Merge tag 'for_linus' of git://git.kernel.org/pub/scm/linux/kernel/git/mst/vhost
[cris-mirror.git] / drivers / net / ethernet / ibm / ibmveth.c
blobf210398200ece183c58470be65f3aa8c606aea9d
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, see <http://www.gnu.org/licenses/>.
17 * Copyright (C) IBM Corporation, 2003, 2010
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>
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 #include <net/tcp.h>
50 #include <net/ip6_checksum.h>
52 #include "ibmveth.h"
54 static irqreturn_t ibmveth_interrupt(int irq, void *dev_instance);
55 static void ibmveth_rxq_harvest_buffer(struct ibmveth_adapter *adapter);
56 static unsigned long ibmveth_get_desired_dma(struct vio_dev *vdev);
58 static struct kobj_type ktype_veth_pool;
61 static const char ibmveth_driver_name[] = "ibmveth";
62 static const char ibmveth_driver_string[] = "IBM Power Virtual Ethernet Driver";
63 #define ibmveth_driver_version "1.06"
65 MODULE_AUTHOR("Santiago Leon <santil@linux.vnet.ibm.com>");
66 MODULE_DESCRIPTION("IBM Power Virtual Ethernet Driver");
67 MODULE_LICENSE("GPL");
68 MODULE_VERSION(ibmveth_driver_version);
70 static unsigned int tx_copybreak __read_mostly = 128;
71 module_param(tx_copybreak, uint, 0644);
72 MODULE_PARM_DESC(tx_copybreak,
73 "Maximum size of packet that is copied to a new buffer on transmit");
75 static unsigned int rx_copybreak __read_mostly = 128;
76 module_param(rx_copybreak, uint, 0644);
77 MODULE_PARM_DESC(rx_copybreak,
78 "Maximum size of packet that is copied to a new buffer on receive");
80 static unsigned int rx_flush __read_mostly = 0;
81 module_param(rx_flush, uint, 0644);
82 MODULE_PARM_DESC(rx_flush, "Flush receive buffers before use");
84 static bool old_large_send __read_mostly;
85 module_param(old_large_send, bool, S_IRUGO);
86 MODULE_PARM_DESC(old_large_send,
87 "Use old large send method on firmware that supports the new method");
89 struct ibmveth_stat {
90 char name[ETH_GSTRING_LEN];
91 int offset;
94 #define IBMVETH_STAT_OFF(stat) offsetof(struct ibmveth_adapter, stat)
95 #define IBMVETH_GET_STAT(a, off) *((u64 *)(((unsigned long)(a)) + off))
97 struct ibmveth_stat ibmveth_stats[] = {
98 { "replenish_task_cycles", IBMVETH_STAT_OFF(replenish_task_cycles) },
99 { "replenish_no_mem", IBMVETH_STAT_OFF(replenish_no_mem) },
100 { "replenish_add_buff_failure",
101 IBMVETH_STAT_OFF(replenish_add_buff_failure) },
102 { "replenish_add_buff_success",
103 IBMVETH_STAT_OFF(replenish_add_buff_success) },
104 { "rx_invalid_buffer", IBMVETH_STAT_OFF(rx_invalid_buffer) },
105 { "rx_no_buffer", IBMVETH_STAT_OFF(rx_no_buffer) },
106 { "tx_map_failed", IBMVETH_STAT_OFF(tx_map_failed) },
107 { "tx_send_failed", IBMVETH_STAT_OFF(tx_send_failed) },
108 { "fw_enabled_ipv4_csum", IBMVETH_STAT_OFF(fw_ipv4_csum_support) },
109 { "fw_enabled_ipv6_csum", IBMVETH_STAT_OFF(fw_ipv6_csum_support) },
110 { "tx_large_packets", IBMVETH_STAT_OFF(tx_large_packets) },
111 { "rx_large_packets", IBMVETH_STAT_OFF(rx_large_packets) },
112 { "fw_enabled_large_send", IBMVETH_STAT_OFF(fw_large_send_support) }
115 /* simple methods of getting data from the current rxq entry */
116 static inline u32 ibmveth_rxq_flags(struct ibmveth_adapter *adapter)
118 return be32_to_cpu(adapter->rx_queue.queue_addr[adapter->rx_queue.index].flags_off);
121 static inline int ibmveth_rxq_toggle(struct ibmveth_adapter *adapter)
123 return (ibmveth_rxq_flags(adapter) & IBMVETH_RXQ_TOGGLE) >>
124 IBMVETH_RXQ_TOGGLE_SHIFT;
127 static inline int ibmveth_rxq_pending_buffer(struct ibmveth_adapter *adapter)
129 return ibmveth_rxq_toggle(adapter) == adapter->rx_queue.toggle;
132 static inline int ibmveth_rxq_buffer_valid(struct ibmveth_adapter *adapter)
134 return ibmveth_rxq_flags(adapter) & IBMVETH_RXQ_VALID;
137 static inline int ibmveth_rxq_frame_offset(struct ibmveth_adapter *adapter)
139 return ibmveth_rxq_flags(adapter) & IBMVETH_RXQ_OFF_MASK;
142 static inline int ibmveth_rxq_large_packet(struct ibmveth_adapter *adapter)
144 return ibmveth_rxq_flags(adapter) & IBMVETH_RXQ_LRG_PKT;
147 static inline int ibmveth_rxq_frame_length(struct ibmveth_adapter *adapter)
149 return be32_to_cpu(adapter->rx_queue.queue_addr[adapter->rx_queue.index].length);
152 static inline int ibmveth_rxq_csum_good(struct ibmveth_adapter *adapter)
154 return ibmveth_rxq_flags(adapter) & IBMVETH_RXQ_CSUM_GOOD;
157 /* setup the initial settings for a buffer pool */
158 static void ibmveth_init_buffer_pool(struct ibmveth_buff_pool *pool,
159 u32 pool_index, u32 pool_size,
160 u32 buff_size, u32 pool_active)
162 pool->size = pool_size;
163 pool->index = pool_index;
164 pool->buff_size = buff_size;
165 pool->threshold = pool_size * 7 / 8;
166 pool->active = pool_active;
169 /* allocate and setup an buffer pool - called during open */
170 static int ibmveth_alloc_buffer_pool(struct ibmveth_buff_pool *pool)
172 int i;
174 pool->free_map = kmalloc(sizeof(u16) * pool->size, GFP_KERNEL);
176 if (!pool->free_map)
177 return -1;
179 pool->dma_addr = kcalloc(pool->size, sizeof(dma_addr_t), GFP_KERNEL);
180 if (!pool->dma_addr) {
181 kfree(pool->free_map);
182 pool->free_map = NULL;
183 return -1;
186 pool->skbuff = kcalloc(pool->size, sizeof(void *), GFP_KERNEL);
188 if (!pool->skbuff) {
189 kfree(pool->dma_addr);
190 pool->dma_addr = NULL;
192 kfree(pool->free_map);
193 pool->free_map = NULL;
194 return -1;
197 for (i = 0; i < pool->size; ++i)
198 pool->free_map[i] = i;
200 atomic_set(&pool->available, 0);
201 pool->producer_index = 0;
202 pool->consumer_index = 0;
204 return 0;
207 static inline void ibmveth_flush_buffer(void *addr, unsigned long length)
209 unsigned long offset;
211 for (offset = 0; offset < length; offset += SMP_CACHE_BYTES)
212 asm("dcbfl %0,%1" :: "b" (addr), "r" (offset));
215 /* replenish the buffers for a pool. note that we don't need to
216 * skb_reserve these since they are used for incoming...
