x86-64, NUMA: make numa_cleanup_meminfo() prettier
[linux-2.6/linux-acpi-2.6/ibm-acpi-2.6.git] / drivers / net / ibmveth.c
blob5522d459654c7a870bf63970ab094e285c040a65
1 /*
2 * IBM Power Virtual Ethernet Device Driver
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.
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.
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.
18 * Copyright (C) IBM Corporation, 2003, 2010
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>
27 #include <linux/module.h>
28 #include <linux/moduleparam.h>
29 #include <linux/types.h>
30 #include <linux/errno.h>
31 #include <linux/dma-mapping.h>
32 #include <linux/kernel.h>
33 #include <linux/netdevice.h>
34 #include <linux/etherdevice.h>
35 #include <linux/skbuff.h>
36 #include <linux/init.h>
37 #include <linux/mm.h>
38 #include <linux/pm.h>
39 #include <linux/ethtool.h>
40 #include <linux/in.h>
41 #include <linux/ip.h>
42 #include <linux/ipv6.h>
43 #include <linux/slab.h>
44 #include <asm/hvcall.h>
45 #include <asm/atomic.h>
46 #include <asm/vio.h>
47 #include <asm/iommu.h>
48 #include <asm/firmware.h>
50 #include "ibmveth.h"
52 static irqreturn_t ibmveth_interrupt(int irq, void *dev_instance);
53 static void ibmveth_rxq_harvest_buffer(struct ibmveth_adapter *adapter);
54 static unsigned long ibmveth_get_desired_dma(struct vio_dev *vdev);
56 static struct kobj_type ktype_veth_pool;
59 static const char ibmveth_driver_name[] = "ibmveth";
60 static const char ibmveth_driver_string[] = "IBM Power Virtual Ethernet Driver";
61 #define ibmveth_driver_version "1.04"
63 MODULE_AUTHOR("Santiago Leon <santil@linux.vnet.ibm.com>");
64 MODULE_DESCRIPTION("IBM Power Virtual Ethernet Driver");
65 MODULE_LICENSE("GPL");
66 MODULE_VERSION(ibmveth_driver_version);
68 static unsigned int tx_copybreak __read_mostly = 128;
69 module_param(tx_copybreak, uint, 0644);
70 MODULE_PARM_DESC(tx_copybreak,
71 "Maximum size of packet that is copied to a new buffer on transmit");
73 static unsigned int rx_copybreak __read_mostly = 128;
74 module_param(rx_copybreak, uint, 0644);
75 MODULE_PARM_DESC(rx_copybreak,
76 "Maximum size of packet that is copied to a new buffer on receive");
78 static unsigned int rx_flush __read_mostly = 0;
79 module_param(rx_flush, uint, 0644);
80 MODULE_PARM_DESC(rx_flush, "Flush receive buffers before use");
82 struct ibmveth_stat {
83 char name[ETH_GSTRING_LEN];
84 int offset;
87 #define IBMVETH_STAT_OFF(stat) offsetof(struct ibmveth_adapter, stat)
88 #define IBMVETH_GET_STAT(a, off) *((u64 *)(((unsigned long)(a)) + off))
90 struct ibmveth_stat ibmveth_stats[] = {
91 { "replenish_task_cycles", IBMVETH_STAT_OFF(replenish_task_cycles) },
92 { "replenish_no_mem", IBMVETH_STAT_OFF(replenish_no_mem) },
93 { "replenish_add_buff_failure",
94 IBMVETH_STAT_OFF(replenish_add_buff_failure) },
95 { "replenish_add_buff_success",
96 IBMVETH_STAT_OFF(replenish_add_buff_success) },
97 { "rx_invalid_buffer", IBMVETH_STAT_OFF(rx_invalid_buffer) },
98 { "rx_no_buffer", IBMVETH_STAT_OFF(rx_no_buffer) },
99 { "tx_map_failed", IBMVETH_STAT_OFF(tx_map_failed) },
100 { "tx_send_failed", IBMVETH_STAT_OFF(tx_send_failed) },
101 { "fw_enabled_ipv4_csum", IBMVETH_STAT_OFF(fw_ipv4_csum_support) },
102 { "fw_enabled_ipv6_csum", IBMVETH_STAT_OFF(fw_ipv6_csum_support) },
105 /* simple methods of getting data from the current rxq entry */
106 static inline u32 ibmveth_rxq_flags(struct ibmveth_adapter *adapter)
108 return adapter->rx_queue.queue_addr[adapter->rx_queue.index].flags_off;
111 static inline int ibmveth_rxq_toggle(struct ibmveth_adapter *adapter)
113 return (ibmveth_rxq_flags(adapter) & IBMVETH_RXQ_TOGGLE) >>
114 IBMVETH_RXQ_TOGGLE_SHIFT;
117 static inline int ibmveth_rxq_pending_buffer(struct ibmveth_adapter *adapter)
119 return ibmveth_rxq_toggle(adapter) == adapter->rx_queue.toggle;
122 static inline int ibmveth_rxq_buffer_valid(struct ibmveth_adapter *adapter)
124 return ibmveth_rxq_flags(adapter) & IBMVETH_RXQ_VALID;
127 static inline int ibmveth_rxq_frame_offset(struct ibmveth_adapter *adapter)
129 return ibmveth_rxq_flags(adapter) & IBMVETH_RXQ_OFF_MASK;
132 static inline int ibmveth_rxq_frame_length(struct ibmveth_adapter *adapter)
134 return adapter->rx_queue.queue_addr[adapter->rx_queue.index].length;
137 static inline int ibmveth_rxq_csum_good(struct ibmveth_adapter *adapter)
139 return ibmveth_rxq_flags(adapter) & IBMVETH_RXQ_CSUM_GOOD;
142 /* setup the initial settings for a buffer pool */
143 static void ibmveth_init_buffer_pool(struct ibmveth_buff_pool *pool,
144 u32 pool_index, u32 pool_size,
145 u32 buff_size, u32 pool_active)
147 pool->size = pool_size;
148 pool->index = pool_index;
149 pool->buff_size = buff_size;
150 pool->threshold = pool_size * 7 / 8;
151 pool->active = pool_active;
154 /* allocate and setup an buffer pool - called during open */
155 static int ibmveth_alloc_buffer_pool(struct ibmveth_buff_pool *pool)
157 int i;
159 pool->free_map = kmalloc(sizeof(u16) * pool->size, GFP_KERNEL);
161 if (!pool->free_map)
162 return -1;
164 pool->dma_addr = kmalloc(sizeof(dma_addr_t) * pool->size, GFP_KERNEL);
165 if (!pool->dma_addr) {
166 kfree(pool->free_map);
167 pool->free_map = NULL;
168 return -1;
171 pool->skbuff = kcalloc(pool->size, sizeof(void *), GFP_KERNEL);
173 if (!pool->skbuff) {
174 kfree(pool->dma_addr);
175 pool->dma_addr = NULL;
177 kfree(pool->free_map);
178 pool->free_map = NULL;
179 return -1;
182 memset(pool->dma_addr, 0, sizeof(dma_addr_t) * pool->size);
184 for (i = 0; i < pool->size; ++i)
185 pool->free_map[i] = i;
187 atomic_set(&pool->available, 0);
188 pool->producer_index = 0;
189 pool->consumer_index = 0;
191 return 0;
194 static inline void ibmveth_flush_buffer(void *addr, unsigned long length)
196 unsigned long offset;
198 for (offset = 0; offset < length; offset += SMP_CACHE_BYTES)
199 asm("dcbfl %0,%1" :: "b" (addr), "r" (offset));
202 /* replenish the buffers for a pool. note that we don't need to
203 * skb_reserve these since they are used for incoming...
