1561
1562#define ATTR(_name, _mode) \
1563 struct attribute veth_##_name##_attr = { \
1564 .name = __stringify(_name), .mode = _mode, \
1565 };
1566
1567static ATTR(active, 0644);
1568static ATTR(num, 0644);
1569static ATTR(size, 0644);
1570
1571static struct attribute *veth_pool_attrs[] = {
1572 &veth_active_attr,
1573 &veth_num_attr,
1574 &veth_size_attr,
1575 NULL,
1576};
1577
1578static const struct sysfs_ops veth_pool_ops = {
1579 .show = veth_pool_show,
1580 .store = veth_pool_store,
1581};
1582
1583static struct kobj_type ktype_veth_pool = {
1584 .release = NULL,
1585 .sysfs_ops = &veth_pool_ops,
1586 .default_attrs = veth_pool_attrs,
1587};
1588
1589static int ibmveth_resume(struct device *dev)
1590{
1591 struct net_device *netdev = dev_get_drvdata(dev);
1592 ibmveth_interrupt(netdev->irq, netdev);
1593 return 0;
1594}
1595
1596static struct vio_device_id ibmveth_device_table[] __devinitdata = {
1597 { "network", "IBM,l-lan"},
1598 { "", "" }
1599};
1600MODULE_DEVICE_TABLE(vio, ibmveth_device_table);
1601
1602static struct dev_pm_ops ibmveth_pm_ops = {
1603 .resume = ibmveth_resume
1604};
1605
1606static struct vio_driver ibmveth_driver = {
1607 .id_table = ibmveth_device_table,
1608 .probe = ibmveth_probe,
1609 .remove = ibmveth_remove,
1610 .get_desired_dma = ibmveth_get_desired_dma,
1611 .name = ibmveth_driver_name,
1612 .pm = &ibmveth_pm_ops,
1613};
1614
1615static int __init ibmveth_module_init(void)
1616{
1617 printk(KERN_DEBUG "%s: %s %s\n", ibmveth_driver_name,
1618 ibmveth_driver_string, ibmveth_driver_version);
1619
1620 return vio_register_driver(&ibmveth_driver);
1621}
1622
1623static void __exit ibmveth_module_exit(void)
1624{
1625 vio_unregister_driver(&ibmveth_driver);
1626}
1627
1628module_init(ibmveth_module_init);
1629module_exit(ibmveth_module_exit);
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, see <http://www.gnu.org/licenses/>.
16 *
17 * Copyright (C) IBM Corporation, 2003, 2010
18 *
19 * Authors: Dave Larson <larson1@us.ibm.com>
20 * Santiago Leon <santil@linux.vnet.ibm.com>
21 * Brian King <brking@linux.vnet.ibm.com>
22 * Robert Jennings <rcj@linux.vnet.ibm.com>
23 * Anton Blanchard <anton@au.ibm.com>
24 */
25
26#include <linux/module.h>
27#include <linux/moduleparam.h>
28#include <linux/types.h>
29#include <linux/errno.h>
30#include <linux/dma-mapping.h>
31#include <linux/kernel.h>
32#include <linux/netdevice.h>
33#include <linux/etherdevice.h>
34#include <linux/skbuff.h>
35#include <linux/init.h>
36#include <linux/interrupt.h>
37#include <linux/mm.h>
38#include <linux/pm.h>
39#include <linux/ethtool.h>
40#include <linux/in.h>
41#include <linux/ip.h>
42#include <linux/ipv6.h>
43#include <linux/slab.h>
44#include <asm/hvcall.h>
45#include <linux/atomic.h>
46#include <asm/vio.h>
47#include <asm/iommu.h>
48#include <asm/firmware.h>
49
50#include "ibmveth.h"
51
52static irqreturn_t ibmveth_interrupt(int irq, void *dev_instance);
53static void ibmveth_rxq_harvest_buffer(struct ibmveth_adapter *adapter);
54static unsigned long ibmveth_get_desired_dma(struct vio_dev *vdev);
55
56static struct kobj_type ktype_veth_pool;
57
58
59static const char ibmveth_driver_name[] = "ibmveth";
60static const char ibmveth_driver_string[] = "IBM Power Virtual Ethernet Driver";
61#define ibmveth_driver_version "1.06"
62
63MODULE_AUTHOR("Santiago Leon <santil@linux.vnet.ibm.com>");
64MODULE_DESCRIPTION("IBM Power Virtual Ethernet Driver");
65MODULE_LICENSE("GPL");
66MODULE_VERSION(ibmveth_driver_version);
67
68static unsigned int tx_copybreak __read_mostly = 128;
69module_param(tx_copybreak, uint, 0644);
70MODULE_PARM_DESC(tx_copybreak,
71 "Maximum size of packet that is copied to a new buffer on transmit");
72
73static unsigned int rx_copybreak __read_mostly = 128;
74module_param(rx_copybreak, uint, 0644);
75MODULE_PARM_DESC(rx_copybreak,
76 "Maximum size of packet that is copied to a new buffer on receive");
77
78static unsigned int rx_flush __read_mostly = 0;
79module_param(rx_flush, uint, 0644);
80MODULE_PARM_DESC(rx_flush, "Flush receive buffers before use");
81
82static bool old_large_send __read_mostly;
83module_param(old_large_send, bool, S_IRUGO);
84MODULE_PARM_DESC(old_large_send,
85 "Use old large send method on firmware that supports the new method");
86
87struct ibmveth_stat {
88 char name[ETH_GSTRING_LEN];
89 int offset;
90};
91
92#define IBMVETH_STAT_OFF(stat) offsetof(struct ibmveth_adapter, stat)
93#define IBMVETH_GET_STAT(a, off) *((u64 *)(((unsigned long)(a)) + off))
94
95struct ibmveth_stat ibmveth_stats[] = {
96 { "replenish_task_cycles", IBMVETH_STAT_OFF(replenish_task_cycles) },
97 { "replenish_no_mem", IBMVETH_STAT_OFF(replenish_no_mem) },
98 { "replenish_add_buff_failure",
99 IBMVETH_STAT_OFF(replenish_add_buff_failure) },
100 { "replenish_add_buff_success",
101 IBMVETH_STAT_OFF(replenish_add_buff_success) },
102 { "rx_invalid_buffer", IBMVETH_STAT_OFF(rx_invalid_buffer) },
103 { "rx_no_buffer", IBMVETH_STAT_OFF(rx_no_buffer) },
104 { "tx_map_failed", IBMVETH_STAT_OFF(tx_map_failed) },
105 { "tx_send_failed", IBMVETH_STAT_OFF(tx_send_failed) },
106 { "fw_enabled_ipv4_csum", IBMVETH_STAT_OFF(fw_ipv4_csum_support) },
107 { "fw_enabled_ipv6_csum", IBMVETH_STAT_OFF(fw_ipv6_csum_support) },
108 { "tx_large_packets", IBMVETH_STAT_OFF(tx_large_packets) },
109 { "rx_large_packets", IBMVETH_STAT_OFF(rx_large_packets) },
110 { "fw_enabled_large_send", IBMVETH_STAT_OFF(fw_large_send_support) }
111};
112
113/* simple methods of getting data from the current rxq entry */
114static inline u32 ibmveth_rxq_flags(struct ibmveth_adapter *adapter)
115{
116 return be32_to_cpu(adapter->rx_queue.queue_addr[adapter->rx_queue.index].flags_off);
117}
118
119static inline int ibmveth_rxq_toggle(struct ibmveth_adapter *adapter)
120{
121 return (ibmveth_rxq_flags(adapter) & IBMVETH_RXQ_TOGGLE) >>
122 IBMVETH_RXQ_TOGGLE_SHIFT;
123}
124
125static inline int ibmveth_rxq_pending_buffer(struct ibmveth_adapter *adapter)
126{
127 return ibmveth_rxq_toggle(adapter) == adapter->rx_queue.toggle;
128}
129
130static inline int ibmveth_rxq_buffer_valid(struct ibmveth_adapter *adapter)
131{
132 return ibmveth_rxq_flags(adapter) & IBMVETH_RXQ_VALID;
133}
134
135static inline int ibmveth_rxq_frame_offset(struct ibmveth_adapter *adapter)
136{
137 return ibmveth_rxq_flags(adapter) & IBMVETH_RXQ_OFF_MASK;
138}
139
140static inline int ibmveth_rxq_large_packet(struct ibmveth_adapter *adapter)
141{
142 return ibmveth_rxq_flags(adapter) & IBMVETH_RXQ_LRG_PKT;
143}
144
145static inline int ibmveth_rxq_frame_length(struct ibmveth_adapter *adapter)
146{
147 return be32_to_cpu(adapter->rx_queue.queue_addr[adapter->rx_queue.index].length);
148}
149
150static inline int ibmveth_rxq_csum_good(struct ibmveth_adapter *adapter)
151{
152 return ibmveth_rxq_flags(adapter) & IBMVETH_RXQ_CSUM_GOOD;
153}
154
155/* setup the initial settings for a buffer pool */
156static void ibmveth_init_buffer_pool(struct ibmveth_buff_pool *pool,
157 u32 pool_index, u32 pool_size,
158 u32 buff_size, u32 pool_active)
159{
160 pool->size = pool_size;
161 pool->index = pool_index;
162 pool->buff_size = buff_size;
163 pool->threshold = pool_size * 7 / 8;
164 pool->active = pool_active;
165}
166
167/* allocate and setup an buffer pool - called during open */
168static int ibmveth_alloc_buffer_pool(struct ibmveth_buff_pool *pool)
169{
170 int i;
171
172 pool->free_map = kmalloc(sizeof(u16) * pool->size, GFP_KERNEL);
173
174 if (!pool->free_map)
175 return -1;
176
177 pool->dma_addr = kcalloc(pool->size, sizeof(dma_addr_t), GFP_KERNEL);
178 if (!pool->dma_addr) {
179 kfree(pool->free_map);
180 pool->free_map = NULL;
181 return -1;
182 }
183
184 pool->skbuff = kcalloc(pool->size, sizeof(void *), GFP_KERNEL);
185
186 if (!pool->skbuff) {
187 kfree(pool->dma_addr);
188 pool->dma_addr = NULL;
189
190 kfree(pool->free_map);
191 pool->free_map = NULL;
192 return -1;
193 }
194
195 for (i = 0; i < pool->size; ++i)
196 pool->free_map[i] = i;
197
198 atomic_set(&pool->available, 0);
199 pool->producer_index = 0;
200 pool->consumer_index = 0;
201
202 return 0;
203}
204
205static inline void ibmveth_flush_buffer(void *addr, unsigned long length)
206{
207 unsigned long offset;
208
209 for (offset = 0; offset < length; offset += SMP_CACHE_BYTES)
210 asm("dcbfl %0,%1" :: "b" (addr), "r" (offset));
211}
212
213/* replenish the buffers for a pool. note that we don't need to
214 * skb_reserve these since they are used for incoming...