218 static void ibmveth_replenish_buffer_pool(struct ibmveth_adapter *adapter,
219 struct ibmveth_buff_pool *pool)
221 u32 i;
222 u32 count = pool->size - atomic_read(&pool->available);
223 u32 buffers_added = 0;
224 struct sk_buff *skb;
225 unsigned int free_index, index;
226 u64 correlator;
227 unsigned long lpar_rc;
228 dma_addr_t dma_addr;
230 mb();
232 for (i = 0; i < count; ++i) {
233 union ibmveth_buf_desc desc;
235 skb = netdev_alloc_skb(adapter->netdev, pool->buff_size);
237 if (!skb) {
238 netdev_dbg(adapter->netdev,
239 "replenish: unable to allocate skb\n");
240 adapter->replenish_no_mem++;
241 break;
244 free_index = pool->consumer_index;
245 pool->consumer_index++;
246 if (pool->consumer_index >= pool->size)
247 pool->consumer_index = 0;
248 index = pool->free_map[free_index];
250 BUG_ON(index == IBM_VETH_INVALID_MAP);
251 BUG_ON(pool->skbuff[index] != NULL);
253 dma_addr = dma_map_single(&adapter->vdev->dev, skb->data,
254 pool->buff_size, DMA_FROM_DEVICE);
256 if (dma_mapping_error(&adapter->vdev->dev, dma_addr))
257 goto failure;
259 pool->free_map[free_index] = IBM_VETH_INVALID_MAP;
260 pool->dma_addr[index] = dma_addr;
261 pool->skbuff[index] = skb;
263 correlator = ((u64)pool->index << 32) | index;
264 *(u64 *)skb->data = correlator;
266 desc.fields.flags_len = IBMVETH_BUF_VALID | pool->buff_size;
267 desc.fields.address = dma_addr;
269 if (rx_flush) {
270 unsigned int len = min(pool->buff_size,
271 adapter->netdev->mtu +
272 IBMVETH_BUFF_OH);
273 ibmveth_flush_buffer(skb->data, len);
275 lpar_rc = h_add_logical_lan_buffer(adapter->vdev->unit_address,
276 desc.desc);
278 if (lpar_rc != H_SUCCESS) {
279 goto failure;
280 } else {
281 buffers_added++;
282 adapter->replenish_add_buff_success++;
286 mb();
287 atomic_add(buffers_added, &(pool->available));
288 return;
290 failure:
291 pool->free_map[free_index] = index;
292 pool->skbuff[index] = NULL;
293 if (pool->consumer_index == 0)
294 pool->consumer_index = pool->size - 1;
295 else
296 pool->consumer_index--;
297 if (!dma_mapping_error(&adapter->vdev->dev, dma_addr))
298 dma_unmap_single(&adapter->vdev->dev,
299 pool->dma_addr[index], pool->buff_size,
300 DMA_FROM_DEVICE);
301 dev_kfree_skb_any(skb);
302 adapter->replenish_add_buff_failure++;
304 mb();
305 atomic_add(buffers_added, &(pool->available));
309 * The final 8 bytes of the buffer list is a counter of frames dropped
310 * because there was not a buffer in the buffer list capable of holding
311 * the frame.
313 static void ibmveth_update_rx_no_buffer(struct ibmveth_adapter *adapter)
315 __be64 *p = adapter->buffer_list_addr + 4096 - 8;
317 adapter->rx_no_buffer = be64_to_cpup(p);
320 /* replenish routine */
321 static void ibmveth_replenish_task(struct ibmveth_adapter *adapter)
323 int i;
325 adapter->replenish_task_cycles++;
327 for (i = (IBMVETH_NUM_BUFF_POOLS - 1); i >= 0; i--) {
328 struct ibmveth_buff_pool *pool = &adapter->rx_buff_pool[i];
330 if (pool->active &&
331 (atomic_read(&pool->available) < pool->threshold))
332 ibmveth_replenish_buffer_pool(adapter, pool);
335 ibmveth_update_rx_no_buffer(adapter);
338 /* empty and free ana buffer pool - also used to do cleanup in error paths */
339 static void ibmveth_free_buffer_pool(struct ibmveth_adapter *adapter,
340 struct ibmveth_buff_pool *pool)
342 int i;
344 kfree(pool->free_map);
345 pool->free_map = NULL;
347 if (pool->skbuff && pool->dma_addr) {
348 for (i = 0; i < pool->size; ++i) {
349 struct sk_buff *skb = pool->skbuff[i];
350 if (skb) {
351 dma_unmap_single(&adapter->vdev->dev,
352 pool->dma_addr[i],
353 pool->buff_size,
354 DMA_FROM_DEVICE);
355 dev_kfree_skb_any(skb);
356 pool->skbuff[i] = NULL;
361 if (pool->dma_addr) {
362 kfree(pool->dma_addr);
363 pool->dma_addr = NULL;
366 if (pool->skbuff) {
367 kfree(pool->skbuff);
368 pool->skbuff = NULL;
372 /* remove a buffer from a pool */
373 static void ibmveth_remove_buffer_from_pool(struct ibmveth_adapter *adapter,
374 u64 correlator)
376 unsigned int pool = correlator >> 32;
377 unsigned int index = correlator & 0xffffffffUL;
378 unsigned int free_index;
379 struct sk_buff *skb;
381 BUG_ON(pool >= IBMVETH_NUM_BUFF_POOLS);
382 BUG_ON(index >= adapter->rx_buff_pool[pool].size);
384 skb = adapter->rx_buff_pool[pool].skbuff[index];
386 BUG_ON(skb == NULL);
388 adapter->rx_buff_pool[pool].skbuff[index] = NULL;
390 dma_unmap_single(&adapter->vdev->dev,
391 adapter->rx_buff_pool[pool].dma_addr[index],