205 static void ibmveth_replenish_buffer_pool(struct ibmveth_adapter *adapter,
206 struct ibmveth_buff_pool *pool)
208 u32 i;
209 u32 count = pool->size - atomic_read(&pool->available);
210 u32 buffers_added = 0;
211 struct sk_buff *skb;
212 unsigned int free_index, index;
213 u64 correlator;
214 unsigned long lpar_rc;
215 dma_addr_t dma_addr;
217 mb();
219 for (i = 0; i < count; ++i) {
220 union ibmveth_buf_desc desc;
222 skb = netdev_alloc_skb(adapter->netdev, pool->buff_size);
224 if (!skb) {
225 netdev_dbg(adapter->netdev,
226 "replenish: unable to allocate skb\n");
227 adapter->replenish_no_mem++;
228 break;
231 free_index = pool->consumer_index;
232 pool->consumer_index++;
233 if (pool->consumer_index >= pool->size)
234 pool->consumer_index = 0;
235 index = pool->free_map[free_index];
237 BUG_ON(index == IBM_VETH_INVALID_MAP);
238 BUG_ON(pool->skbuff[index] != NULL);
240 dma_addr = dma_map_single(&adapter->vdev->dev, skb->data,
241 pool->buff_size, DMA_FROM_DEVICE);
243 if (dma_mapping_error(&adapter->vdev->dev, dma_addr))
244 goto failure;
246 pool->free_map[free_index] = IBM_VETH_INVALID_MAP;
247 pool->dma_addr[index] = dma_addr;
248 pool->skbuff[index] = skb;
250 correlator = ((u64)pool->index << 32) | index;
251 *(u64 *)skb->data = correlator;
253 desc.fields.flags_len = IBMVETH_BUF_VALID | pool->buff_size;
254 desc.fields.address = dma_addr;
256 if (rx_flush) {
257 unsigned int len = min(pool->buff_size,
258 adapter->netdev->mtu +
259 IBMVETH_BUFF_OH);
260 ibmveth_flush_buffer(skb->data, len);
262 lpar_rc = h_add_logical_lan_buffer(adapter->vdev->unit_address,
263 desc.desc);
265 if (lpar_rc != H_SUCCESS) {
266 goto failure;
267 } else {
268 buffers_added++;
269 adapter->replenish_add_buff_success++;
273 mb();
274 atomic_add(buffers_added, &(pool->available));
275 return;
277 failure:
278 pool->free_map[free_index] = index;
279 pool->skbuff[index] = NULL;
280 if (pool->consumer_index == 0)
281 pool->consumer_index = pool->size - 1;
282 else
283 pool->consumer_index--;
284 if (!dma_mapping_error(&adapter->vdev->dev, dma_addr))
285 dma_unmap_single(&adapter->vdev->dev,
286 pool->dma_addr[index], pool->buff_size,
287 DMA_FROM_DEVICE);
288 dev_kfree_skb_any(skb);
289 adapter->replenish_add_buff_failure++;
291 mb();
292 atomic_add(buffers_added, &(pool->available));
295 /* replenish routine */
296 static void ibmveth_replenish_task(struct ibmveth_adapter *adapter)
298 int i;
300 adapter->replenish_task_cycles++;
302 for (i = (IBMVETH_NUM_BUFF_POOLS - 1); i >= 0; i--) {
303 struct ibmveth_buff_pool *pool = &adapter->rx_buff_pool[i];
305 if (pool->active &&
306 (atomic_read(&pool->available) < pool->threshold))
307 ibmveth_replenish_buffer_pool(adapter, pool);
310 adapter->rx_no_buffer = *(u64 *)(((char*)adapter->buffer_list_addr) +
311 4096 - 8);
314 /* empty and free ana buffer pool - also used to do cleanup in error paths */
315 static void ibmveth_free_buffer_pool(struct ibmveth_adapter *adapter,
316 struct ibmveth_buff_pool *pool)
318 int i;
320 kfree(pool->free_map);
321 pool->free_map = NULL;
323 if (pool->skbuff && pool->dma_addr) {
324 for (i = 0; i < pool->size; ++i) {
325 struct sk_buff *skb = pool->skbuff[i];
326 if (skb) {
327 dma_unmap_single(&adapter->vdev->dev,
328 pool->dma_addr[i],
329 pool->buff_size,
330 DMA_FROM_DEVICE);
331 dev_kfree_skb_any(skb);
332 pool->skbuff[i] = NULL;
337 if (pool->dma_addr) {
338 kfree(pool->dma_addr);
339 pool->dma_addr = NULL;
342 if (pool->skbuff) {
343 kfree(pool->skbuff);
344 pool->skbuff = NULL;
348 /* remove a buffer from a pool */
349 static void ibmveth_remove_buffer_from_pool(struct ibmveth_adapter *adapter,
350 u64 correlator)
352 unsigned int pool = correlator >> 32;
353 unsigned int index = correlator & 0xffffffffUL;
354 unsigned int free_index;
355 struct sk_buff *skb;
357 BUG_ON(pool >= IBMVETH_NUM_BUFF_POOLS);
358 BUG_ON(index >= adapter->rx_buff_pool[pool].size);
360 skb = adapter->rx_buff_pool[pool].skbuff[index];
362 BUG_ON(skb == NULL);
364 adapter->rx_buff_pool[pool].skbuff[index] = NULL;
366 dma_unmap_single(&adapter->vdev->dev,
367 adapter->rx_buff_pool[pool].dma_addr[index],
368 adapter->rx_buff_pool[pool].buff_size,
369 DMA_FROM_DEVICE);
371 free_index = adapter->rx_buff_pool[pool].producer_index;
372 adapter->rx_buff_pool[pool].producer_index++;
373 if (adapter->rx_buff_pool[pool].producer_index >=
374 adapter->rx_buff_pool[pool].size)
375 adapter->rx_buff_pool[pool].producer_index = 0;
376 adapter->rx_buff_pool[pool].free_map[free_index] = index;
378 mb();
380 atomic_dec(&(adapter->rx_buff_pool[pool].available));
383 /* get the current buffer on the rx queue */
384 static inline struct sk_buff *ibmveth_rxq_get_buffer(struct ibmveth_adapter *adapter)
386 u64 correlator = adapter->rx_queue.queue_addr[adapter->rx_queue.index].correlator;
387 unsigned int pool = correlator >> 32;
388 unsigned int index = correlator & 0xffffffffUL;
390 BUG_ON(pool >= IBMVETH_NUM_BUFF_POOLS);
391 BUG_ON(index >= adapter->rx_buff_pool[pool].size);
393 return adapter->rx_buff_pool[pool].skbuff[index];
396 /* recycle the current buffer on the rx queue */
397 static void ibmveth_rxq_recycle_buffer(struct ibmveth_adapter *adapter)
399 u32 q_index = adapter->rx_queue.index;
400 u64 correlator = adapter->rx_queue.queue_addr[q_index].correlator;
401 unsigned int pool = correlator >> 32;
402 unsigned int index = correlator & 0xffffffffUL;
403 union ibmveth_buf_desc desc;
404 unsigned long lpar_rc;
406 BUG_ON(pool >= IBMVETH_NUM_BUFF_POOLS);
407 BUG_ON(index >= adapter->rx_buff_pool[pool].size);
409 if (!adapter->rx_buff_pool[pool].active) {
410 ibmveth_rxq_harvest_buffer(adapter);
411 ibmveth_free_buffer_pool(adapter, &adapter->rx_buff_pool[pool]);
412 return;
415 desc.fields.flags_len = IBMVETH_BUF_VALID |
416 adapter->rx_buff_pool[pool].buff_size;
417 desc.fields.address = adapter->rx_buff_pool[pool].dma_addr[index];
419 lpar_rc = h_add_logical_lan_buffer(adapter->vdev->unit_address, desc.desc);
421 if (lpar_rc != H_SUCCESS) {