215 */
216static void ibmveth_replenish_buffer_pool(struct ibmveth_adapter *adapter,
217 struct ibmveth_buff_pool *pool)
218{
219 u32 i;
220 u32 count = pool->size - atomic_read(&pool->available);
221 u32 buffers_added = 0;
222 struct sk_buff *skb;
223 unsigned int free_index, index;
224 u64 correlator;
225 unsigned long lpar_rc;
226 dma_addr_t dma_addr;
227
228 mb();
229
230 for (i = 0; i < count; ++i) {
231 union ibmveth_buf_desc desc;
232
233 skb = netdev_alloc_skb(adapter->netdev, pool->buff_size);
234
235 if (!skb) {
236 netdev_dbg(adapter->netdev,
237 "replenish: unable to allocate skb\n");
238 adapter->replenish_no_mem++;
239 break;
240 }
241
242 free_index = pool->consumer_index;
243 pool->consumer_index++;
244 if (pool->consumer_index >= pool->size)
245 pool->consumer_index = 0;
246 index = pool->free_map[free_index];
247
248 BUG_ON(index == IBM_VETH_INVALID_MAP);
249 BUG_ON(pool->skbuff[index] != NULL);
250
251 dma_addr = dma_map_single(&adapter->vdev->dev, skb->data,
252 pool->buff_size, DMA_FROM_DEVICE);
253
254 if (dma_mapping_error(&adapter->vdev->dev, dma_addr))
255 goto failure;
256
257 pool->free_map[free_index] = IBM_VETH_INVALID_MAP;
258 pool->dma_addr[index] = dma_addr;
259 pool->skbuff[index] = skb;
260
261 correlator = ((u64)pool->index << 32) | index;
262 *(u64 *)skb->data = correlator;
263
264 desc.fields.flags_len = IBMVETH_BUF_VALID | pool->buff_size;
265 desc.fields.address = dma_addr;
266
267 if (rx_flush) {
268 unsigned int len = min(pool->buff_size,
269 adapter->netdev->mtu +
270 IBMVETH_BUFF_OH);
271 ibmveth_flush_buffer(skb->data, len);
272 }
273 lpar_rc = h_add_logical_lan_buffer(adapter->vdev->unit_address,
274 desc.desc);
275
276 if (lpar_rc != H_SUCCESS) {
277 goto failure;
278 } else {
279 buffers_added++;
280 adapter->replenish_add_buff_success++;
281 }
282 }
283
284 mb();
285 atomic_add(buffers_added, &(pool->available));
286 return;
287
288failure:
289 pool->free_map[free_index] = index;
290 pool->skbuff[index] = NULL;
291 if (pool->consumer_index == 0)
292 pool->consumer_index = pool->size - 1;
293 else
294 pool->consumer_index--;
295 if (!dma_mapping_error(&adapter->vdev->dev, dma_addr))
296 dma_unmap_single(&adapter->vdev->dev,
297 pool->dma_addr[index], pool->buff_size,
298 DMA_FROM_DEVICE);
299 dev_kfree_skb_any(skb);
300 adapter->replenish_add_buff_failure++;
301
302 mb();
303 atomic_add(buffers_added, &(pool->available));
304}
305
306/*
307 * The final 8 bytes of the buffer list is a counter of frames dropped
308 * because there was not a buffer in the buffer list capable of holding
309 * the frame.
310 */
311static void ibmveth_update_rx_no_buffer(struct ibmveth_adapter *adapter)
312{
313 __be64 *p = adapter->buffer_list_addr + 4096 - 8;
314
315 adapter->rx_no_buffer = be64_to_cpup(p);
316}
317
318/* replenish routine */
319static void ibmveth_replenish_task(struct ibmveth_adapter *adapter)
320{
321 int i;
322
323 adapter->replenish_task_cycles++;
324
325 for (i = (IBMVETH_NUM_BUFF_POOLS - 1); i >= 0; i--) {
326 struct ibmveth_buff_pool *pool = &adapter->rx_buff_pool[i];
327
328 if (pool->active &&
329 (atomic_read(&pool->available) < pool->threshold))
330 ibmveth_replenish_buffer_pool(adapter, pool);
331 }
332
333 ibmveth_update_rx_no_buffer(adapter);
334}
335
336/* empty and free ana buffer pool - also used to do cleanup in error paths */
337static void ibmveth_free_buffer_pool(struct ibmveth_adapter *adapter,
338 struct ibmveth_buff_pool *pool)
339{
340 int i;
341
342 kfree(pool->free_map);
343 pool->free_map = NULL;
344
345 if (pool->skbuff && pool->dma_addr) {
346 for (i = 0; i < pool->size; ++i) {
347 struct sk_buff *skb = pool->skbuff[i];
348 if (skb) {
349 dma_unmap_single(&adapter->vdev->dev,
350 pool->dma_addr[i],
351 pool->buff_size,
352 DMA_FROM_DEVICE);
353 dev_kfree_skb_any(skb);
354 pool->skbuff[i] = NULL;
355 }
356 }
357 }
358
359 if (pool->dma_addr) {
360 kfree(pool->dma_addr);
361 pool->dma_addr = NULL;
362 }
363
364 if (pool->skbuff) {
365 kfree(pool->skbuff);
366 pool->skbuff = NULL;
367 }
368}
369
370/* remove a buffer from a pool */
371static void ibmveth_remove_buffer_from_pool(struct ibmveth_adapter *adapter,
372 u64 correlator)
373{
374 unsigned int pool = correlator >> 32;
375 unsigned int index = correlator & 0xffffffffUL;
376 unsigned int free_index;
377 struct sk_buff *skb;
378
379 BUG_ON(pool >= IBMVETH_NUM_BUFF_POOLS);
380 BUG_ON(index >= adapter->rx_buff_pool[pool].size);
381
382 skb = adapter->rx_buff_pool[pool].skbuff[index];
383
384 BUG_ON(skb == NULL);
385
386 adapter->rx_buff_pool[pool].skbuff[index] = NULL;
387
388 dma_unmap_single(&adapter->vdev->dev,
389 adapter->rx_buff_pool[pool].dma_addr[index],
390 adapter->rx_buff_pool[pool].buff_size,
391 DMA_FROM_DEVICE);
392
393 free_index = adapter->rx_buff_pool[pool].producer_index;
394 adapter->rx_buff_pool[pool].producer_index++;
395 if (adapter->rx_buff_pool[pool].producer_index >=
396 adapter->rx_buff_pool[pool].size)
397 adapter->rx_buff_pool[pool].producer_index = 0;
398 adapter->rx_buff_pool[pool].free_map[free_index] = index;
399
400 mb();
401
402 atomic_dec(&(adapter->rx_buff_pool[pool].available));
403}
404
405/* get the current buffer on the rx queue */
406static inline struct sk_buff *ibmveth_rxq_get_buffer(struct ibmveth_adapter *adapter)
407{
408 u64 correlator = adapter->rx_queue.queue_addr[adapter->rx_queue.index].correlator;
409 unsigned int pool = correlator >> 32;
410 unsigned int index = correlator & 0xffffffffUL;
411
412 BUG_ON(pool >= IBMVETH_NUM_BUFF_POOLS);
413 BUG_ON(index >= adapter->rx_buff_pool[pool].size);
414
415 return adapter->rx_buff_pool[pool].skbuff[index];
416}
417
418/* recycle the current buffer on the rx queue */
419static int ibmveth_rxq_recycle_buffer(struct ibmveth_adapter *adapter)
420{
421 u32 q_index = adapter->rx_queue.index;
422 u64 correlator = adapter->rx_queue.queue_addr[q_index].correlator;
423 unsigned int pool = correlator >> 32;
424 unsigned int index = correlator & 0xffffffffUL;
425 union ibmveth_buf_desc desc;
426 unsigned long lpar_rc;
427 int ret = 1;
428
429 BUG_ON(pool >= IBMVETH_NUM_BUFF_POOLS);
430 BUG_ON(index >= adapter->rx_buff_pool[pool].size);
431