392 adapter->rx_buff_pool[pool].buff_size,
393 DMA_FROM_DEVICE);
395 free_index = adapter->rx_buff_pool[pool].producer_index;
396 adapter->rx_buff_pool[pool].producer_index++;
397 if (adapter->rx_buff_pool[pool].producer_index >=
398 adapter->rx_buff_pool[pool].size)
399 adapter->rx_buff_pool[pool].producer_index = 0;
400 adapter->rx_buff_pool[pool].free_map[free_index] = index;
402 mb();
404 atomic_dec(&(adapter->rx_buff_pool[pool].available));
407 /* get the current buffer on the rx queue */
408 static inline struct sk_buff *ibmveth_rxq_get_buffer(struct ibmveth_adapter *adapter)
410 u64 correlator = adapter->rx_queue.queue_addr[adapter->rx_queue.index].correlator;
411 unsigned int pool = correlator >> 32;
412 unsigned int index = correlator & 0xffffffffUL;
414 BUG_ON(pool >= IBMVETH_NUM_BUFF_POOLS);
415 BUG_ON(index >= adapter->rx_buff_pool[pool].size);
417 return adapter->rx_buff_pool[pool].skbuff[index];
420 /* recycle the current buffer on the rx queue */
421 static int ibmveth_rxq_recycle_buffer(struct ibmveth_adapter *adapter)
423 u32 q_index = adapter->rx_queue.index;
424 u64 correlator = adapter->rx_queue.queue_addr[q_index].correlator;
425 unsigned int pool = correlator >> 32;
426 unsigned int index = correlator & 0xffffffffUL;
427 union ibmveth_buf_desc desc;
428 unsigned long lpar_rc;
429 int ret = 1;
431 BUG_ON(pool >= IBMVETH_NUM_BUFF_POOLS);
432 BUG_ON(index >= adapter->rx_buff_pool[pool].size);
434 if (!adapter->rx_buff_pool[pool].active) {
435 ibmveth_rxq_harvest_buffer(adapter);
436 ibmveth_free_buffer_pool(adapter, &adapter->rx_buff_pool[pool]);
437 goto out;
440 desc.fields.flags_len = IBMVETH_BUF_VALID |
441 adapter->rx_buff_pool[pool].buff_size;
442 desc.fields.address = adapter->rx_buff_pool[pool].dma_addr[index];
444 lpar_rc = h_add_logical_lan_buffer(adapter->vdev->unit_address, desc.desc);
446 if (lpar_rc != H_SUCCESS) {
447 netdev_dbg(adapter->netdev, "h_add_logical_lan_buffer failed "
448 "during recycle rc=%ld", lpar_rc);
449 ibmveth_remove_buffer_from_pool(adapter, adapter->rx_queue.queue_addr[adapter->rx_queue.index].correlator);
450 ret = 0;
453 if (++adapter->rx_queue.index == adapter->rx_queue.num_slots) {
454 adapter->rx_queue.index = 0;
455 adapter->rx_queue.toggle = !adapter->rx_queue.toggle;
458 out:
459 return ret;
462 static void ibmveth_rxq_harvest_buffer(struct ibmveth_adapter *adapter)
464 ibmveth_remove_buffer_from_pool(adapter, adapter->rx_queue.queue_addr[adapter->rx_queue.index].correlator);
466 if (++adapter->rx_queue.index == adapter->rx_queue.num_slots) {
467 adapter->rx_queue.index = 0;
468 adapter->rx_queue.toggle = !adapter->rx_queue.toggle;
472 static int ibmveth_register_logical_lan(struct ibmveth_adapter *adapter,
473 union ibmveth_buf_desc rxq_desc, u64 mac_address)
475 int rc, try_again = 1;
478 * After a kexec the adapter will still be open, so our attempt to
479 * open it will fail. So if we get a failure we free the adapter and
480 * try again, but only once.
482 retry:
483 rc = h_register_logical_lan(adapter->vdev->unit_address,
484 adapter->buffer_list_dma, rxq_desc.desc,
485 adapter->filter_list_dma, mac_address);
487 if (rc != H_SUCCESS && try_again) {
488 do {
489 rc = h_free_logical_lan(adapter->vdev->unit_address);
490 } while (H_IS_LONG_BUSY(rc) || (rc == H_BUSY));
492 try_again = 0;
493 goto retry;
496 return rc;
499 static u64 ibmveth_encode_mac_addr(u8 *mac)
501 int i;
502 u64 encoded = 0;
504 for (i = 0; i < ETH_ALEN; i++)
505 encoded = (encoded << 8) | mac[i];
507 return encoded;
510 static int ibmveth_open(struct net_device *netdev)
512 struct ibmveth_adapter *adapter = netdev_priv(netdev);
513 u64 mac_address;
514 int rxq_entries = 1;
515 unsigned long lpar_rc;
516 int rc;
517 union ibmveth_buf_desc rxq_desc;
518 int i;
519 struct device *dev;
521 netdev_dbg(netdev, "open starting\n");
523 napi_enable(&adapter->napi);
525 for(i = 0; i < IBMVETH_NUM_BUFF_POOLS; i++)
526 rxq_entries += adapter->rx_buff_pool[i].size;
528 rc = -ENOMEM;
529 adapter->buffer_list_addr = (void*) get_zeroed_page(GFP_KERNEL);
530 if (!adapter->buffer_list_addr) {
531 netdev_err(netdev, "unable to allocate list pages\n");
532 goto out;
535 adapter->filter_list_addr = (void*) get_zeroed_page(GFP_KERNEL);
536 if (!adapter->filter_list_addr) {
537 netdev_err(netdev, "unable to allocate filter pages\n");
538 goto out_free_buffer_list;
541 dev = &adapter->vdev->dev;
543 adapter->rx_queue.queue_len = sizeof(struct ibmveth_rx_q_entry) *