422 netdev_dbg(adapter->netdev, "h_add_logical_lan_buffer failed "
423 "during recycle rc=%ld", lpar_rc);
424 ibmveth_remove_buffer_from_pool(adapter, adapter->rx_queue.queue_addr[adapter->rx_queue.index].correlator);
427 if (++adapter->rx_queue.index == adapter->rx_queue.num_slots) {
428 adapter->rx_queue.index = 0;
429 adapter->rx_queue.toggle = !adapter->rx_queue.toggle;
433 static void ibmveth_rxq_harvest_buffer(struct ibmveth_adapter *adapter)
435 ibmveth_remove_buffer_from_pool(adapter, adapter->rx_queue.queue_addr[adapter->rx_queue.index].correlator);
437 if (++adapter->rx_queue.index == adapter->rx_queue.num_slots) {
438 adapter->rx_queue.index = 0;
439 adapter->rx_queue.toggle = !adapter->rx_queue.toggle;
443 static void ibmveth_cleanup(struct ibmveth_adapter *adapter)
445 int i;
446 struct device *dev = &adapter->vdev->dev;
448 if (adapter->buffer_list_addr != NULL) {
449 if (!dma_mapping_error(dev, adapter->buffer_list_dma)) {
450 dma_unmap_single(dev, adapter->buffer_list_dma, 4096,
451 DMA_BIDIRECTIONAL);
452 adapter->buffer_list_dma = DMA_ERROR_CODE;
454 free_page((unsigned long)adapter->buffer_list_addr);
455 adapter->buffer_list_addr = NULL;
458 if (adapter->filter_list_addr != NULL) {
459 if (!dma_mapping_error(dev, adapter->filter_list_dma)) {
460 dma_unmap_single(dev, adapter->filter_list_dma, 4096,
461 DMA_BIDIRECTIONAL);
462 adapter->filter_list_dma = DMA_ERROR_CODE;
464 free_page((unsigned long)adapter->filter_list_addr);
465 adapter->filter_list_addr = NULL;
468 if (adapter->rx_queue.queue_addr != NULL) {
469 if (!dma_mapping_error(dev, adapter->rx_queue.queue_dma)) {
470 dma_unmap_single(dev,
471 adapter->rx_queue.queue_dma,
472 adapter->rx_queue.queue_len,
473 DMA_BIDIRECTIONAL);
474 adapter->rx_queue.queue_dma = DMA_ERROR_CODE;
476 kfree(adapter->rx_queue.queue_addr);
477 adapter->rx_queue.queue_addr = NULL;
480 for (i = 0; i < IBMVETH_NUM_BUFF_POOLS; i++)
481 if (adapter->rx_buff_pool[i].active)
482 ibmveth_free_buffer_pool(adapter,
483 &adapter->rx_buff_pool[i]);
485 if (adapter->bounce_buffer != NULL) {
486 if (!dma_mapping_error(dev, adapter->bounce_buffer_dma)) {
487 dma_unmap_single(&adapter->vdev->dev,
488 adapter->bounce_buffer_dma,
489 adapter->netdev->mtu + IBMVETH_BUFF_OH,
490 DMA_BIDIRECTIONAL);
491 adapter->bounce_buffer_dma = DMA_ERROR_CODE;
493 kfree(adapter->bounce_buffer);
494 adapter->bounce_buffer = NULL;
498 static int ibmveth_register_logical_lan(struct ibmveth_adapter *adapter,
499 union ibmveth_buf_desc rxq_desc, u64 mac_address)
501 int rc, try_again = 1;
504 * After a kexec the adapter will still be open, so our attempt to
505 * open it will fail. So if we get a failure we free the adapter and
506 * try again, but only once.
508 retry:
509 rc = h_register_logical_lan(adapter->vdev->unit_address,
510 adapter->buffer_list_dma, rxq_desc.desc,
511 adapter->filter_list_dma, mac_address);
513 if (rc != H_SUCCESS && try_again) {
514 do {
515 rc = h_free_logical_lan(adapter->vdev->unit_address);
516 } while (H_IS_LONG_BUSY(rc) || (rc == H_BUSY));
518 try_again = 0;
519 goto retry;
522 return rc;
525 static int ibmveth_open(struct net_device *netdev)
527 struct ibmveth_adapter *adapter = netdev_priv(netdev);
528 u64 mac_address = 0;
529 int rxq_entries = 1;
530 unsigned long lpar_rc;
531 int rc;
532 union ibmveth_buf_desc rxq_desc;
533 int i;
534 struct device *dev;
536 netdev_dbg(netdev, "open starting\n");
538 napi_enable(&adapter->napi);
540 for(i = 0; i < IBMVETH_NUM_BUFF_POOLS; i++)
541 rxq_entries += adapter->rx_buff_pool[i].size;
543 adapter->buffer_list_addr = (void*) get_zeroed_page(GFP_KERNEL);
544 adapter->filter_list_addr = (void*) get_zeroed_page(GFP_KERNEL);
546 if (!adapter->buffer_list_addr || !adapter->filter_list_addr) {
547 netdev_err(netdev, "unable to allocate filter or buffer list "
548 "pages\n");
549 rc = -ENOMEM;
550 goto err_out;
553 adapter->rx_queue.queue_len = sizeof(struct ibmveth_rx_q_entry) *
554 rxq_entries;
555 adapter->rx_queue.queue_addr = kmalloc(adapter->rx_queue.queue_len,
556 GFP_KERNEL);
558 if (!adapter->rx_queue.queue_addr) {
559 netdev_err(netdev, "unable to allocate rx queue pages\n");
560 rc = -ENOMEM;
561 goto err_out;
564 dev = &adapter->vdev->dev;
566 adapter->buffer_list_dma = dma_map_single(dev,
567 adapter->buffer_list_addr, 4096, DMA_BIDIRECTIONAL);
568 adapter->filter_list_dma = dma_map_single(dev,
569 adapter->filter_list_addr, 4096, DMA_BIDIRECTIONAL);
570 adapter->rx_queue.queue_dma = dma_map_single(dev,
571 adapter->rx_queue.queue_addr,
572 adapter->rx_queue.queue_len, DMA_BIDIRECTIONAL);
574 if ((dma_mapping_error(dev, adapter->buffer_list_dma)) ||
575 (dma_mapping_error(dev, adapter->filter_list_dma)) ||
576 (dma_mapping_error(dev, adapter->rx_queue.queue_dma))) {
577 netdev_err(netdev, "unable to map filter or buffer list "
578 "pages\n");
579 rc = -ENOMEM;
580 goto err_out;
583 adapter->rx_queue.index = 0;
584 adapter->rx_queue.num_slots = rxq_entries;
585 adapter->rx_queue.toggle = 1;
587 memcpy(&mac_address, netdev->dev_addr, netdev->addr_len);
588 mac_address = mac_address >> 16;
590 rxq_desc.fields.flags_len = IBMVETH_BUF_VALID |
591 adapter->rx_queue.queue_len;
592 rxq_desc.fields.address = adapter->rx_queue.queue_dma;
594 netdev_dbg(netdev, "buffer list @ 0x%p\n", adapter->buffer_list_addr);
595 netdev_dbg(netdev, "filter list @ 0x%p\n", adapter->filter_list_addr);
596 netdev_dbg(netdev, "receive q @ 0x%p\n", adapter->rx_queue.queue_addr);
598 h_vio_signal(adapter->vdev->unit_address, VIO_IRQ_DISABLE);
600 lpar_rc = ibmveth_register_logical_lan(adapter, rxq_desc, mac_address);
602 if (lpar_rc != H_SUCCESS) {
603 netdev_err(netdev, "h_register_logical_lan failed with %ld\n",
604 lpar_rc);
605 netdev_err(netdev, "buffer TCE:0x%llx filter TCE:0x%llx rxq "
606 "desc:0x%llx MAC:0x%llx\n",
607 adapter->buffer_list_dma,
608 adapter->filter_list_dma,
609 rxq_desc.desc,
610 mac_address);
611 rc = -ENONET;
612 goto err_out;
615 for (i = 0; i < IBMVETH_NUM_BUFF_POOLS; i++) {
616 if (!adapter->rx_buff_pool[i].active)