432 if (!adapter->rx_buff_pool[pool].active) {
433 ibmveth_rxq_harvest_buffer(adapter);
434 ibmveth_free_buffer_pool(adapter, &adapter->rx_buff_pool[pool]);
435 goto out;
436 }
437
438 desc.fields.flags_len = IBMVETH_BUF_VALID |
439 adapter->rx_buff_pool[pool].buff_size;
440 desc.fields.address = adapter->rx_buff_pool[pool].dma_addr[index];
441
442 lpar_rc = h_add_logical_lan_buffer(adapter->vdev->unit_address, desc.desc);
443
444 if (lpar_rc != H_SUCCESS) {
445 netdev_dbg(adapter->netdev, "h_add_logical_lan_buffer failed "
446 "during recycle rc=%ld", lpar_rc);
447 ibmveth_remove_buffer_from_pool(adapter, adapter->rx_queue.queue_addr[adapter->rx_queue.index].correlator);
448 ret = 0;
449 }
450
451 if (++adapter->rx_queue.index == adapter->rx_queue.num_slots) {
452 adapter->rx_queue.index = 0;
453 adapter->rx_queue.toggle = !adapter->rx_queue.toggle;
454 }
455
456out:
457 return ret;
458}
459
460static void ibmveth_rxq_harvest_buffer(struct ibmveth_adapter *adapter)
461{
462 ibmveth_remove_buffer_from_pool(adapter, adapter->rx_queue.queue_addr[adapter->rx_queue.index].correlator);
463
464 if (++adapter->rx_queue.index == adapter->rx_queue.num_slots) {
465 adapter->rx_queue.index = 0;
466 adapter->rx_queue.toggle = !adapter->rx_queue.toggle;
467 }
468}
469
470static void ibmveth_cleanup(struct ibmveth_adapter *adapter)
471{
472 int i;
473 struct device *dev = &adapter->vdev->dev;
474
475 if (adapter->buffer_list_addr != NULL) {
476 if (!dma_mapping_error(dev, adapter->buffer_list_dma)) {
477 dma_unmap_single(dev, adapter->buffer_list_dma, 4096,
478 DMA_BIDIRECTIONAL);
479 adapter->buffer_list_dma = DMA_ERROR_CODE;
480 }
481 free_page((unsigned long)adapter->buffer_list_addr);
482 adapter->buffer_list_addr = NULL;
483 }
484
485 if (adapter->filter_list_addr != NULL) {
486 if (!dma_mapping_error(dev, adapter->filter_list_dma)) {
487 dma_unmap_single(dev, adapter->filter_list_dma, 4096,
488 DMA_BIDIRECTIONAL);
489 adapter->filter_list_dma = DMA_ERROR_CODE;
490 }
491 free_page((unsigned long)adapter->filter_list_addr);
492 adapter->filter_list_addr = NULL;
493 }
494
495 if (adapter->rx_queue.queue_addr != NULL) {
496 dma_free_coherent(dev, adapter->rx_queue.queue_len,
497 adapter->rx_queue.queue_addr,
498 adapter->rx_queue.queue_dma);
499 adapter->rx_queue.queue_addr = NULL;
500 }
501
502 for (i = 0; i < IBMVETH_NUM_BUFF_POOLS; i++)
503 if (adapter->rx_buff_pool[i].active)
504 ibmveth_free_buffer_pool(adapter,
505 &adapter->rx_buff_pool[i]);
506
507 if (adapter->bounce_buffer != NULL) {
508 if (!dma_mapping_error(dev, adapter->bounce_buffer_dma)) {
509 dma_unmap_single(&adapter->vdev->dev,
510 adapter->bounce_buffer_dma,
511 adapter->netdev->mtu + IBMVETH_BUFF_OH,
512 DMA_BIDIRECTIONAL);
513 adapter->bounce_buffer_dma = DMA_ERROR_CODE;
514 }
515 kfree(adapter->bounce_buffer);
516 adapter->bounce_buffer = NULL;
517 }
518}
519
520static int ibmveth_register_logical_lan(struct ibmveth_adapter *adapter,
521 union ibmveth_buf_desc rxq_desc, u64 mac_address)
522{
523 int rc, try_again = 1;
524
525 /*
526 * After a kexec the adapter will still be open, so our attempt to
527 * open it will fail. So if we get a failure we free the adapter and
528 * try again, but only once.
529 */
530retry:
531 rc = h_register_logical_lan(adapter->vdev->unit_address,
532 adapter->buffer_list_dma, rxq_desc.desc,
533 adapter->filter_list_dma, mac_address);
534
535 if (rc != H_SUCCESS && try_again) {
536 do {
537 rc = h_free_logical_lan(adapter->vdev->unit_address);
538 } while (H_IS_LONG_BUSY(rc) || (rc == H_BUSY));
539
540 try_again = 0;
541 goto retry;
542 }
543
544 return rc;
545}
546
547static u64 ibmveth_encode_mac_addr(u8 *mac)
548{
549 int i;
550 u64 encoded = 0;
551
552 for (i = 0; i < ETH_ALEN; i++)
553 encoded = (encoded << 8) | mac[i];
554
555 return encoded;
556}
557
558static int ibmveth_open(struct net_device *netdev)
559{
560 struct ibmveth_adapter *adapter = netdev_priv(netdev);
561 u64 mac_address;
562 int rxq_entries = 1;
563 unsigned long lpar_rc;
564 int rc;
565 union ibmveth_buf_desc rxq_desc;
566 int i;
567 struct device *dev;
568
569 netdev_dbg(netdev, "open starting\n");
570
571 napi_enable(&adapter->napi);
572
573 for(i = 0; i < IBMVETH_NUM_BUFF_POOLS; i++)
574 rxq_entries += adapter->rx_buff_pool[i].size;
575
576 adapter->buffer_list_addr = (void*) get_zeroed_page(GFP_KERNEL);
577 adapter->filter_list_addr = (void*) get_zeroed_page(GFP_KERNEL);
578
579 if (!adapter->buffer_list_addr || !adapter->filter_list_addr) {
580 netdev_err(netdev, "unable to allocate filter or buffer list "
581 "pages\n");
582 rc = -ENOMEM;
583 goto err_out;
584 }
585
586 dev = &adapter->vdev->dev;
587
588 adapter->rx_queue.queue_len = sizeof(struct ibmveth_rx_q_entry) *
589 rxq_entries;
590 adapter->rx_queue.queue_addr =
591 dma_alloc_coherent(dev, adapter->rx_queue.queue_len,
592 &adapter->rx_queue.queue_dma, GFP_KERNEL);
593 if (!adapter->rx_queue.queue_addr) {
594 rc = -ENOMEM;
595 goto err_out;
596 }
597
598 adapter->buffer_list_dma = dma_map_single(dev,
599 adapter->buffer_list_addr, 4096, DMA_BIDIRECTIONAL);
600 adapter->filter_list_dma = dma_map_single(dev,
601 adapter->filter_list_addr, 4096, DMA_BIDIRECTIONAL);
602
603 if ((dma_mapping_error(dev, adapter->buffer_list_dma)) ||
604 (dma_mapping_error(dev, adapter->filter_list_dma))) {
605 netdev_err(netdev, "unable to map filter or buffer list "
606 "pages\n");
607 rc = -ENOMEM;
608 goto err_out;
609 }
610
611 adapter->rx_queue.index = 0;
612 adapter->rx_queue.num_slots = rxq_entries;
613 adapter->rx_queue.toggle = 1;
614
615 mac_address = ibmveth_encode_mac_addr(netdev->dev_addr);
616
617 rxq_desc.fields.flags_len = IBMVETH_BUF_VALID |
618 adapter->rx_queue.queue_len;
619 rxq_desc.fields.address = adapter->rx_queue.queue_dma;
620
621 netdev_dbg(netdev, "buffer list @ 0x%p\n", adapter->buffer_list_addr);
622 netdev_dbg(netdev, "filter list @ 0x%p\n", adapter->filter_list_addr);
623 netdev_dbg(netdev, "receive q @ 0x%p\n", adapter->rx_queue.queue_addr);
624
625 h_vio_signal(adapter->vdev->unit_address, VIO_IRQ_DISABLE);
626
627 lpar_rc = ibmveth_register_logical_lan(adapter, rxq_desc, mac_address);
628
629 if (lpar_rc != H_SUCCESS) {