544 rxq_entries;
545 adapter->rx_queue.queue_addr =
546 dma_alloc_coherent(dev, adapter->rx_queue.queue_len,
547 &adapter->rx_queue.queue_dma, GFP_KERNEL);
548 if (!adapter->rx_queue.queue_addr)
549 goto out_free_filter_list;
551 adapter->buffer_list_dma = dma_map_single(dev,
552 adapter->buffer_list_addr, 4096, DMA_BIDIRECTIONAL);
553 if (dma_mapping_error(dev, adapter->buffer_list_dma)) {
554 netdev_err(netdev, "unable to map buffer list pages\n");
555 goto out_free_queue_mem;
558 adapter->filter_list_dma = dma_map_single(dev,
559 adapter->filter_list_addr, 4096, DMA_BIDIRECTIONAL);
560 if (dma_mapping_error(dev, adapter->filter_list_dma)) {
561 netdev_err(netdev, "unable to map filter list pages\n");
562 goto out_unmap_buffer_list;
565 adapter->rx_queue.index = 0;
566 adapter->rx_queue.num_slots = rxq_entries;
567 adapter->rx_queue.toggle = 1;
569 mac_address = ibmveth_encode_mac_addr(netdev->dev_addr);
571 rxq_desc.fields.flags_len = IBMVETH_BUF_VALID |
572 adapter->rx_queue.queue_len;
573 rxq_desc.fields.address = adapter->rx_queue.queue_dma;
575 netdev_dbg(netdev, "buffer list @ 0x%p\n", adapter->buffer_list_addr);
576 netdev_dbg(netdev, "filter list @ 0x%p\n", adapter->filter_list_addr);
577 netdev_dbg(netdev, "receive q @ 0x%p\n", adapter->rx_queue.queue_addr);
579 h_vio_signal(adapter->vdev->unit_address, VIO_IRQ_DISABLE);
581 lpar_rc = ibmveth_register_logical_lan(adapter, rxq_desc, mac_address);
583 if (lpar_rc != H_SUCCESS) {
584 netdev_err(netdev, "h_register_logical_lan failed with %ld\n",
585 lpar_rc);
586 netdev_err(netdev, "buffer TCE:0x%llx filter TCE:0x%llx rxq "
587 "desc:0x%llx MAC:0x%llx\n",
588 adapter->buffer_list_dma,
589 adapter->filter_list_dma,
590 rxq_desc.desc,
591 mac_address);
592 rc = -ENONET;
593 goto out_unmap_filter_list;
596 for (i = 0; i < IBMVETH_NUM_BUFF_POOLS; i++) {
597 if (!adapter->rx_buff_pool[i].active)
598 continue;
599 if (ibmveth_alloc_buffer_pool(&adapter->rx_buff_pool[i])) {
600 netdev_err(netdev, "unable to alloc pool\n");
601 adapter->rx_buff_pool[i].active = 0;
602 rc = -ENOMEM;
603 goto out_free_buffer_pools;
607 netdev_dbg(netdev, "registering irq 0x%x\n", netdev->irq);
608 rc = request_irq(netdev->irq, ibmveth_interrupt, 0, netdev->name,
609 netdev);
610 if (rc != 0) {
611 netdev_err(netdev, "unable to request irq 0x%x, rc %d\n",
612 netdev->irq, rc);
613 do {
614 lpar_rc = h_free_logical_lan(adapter->vdev->unit_address);
615 } while (H_IS_LONG_BUSY(lpar_rc) || (lpar_rc == H_BUSY));
617 goto out_free_buffer_pools;
620 rc = -ENOMEM;
621 adapter->bounce_buffer =
622 kmalloc(netdev->mtu + IBMVETH_BUFF_OH, GFP_KERNEL);
623 if (!adapter->bounce_buffer)
624 goto out_free_irq;
626 adapter->bounce_buffer_dma =
627 dma_map_single(&adapter->vdev->dev, adapter->bounce_buffer,
628 netdev->mtu + IBMVETH_BUFF_OH, DMA_BIDIRECTIONAL);
629 if (dma_mapping_error(dev, adapter->bounce_buffer_dma)) {
630 netdev_err(netdev, "unable to map bounce buffer\n");
631 goto out_free_bounce_buffer;
634 netdev_dbg(netdev, "initial replenish cycle\n");
635 ibmveth_interrupt(netdev->irq, netdev);
637 netif_start_queue(netdev);
639 netdev_dbg(netdev, "open complete\n");
641 return 0;
643 out_free_bounce_buffer:
644 kfree(adapter->bounce_buffer);
645 out_free_irq:
646 free_irq(netdev->irq, netdev);
647 out_free_buffer_pools:
648 while (--i >= 0) {
649 if (adapter->rx_buff_pool[i].active)
650 ibmveth_free_buffer_pool(adapter,
651 &adapter->rx_buff_pool[i]);
653 out_unmap_filter_list:
654 dma_unmap_single(dev, adapter->filter_list_dma, 4096,
655 DMA_BIDIRECTIONAL);
656 out_unmap_buffer_list:
657 dma_unmap_single(dev, adapter->buffer_list_dma, 4096,
658 DMA_BIDIRECTIONAL);
659 out_free_queue_mem:
660 dma_free_coherent(dev, adapter->rx_queue.queue_len,
661 adapter->rx_queue.queue_addr,
662 adapter->rx_queue.queue_dma);
663 out_free_filter_list:
664 free_page((unsigned long)adapter->filter_list_addr);
665 out_free_buffer_list:
666 free_page((unsigned long)adapter->buffer_list_addr);
667 out:
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 struct device *dev = &adapter->vdev->dev;
676 long lpar_rc;
677 int i;
679 netdev_dbg(netdev, "close starting\n");
681 napi_disable(&adapter->napi);
683 if (!adapter->pool_config)
684 netif_stop_queue(netdev);
686 h_vio_signal(adapter->vdev->unit_address, VIO_IRQ_DISABLE);
688 do {
689 lpar_rc = h_free_logical_lan(adapter->vdev->unit_address);