617 continue;
618 if (ibmveth_alloc_buffer_pool(&adapter->rx_buff_pool[i])) {
619 netdev_err(netdev, "unable to alloc pool\n");
620 adapter->rx_buff_pool[i].active = 0;
621 rc = -ENOMEM;
622 goto err_out;
626 netdev_dbg(netdev, "registering irq 0x%x\n", netdev->irq);
627 rc = request_irq(netdev->irq, ibmveth_interrupt, 0, netdev->name,
628 netdev);
629 if (rc != 0) {
630 netdev_err(netdev, "unable to request irq 0x%x, rc %d\n",
631 netdev->irq, rc);
632 do {
633 rc = h_free_logical_lan(adapter->vdev->unit_address);
634 } while (H_IS_LONG_BUSY(rc) || (rc == H_BUSY));
636 goto err_out;
639 adapter->bounce_buffer =
640 kmalloc(netdev->mtu + IBMVETH_BUFF_OH, GFP_KERNEL);
641 if (!adapter->bounce_buffer) {
642 netdev_err(netdev, "unable to allocate bounce buffer\n");
643 rc = -ENOMEM;
644 goto err_out_free_irq;
646 adapter->bounce_buffer_dma =
647 dma_map_single(&adapter->vdev->dev, adapter->bounce_buffer,
648 netdev->mtu + IBMVETH_BUFF_OH, DMA_BIDIRECTIONAL);
649 if (dma_mapping_error(dev, adapter->bounce_buffer_dma)) {
650 netdev_err(netdev, "unable to map bounce buffer\n");
651 rc = -ENOMEM;
652 goto err_out_free_irq;
655 netdev_dbg(netdev, "initial replenish cycle\n");
656 ibmveth_interrupt(netdev->irq, netdev);
658 netif_start_queue(netdev);
660 netdev_dbg(netdev, "open complete\n");
662 return 0;
664 err_out_free_irq:
665 free_irq(netdev->irq, netdev);
666 err_out:
667 ibmveth_cleanup(adapter);
668 napi_disable(&adapter->napi);
669 return rc;
672 static int ibmveth_close(struct net_device *netdev)
674 struct ibmveth_adapter *adapter = netdev_priv(netdev);
675 long lpar_rc;
677 netdev_dbg(netdev, "close starting\n");
679 napi_disable(&adapter->napi);
681 if (!adapter->pool_config)
682 netif_stop_queue(netdev);
684 h_vio_signal(adapter->vdev->unit_address, VIO_IRQ_DISABLE);
686 do {
687 lpar_rc = h_free_logical_lan(adapter->vdev->unit_address);
688 } while (H_IS_LONG_BUSY(lpar_rc) || (lpar_rc == H_BUSY));
690 if (lpar_rc != H_SUCCESS) {
691 netdev_err(netdev, "h_free_logical_lan failed with %lx, "
692 "continuing with close\n", lpar_rc);
695 free_irq(netdev->irq, netdev);
697 adapter->rx_no_buffer = *(u64 *)(((char *)adapter->buffer_list_addr) +
698 4096 - 8);
700 ibmveth_cleanup(adapter);
702 netdev_dbg(netdev, "close complete\n");
704 return 0;
707 static int netdev_get_settings(struct net_device *dev, struct ethtool_cmd *cmd)
709 cmd->supported = (SUPPORTED_1000baseT_Full | SUPPORTED_Autoneg |
710 SUPPORTED_FIBRE);
711 cmd->advertising = (ADVERTISED_1000baseT_Full | ADVERTISED_Autoneg |
712 ADVERTISED_FIBRE);
713 cmd->speed = SPEED_1000;
714 cmd->duplex = DUPLEX_FULL;
715 cmd->port = PORT_FIBRE;
716 cmd->phy_address = 0;
717 cmd->transceiver = XCVR_INTERNAL;
718 cmd->autoneg = AUTONEG_ENABLE;
719 cmd->maxtxpkt = 0;
720 cmd->maxrxpkt = 1;
721 return 0;
724 static void netdev_get_drvinfo(struct net_device *dev,
725 struct ethtool_drvinfo *info)
727 strncpy(info->driver, ibmveth_driver_name, sizeof(info->driver) - 1);
728 strncpy(info->version, ibmveth_driver_version,
729 sizeof(info->version) - 1);
732 static void ibmveth_set_rx_csum_flags(struct net_device *dev, u32 data)
734 struct ibmveth_adapter *adapter = netdev_priv(dev);
736 if (data) {
737 adapter->rx_csum = 1;
738 } else {
740 * Since the ibmveth firmware interface does not have the
741 * concept of separate tx/rx checksum offload enable, if rx
742 * checksum is disabled we also have to disable tx checksum
743 * offload. Once we disable rx checksum offload, we are no
744 * longer allowed to send tx buffers that are not properly
745 * checksummed.
747 adapter->rx_csum = 0;
748 dev->features &= ~NETIF_F_IP_CSUM;
749 dev->features &= ~NETIF_F_IPV6_CSUM;
753 static void ibmveth_set_tx_csum_flags(struct net_device *dev, u32 data)
755 struct ibmveth_adapter *adapter = netdev_priv(dev);
757 if (data) {
758 if (adapter->fw_ipv4_csum_support)
759 dev->features |= NETIF_F_IP_CSUM;
760 if (adapter->fw_ipv6_csum_support)
761 dev->features |= NETIF_F_IPV6_CSUM;
762 adapter->rx_csum = 1;
763 } else {
764 dev->features &= ~NETIF_F_IP_CSUM;
765 dev->features &= ~NETIF_F_IPV6_CSUM;
769 static int ibmveth_set_csum_offload(struct net_device *dev, u32 data,
770 void (*done) (struct net_device *, u32))
772 struct ibmveth_adapter *adapter = netdev_priv(dev);
773 unsigned long set_attr, clr_attr, ret_attr;
774 unsigned long set_attr6, clr_attr6;
775 long ret, ret6;
776 int rc1 = 0, rc2 = 0;
777 int restart = 0;
779 if (netif_running(dev)) {
780 restart = 1;
781 adapter->pool_config = 1;
782 ibmveth_close(dev);
783 adapter->pool_config = 0;
786 set_attr = 0;
787 clr_attr = 0;
789 if (data) {
790 set_attr = IBMVETH_ILLAN_IPV4_TCP_CSUM;
791 set_attr6 = IBMVETH_ILLAN_IPV6_TCP_CSUM;
792 } else {
793 clr_attr = IBMVETH_ILLAN_IPV4_TCP_CSUM;
794 clr_attr6 = IBMVETH_ILLAN_IPV6_TCP_CSUM;
797 ret = h_illan_attributes(adapter->vdev->unit_address, 0, 0, &ret_attr);
799 if (ret == H_SUCCESS && !(ret_attr & IBMVETH_ILLAN_ACTIVE_TRUNK) &&
800 !(ret_attr & IBMVETH_ILLAN_TRUNK_PRI_MASK) &&
801 (ret_attr & IBMVETH_ILLAN_PADDED_PKT_CSUM)) {
802 ret = h_illan_attributes(adapter->vdev->unit_address, clr_attr,
803 set_attr, &ret_attr);
805 if (ret != H_SUCCESS) {
806 netdev_err(dev, "unable to change IPv4 checksum "
807 "offload settings. %d rc=%ld\n",
808 data, ret);
810 ret = h_illan_attributes(adapter->vdev->unit_address,
811 set_attr, clr_attr, &ret_attr);
812 } else {
813 adapter->fw_ipv4_csum_support = data;
816 ret6 = h_illan_attributes(adapter->vdev->unit_address,
817 clr_attr6, set_attr6, &ret_attr);
819 if (ret6 != H_SUCCESS) {
820 netdev_err(dev, "unable to change IPv6 checksum "
821 "offload settings. %d rc=%ld\n",
822 data, ret);
824 ret = h_illan_attributes(adapter->vdev->unit_address,
825 set_attr6, clr_attr6,
826 &ret_attr);
827 } else
828 adapter->fw_ipv6_csum_support = data;
830 if (ret == H_SUCCESS || ret6 == H_SUCCESS)
831 done(dev, data);
832 else
833 rc1 = -EIO;
834 } else {
835 rc1 = -EIO;
836 netdev_err(dev, "unable to change checksum offload settings."