630 netdev_err(netdev, "h_register_logical_lan failed with %ld\n",
631 lpar_rc);
632 netdev_err(netdev, "buffer TCE:0x%llx filter TCE:0x%llx rxq "
633 "desc:0x%llx MAC:0x%llx\n",
634 adapter->buffer_list_dma,
635 adapter->filter_list_dma,
636 rxq_desc.desc,
637 mac_address);
638 rc = -ENONET;
639 goto err_out;
640 }
641
642 for (i = 0; i < IBMVETH_NUM_BUFF_POOLS; i++) {
643 if (!adapter->rx_buff_pool[i].active)
644 continue;
645 if (ibmveth_alloc_buffer_pool(&adapter->rx_buff_pool[i])) {
646 netdev_err(netdev, "unable to alloc pool\n");
647 adapter->rx_buff_pool[i].active = 0;
648 rc = -ENOMEM;
649 goto err_out;
650 }
651 }
652
653 netdev_dbg(netdev, "registering irq 0x%x\n", netdev->irq);
654 rc = request_irq(netdev->irq, ibmveth_interrupt, 0, netdev->name,
655 netdev);
656 if (rc != 0) {
657 netdev_err(netdev, "unable to request irq 0x%x, rc %d\n",
658 netdev->irq, rc);
659 do {
660 lpar_rc = h_free_logical_lan(adapter->vdev->unit_address);
661 } while (H_IS_LONG_BUSY(lpar_rc) || (lpar_rc == H_BUSY));
662
663 goto err_out;
664 }
665
666 adapter->bounce_buffer =
667 kmalloc(netdev->mtu + IBMVETH_BUFF_OH, GFP_KERNEL);
668 if (!adapter->bounce_buffer) {
669 rc = -ENOMEM;
670 goto err_out_free_irq;
671 }
672 adapter->bounce_buffer_dma =
673 dma_map_single(&adapter->vdev->dev, adapter->bounce_buffer,
674 netdev->mtu + IBMVETH_BUFF_OH, DMA_BIDIRECTIONAL);
675 if (dma_mapping_error(dev, adapter->bounce_buffer_dma)) {
676 netdev_err(netdev, "unable to map bounce buffer\n");
677 rc = -ENOMEM;
678 goto err_out_free_irq;
679 }
680
681 netdev_dbg(netdev, "initial replenish cycle\n");
682 ibmveth_interrupt(netdev->irq, netdev);
683
684 netif_start_queue(netdev);
685
686 netdev_dbg(netdev, "open complete\n");
687
688 return 0;
689
690err_out_free_irq:
691 free_irq(netdev->irq, netdev);
692err_out:
693 ibmveth_cleanup(adapter);
694 napi_disable(&adapter->napi);
695 return rc;
696}
697
698static int ibmveth_close(struct net_device *netdev)
699{
700 struct ibmveth_adapter *adapter = netdev_priv(netdev);
701 long lpar_rc;
702
703 netdev_dbg(netdev, "close starting\n");
704
705 napi_disable(&adapter->napi);
706
707 if (!adapter->pool_config)
708 netif_stop_queue(netdev);
709
710 h_vio_signal(adapter->vdev->unit_address, VIO_IRQ_DISABLE);
711
712 do {
713 lpar_rc = h_free_logical_lan(adapter->vdev->unit_address);
714 } while (H_IS_LONG_BUSY(lpar_rc) || (lpar_rc == H_BUSY));
715
716 if (lpar_rc != H_SUCCESS) {
717 netdev_err(netdev, "h_free_logical_lan failed with %lx, "
718 "continuing with close\n", lpar_rc);
719 }
720
721 free_irq(netdev->irq, netdev);
722
723 ibmveth_update_rx_no_buffer(adapter);
724
725 ibmveth_cleanup(adapter);
726
727 netdev_dbg(netdev, "close complete\n");
728
729 return 0;
730}
731
732static int netdev_get_settings(struct net_device *dev, struct ethtool_cmd *cmd)
733{
734 cmd->supported = (SUPPORTED_1000baseT_Full | SUPPORTED_Autoneg |
735 SUPPORTED_FIBRE);
736 cmd->advertising = (ADVERTISED_1000baseT_Full | ADVERTISED_Autoneg |
737 ADVERTISED_FIBRE);
738 ethtool_cmd_speed_set(cmd, SPEED_1000);
739 cmd->duplex = DUPLEX_FULL;
740 cmd->port = PORT_FIBRE;
741 cmd->phy_address = 0;
742 cmd->transceiver = XCVR_INTERNAL;
743 cmd->autoneg = AUTONEG_ENABLE;
744 cmd->maxtxpkt = 0;
745 cmd->maxrxpkt = 1;
746 return 0;
747}
748
749static void netdev_get_drvinfo(struct net_device *dev,
750 struct ethtool_drvinfo *info)
751{
752 strlcpy(info->driver, ibmveth_driver_name, sizeof(info->driver));
753 strlcpy(info->version, ibmveth_driver_version, sizeof(info->version));
754}
755
756static netdev_features_t ibmveth_fix_features(struct net_device *dev,
757 netdev_features_t features)
758{
759 /*
760 * Since the ibmveth firmware interface does not have the
761 * concept of separate tx/rx checksum offload enable, if rx
762 * checksum is disabled we also have to disable tx checksum
763 * offload. Once we disable rx checksum offload, we are no
764 * longer allowed to send tx buffers that are not properly
765 * checksummed.
766 */
767
768 if (!(features & NETIF_F_RXCSUM))
769 features &= ~NETIF_F_CSUM_MASK;
770
771 return features;
772}
773
774static int ibmveth_set_csum_offload(struct net_device *dev, u32 data)
775{
776 struct ibmveth_adapter *adapter = netdev_priv(dev);
777 unsigned long set_attr, clr_attr, ret_attr;
778 unsigned long set_attr6, clr_attr6;
779 long ret, ret4, ret6;
780 int rc1 = 0, rc2 = 0;
781 int restart = 0;
782
783 if (netif_running(dev)) {
784 restart = 1;
785 adapter->pool_config = 1;
786 ibmveth_close(dev);
787 adapter->pool_config = 0;
788 }
789
790 set_attr = 0;
791 clr_attr = 0;
792 set_attr6 = 0;
793 clr_attr6 = 0;
794
795 if (data) {
796 set_attr = IBMVETH_ILLAN_IPV4_TCP_CSUM;
797 set_attr6 = IBMVETH_ILLAN_IPV6_TCP_CSUM;
798 } else {
799 clr_attr = IBMVETH_ILLAN_IPV4_TCP_CSUM;
800 clr_attr6 = IBMVETH_ILLAN_IPV6_TCP_CSUM;
801 }
802
803 ret = h_illan_attributes(adapter->vdev->unit_address, 0, 0, &ret_attr);
804
805 if (ret == H_SUCCESS && !(ret_attr & IBMVETH_ILLAN_ACTIVE_TRUNK) &&
806 !(ret_attr & IBMVETH_ILLAN_TRUNK_PRI_MASK) &&
807 (ret_attr & IBMVETH_ILLAN_PADDED_PKT_CSUM)) {
808 ret4 = h_illan_attributes(adapter->vdev->unit_address, clr_attr,
809 set_attr, &ret_attr);
810
811 if (ret4 != H_SUCCESS) {
812 netdev_err(dev, "unable to change IPv4 checksum "
813 "offload settings. %d rc=%ld\n",
814 data, ret4);
815
816 h_illan_attributes(adapter->vdev->unit_address,
817 set_attr, clr_attr, &ret_attr);
818
819 if (data == 1)
820 dev->features &= ~NETIF_F_IP_CSUM;
821
822 } else {
823 adapter->fw_ipv4_csum_support = data;
824 }
825
826 ret6 = h_illan_attributes(adapter->vdev->unit_address,
827 clr_attr6, set_attr6, &ret_attr);
828
829 if (ret6 != H_SUCCESS) {
830 netdev_err(dev, "unable to change IPv6 checksum "
831 "offload settings. %d rc=%ld\n",
832 data, ret6);
833
834 h_illan_attributes(adapter->vdev->unit_address,
835 set_attr6, clr_attr6, &ret_attr);
836
837 if (data == 1)
838 dev->features &= ~NETIF_F_IPV6_CSUM;
839
840 } else
841 adapter->fw_ipv6_csum_support = data;
842
843 if (ret4 == H_SUCCESS || ret6 == H_SUCCESS)
844 adapter->rx_csum = data;
845 else
846 rc1 = -EIO;
847 } else {
848 rc1 = -EIO;
849 netdev_err(dev, "unable to change checksum offload settings."