690 } while (H_IS_LONG_BUSY(lpar_rc) || (lpar_rc == H_BUSY));
692 if (lpar_rc != H_SUCCESS) {
693 netdev_err(netdev, "h_free_logical_lan failed with %lx, "
694 "continuing with close\n", lpar_rc);
697 free_irq(netdev->irq, netdev);
699 ibmveth_update_rx_no_buffer(adapter);
701 dma_unmap_single(dev, adapter->buffer_list_dma, 4096,
702 DMA_BIDIRECTIONAL);
703 free_page((unsigned long)adapter->buffer_list_addr);
705 dma_unmap_single(dev, adapter->filter_list_dma, 4096,
706 DMA_BIDIRECTIONAL);
707 free_page((unsigned long)adapter->filter_list_addr);
709 dma_free_coherent(dev, adapter->rx_queue.queue_len,
710 adapter->rx_queue.queue_addr,
711 adapter->rx_queue.queue_dma);
713 for (i = 0; i < IBMVETH_NUM_BUFF_POOLS; i++)
714 if (adapter->rx_buff_pool[i].active)
715 ibmveth_free_buffer_pool(adapter,
716 &adapter->rx_buff_pool[i]);
718 dma_unmap_single(&adapter->vdev->dev, adapter->bounce_buffer_dma,
719 adapter->netdev->mtu + IBMVETH_BUFF_OH,
720 DMA_BIDIRECTIONAL);
721 kfree(adapter->bounce_buffer);
723 netdev_dbg(netdev, "close complete\n");
725 return 0;
728 static int netdev_get_link_ksettings(struct net_device *dev,
729 struct ethtool_link_ksettings *cmd)
731 u32 supported, advertising;
733 supported = (SUPPORTED_1000baseT_Full | SUPPORTED_Autoneg |
734 SUPPORTED_FIBRE);
735 advertising = (ADVERTISED_1000baseT_Full | ADVERTISED_Autoneg |
736 ADVERTISED_FIBRE);
737 cmd->base.speed = SPEED_1000;
738 cmd->base.duplex = DUPLEX_FULL;
739 cmd->base.port = PORT_FIBRE;
740 cmd->base.phy_address = 0;
741 cmd->base.autoneg = AUTONEG_ENABLE;
743 ethtool_convert_legacy_u32_to_link_mode(cmd->link_modes.supported,
744 supported);
745 ethtool_convert_legacy_u32_to_link_mode(cmd->link_modes.advertising,
746 advertising);
748 return 0;
751 static void netdev_get_drvinfo(struct net_device *dev,
752 struct ethtool_drvinfo *info)
754 strlcpy(info->driver, ibmveth_driver_name, sizeof(info->driver));
755 strlcpy(info->version, ibmveth_driver_version, sizeof(info->version));
758 static netdev_features_t ibmveth_fix_features(struct net_device *dev,
759 netdev_features_t features)
762 * Since the ibmveth firmware interface does not have the
763 * concept of separate tx/rx checksum offload enable, if rx
764 * checksum is disabled we also have to disable tx checksum
765 * offload. Once we disable rx checksum offload, we are no
766 * longer allowed to send tx buffers that are not properly
767 * checksummed.
770 if (!(features & NETIF_F_RXCSUM))
771 features &= ~NETIF_F_CSUM_MASK;
773 return features;
776 static int ibmveth_set_csum_offload(struct net_device *dev, u32 data)
778 struct ibmveth_adapter *adapter = netdev_priv(dev);
779 unsigned long set_attr, clr_attr, ret_attr;
780 unsigned long set_attr6, clr_attr6;
781 long ret, ret4, ret6;
782 int rc1 = 0, rc2 = 0;
783 int restart = 0;
785 if (netif_running(dev)) {
786 restart = 1;
787 adapter->pool_config = 1;
788 ibmveth_close(dev);
789 adapter->pool_config = 0;
792 set_attr = 0;
793 clr_attr = 0;
794 set_attr6 = 0;
795 clr_attr6 = 0;
797 if (data) {
798 set_attr = IBMVETH_ILLAN_IPV4_TCP_CSUM;
799 set_attr6 = IBMVETH_ILLAN_IPV6_TCP_CSUM;
800 } else {
801 clr_attr = IBMVETH_ILLAN_IPV4_TCP_CSUM;
802 clr_attr6 = IBMVETH_ILLAN_IPV6_TCP_CSUM;
805 ret = h_illan_attributes(adapter->vdev->unit_address, 0, 0, &ret_attr);
807 if (ret == H_SUCCESS &&
808 (ret_attr & IBMVETH_ILLAN_PADDED_PKT_CSUM)) {
809 ret4 = h_illan_attributes(adapter->vdev->unit_address, clr_attr,
810 set_attr, &ret_attr);
812 if (ret4 != H_SUCCESS) {
813 netdev_err(dev, "unable to change IPv4 checksum "
814 "offload settings. %d rc=%ld\n",
815 data, ret4);
817 h_illan_attributes(adapter->vdev->unit_address,
818 set_attr, clr_attr, &ret_attr);
820 if (data == 1)
821 dev->features &= ~NETIF_F_IP_CSUM;
823 } else {
824 adapter->fw_ipv4_csum_support = data;
827 ret6 = h_illan_attributes(adapter->vdev->unit_address,
828 clr_attr6, set_attr6, &ret_attr);
830 if (ret6 != H_SUCCESS) {
831 netdev_err(dev, "unable to change IPv6 checksum "
832 "offload settings. %d rc=%ld\n",
833 data, ret6);
835 h_illan_attributes(adapter->vdev->unit_address,
836 set_attr6, clr_attr6, &ret_attr);
838 if (data == 1)
839 dev->features &= ~NETIF_F_IPV6_CSUM;
841 } else
842 adapter->fw_ipv6_csum_support = data;
844 if (ret4 == H_SUCCESS || ret6 == H_SUCCESS)
845 adapter->rx_csum = data;
846 else
847 rc1 = -EIO;
848 } else {
849 rc1 = -EIO;
850 netdev_err(dev, "unable to change checksum offload settings."