837 " %d rc=%ld ret_attr=%lx\n", data, ret,
838 ret_attr);
841 if (restart)
842 rc2 = ibmveth_open(dev);
844 return rc1 ? rc1 : rc2;
847 static int ibmveth_set_rx_csum(struct net_device *dev, u32 data)
849 struct ibmveth_adapter *adapter = netdev_priv(dev);
851 if ((data && adapter->rx_csum) || (!data && !adapter->rx_csum))
852 return 0;
854 return ibmveth_set_csum_offload(dev, data, ibmveth_set_rx_csum_flags);
857 static int ibmveth_set_tx_csum(struct net_device *dev, u32 data)
859 struct ibmveth_adapter *adapter = netdev_priv(dev);
860 int rc = 0;
862 if (data && (dev->features & (NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM)))
863 return 0;
864 if (!data && !(dev->features & (NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM)))
865 return 0;
867 if (data && !adapter->rx_csum)
868 rc = ibmveth_set_csum_offload(dev, data,
869 ibmveth_set_tx_csum_flags);
870 else
871 ibmveth_set_tx_csum_flags(dev, data);
873 return rc;
876 static u32 ibmveth_get_rx_csum(struct net_device *dev)
878 struct ibmveth_adapter *adapter = netdev_priv(dev);
879 return adapter->rx_csum;
882 static void ibmveth_get_strings(struct net_device *dev, u32 stringset, u8 *data)
884 int i;
886 if (stringset != ETH_SS_STATS)
887 return;
889 for (i = 0; i < ARRAY_SIZE(ibmveth_stats); i++, data += ETH_GSTRING_LEN)
890 memcpy(data, ibmveth_stats[i].name, ETH_GSTRING_LEN);
893 static int ibmveth_get_sset_count(struct net_device *dev, int sset)
895 switch (sset) {
896 case ETH_SS_STATS:
897 return ARRAY_SIZE(ibmveth_stats);
898 default:
899 return -EOPNOTSUPP;
903 static void ibmveth_get_ethtool_stats(struct net_device *dev,
904 struct ethtool_stats *stats, u64 *data)
906 int i;
907 struct ibmveth_adapter *adapter = netdev_priv(dev);
909 for (i = 0; i < ARRAY_SIZE(ibmveth_stats); i++)
910 data[i] = IBMVETH_GET_STAT(adapter, ibmveth_stats[i].offset);
913 static const struct ethtool_ops netdev_ethtool_ops = {
914 .get_drvinfo = netdev_get_drvinfo,
915 .get_settings = netdev_get_settings,
916 .get_link = ethtool_op_get_link,
917 .set_tx_csum = ibmveth_set_tx_csum,
918 .get_rx_csum = ibmveth_get_rx_csum,
919 .set_rx_csum = ibmveth_set_rx_csum,
920 .get_strings = ibmveth_get_strings,
921 .get_sset_count = ibmveth_get_sset_count,
922 .get_ethtool_stats = ibmveth_get_ethtool_stats,
923 .set_sg = ethtool_op_set_sg,
926 static int ibmveth_ioctl(struct net_device *dev, struct ifreq *ifr, int cmd)
928 return -EOPNOTSUPP;
931 #define page_offset(v) ((unsigned long)(v) & ((1 << 12) - 1))
933 static int ibmveth_send(struct ibmveth_adapter *adapter,
934 union ibmveth_buf_desc *descs)
936 unsigned long correlator;
937 unsigned int retry_count;
938 unsigned long ret;
941 * The retry count sets a maximum for the number of broadcast and
942 * multicast destinations within the system.
944 retry_count = 1024;
945 correlator = 0;
946 do {
947 ret = h_send_logical_lan(adapter->vdev->unit_address,
948 descs[0].desc, descs[1].desc,
949 descs[2].desc, descs[3].desc,
950 descs[4].desc, descs[5].desc,
951 correlator, &correlator);
952 } while ((ret == H_BUSY) && (retry_count--));
954 if (ret != H_SUCCESS && ret != H_DROPPED) {
955 netdev_err(adapter->netdev, "tx: h_send_logical_lan failed "
956 "with rc=%ld\n", ret);
957 return 1;
960 return 0;
963 static netdev_tx_t ibmveth_start_xmit(struct sk_buff *skb,
964 struct net_device *netdev)
966 struct ibmveth_adapter *adapter = netdev_priv(netdev);
967 unsigned int desc_flags;
968 union ibmveth_buf_desc descs[6];
969 int last, i;
970 int force_bounce = 0;
973 * veth handles a maximum of 6 segments including the header, so
974 * we have to linearize the skb if there are more than this.
976 if (skb_shinfo(skb)->nr_frags > 5 && __skb_linearize(skb)) {
977 netdev->stats.tx_dropped++;
978 goto out;
981 /* veth can't checksum offload UDP */
982 if (skb->ip_summed == CHECKSUM_PARTIAL &&
983 ((skb->protocol == htons(ETH_P_IP) &&
984 ip_hdr(skb)->protocol != IPPROTO_TCP) ||
985 (skb->protocol == htons(ETH_P_IPV6) &&
986 ipv6_hdr(skb)->nexthdr != IPPROTO_TCP)) &&
987 skb_checksum_help(skb)) {
989 netdev_err(netdev, "tx: failed to checksum packet\n");
990 netdev->stats.tx_dropped++;
991 goto out;
994 desc_flags = IBMVETH_BUF_VALID;
996 if (skb->ip_summed == CHECKSUM_PARTIAL) {
997 unsigned char *buf = skb_transport_header(skb) +
998 skb->csum_offset;
1000 desc_flags |= (IBMVETH_BUF_NO_CSUM | IBMVETH_BUF_CSUM_GOOD);
1002 /* Need to zero out the checksum */
1003 buf[0] = 0;
1004 buf[1] = 0;
1007 retry_bounce:
1008 memset(descs, 0, sizeof(descs));
1011 * If a linear packet is below the rx threshold then
1012 * copy it into the static bounce buffer. This avoids the
1013 * cost of a TCE insert and remove.