850 " %d rc=%ld ret_attr=%lx\n", data, ret,
851 ret_attr);
852 }
853
854 if (restart)
855 rc2 = ibmveth_open(dev);
856
857 return rc1 ? rc1 : rc2;
858}
859
860static int ibmveth_set_tso(struct net_device *dev, u32 data)
861{
862 struct ibmveth_adapter *adapter = netdev_priv(dev);
863 unsigned long set_attr, clr_attr, ret_attr;
864 long ret1, ret2;
865 int rc1 = 0, rc2 = 0;
866 int restart = 0;
867
868 if (netif_running(dev)) {
869 restart = 1;
870 adapter->pool_config = 1;
871 ibmveth_close(dev);
872 adapter->pool_config = 0;
873 }
874
875 set_attr = 0;
876 clr_attr = 0;
877
878 if (data)
879 set_attr = IBMVETH_ILLAN_LRG_SR_ENABLED;
880 else
881 clr_attr = IBMVETH_ILLAN_LRG_SR_ENABLED;
882
883 ret1 = h_illan_attributes(adapter->vdev->unit_address, 0, 0, &ret_attr);
884
885 if (ret1 == H_SUCCESS && (ret_attr & IBMVETH_ILLAN_LRG_SND_SUPPORT) &&
886 !old_large_send) {
887 ret2 = h_illan_attributes(adapter->vdev->unit_address, clr_attr,
888 set_attr, &ret_attr);
889
890 if (ret2 != H_SUCCESS) {
891 netdev_err(dev, "unable to change tso settings. %d rc=%ld\n",
892 data, ret2);
893
894 h_illan_attributes(adapter->vdev->unit_address,
895 set_attr, clr_attr, &ret_attr);
896
897 if (data == 1)
898 dev->features &= ~(NETIF_F_TSO | NETIF_F_TSO6);
899 rc1 = -EIO;
900
901 } else {
902 adapter->fw_large_send_support = data;
903 adapter->large_send = data;
904 }
905 } else {
906 /* Older firmware version of large send offload does not
907 * support tcp6/ipv6
908 */
909 if (data == 1) {
910 dev->features &= ~NETIF_F_TSO6;
911 netdev_info(dev, "TSO feature requires all partitions to have updated driver");
912 }
913 adapter->large_send = data;
914 }
915
916 if (restart)
917 rc2 = ibmveth_open(dev);
918
919 return rc1 ? rc1 : rc2;
920}
921
922static int ibmveth_set_features(struct net_device *dev,
923 netdev_features_t features)
924{
925 struct ibmveth_adapter *adapter = netdev_priv(dev);
926 int rx_csum = !!(features & NETIF_F_RXCSUM);
927 int large_send = !!(features & (NETIF_F_TSO | NETIF_F_TSO6));
928 int rc1 = 0, rc2 = 0;
929
930 if (rx_csum != adapter->rx_csum) {
931 rc1 = ibmveth_set_csum_offload(dev, rx_csum);
932 if (rc1 && !adapter->rx_csum)
933 dev->features =
934 features & ~(NETIF_F_CSUM_MASK |
935 NETIF_F_RXCSUM);
936 }
937
938 if (large_send != adapter->large_send) {
939 rc2 = ibmveth_set_tso(dev, large_send);
940 if (rc2 && !adapter->large_send)
941 dev->features =
942 features & ~(NETIF_F_TSO | NETIF_F_TSO6);
943 }
944
945 return rc1 ? rc1 : rc2;
946}
947
948static void ibmveth_get_strings(struct net_device *dev, u32 stringset, u8 *data)
949{
950 int i;
951
952 if (stringset != ETH_SS_STATS)
953 return;
954
955 for (i = 0; i < ARRAY_SIZE(ibmveth_stats); i++, data += ETH_GSTRING_LEN)
956 memcpy(data, ibmveth_stats[i].name, ETH_GSTRING_LEN);
957}
958
959static int ibmveth_get_sset_count(struct net_device *dev, int sset)
960{
961 switch (sset) {
962 case ETH_SS_STATS:
963 return ARRAY_SIZE(ibmveth_stats);
964 default:
965 return -EOPNOTSUPP;
966 }
967}
968
969static void ibmveth_get_ethtool_stats(struct net_device *dev,
970 struct ethtool_stats *stats, u64 *data)
971{
972 int i;
973 struct ibmveth_adapter *adapter = netdev_priv(dev);
974
975 for (i = 0; i < ARRAY_SIZE(ibmveth_stats); i++)
976 data[i] = IBMVETH_GET_STAT(adapter, ibmveth_stats[i].offset);
977}
978
979static const struct ethtool_ops netdev_ethtool_ops = {
980 .get_drvinfo = netdev_get_drvinfo,
981 .get_settings = netdev_get_settings,
982 .get_link = ethtool_op_get_link,
983 .get_strings = ibmveth_get_strings,
984 .get_sset_count = ibmveth_get_sset_count,
985 .get_ethtool_stats = ibmveth_get_ethtool_stats,
986};
987
988static int ibmveth_ioctl(struct net_device *dev, struct ifreq *ifr, int cmd)
989{
990 return -EOPNOTSUPP;
991}
992
993#define page_offset(v) ((unsigned long)(v) & ((1 << 12) - 1))
994
995static int ibmveth_send(struct ibmveth_adapter *adapter,
996 union ibmveth_buf_desc *descs, unsigned long mss)
997{
998 unsigned long correlator;
999 unsigned int retry_count;
1000 unsigned long ret;
1001
1002 /*
1003 * The retry count sets a maximum for the number of broadcast and
1004 * multicast destinations within the system.
1005 */
1006 retry_count = 1024;
1007 correlator = 0;
1008 do {
1009 ret = h_send_logical_lan(adapter->vdev->unit_address,
1010 descs[0].desc, descs[1].desc,
1011 descs[2].desc, descs[3].desc,
1012 descs[4].desc, descs[5].desc,
1013 correlator, &correlator, mss,
1014 adapter->fw_large_send_support);
1015 } while ((ret == H_BUSY) && (retry_count--));
1016
1017 if (ret != H_SUCCESS && ret != H_DROPPED) {
1018 netdev_err(adapter->netdev, "tx: h_send_logical_lan failed "
1019 "with rc=%ld\n", ret);
1020 return 1;
1021 }
1022
1023 return 0;
1024}
1025
1026static netdev_tx_t ibmveth_start_xmit(struct sk_buff *skb,
1027 struct net_device *netdev)
1028{
1029 struct ibmveth_adapter *adapter = netdev_priv(netdev);
1030 unsigned int desc_flags;
1031 union ibmveth_buf_desc descs[6];
1032 int last, i;
1033 int force_bounce = 0;
1034 dma_addr_t dma_addr;
1035 unsigned long mss = 0;
1036
1037 /*
1038 * veth handles a maximum of 6 segments including the header, so
1039 * we have to linearize the skb if there are more than this.
1040 */
1041 if (skb_shinfo(skb)->nr_frags > 5 && __skb_linearize(skb)) {
1042 netdev->stats.tx_dropped++;
1043 goto out;
1044 }
1045
1046 /* veth can't checksum offload UDP */
1047 if (skb->ip_summed == CHECKSUM_PARTIAL &&
1048 ((skb->protocol == htons(ETH_P_IP) &&
1049 ip_hdr(skb)->protocol != IPPROTO_TCP) ||
1050 (skb->protocol == htons(ETH_P_IPV6) &&
1051 ipv6_hdr(skb)->nexthdr != IPPROTO_TCP)) &&
1052 skb_checksum_help(skb)) {
1053
1054 netdev_err(netdev, "tx: failed to checksum packet\n");
1055 netdev->stats.tx_dropped++;
1056 goto out;
1057 }
1058
1059 desc_flags = IBMVETH_BUF_VALID;
1060
1061 if (skb_is_gso(skb) && adapter->fw_large_send_support)
1062 desc_flags |= IBMVETH_BUF_LRG_SND;
1063
1064 if (skb->ip_summed == CHECKSUM_PARTIAL) {
1065 unsigned char *buf = skb_transport_header(skb) +
1066 skb->csum_offset;
1067
1068 desc_flags |= (IBMVETH_BUF_NO_CSUM | IBMVETH_BUF_CSUM_GOOD);
1069
1070 /* Need to zero out the checksum */
1071 buf[0] = 0;
1072 buf[1] = 0;
1073 }
1074
1075retry_bounce:
1076 memset(descs, 0, sizeof(descs));
1077
1078 /*
1079 * If a linear packet is below the rx threshold then
1080 * copy it into the static bounce buffer. This avoids the
1081 * cost of a TCE insert and remove.
1082 */
1083 if (force_bounce || (!skb_is_nonlinear(skb) &&
1084 (skb->len < tx_copybreak))) {
1085 skb_copy_from_linear_data(skb, adapter->bounce_buffer,
1086 skb->len);
1087
1088 descs[0].fields.flags_len = desc_flags | skb->len;
1089 descs[0].fields.address = adapter->bounce_buffer_dma;
1090
1091 if (ibmveth_send(adapter, descs, 0)) {
1092 adapter->tx_send_failed++;
1093 netdev->stats.tx_dropped++;
1094 } else {
1095 netdev->stats.tx_packets++;
1096 netdev->stats.tx_bytes += skb->len;
1097 }
1098
1099 goto out;
1100 }
1101
1102 /* Map the header */
1103 dma_addr = dma_map_single(&adapter->vdev->dev, skb->data,
1104 skb_headlen(skb), DMA_TO_DEVICE);
1105 if (dma_mapping_error(&adapter->vdev->dev, dma_addr))
1106 goto map_failed;
1107
1108 descs[0].fields.flags_len = desc_flags | skb_headlen(skb);
1109 descs[0].fields.address = dma_addr;
1110
1111 /* Map the frags */
1112 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
1113 const skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
1114
1115 dma_addr = skb_frag_dma_map(&adapter->vdev->dev, frag, 0,
1116 skb_frag_size(frag), DMA_TO_DEVICE);
1117
1118 if (dma_mapping_error(&adapter->vdev->dev, dma_addr))
1119 goto map_failed_frags;
1120
1121 descs[i+1].fields.flags_len = desc_flags | skb_frag_size(frag);
1122 descs[i+1].fields.address = dma_addr;
1123 }
1124
1125 if (skb_is_gso(skb)) {
1126 if (adapter->fw_large_send_support) {
1127 mss = (unsigned long)skb_shinfo(skb)->gso_size;
1128 adapter->tx_large_packets++;
1129 } else if (!skb_is_gso_v6(skb)) {
1130 /* Put -1 in the IP checksum to tell phyp it
1131 * is a largesend packet. Put the mss in
1132 * the TCP checksum.