851 " %d rc=%ld ret_attr=%lx\n", data, ret,
852 ret_attr);
855 if (restart)
856 rc2 = ibmveth_open(dev);
858 return rc1 ? rc1 : rc2;
861 static int ibmveth_set_tso(struct net_device *dev, u32 data)
863 struct ibmveth_adapter *adapter = netdev_priv(dev);
864 unsigned long set_attr, clr_attr, ret_attr;
865 long ret1, ret2;
866 int rc1 = 0, rc2 = 0;
867 int restart = 0;
869 if (netif_running(dev)) {
870 restart = 1;
871 adapter->pool_config = 1;
872 ibmveth_close(dev);
873 adapter->pool_config = 0;
876 set_attr = 0;
877 clr_attr = 0;
879 if (data)
880 set_attr = IBMVETH_ILLAN_LRG_SR_ENABLED;
881 else
882 clr_attr = IBMVETH_ILLAN_LRG_SR_ENABLED;
884 ret1 = h_illan_attributes(adapter->vdev->unit_address, 0, 0, &ret_attr);
886 if (ret1 == H_SUCCESS && (ret_attr & IBMVETH_ILLAN_LRG_SND_SUPPORT) &&
887 !old_large_send) {
888 ret2 = h_illan_attributes(adapter->vdev->unit_address, clr_attr,
889 set_attr, &ret_attr);
891 if (ret2 != H_SUCCESS) {
892 netdev_err(dev, "unable to change tso settings. %d rc=%ld\n",
893 data, ret2);
895 h_illan_attributes(adapter->vdev->unit_address,
896 set_attr, clr_attr, &ret_attr);
898 if (data == 1)
899 dev->features &= ~(NETIF_F_TSO | NETIF_F_TSO6);
900 rc1 = -EIO;
902 } else {
903 adapter->fw_large_send_support = data;
904 adapter->large_send = data;
906 } else {
907 /* Older firmware version of large send offload does not
908 * support tcp6/ipv6
910 if (data == 1) {
911 dev->features &= ~NETIF_F_TSO6;
912 netdev_info(dev, "TSO feature requires all partitions to have updated driver");
914 adapter->large_send = data;
917 if (restart)
918 rc2 = ibmveth_open(dev);
920 return rc1 ? rc1 : rc2;
923 static int ibmveth_set_features(struct net_device *dev,
924 netdev_features_t features)
926 struct ibmveth_adapter *adapter = netdev_priv(dev);
927 int rx_csum = !!(features & NETIF_F_RXCSUM);
928 int large_send = !!(features & (NETIF_F_TSO | NETIF_F_TSO6));
929 int rc1 = 0, rc2 = 0;
931 if (rx_csum != adapter->rx_csum) {
932 rc1 = ibmveth_set_csum_offload(dev, rx_csum);
933 if (rc1 && !adapter->rx_csum)
934 dev->features =
935 features & ~(NETIF_F_CSUM_MASK |
936 NETIF_F_RXCSUM);
939 if (large_send != adapter->large_send) {
940 rc2 = ibmveth_set_tso(dev, large_send);
941 if (rc2 && !adapter->large_send)
942 dev->features =
943 features & ~(NETIF_F_TSO | NETIF_F_TSO6);
946 return rc1 ? rc1 : rc2;
949 static void ibmveth_get_strings(struct net_device *dev, u32 stringset, u8 *data)
951 int i;
953 if (stringset != ETH_SS_STATS)
954 return;
956 for (i = 0; i < ARRAY_SIZE(ibmveth_stats); i++, data += ETH_GSTRING_LEN)
957 memcpy(data, ibmveth_stats[i].name, ETH_GSTRING_LEN);
960 static int ibmveth_get_sset_count(struct net_device *dev, int sset)
962 switch (sset) {
963 case ETH_SS_STATS:
964 return ARRAY_SIZE(ibmveth_stats);
965 default:
966 return -EOPNOTSUPP;
970 static void ibmveth_get_ethtool_stats(struct net_device *dev,
971 struct ethtool_stats *stats, u64 *data)
973 int i;
974 struct ibmveth_adapter *adapter = netdev_priv(dev);
976 for (i = 0; i < ARRAY_SIZE(ibmveth_stats); i++)
977 data[i] = IBMVETH_GET_STAT(adapter, ibmveth_stats[i].offset);
980 static const struct ethtool_ops netdev_ethtool_ops = {
981 .get_drvinfo = netdev_get_drvinfo,
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 .get_link_ksettings = netdev_get_link_ksettings,
989 static int ibmveth_ioctl(struct net_device *dev, struct ifreq *ifr, int cmd)
991 return -EOPNOTSUPP;
994 #define page_offset(v) ((unsigned long)(v) & ((1 << 12) - 1))
996 static int ibmveth_send(struct ibmveth_adapter *adapter,
997 union ibmveth_buf_desc *descs, unsigned long mss)
999 unsigned long correlator;
1000 unsigned int retry_count;
1001 unsigned long ret;
1004 * The retry count sets a maximum for the number of broadcast and
1005 * multicast destinations within the system.
1007 retry_count = 1024;
1008 correlator = 0;
1009 do {
1010 ret = h_send_logical_lan(adapter->vdev->unit_address,
1011 descs[0].desc, descs[1].desc,
1012 descs[2].desc, descs[3].desc,
1013 descs[4].desc, descs[5].desc,
1014 correlator, &correlator, mss,
1015 adapter->fw_large_send_support);
1016 } while ((ret == H_BUSY) && (retry_count--));
1018 if (ret != H_SUCCESS && ret != H_DROPPED) {
1019 netdev_err(adapter->netdev, "tx: h_send_logical_lan failed "
1020 "with rc=%ld\n", ret);
1021 return 1;
1024 return 0;
1027 static netdev_tx_t ibmveth_start_xmit(struct sk_buff *skb,
1028 struct net_device *netdev)
1030 struct ibmveth_adapter *adapter = netdev_priv(netdev);
1031 unsigned int desc_flags;
1032 union ibmveth_buf_desc descs[6];
1033 int last, i;
1034 int force_bounce = 0;
1035 dma_addr_t dma_addr;
1036 unsigned long mss = 0;
1038 /* veth doesn't handle frag_list, so linearize the skb.
1039 * When GRO is enabled SKB's can have frag_list.
1041 if (adapter->is_active_trunk &&
1042 skb_has_frag_list(skb) && __skb_linearize(skb)) {
1043 netdev->stats.tx_dropped++;
1044 goto out;
1048 * veth handles a maximum of 6 segments including the header, so
1049 * we have to linearize the skb if there are more than this.
1051 if (skb_shinfo(skb)->nr_frags > 5 && __skb_linearize(skb)) {
1052 netdev->stats.tx_dropped++;
1053 goto out;
1056 /* veth can't checksum offload UDP */
1057 if (skb->ip_summed == CHECKSUM_PARTIAL &&
1058 ((skb->protocol == htons(ETH_P_IP) &&
1059 ip_hdr(skb)->protocol != IPPROTO_TCP) ||
1060 (skb->protocol == htons(ETH_P_IPV6) &&
1061 ipv6_hdr(skb)->nexthdr != IPPROTO_TCP)) &&
1062 skb_checksum_help(skb)) {
1064 netdev_err(netdev, "tx: failed to checksum packet\n");
1065 netdev->stats.tx_dropped++;
1066 goto out;
1069 desc_flags = IBMVETH_BUF_VALID;
1071 if (skb->ip_summed == CHECKSUM_PARTIAL) {
1072 unsigned char *buf = skb_transport_header(skb) +
1073 skb->csum_offset;
1075 desc_flags |= (IBMVETH_BUF_NO_CSUM | IBMVETH_BUF_CSUM_GOOD);
1077 /* Need to zero out the checksum */
1078 buf[0] = 0;
1079 buf[1] = 0;
1081 if (skb_is_gso(skb) && adapter->fw_large_send_support)
1082 desc_flags |= IBMVETH_BUF_LRG_SND;
1085 retry_bounce:
1086 memset(descs, 0, sizeof(descs));
1089 * If a linear packet is below the rx threshold then
1090 * copy it into the static bounce buffer. This avoids the
1091 * cost of a TCE insert and remove.