1015 if (force_bounce || (!skb_is_nonlinear(skb) &&
1016 (skb->len < tx_copybreak))) {
1017 skb_copy_from_linear_data(skb, adapter->bounce_buffer,
1018 skb->len);
1020 descs[0].fields.flags_len = desc_flags | skb->len;
1021 descs[0].fields.address = adapter->bounce_buffer_dma;
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;
1031 goto out;
1034 /* Map the header */
1035 descs[0].fields.address = dma_map_single(&adapter->vdev->dev, skb->data,
1036 skb_headlen(skb),
1037 DMA_TO_DEVICE);
1038 if (dma_mapping_error(&adapter->vdev->dev, descs[0].fields.address))
1039 goto map_failed;
1041 descs[0].fields.flags_len = desc_flags | skb_headlen(skb);
1043 /* Map the frags */
1044 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
1045 unsigned long dma_addr;
1046 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
1048 dma_addr = dma_map_page(&adapter->vdev->dev, frag->page,
1049 frag->page_offset, frag->size,
1050 DMA_TO_DEVICE);
1052 if (dma_mapping_error(&adapter->vdev->dev, dma_addr))
1053 goto map_failed_frags;
1055 descs[i+1].fields.flags_len = desc_flags | frag->size;
1056 descs[i+1].fields.address = dma_addr;
1059 if (ibmveth_send(adapter, descs)) {
1060 adapter->tx_send_failed++;
1061 netdev->stats.tx_dropped++;
1062 } else {
1063 netdev->stats.tx_packets++;
1064 netdev->stats.tx_bytes += skb->len;
1067 for (i = 0; i < skb_shinfo(skb)->nr_frags + 1; i++)
1068 dma_unmap_page(&adapter->vdev->dev, descs[i].fields.address,
1069 descs[i].fields.flags_len & IBMVETH_BUF_LEN_MASK,
1070 DMA_TO_DEVICE);
1072 out:
1073 dev_kfree_skb(skb);
1074 return NETDEV_TX_OK;
1076 map_failed_frags:
1077 last = i+1;
1078 for (i = 0; i < last; i++)
1079 dma_unmap_page(&adapter->vdev->dev, descs[i].fields.address,
1080 descs[i].fields.flags_len & IBMVETH_BUF_LEN_MASK,
1081 DMA_TO_DEVICE);
1083 map_failed:
1084 if (!firmware_has_feature(FW_FEATURE_CMO))
1085 netdev_err(netdev, "tx: unable to map xmit buffer\n");
1086 adapter->tx_map_failed++;
1087 skb_linearize(skb);
1088 force_bounce = 1;
1089 goto retry_bounce;
1092 static int ibmveth_poll(struct napi_struct *napi, int budget)
1094 struct ibmveth_adapter *adapter =
1095 container_of(napi, struct ibmveth_adapter, napi);
1096 struct net_device *netdev = adapter->netdev;
1097 int frames_processed = 0;
1098 unsigned long lpar_rc;
1100 restart_poll:
1101 do {
1102 if (!ibmveth_rxq_pending_buffer(adapter))
1103 break;
1105 smp_rmb();
1106 if (!ibmveth_rxq_buffer_valid(adapter)) {
1107 wmb(); /* suggested by larson1 */
1108 adapter->rx_invalid_buffer++;
1109 netdev_dbg(netdev, "recycling invalid buffer\n");
1110 ibmveth_rxq_recycle_buffer(adapter);
1111 } else {
1112 struct sk_buff *skb, *new_skb;
1113 int length = ibmveth_rxq_frame_length(adapter);
1114 int offset = ibmveth_rxq_frame_offset(adapter);
1115 int csum_good = ibmveth_rxq_csum_good(adapter);
1117 skb = ibmveth_rxq_get_buffer(adapter);
1119 new_skb = NULL;
1120 if (length < rx_copybreak)
1121 new_skb = netdev_alloc_skb(netdev, length);
1123 if (new_skb) {
1124 skb_copy_to_linear_data(new_skb,
1125 skb->data + offset,
1126 length);
1127 if (rx_flush)
1128 ibmveth_flush_buffer(skb->data,
1129 length + offset);
1130 skb = new_skb;
1131 ibmveth_rxq_recycle_buffer(adapter);
1132 } else {
1133 ibmveth_rxq_harvest_buffer(adapter);
1134 skb_reserve(skb, offset);
1137 skb_put(skb, length);
1138 skb->protocol = eth_type_trans(skb, netdev);
1140 if (csum_good)
1141 skb->ip_summed = CHECKSUM_UNNECESSARY;
1143 netif_receive_skb(skb); /* send it up */
1145 netdev->stats.rx_packets++;
1146 netdev->stats.rx_bytes += length;
1147 frames_processed++;
1149 } while (frames_processed < budget);
1151 ibmveth_replenish_task(adapter);
1153 if (frames_processed < budget) {
1154 /* We think we are done - reenable interrupts,
1155 * then check once more to make sure we are done.
1157 lpar_rc = h_vio_signal(adapter->vdev->unit_address,
1158 VIO_IRQ_ENABLE);
1160 BUG_ON(lpar_rc != H_SUCCESS);
1162 napi_complete(napi);
1164 if (ibmveth_rxq_pending_buffer(adapter) &&
1165 napi_reschedule(napi)) {
1166 lpar_rc = h_vio_signal(adapter->vdev->unit_address,
1167 VIO_IRQ_DISABLE);
1168 goto restart_poll;
1172 return frames_processed;
1175 static irqreturn_t ibmveth_interrupt(int irq, void *dev_instance)
1177 struct net_device *netdev = dev_instance;
1178 struct ibmveth_adapter *adapter = netdev_priv(netdev);
1179 unsigned long lpar_rc;
1181 if (napi_schedule_prep(&adapter->napi)) {
1182 lpar_rc = h_vio_signal(adapter->vdev->unit_address,
1183 VIO_IRQ_DISABLE);
1184 BUG_ON(lpar_rc != H_SUCCESS);
1185 __napi_schedule(&adapter->napi);
1187 return IRQ_HANDLED;
1190 static void ibmveth_set_multicast_list(struct net_device *netdev)
1192 struct ibmveth_adapter *adapter = netdev_priv(netdev);
1193 unsigned long lpar_rc;
1195 if ((netdev->flags & IFF_PROMISC) ||
1196 (netdev_mc_count(netdev) > adapter->mcastFilterSize)) {
1197 lpar_rc = h_multicast_ctrl(adapter->vdev->unit_address,
1198 IbmVethMcastEnableRecv |
1199 IbmVethMcastDisableFiltering,
1201 if (lpar_rc != H_SUCCESS) {
1202 netdev_err(netdev, "h_multicast_ctrl rc=%ld when "
1203 "entering promisc mode\n", lpar_rc);
1205 } else {
1206 struct netdev_hw_addr *ha;
1207 /* clear the filter table & disable filtering */
1208 lpar_rc = h_multicast_ctrl(adapter->vdev->unit_address,
1209 IbmVethMcastEnableRecv |
1210 IbmVethMcastDisableFiltering |
1211 IbmVethMcastClearFilterTable,
1213 if (lpar_rc != H_SUCCESS) {
1214 netdev_err(netdev, "h_multicast_ctrl rc=%ld when "
1215 "attempting to clear filter table\n",