1133 */
1134 ip_hdr(skb)->check = 0xffff;
1135 tcp_hdr(skb)->check =
1136 cpu_to_be16(skb_shinfo(skb)->gso_size);
1137 adapter->tx_large_packets++;
1138 }
1139 }
1140
1141 if (ibmveth_send(adapter, descs, mss)) {
1142 adapter->tx_send_failed++;
1143 netdev->stats.tx_dropped++;
1144 } else {
1145 netdev->stats.tx_packets++;
1146 netdev->stats.tx_bytes += skb->len;
1147 }
1148
1149 dma_unmap_single(&adapter->vdev->dev,
1150 descs[0].fields.address,
1151 descs[0].fields.flags_len & IBMVETH_BUF_LEN_MASK,
1152 DMA_TO_DEVICE);
1153
1154 for (i = 1; i < skb_shinfo(skb)->nr_frags + 1; i++)
1155 dma_unmap_page(&adapter->vdev->dev, descs[i].fields.address,
1156 descs[i].fields.flags_len & IBMVETH_BUF_LEN_MASK,
1157 DMA_TO_DEVICE);
1158
1159out:
1160 dev_consume_skb_any(skb);
1161 return NETDEV_TX_OK;
1162
1163map_failed_frags:
1164 last = i+1;
1165 for (i = 0; i < last; i++)
1166 dma_unmap_page(&adapter->vdev->dev, descs[i].fields.address,
1167 descs[i].fields.flags_len & IBMVETH_BUF_LEN_MASK,
1168 DMA_TO_DEVICE);
1169
1170map_failed:
1171 if (!firmware_has_feature(FW_FEATURE_CMO))
1172 netdev_err(netdev, "tx: unable to map xmit buffer\n");
1173 adapter->tx_map_failed++;
1174 if (skb_linearize(skb)) {
1175 netdev->stats.tx_dropped++;
1176 goto out;
1177 }
1178 force_bounce = 1;
1179 goto retry_bounce;
1180}
1181
1182static void ibmveth_rx_mss_helper(struct sk_buff *skb, u16 mss, int lrg_pkt)
1183{
1184 struct tcphdr *tcph;
1185 int offset = 0;
1186 int hdr_len;
1187
1188 /* only TCP packets will be aggregated */
1189 if (skb->protocol == htons(ETH_P_IP)) {
1190 struct iphdr *iph = (struct iphdr *)skb->data;
1191
1192 if (iph->protocol == IPPROTO_TCP) {
1193 offset = iph->ihl * 4;
1194 skb_shinfo(skb)->gso_type = SKB_GSO_TCPV4;
1195 } else {
1196 return;
1197 }
1198 } else if (skb->protocol == htons(ETH_P_IPV6)) {
1199 struct ipv6hdr *iph6 = (struct ipv6hdr *)skb->data;
1200
1201 if (iph6->nexthdr == IPPROTO_TCP) {
1202 offset = sizeof(struct ipv6hdr);
1203 skb_shinfo(skb)->gso_type = SKB_GSO_TCPV6;
1204 } else {
1205 return;
1206 }
1207 } else {
1208 return;
1209 }
1210 /* if mss is not set through Large Packet bit/mss in rx buffer,
1211 * expect that the mss will be written to the tcp header checksum.
1212 */
1213 tcph = (struct tcphdr *)(skb->data + offset);
1214 if (lrg_pkt) {
1215 skb_shinfo(skb)->gso_size = mss;
1216 } else if (offset) {
1217 skb_shinfo(skb)->gso_size = ntohs(tcph->check);
1218 tcph->check = 0;
1219 }
1220
1221 if (skb_shinfo(skb)->gso_size) {
1222 hdr_len = offset + tcph->doff * 4;
1223 skb_shinfo(skb)->gso_segs =
1224 DIV_ROUND_UP(skb->len - hdr_len,
1225 skb_shinfo(skb)->gso_size);
1226 }
1227}
1228
1229static int ibmveth_poll(struct napi_struct *napi, int budget)
1230{
1231 struct ibmveth_adapter *adapter =
1232 container_of(napi, struct ibmveth_adapter, napi);
1233 struct net_device *netdev = adapter->netdev;
1234 int frames_processed = 0;
1235 unsigned long lpar_rc;
1236 struct iphdr *iph;
1237 u16 mss = 0;
1238
1239restart_poll:
1240 while (frames_processed < budget) {
1241 if (!ibmveth_rxq_pending_buffer(adapter))
1242 break;
1243
1244 smp_rmb();
1245 if (!ibmveth_rxq_buffer_valid(adapter)) {
1246 wmb(); /* suggested by larson1 */
1247 adapter->rx_invalid_buffer++;
1248 netdev_dbg(netdev, "recycling invalid buffer\n");
1249 ibmveth_rxq_recycle_buffer(adapter);
1250 } else {
1251 struct sk_buff *skb, *new_skb;
1252 int length = ibmveth_rxq_frame_length(adapter);
1253 int offset = ibmveth_rxq_frame_offset(adapter);
1254 int csum_good = ibmveth_rxq_csum_good(adapter);
1255 int lrg_pkt = ibmveth_rxq_large_packet(adapter);
1256
1257 skb = ibmveth_rxq_get_buffer(adapter);
1258
1259 /* if the large packet bit is set in the rx queue
1260 * descriptor, the mss will be written by PHYP eight
1261 * bytes from the start of the rx buffer, which is
1262 * skb->data at this stage
1263 */
1264 if (lrg_pkt) {
1265 __be64 *rxmss = (__be64 *)(skb->data + 8);
1266
1267 mss = (u16)be64_to_cpu(*rxmss);
1268 }
1269
1270 new_skb = NULL;
1271 if (length < rx_copybreak)
1272 new_skb = netdev_alloc_skb(netdev, length);
1273
1274 if (new_skb) {
1275 skb_copy_to_linear_data(new_skb,
1276 skb->data + offset,
1277 length);
1278 if (rx_flush)
1279 ibmveth_flush_buffer(skb->data,
1280 length + offset);
1281 if (!ibmveth_rxq_recycle_buffer(adapter))
1282 kfree_skb(skb);
1283 skb = new_skb;
1284 } else {
1285 ibmveth_rxq_harvest_buffer(adapter);
1286 skb_reserve(skb, offset);
1287 }
1288
1289 skb_put(skb, length);
1290 skb->protocol = eth_type_trans(skb, netdev);
1291
1292 if (csum_good) {
1293 skb->ip_summed = CHECKSUM_UNNECESSARY;
1294 if (be16_to_cpu(skb->protocol) == ETH_P_IP) {
1295 iph = (struct iphdr *)skb->data;
1296
1297 /* If the IP checksum is not offloaded and if the packet
1298 * is large send, the checksum must be rebuilt.
1299 */
1300 if (iph->check == 0xffff) {
1301 iph->check = 0;
1302 iph->check = ip_fast_csum((unsigned char *)iph, iph->ihl);
1303 }
1304 }
1305 }
1306
1307 if (length > netdev->mtu + ETH_HLEN) {
1308 ibmveth_rx_mss_helper(skb, mss, lrg_pkt);
1309 adapter->rx_large_packets++;
1310 }
1311
1312 napi_gro_receive(napi, skb); /* send it up */
1313
1314 netdev->stats.rx_packets++;
1315 netdev->stats.rx_bytes += length;
1316 frames_processed++;
1317 }
1318 }
1319
1320 ibmveth_replenish_task(adapter);
1321
1322 if (frames_processed < budget) {
1323 napi_complete(napi);
1324
1325 /* We think we are done - reenable interrupts,
1326 * then check once more to make sure we are done.