1093 if (force_bounce || (!skb_is_nonlinear(skb) &&
1094 (skb->len < tx_copybreak))) {
1095 skb_copy_from_linear_data(skb, adapter->bounce_buffer,
1096 skb->len);
1098 descs[0].fields.flags_len = desc_flags | skb->len;
1099 descs[0].fields.address = adapter->bounce_buffer_dma;
1101 if (ibmveth_send(adapter, descs, 0)) {
1102 adapter->tx_send_failed++;
1103 netdev->stats.tx_dropped++;
1104 } else {
1105 netdev->stats.tx_packets++;
1106 netdev->stats.tx_bytes += skb->len;
1109 goto out;
1112 /* Map the header */
1113 dma_addr = dma_map_single(&adapter->vdev->dev, skb->data,
1114 skb_headlen(skb), DMA_TO_DEVICE);
1115 if (dma_mapping_error(&adapter->vdev->dev, dma_addr))
1116 goto map_failed;
1118 descs[0].fields.flags_len = desc_flags | skb_headlen(skb);
1119 descs[0].fields.address = dma_addr;
1121 /* Map the frags */
1122 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
1123 const skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
1125 dma_addr = skb_frag_dma_map(&adapter->vdev->dev, frag, 0,
1126 skb_frag_size(frag), DMA_TO_DEVICE);
1128 if (dma_mapping_error(&adapter->vdev->dev, dma_addr))
1129 goto map_failed_frags;
1131 descs[i+1].fields.flags_len = desc_flags | skb_frag_size(frag);
1132 descs[i+1].fields.address = dma_addr;
1135 if (skb->ip_summed == CHECKSUM_PARTIAL && skb_is_gso(skb)) {
1136 if (adapter->fw_large_send_support) {
1137 mss = (unsigned long)skb_shinfo(skb)->gso_size;
1138 adapter->tx_large_packets++;
1139 } else if (!skb_is_gso_v6(skb)) {
1140 /* Put -1 in the IP checksum to tell phyp it
1141 * is a largesend packet. Put the mss in
1142 * the TCP checksum.
1144 ip_hdr(skb)->check = 0xffff;
1145 tcp_hdr(skb)->check =
1146 cpu_to_be16(skb_shinfo(skb)->gso_size);
1147 adapter->tx_large_packets++;
1151 if (ibmveth_send(adapter, descs, mss)) {
1152 adapter->tx_send_failed++;
1153 netdev->stats.tx_dropped++;
1154 } else {
1155 netdev->stats.tx_packets++;
1156 netdev->stats.tx_bytes += skb->len;
1159 dma_unmap_single(&adapter->vdev->dev,
1160 descs[0].fields.address,
1161 descs[0].fields.flags_len & IBMVETH_BUF_LEN_MASK,
1162 DMA_TO_DEVICE);
1164 for (i = 1; i < skb_shinfo(skb)->nr_frags + 1; i++)
1165 dma_unmap_page(&adapter->vdev->dev, descs[i].fields.address,
1166 descs[i].fields.flags_len & IBMVETH_BUF_LEN_MASK,
1167 DMA_TO_DEVICE);
1169 out:
1170 dev_consume_skb_any(skb);
1171 return NETDEV_TX_OK;
1173 map_failed_frags:
1174 last = i+1;
1175 for (i = 0; i < last; i++)
1176 dma_unmap_page(&adapter->vdev->dev, descs[i].fields.address,
1177 descs[i].fields.flags_len & IBMVETH_BUF_LEN_MASK,
1178 DMA_TO_DEVICE);
1180 map_failed:
1181 if (!firmware_has_feature(FW_FEATURE_CMO))
1182 netdev_err(netdev, "tx: unable to map xmit buffer\n");
1183 adapter->tx_map_failed++;
1184 if (skb_linearize(skb)) {
1185 netdev->stats.tx_dropped++;
1186 goto out;
1188 force_bounce = 1;
1189 goto retry_bounce;
1192 static void ibmveth_rx_mss_helper(struct sk_buff *skb, u16 mss, int lrg_pkt)
1194 struct tcphdr *tcph;
1195 int offset = 0;
1196 int hdr_len;
1198 /* only TCP packets will be aggregated */
1199 if (skb->protocol == htons(ETH_P_IP)) {
1200 struct iphdr *iph = (struct iphdr *)skb->data;
1202 if (iph->protocol == IPPROTO_TCP) {
1203 offset = iph->ihl * 4;
1204 skb_shinfo(skb)->gso_type = SKB_GSO_TCPV4;
1205 } else {
1206 return;
1208 } else if (skb->protocol == htons(ETH_P_IPV6)) {
1209 struct ipv6hdr *iph6 = (struct ipv6hdr *)skb->data;
1211 if (iph6->nexthdr == IPPROTO_TCP) {
1212 offset = sizeof(struct ipv6hdr);
1213 skb_shinfo(skb)->gso_type = SKB_GSO_TCPV6;
1214 } else {
1215 return;
1217 } else {
1218 return;
1220 /* if mss is not set through Large Packet bit/mss in rx buffer,
1221 * expect that the mss will be written to the tcp header checksum.
1223 tcph = (struct tcphdr *)(skb->data + offset);
1224 if (lrg_pkt) {
1225 skb_shinfo(skb)->gso_size = mss;
1226 } else if (offset) {
1227 skb_shinfo(skb)->gso_size = ntohs(tcph->check);
1228 tcph->check = 0;
1231 if (skb_shinfo(skb)->gso_size) {
1232 hdr_len = offset + tcph->doff * 4;
1233 skb_shinfo(skb)->gso_segs =
1234 DIV_ROUND_UP(skb->len - hdr_len,
1235 skb_shinfo(skb)->gso_size);
1239 static void ibmveth_rx_csum_helper(struct sk_buff *skb,
1240 struct ibmveth_adapter *adapter)
1242 struct iphdr *iph = NULL;
1243 struct ipv6hdr *iph6 = NULL;
1244 __be16 skb_proto = 0;
1245 u16 iphlen = 0;
1246 u16 iph_proto = 0;
1247 u16 tcphdrlen = 0;
1249 skb_proto = be16_to_cpu(skb->protocol);
1251 if (skb_proto == ETH_P_IP) {
1252 iph = (struct iphdr *)skb->data;
1254 /* If the IP checksum is not offloaded and if the packet
1255 * is large send, the checksum must be rebuilt.