1216 lpar_rc);
1218 /* add the addresses to the filter table */
1219 netdev_for_each_mc_addr(ha, netdev) {
1220 /* add the multicast address to the filter table */
1221 unsigned long mcast_addr = 0;
1222 memcpy(((char *)&mcast_addr)+2, ha->addr, 6);
1223 lpar_rc = h_multicast_ctrl(adapter->vdev->unit_address,
1224 IbmVethMcastAddFilter,
1225 mcast_addr);
1226 if (lpar_rc != H_SUCCESS) {
1227 netdev_err(netdev, "h_multicast_ctrl rc=%ld "
1228 "when adding an entry to the filter "
1229 "table\n", lpar_rc);
1233 /* re-enable filtering */
1234 lpar_rc = h_multicast_ctrl(adapter->vdev->unit_address,
1235 IbmVethMcastEnableFiltering,
1237 if (lpar_rc != H_SUCCESS) {
1238 netdev_err(netdev, "h_multicast_ctrl rc=%ld when "
1239 "enabling filtering\n", lpar_rc);
1244 static int ibmveth_change_mtu(struct net_device *dev, int new_mtu)
1246 struct ibmveth_adapter *adapter = netdev_priv(dev);
1247 struct vio_dev *viodev = adapter->vdev;
1248 int new_mtu_oh = new_mtu + IBMVETH_BUFF_OH;
1249 int i, rc;
1250 int need_restart = 0;
1252 if (new_mtu < IBMVETH_MIN_MTU)
1253 return -EINVAL;
1255 for (i = 0; i < IBMVETH_NUM_BUFF_POOLS; i++)
1256 if (new_mtu_oh < adapter->rx_buff_pool[i].buff_size)
1257 break;
1259 if (i == IBMVETH_NUM_BUFF_POOLS)
1260 return -EINVAL;
1262 /* Deactivate all the buffer pools so that the next loop can activate
1263 only the buffer pools necessary to hold the new MTU */
1264 if (netif_running(adapter->netdev)) {
1265 need_restart = 1;
1266 adapter->pool_config = 1;
1267 ibmveth_close(adapter->netdev);
1268 adapter->pool_config = 0;
1271 /* Look for an active buffer pool that can hold the new MTU */
1272 for (i = 0; i < IBMVETH_NUM_BUFF_POOLS; i++) {
1273 adapter->rx_buff_pool[i].active = 1;
1275 if (new_mtu_oh < adapter->rx_buff_pool[i].buff_size) {
1276 dev->mtu = new_mtu;
1277 vio_cmo_set_dev_desired(viodev,
1278 ibmveth_get_desired_dma
1279 (viodev));
1280 if (need_restart) {
1281 return ibmveth_open(adapter->netdev);
1283 return 0;
1287 if (need_restart && (rc = ibmveth_open(adapter->netdev)))
1288 return rc;
1290 return -EINVAL;
1293 #ifdef CONFIG_NET_POLL_CONTROLLER
1294 static void ibmveth_poll_controller(struct net_device *dev)
1296 ibmveth_replenish_task(netdev_priv(dev));
1297 ibmveth_interrupt(dev->irq, dev);
1299 #endif
1302 * ibmveth_get_desired_dma - Calculate IO memory desired by the driver
1304 * @vdev: struct vio_dev for the device whose desired IO mem is to be returned
1306 * Return value:
1307 * Number of bytes of IO data the driver will need to perform well.
1309 static unsigned long ibmveth_get_desired_dma(struct vio_dev *vdev)
1311 struct net_device *netdev = dev_get_drvdata(&vdev->dev);
1312 struct ibmveth_adapter *adapter;
1313 unsigned long ret;
1314 int i;
1315 int rxqentries = 1;
1317 /* netdev inits at probe time along with the structures we need below*/
1318 if (netdev == NULL)
1319 return IOMMU_PAGE_ALIGN(IBMVETH_IO_ENTITLEMENT_DEFAULT);
1321 adapter = netdev_priv(netdev);
1323 ret = IBMVETH_BUFF_LIST_SIZE + IBMVETH_FILT_LIST_SIZE;
1324 ret += IOMMU_PAGE_ALIGN(netdev->mtu);
1326 for (i = 0; i < IBMVETH_NUM_BUFF_POOLS; i++) {
1327 /* add the size of the active receive buffers */
1328 if (adapter->rx_buff_pool[i].active)
1329 ret +=
1330 adapter->rx_buff_pool[i].size *
1331 IOMMU_PAGE_ALIGN(adapter->rx_buff_pool[i].
1332 buff_size);
1333 rxqentries += adapter->rx_buff_pool[i].size;
1335 /* add the size of the receive queue entries */
1336 ret += IOMMU_PAGE_ALIGN(rxqentries * sizeof(struct ibmveth_rx_q_entry));
1338 return ret;
1341 static const struct net_device_ops ibmveth_netdev_ops = {
1342 .ndo_open = ibmveth_open,
1343 .ndo_stop = ibmveth_close,
1344 .ndo_start_xmit = ibmveth_start_xmit,
1345 .ndo_set_multicast_list = ibmveth_set_multicast_list,
1346 .ndo_do_ioctl = ibmveth_ioctl,
1347 .ndo_change_mtu = ibmveth_change_mtu,
1348 .ndo_validate_addr = eth_validate_addr,
1349 .ndo_set_mac_address = eth_mac_addr,
1350 #ifdef CONFIG_NET_POLL_CONTROLLER
1351 .ndo_poll_controller = ibmveth_poll_controller,
1352 #endif
1355 static int __devinit ibmveth_probe(struct vio_dev *dev,
1356 const struct vio_device_id *id)
1358 int rc, i;
1359 struct net_device *netdev;
1360 struct ibmveth_adapter *adapter;
1361 unsigned char *mac_addr_p;
1362 unsigned int *mcastFilterSize_p;
1364 dev_dbg(&dev->dev, "entering ibmveth_probe for UA 0x%x\n",
1365 dev->unit_address);
1367 mac_addr_p = (unsigned char *)vio_get_attribute(dev, VETH_MAC_ADDR,
1368 NULL);
1369 if (!mac_addr_p) {
1370 dev_err(&dev->dev, "Can't find VETH_MAC_ADDR attribute\n");
1371 return -EINVAL;
1374 mcastFilterSize_p = (unsigned int *)vio_get_attribute(dev,
1375 VETH_MCAST_FILTER_SIZE, NULL);
1376 if (!mcastFilterSize_p) {
1377 dev_err(&dev->dev, "Can't find VETH_MCAST_FILTER_SIZE "
1378 "attribute\n");
1379 return -EINVAL;
1382 netdev = alloc_etherdev(sizeof(struct ibmveth_adapter));
1384 if (!netdev)
1385 return -ENOMEM;
1387 adapter = netdev_priv(netdev);
1388 dev_set_drvdata(&dev->dev, netdev);
1390 adapter->vdev = dev;
1391 adapter->netdev = netdev;
1392 adapter->mcastFilterSize = *mcastFilterSize_p;
1393 adapter->pool_config = 0;
1395 netif_napi_add(netdev, &adapter->napi, ibmveth_poll, 16);
1398 * Some older boxes running PHYP non-natively have an OF that returns
1399 * a 8-byte local-mac-address field (and the first 2 bytes have to be
1400 * ignored) while newer boxes' OF return a 6-byte field. Note that
1401 * IEEE 1275 specifies that local-mac-address must be a 6-byte field.