1327 */
1328 lpar_rc = h_vio_signal(adapter->vdev->unit_address,
1329 VIO_IRQ_ENABLE);
1330
1331 BUG_ON(lpar_rc != H_SUCCESS);
1332
1333 if (ibmveth_rxq_pending_buffer(adapter) &&
1334 napi_reschedule(napi)) {
1335 lpar_rc = h_vio_signal(adapter->vdev->unit_address,
1336 VIO_IRQ_DISABLE);
1337 goto restart_poll;
1338 }
1339 }
1340
1341 return frames_processed;
1342}
1343
1344static irqreturn_t ibmveth_interrupt(int irq, void *dev_instance)
1345{
1346 struct net_device *netdev = dev_instance;
1347 struct ibmveth_adapter *adapter = netdev_priv(netdev);
1348 unsigned long lpar_rc;
1349
1350 if (napi_schedule_prep(&adapter->napi)) {
1351 lpar_rc = h_vio_signal(adapter->vdev->unit_address,
1352 VIO_IRQ_DISABLE);
1353 BUG_ON(lpar_rc != H_SUCCESS);
1354 __napi_schedule(&adapter->napi);
1355 }
1356 return IRQ_HANDLED;
1357}
1358
1359static void ibmveth_set_multicast_list(struct net_device *netdev)
1360{
1361 struct ibmveth_adapter *adapter = netdev_priv(netdev);
1362 unsigned long lpar_rc;
1363
1364 if ((netdev->flags & IFF_PROMISC) ||
1365 (netdev_mc_count(netdev) > adapter->mcastFilterSize)) {
1366 lpar_rc = h_multicast_ctrl(adapter->vdev->unit_address,
1367 IbmVethMcastEnableRecv |
1368 IbmVethMcastDisableFiltering,
1369 0);
1370 if (lpar_rc != H_SUCCESS) {
1371 netdev_err(netdev, "h_multicast_ctrl rc=%ld when "
1372 "entering promisc mode\n", lpar_rc);
1373 }
1374 } else {
1375 struct netdev_hw_addr *ha;
1376 /* clear the filter table & disable filtering */
1377 lpar_rc = h_multicast_ctrl(adapter->vdev->unit_address,
1378 IbmVethMcastEnableRecv |
1379 IbmVethMcastDisableFiltering |
1380 IbmVethMcastClearFilterTable,
1381 0);
1382 if (lpar_rc != H_SUCCESS) {
1383 netdev_err(netdev, "h_multicast_ctrl rc=%ld when "
1384 "attempting to clear filter table\n",
1385 lpar_rc);
1386 }
1387 /* add the addresses to the filter table */
1388 netdev_for_each_mc_addr(ha, netdev) {
1389 /* add the multicast address to the filter table */
1390 u64 mcast_addr;
1391 mcast_addr = ibmveth_encode_mac_addr(ha->addr);
1392 lpar_rc = h_multicast_ctrl(adapter->vdev->unit_address,
1393 IbmVethMcastAddFilter,
1394 mcast_addr);
1395 if (lpar_rc != H_SUCCESS) {
1396 netdev_err(netdev, "h_multicast_ctrl rc=%ld "
1397 "when adding an entry to the filter "
1398 "table\n", lpar_rc);
1399 }
1400 }
1401
1402 /* re-enable filtering */
1403 lpar_rc = h_multicast_ctrl(adapter->vdev->unit_address,
1404 IbmVethMcastEnableFiltering,
1405 0);
1406 if (lpar_rc != H_SUCCESS) {
1407 netdev_err(netdev, "h_multicast_ctrl rc=%ld when "
1408 "enabling filtering\n", lpar_rc);
1409 }
1410 }
1411}
1412
1413static int ibmveth_change_mtu(struct net_device *dev, int new_mtu)
1414{
1415 struct ibmveth_adapter *adapter = netdev_priv(dev);
1416 struct vio_dev *viodev = adapter->vdev;
1417 int new_mtu_oh = new_mtu + IBMVETH_BUFF_OH;
1418 int i, rc;
1419 int need_restart = 0;
1420
1421 for (i = 0; i < IBMVETH_NUM_BUFF_POOLS; i++)
1422 if (new_mtu_oh <= adapter->rx_buff_pool[i].buff_size)
1423 break;
1424
1425 if (i == IBMVETH_NUM_BUFF_POOLS)
1426 return -EINVAL;
1427
1428 /* Deactivate all the buffer pools so that the next loop can activate
1429 only the buffer pools necessary to hold the new MTU */
1430 if (netif_running(adapter->netdev)) {
1431 need_restart = 1;
1432 adapter->pool_config = 1;
1433 ibmveth_close(adapter->netdev);
1434 adapter->pool_config = 0;
1435 }
1436
1437 /* Look for an active buffer pool that can hold the new MTU */
1438 for (i = 0; i < IBMVETH_NUM_BUFF_POOLS; i++) {
1439 adapter->rx_buff_pool[i].active = 1;
1440
1441 if (new_mtu_oh <= adapter->rx_buff_pool[i].buff_size) {
1442 dev->mtu = new_mtu;
1443 vio_cmo_set_dev_desired(viodev,
1444 ibmveth_get_desired_dma
1445 (viodev));
1446 if (need_restart) {
1447 return ibmveth_open(adapter->netdev);
1448 }
1449 return 0;
1450 }
1451 }
1452
1453 if (need_restart && (rc = ibmveth_open(adapter->netdev)))
1454 return rc;
1455
1456 return -EINVAL;
1457}
1458
1459#ifdef CONFIG_NET_POLL_CONTROLLER
1460static void ibmveth_poll_controller(struct net_device *dev)
1461{
1462 ibmveth_replenish_task(netdev_priv(dev));
1463 ibmveth_interrupt(dev->irq, dev);
1464}
1465#endif
1466
1467/**
1468 * ibmveth_get_desired_dma - Calculate IO memory desired by the driver
1469 *
1470 * @vdev: struct vio_dev for the device whose desired IO mem is to be returned
1471 *
1472 * Return value:
1473 * Number of bytes of IO data the driver will need to perform well.
1474 */
1475static unsigned long ibmveth_get_desired_dma(struct vio_dev *vdev)
1476{
1477 struct net_device *netdev = dev_get_drvdata(&vdev->dev);
1478 struct ibmveth_adapter *adapter;
1479 struct iommu_table *tbl;
1480 unsigned long ret;
1481 int i;
1482 int rxqentries = 1;
1483
1484 tbl = get_iommu_table_base(&vdev->dev);
1485
1486 /* netdev inits at probe time along with the structures we need below*/
1487 if (netdev == NULL)
1488 return IOMMU_PAGE_ALIGN(IBMVETH_IO_ENTITLEMENT_DEFAULT, tbl);
1489
1490 adapter = netdev_priv(netdev);
1491
1492 ret = IBMVETH_BUFF_LIST_SIZE + IBMVETH_FILT_LIST_SIZE;
1493 ret += IOMMU_PAGE_ALIGN(netdev->mtu, tbl);
1494
1495 for (i = 0; i < IBMVETH_NUM_BUFF_POOLS; i++) {
1496 /* add the size of the active receive buffers */
1497 if (adapter->rx_buff_pool[i].active)
1498 ret +=
1499 adapter->rx_buff_pool[i].size *
1500 IOMMU_PAGE_ALIGN(adapter->rx_buff_pool[i].
1501 buff_size, tbl);
1502 rxqentries += adapter->rx_buff_pool[i].size;
1503 }
1504 /* add the size of the receive queue entries */
1505 ret += IOMMU_PAGE_ALIGN(
1506 rxqentries * sizeof(struct ibmveth_rx_q_entry), tbl);
1507
1508 return ret;
1509}
1510
1511static int ibmveth_set_mac_addr(struct net_device *dev, void *p)
1512{
1513 struct ibmveth_adapter *adapter = netdev_priv(dev);
1514 struct sockaddr *addr = p;
1515 u64 mac_address;
1516 int rc;
1517
1518 if (!is_valid_ether_addr(addr->sa_data))
1519 return -EADDRNOTAVAIL;
1520
1521 mac_address = ibmveth_encode_mac_addr(addr->sa_data);
1522 rc = h_change_logical_lan_mac(adapter->vdev->unit_address, mac_address);
1523 if (rc) {
1524 netdev_err(adapter->netdev, "h_change_logical_lan_mac failed with rc=%d\n", rc);
1525 return rc;
1526 }
1527
1528 ether_addr_copy(dev->dev_addr, addr->sa_data);
1529
1530 return 0;
1531}
1532
1533static const struct net_device_ops ibmveth_netdev_ops = {
1534 .ndo_open = ibmveth_open,
1535 .ndo_stop = ibmveth_close,
1536 .ndo_start_xmit = ibmveth_start_xmit,
1537 .ndo_set_rx_mode = ibmveth_set_multicast_list,
1538 .ndo_do_ioctl = ibmveth_ioctl,
1539 .ndo_change_mtu = ibmveth_change_mtu,
1540 .ndo_fix_features = ibmveth_fix_features,
1541 .ndo_set_features = ibmveth_set_features,
1542 .ndo_validate_addr = eth_validate_addr,
1543 .ndo_set_mac_address = ibmveth_set_mac_addr,
1544#ifdef CONFIG_NET_POLL_CONTROLLER
1545 .ndo_poll_controller = ibmveth_poll_controller,
1546#endif
1547};
1548
1549static int ibmveth_probe(struct vio_dev *dev, const struct vio_device_id *id)
1550{
1551 int rc, i, mac_len;
1552 struct net_device *netdev;
1553 struct ibmveth_adapter *adapter;
1554 unsigned char *mac_addr_p;
1555 unsigned int *mcastFilterSize_p;
1556 long ret;
1557 unsigned long ret_attr;
1558
1559 dev_dbg(&dev->dev, "entering ibmveth_probe for UA 0x%x\n",
1560 dev->unit_address);
1561
1562 mac_addr_p = (unsigned char *)vio_get_attribute(dev, VETH_MAC_ADDR,
1563 &mac_len);
1564 if (!mac_addr_p) {
1565 dev_err(&dev->dev, "Can't find VETH_MAC_ADDR attribute\n");
1566 return -EINVAL;
1567 }
1568 /* Workaround for old/broken pHyp */
1569 if (mac_len == 8)
1570 mac_addr_p += 2;
1571 else if (mac_len != 6) {
1572 dev_err(&dev->dev, "VETH_MAC_ADDR attribute wrong len %d\n",
1573 mac_len);
1574 return -EINVAL;
1575 }
1576
1577 mcastFilterSize_p = (unsigned int *)vio_get_attribute(dev,
1578 VETH_MCAST_FILTER_SIZE, NULL);
1579 if (!mcastFilterSize_p) {
1580 dev_err(&dev->dev, "Can't find VETH_MCAST_FILTER_SIZE "
1581 "attribute\n");
1582 return -EINVAL;
1583 }
1584
1585 netdev = alloc_etherdev(sizeof(struct ibmveth_adapter));
1586
1587 if (!netdev)
1588 return -ENOMEM;
1589
1590 adapter = netdev_priv(netdev);
1591 dev_set_drvdata(&dev->dev, netdev);
1592
1593 adapter->vdev = dev;
1594 adapter->netdev = netdev;
1595 adapter->mcastFilterSize = *mcastFilterSize_p;
1596 adapter->pool_config = 0;
1597
1598 netif_napi_add(netdev, &adapter->napi, ibmveth_poll, 16);
1599
1600 netdev->irq = dev->irq;
1601 netdev->netdev_ops = &ibmveth_netdev_ops;
1602 netdev->ethtool_ops = &netdev_ethtool_ops;
1603 SET_NETDEV_DEV(netdev, &dev->dev);
1604 netdev->hw_features = NETIF_F_SG;
1605 if (vio_get_attribute(dev, "ibm,illan-options", NULL) != NULL) {
1606 netdev->hw_features |= NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM |
1607 NETIF_F_RXCSUM;
1608 }
1609
1610 netdev->features |= netdev->hw_features;
1611
1612 ret = h_illan_attributes(adapter->vdev->unit_address, 0, 0, &ret_attr);
1613
1614 /* If running older firmware, TSO should not be enabled by default */
1615 if (ret == H_SUCCESS && (ret_attr & IBMVETH_ILLAN_LRG_SND_SUPPORT) &&
1616 !old_large_send) {
1617 netdev->hw_features |= NETIF_F_TSO | NETIF_F_TSO6;
1618 netdev->features |= netdev->hw_features;
1619 } else {
1620 netdev->hw_features |= NETIF_F_TSO;
1621 }
1622
1623 netdev->min_mtu = IBMVETH_MIN_MTU;
1624 netdev->max_mtu = ETH_MAX_MTU;
1625
1626 memcpy(netdev->dev_addr, mac_addr_p, ETH_ALEN);
1627
1628 if (firmware_has_feature(FW_FEATURE_CMO))
1629 memcpy(pool_count, pool_count_cmo, sizeof(pool_count));
1630
1631 for (i = 0; i < IBMVETH_NUM_BUFF_POOLS; i++) {
1632 struct kobject *kobj = &adapter->rx_buff_pool[i].kobj;
1633 int error;
1634
1635 ibmveth_init_buffer_pool(&adapter->rx_buff_pool[i], i,
1636 pool_count[i], pool_size[i],
1637 pool_active[i]);
1638 error = kobject_init_and_add(kobj, &ktype_veth_pool,
1639 &dev->dev.kobj, "pool%d", i);
1640 if (!error)
1641 kobject_uevent(kobj, KOBJ_ADD);
1642 }
1643
1644 netdev_dbg(netdev, "adapter @ 0x%p\n", adapter);
1645
1646 adapter->buffer_list_dma = DMA_ERROR_CODE;
1647 adapter->filter_list_dma = DMA_ERROR_CODE;
1648 adapter->rx_queue.queue_dma = DMA_ERROR_CODE;
1649
1650 netdev_dbg(netdev, "registering netdev...\n");
1651
1652 ibmveth_set_features(netdev, netdev->features);
1653
1654 rc = register_netdev(netdev);
1655
1656 if (rc) {
1657 netdev_dbg(netdev, "failed to register netdev rc=%d\n", rc);
1658 free_netdev(netdev);
1659 return rc;
1660 }
1661
1662 netdev_dbg(netdev, "registered\n");
1663
1664 return 0;
1665}
1666
1667static int ibmveth_remove(struct vio_dev *dev)
1668{
1669 struct net_device *netdev = dev_get_drvdata(&dev->dev);
1670 struct ibmveth_adapter *adapter = netdev_priv(netdev);
1671 int i;
1672
1673 for (i = 0; i < IBMVETH_NUM_BUFF_POOLS; i++)
1674 kobject_put(&adapter->rx_buff_pool[i].kobj);
1675
1676 unregister_netdev(netdev);
1677
1678 free_netdev(netdev);
1679 dev_set_drvdata(&dev->dev, NULL);
1680
1681 return 0;
1682}
1683
1684static struct attribute veth_active_attr;
1685static struct attribute veth_num_attr;
1686static struct attribute veth_size_attr;
1687
1688static ssize_t veth_pool_show(struct kobject *kobj,
1689 struct attribute *attr, char *buf)
1690{
1691 struct ibmveth_buff_pool *pool = container_of(kobj,
1692 struct ibmveth_buff_pool,
1693 kobj);
1694
1695 if (attr == &veth_active_attr)
1696 return sprintf(buf, "%d\n", pool->active);
1697 else if (attr == &veth_num_attr)
1698 return sprintf(buf, "%d\n", pool->size);
1699 else if (attr == &veth_size_attr)
1700 return sprintf(buf, "%d\n", pool->buff_size);
1701 return 0;
1702}
1703
1704static ssize_t veth_pool_store(struct kobject *kobj, struct attribute *attr,
1705 const char *buf, size_t count)
1706{
1707 struct ibmveth_buff_pool *pool = container_of(kobj,
1708 struct ibmveth_buff_pool,
1709 kobj);
1710 struct net_device *netdev = dev_get_drvdata(
1711 container_of(kobj->parent, struct device, kobj));
1712 struct ibmveth_adapter *adapter = netdev_priv(netdev);
1713 long value = simple_strtol(buf, NULL, 10);
1714 long rc;
1715
1716 if (attr == &veth_active_attr) {
1717 if (value && !pool->active) {
1718 if (netif_running(netdev)) {
1719 if (ibmveth_alloc_buffer_pool(pool)) {
1720 netdev_err(netdev,
1721 "unable to alloc pool\n");
1722 return -ENOMEM;
1723 }
1724 pool->active = 1;
1725 adapter->pool_config = 1;
1726 ibmveth_close(netdev);
1727 adapter->pool_config = 0;
1728 if ((rc = ibmveth_open(netdev)))
1729 return rc;
1730 } else {
1731 pool->active = 1;
1732 }
1733 } else if (!value && pool->active) {
1734 int mtu = netdev->mtu + IBMVETH_BUFF_OH;
1735 int i;
1736 /* Make sure there is a buffer pool with buffers that
1737 can hold a packet of the size of the MTU */
1738 for (i = 0; i < IBMVETH_NUM_BUFF_POOLS; i++) {
1739 if (pool == &adapter->rx_buff_pool[i])
1740 continue;
1741 if (!adapter->rx_buff_pool[i].active)
1742 continue;
1743 if (mtu <= adapter->rx_buff_pool[i].buff_size)
1744 break;
1745 }
1746
1747 if (i == IBMVETH_NUM_BUFF_POOLS) {
1748 netdev_err(netdev, "no active pool >= MTU\n");
1749 return -EPERM;
1750 }
1751
1752 if (netif_running(netdev)) {
1753 adapter->pool_config = 1;
1754 ibmveth_close(netdev);
1755 pool->active = 0;
1756 adapter->pool_config = 0;
1757 if ((rc = ibmveth_open(netdev)))
1758 return rc;
1759 }
1760 pool->active = 0;
1761 }
1762 } else if (attr == &veth_num_attr) {
1763 if (value <= 0 || value > IBMVETH_MAX_POOL_COUNT) {
1764 return -EINVAL;
1765 } else {
1766 if (netif_running(netdev)) {
1767 adapter->pool_config = 1;
1768 ibmveth_close(netdev);
1769 adapter->pool_config = 0;
1770 pool->size = value;
1771 if ((rc = ibmveth_open(netdev)))
1772 return rc;
1773 } else {
1774 pool->size = value;
1775 }
1776 }
1777 } else if (attr == &veth_size_attr) {
1778 if (value <= IBMVETH_BUFF_OH || value > IBMVETH_MAX_BUF_SIZE) {
1779 return -EINVAL;
1780 } else {
1781 if (netif_running(netdev)) {
1782 adapter->pool_config = 1;
1783 ibmveth_close(netdev);
1784 adapter->pool_config = 0;
1785 pool->buff_size = value;
1786 if ((rc = ibmveth_open(netdev)))
1787 return rc;
1788 } else {
1789 pool->buff_size = value;
1790 }
1791 }
1792 }
1793
1794 /* kick the interrupt handler to allocate/deallocate pools */
1795 ibmveth_interrupt(netdev->irq, netdev);
1796 return count;
1797}
1798
1799
1800#define ATTR(_name, _mode) \
1801 struct attribute veth_##_name##_attr = { \
1802 .name = __stringify(_name), .mode = _mode, \
1803 };
1804
1805static ATTR(active, 0644);
1806static ATTR(num, 0644);
1807static ATTR(size, 0644);
1808
1809static struct attribute *veth_pool_attrs[] = {
1810 &veth_active_attr,
1811 &veth_num_attr,
1812 &veth_size_attr,
1813 NULL,
1814};
1815
1816static const struct sysfs_ops veth_pool_ops = {
1817 .show = veth_pool_show,
1818 .store = veth_pool_store,
1819};
1820
1821static struct kobj_type ktype_veth_pool = {
1822 .release = NULL,
1823 .sysfs_ops = &veth_pool_ops,
1824 .default_attrs = veth_pool_attrs,
1825};
1826
1827static int ibmveth_resume(struct device *dev)
1828{
1829 struct net_device *netdev = dev_get_drvdata(dev);
1830 ibmveth_interrupt(netdev->irq, netdev);
1831 return 0;
1832}
1833
1834static struct vio_device_id ibmveth_device_table[] = {
1835 { "network", "IBM,l-lan"},
1836 { "", "" }
1837};
1838MODULE_DEVICE_TABLE(vio, ibmveth_device_table);
1839
1840static struct dev_pm_ops ibmveth_pm_ops = {
1841 .resume = ibmveth_resume
1842};
1843
1844static struct vio_driver ibmveth_driver = {
1845 .id_table = ibmveth_device_table,
1846 .probe = ibmveth_probe,
1847 .remove = ibmveth_remove,
1848 .get_desired_dma = ibmveth_get_desired_dma,
1849 .name = ibmveth_driver_name,
1850 .pm = &ibmveth_pm_ops,
1851};
1852
1853static int __init ibmveth_module_init(void)
1854{
1855 printk(KERN_DEBUG "%s: %s %s\n", ibmveth_driver_name,
1856 ibmveth_driver_string, ibmveth_driver_version);
1857
1858 return vio_register_driver(&ibmveth_driver);
1859}
1860
1861static void __exit ibmveth_module_exit(void)
1862{
1863 vio_unregister_driver(&ibmveth_driver);
1864}
1865
1866module_init(ibmveth_module_init);
1867module_exit(ibmveth_module_exit);