1402 * The RPA doc specifies that the first byte must be 10b, so we'll
1403 * just look for it to solve this 8 vs. 6 byte field issue
1405 if ((*mac_addr_p & 0x3) != 0x02)
1406 mac_addr_p += 2;
1408 adapter->mac_addr = 0;
1409 memcpy(&adapter->mac_addr, mac_addr_p, 6);
1411 netdev->irq = dev->irq;
1412 netdev->netdev_ops = &ibmveth_netdev_ops;
1413 netdev->ethtool_ops = &netdev_ethtool_ops;
1414 SET_NETDEV_DEV(netdev, &dev->dev);
1415 netdev->features |= NETIF_F_SG;
1417 memcpy(netdev->dev_addr, &adapter->mac_addr, netdev->addr_len);
1419 for (i = 0; i < IBMVETH_NUM_BUFF_POOLS; i++) {
1420 struct kobject *kobj = &adapter->rx_buff_pool[i].kobj;
1421 int error;
1423 ibmveth_init_buffer_pool(&adapter->rx_buff_pool[i], i,
1424 pool_count[i], pool_size[i],
1425 pool_active[i]);
1426 error = kobject_init_and_add(kobj, &ktype_veth_pool,
1427 &dev->dev.kobj, "pool%d", i);
1428 if (!error)
1429 kobject_uevent(kobj, KOBJ_ADD);
1432 netdev_dbg(netdev, "adapter @ 0x%p\n", adapter);
1434 adapter->buffer_list_dma = DMA_ERROR_CODE;
1435 adapter->filter_list_dma = DMA_ERROR_CODE;
1436 adapter->rx_queue.queue_dma = DMA_ERROR_CODE;
1438 netdev_dbg(netdev, "registering netdev...\n");
1440 ibmveth_set_csum_offload(netdev, 1, ibmveth_set_tx_csum_flags);
1442 rc = register_netdev(netdev);
1444 if (rc) {
1445 netdev_dbg(netdev, "failed to register netdev rc=%d\n", rc);
1446 free_netdev(netdev);
1447 return rc;
1450 netdev_dbg(netdev, "registered\n");
1452 return 0;
1455 static int __devexit ibmveth_remove(struct vio_dev *dev)
1457 struct net_device *netdev = dev_get_drvdata(&dev->dev);
1458 struct ibmveth_adapter *adapter = netdev_priv(netdev);
1459 int i;
1461 for (i = 0; i < IBMVETH_NUM_BUFF_POOLS; i++)
1462 kobject_put(&adapter->rx_buff_pool[i].kobj);
1464 unregister_netdev(netdev);
1466 free_netdev(netdev);
1467 dev_set_drvdata(&dev->dev, NULL);
1469 return 0;
1472 static struct attribute veth_active_attr;
1473 static struct attribute veth_num_attr;
1474 static struct attribute veth_size_attr;
1476 static ssize_t veth_pool_show(struct kobject *kobj,
1477 struct attribute *attr, char *buf)
1479 struct ibmveth_buff_pool *pool = container_of(kobj,
1480 struct ibmveth_buff_pool,
1481 kobj);
1483 if (attr == &veth_active_attr)
1484 return sprintf(buf, "%d\n", pool->active);
1485 else if (attr == &veth_num_attr)
1486 return sprintf(buf, "%d\n", pool->size);
1487 else if (attr == &veth_size_attr)
1488 return sprintf(buf, "%d\n", pool->buff_size);
1489 return 0;
1492 static ssize_t veth_pool_store(struct kobject *kobj, struct attribute *attr,
1493 const char *buf, size_t count)
1495 struct ibmveth_buff_pool *pool = container_of(kobj,
1496 struct ibmveth_buff_pool,
1497 kobj);
1498 struct net_device *netdev = dev_get_drvdata(
1499 container_of(kobj->parent, struct device, kobj));
1500 struct ibmveth_adapter *adapter = netdev_priv(netdev);
1501 long value = simple_strtol(buf, NULL, 10);
1502 long rc;
1504 if (attr == &veth_active_attr) {
1505 if (value && !pool->active) {
1506 if (netif_running(netdev)) {
1507 if (ibmveth_alloc_buffer_pool(pool)) {
1508 netdev_err(netdev,
1509 "unable to alloc pool\n");
1510 return -ENOMEM;
1512 pool->active = 1;
1513 adapter->pool_config = 1;
1514 ibmveth_close(netdev);
1515 adapter->pool_config = 0;
1516 if ((rc = ibmveth_open(netdev)))
1517 return rc;
1518 } else {
1519 pool->active = 1;
1521 } else if (!value && pool->active) {
1522 int mtu = netdev->mtu + IBMVETH_BUFF_OH;
1523 int i;
1524 /* Make sure there is a buffer pool with buffers that
1525 can hold a packet of the size of the MTU */
1526 for (i = 0; i < IBMVETH_NUM_BUFF_POOLS; i++) {
1527 if (pool == &adapter->rx_buff_pool[i])
1528 continue;
1529 if (!adapter->rx_buff_pool[i].active)
1530 continue;
1531 if (mtu <= adapter->rx_buff_pool[i].buff_size)
1532 break;
1535 if (i == IBMVETH_NUM_BUFF_POOLS) {
1536 netdev_err(netdev, "no active pool >= MTU\n");
1537 return -EPERM;
1540 if (netif_running(netdev)) {
1541 adapter->pool_config = 1;
1542 ibmveth_close(netdev);
1543 pool->active = 0;
1544 adapter->pool_config = 0;
1545 if ((rc = ibmveth_open(netdev)))
1546 return rc;
1548 pool->active = 0;
1550 } else if (attr == &veth_num_attr) {
1551 if (value <= 0 || value > IBMVETH_MAX_POOL_COUNT) {
1552 return -EINVAL;
1553 } else {
1554 if (netif_running(netdev)) {
1555 adapter->pool_config = 1;
1556 ibmveth_close(netdev);
1557 adapter->pool_config = 0;
1558 pool->size = value;
1559 if ((rc = ibmveth_open(netdev)))
1560 return rc;
1561 } else {
1562 pool->size = value;
1565 } else if (attr == &veth_size_attr) {
1566 if (value <= IBMVETH_BUFF_OH || value > IBMVETH_MAX_BUF_SIZE) {
1567 return -EINVAL;
1568 } else {
1569 if (netif_running(netdev)) {
1570 adapter->pool_config = 1;
1571 ibmveth_close(netdev);
1572 adapter->pool_config = 0;
1573 pool->buff_size = value;
1574 if ((rc = ibmveth_open(netdev)))
1575 return rc;
1576 } else {
1577 pool->buff_size = value;
1582 /* kick the interrupt handler to allocate/deallocate pools */
1583 ibmveth_interrupt(netdev->irq, netdev);
1584 return count;
1588 #define ATTR(_name, _mode) \
1589 struct attribute veth_##_name##_attr = { \
1590 .name = __stringify(_name), .mode = _mode, \
1593 static ATTR(active, 0644);
1594 static ATTR(num, 0644);
1595 static ATTR(size, 0644);
1597 static struct attribute *veth_pool_attrs[] = {
1598 &veth_active_attr,
1599 &veth_num_attr,
1600 &veth_size_attr,
1601 NULL,
1604 static const struct sysfs_ops veth_pool_ops = {
1605 .show = veth_pool_show,
1606 .store = veth_pool_store,
1609 static struct kobj_type ktype_veth_pool = {
1610 .release = NULL,
1611 .sysfs_ops = &veth_pool_ops,
1612 .default_attrs = veth_pool_attrs,
1615 static int ibmveth_resume(struct device *dev)
1617 struct net_device *netdev = dev_get_drvdata(dev);
1618 ibmveth_interrupt(netdev->irq, netdev);
1619 return 0;
1622 static struct vio_device_id ibmveth_device_table[] __devinitdata = {
1623 { "network", "IBM,l-lan"},
1624 { "", "" }
1626 MODULE_DEVICE_TABLE(vio, ibmveth_device_table);
1628 static struct dev_pm_ops ibmveth_pm_ops = {
1629 .resume = ibmveth_resume
1632 static struct vio_driver ibmveth_driver = {
1633 .id_table = ibmveth_device_table,
1634 .probe = ibmveth_probe,
1635 .remove = ibmveth_remove,
1636 .get_desired_dma = ibmveth_get_desired_dma,
1637 .driver = {
1638 .name = ibmveth_driver_name,
1639 .owner = THIS_MODULE,
1640 .pm = &ibmveth_pm_ops,
1644 static int __init ibmveth_module_init(void)
1646 printk(KERN_DEBUG "%s: %s %s\n", ibmveth_driver_name,
1647 ibmveth_driver_string, ibmveth_driver_version);
1649 return vio_register_driver(&ibmveth_driver);
1652 static void __exit ibmveth_module_exit(void)
1654 vio_unregister_driver(&ibmveth_driver);
1657 module_init(ibmveth_module_init);
1658 module_exit(ibmveth_module_exit);