1257 if (iph->check == 0xffff) {
1258 iph->check = 0;
1259 iph->check = ip_fast_csum((unsigned char *)iph,
1260 iph->ihl);
1263 iphlen = iph->ihl * 4;
1264 iph_proto = iph->protocol;
1265 } else if (skb_proto == ETH_P_IPV6) {
1266 iph6 = (struct ipv6hdr *)skb->data;
1267 iphlen = sizeof(struct ipv6hdr);
1268 iph_proto = iph6->nexthdr;
1271 /* In OVS environment, when a flow is not cached, specifically for a
1272 * new TCP connection, the first packet information is passed up
1273 * the user space for finding a flow. During this process, OVS computes
1274 * checksum on the first packet when CHECKSUM_PARTIAL flag is set.
1276 * Given that we zeroed out TCP checksum field in transmit path
1277 * (refer ibmveth_start_xmit routine) as we set "no checksum bit",
1278 * OVS computed checksum will be incorrect w/o TCP pseudo checksum
1279 * in the packet. This leads to OVS dropping the packet and hence
1280 * TCP retransmissions are seen.
1282 * So, re-compute TCP pseudo header checksum.
1284 if (iph_proto == IPPROTO_TCP && adapter->is_active_trunk) {
1285 struct tcphdr *tcph = (struct tcphdr *)(skb->data + iphlen);
1287 tcphdrlen = skb->len - iphlen;
1289 /* Recompute TCP pseudo header checksum */
1290 if (skb_proto == ETH_P_IP)
1291 tcph->check = ~csum_tcpudp_magic(iph->saddr,
1292 iph->daddr, tcphdrlen, iph_proto, 0);
1293 else if (skb_proto == ETH_P_IPV6)
1294 tcph->check = ~csum_ipv6_magic(&iph6->saddr,
1295 &iph6->daddr, tcphdrlen, iph_proto, 0);
1297 /* Setup SKB fields for checksum offload */
1298 skb_partial_csum_set(skb, iphlen,
1299 offsetof(struct tcphdr, check));
1300 skb_reset_network_header(skb);
1304 static int ibmveth_poll(struct napi_struct *napi, int budget)
1306 struct ibmveth_adapter *adapter =
1307 container_of(napi, struct ibmveth_adapter, napi);
1308 struct net_device *netdev = adapter->netdev;
1309 int frames_processed = 0;
1310 unsigned long lpar_rc;
1311 u16 mss = 0;
1313 restart_poll:
1314 while (frames_processed < budget) {
1315 if (!ibmveth_rxq_pending_buffer(adapter))
1316 break;
1318 smp_rmb();
1319 if (!ibmveth_rxq_buffer_valid(adapter)) {
1320 wmb(); /* suggested by larson1 */
1321 adapter->rx_invalid_buffer++;
1322 netdev_dbg(netdev, "recycling invalid buffer\n");
1323 ibmveth_rxq_recycle_buffer(adapter);
1324 } else {
1325 struct sk_buff *skb, *new_skb;
1326 int length = ibmveth_rxq_frame_length(adapter);
1327 int offset = ibmveth_rxq_frame_offset(adapter);
1328 int csum_good = ibmveth_rxq_csum_good(adapter);
1329 int lrg_pkt = ibmveth_rxq_large_packet(adapter);
1331 skb = ibmveth_rxq_get_buffer(adapter);
1333 /* if the large packet bit is set in the rx queue
1334 * descriptor, the mss will be written by PHYP eight
1335 * bytes from the start of the rx buffer, which is
1336 * skb->data at this stage
1338 if (lrg_pkt) {
1339 __be64 *rxmss = (__be64 *)(skb->data + 8);
1341 mss = (u16)be64_to_cpu(*rxmss);
1344 new_skb = NULL;
1345 if (length < rx_copybreak)
1346 new_skb = netdev_alloc_skb(netdev, length);
1348 if (new_skb) {
1349 skb_copy_to_linear_data(new_skb,
1350 skb->data + offset,
1351 length);
1352 if (rx_flush)
1353 ibmveth_flush_buffer(skb->data,
1354 length + offset);
1355 if (!ibmveth_rxq_recycle_buffer(adapter))
1356 kfree_skb(skb);
1357 skb = new_skb;
1358 } else {
1359 ibmveth_rxq_harvest_buffer(adapter);
1360 skb_reserve(skb, offset);
1363 skb_put(skb, length);
1364 skb->protocol = eth_type_trans(skb, netdev);
1366 if (csum_good) {
1367 skb->ip_summed = CHECKSUM_UNNECESSARY;
1368 ibmveth_rx_csum_helper(skb, adapter);
1371 if (length > netdev->mtu + ETH_HLEN) {
1372 ibmveth_rx_mss_helper(skb, mss, lrg_pkt);
1373 adapter->rx_large_packets++;
1376 napi_gro_receive(napi, skb); /* send it up */
1378 netdev->stats.rx_packets++;
1379 netdev->stats.rx_bytes += length;
1380 frames_processed++;
1384 ibmveth_replenish_task(adapter);
1386 if (frames_processed < budget) {
1387 napi_complete_done(napi, frames_processed);
1389 /* We think we are done - reenable interrupts,
1390 * then check once more to make sure we are done.
1392 lpar_rc = h_vio_signal(adapter->vdev->unit_address,
1393 VIO_IRQ_ENABLE);
1395 BUG_ON(lpar_rc != H_SUCCESS);
1397 if (ibmveth_rxq_pending_buffer(adapter) &&
1398 napi_reschedule(napi)) {
1399 lpar_rc = h_vio_signal(adapter->vdev->unit_address,
1400 VIO_IRQ_DISABLE);
1401 goto restart_poll;
1405 return frames_processed;
1408 static irqreturn_t ibmveth_interrupt(int irq, void *dev_instance)
1410 struct net_device *netdev = dev_instance;
1411 struct ibmveth_adapter *adapter = netdev_priv(netdev);
1412 unsigned long lpar_rc;
1414 if (napi_schedule_prep(&adapter->napi)) {
1415 lpar_rc = h_vio_signal(adapter->vdev->unit_address,
1416 VIO_IRQ_DISABLE);
1417 BUG_ON(lpar_rc != H_SUCCESS);
1418 __napi_schedule(&adapter->napi);
1420 return IRQ_HANDLED;
1423 static void ibmveth_set_multicast_list(struct net_device *netdev)
1425 struct ibmveth_adapter *adapter = netdev_priv(netdev);
1426 unsigned long lpar_rc;
1428 if ((netdev->flags & IFF_PROMISC) ||
1429 (netdev_mc_count(netdev) > adapter->mcastFilterSize)) {
1430 lpar_rc = h_multicast_ctrl(adapter->vdev->unit_address,
1431 IbmVethMcastEnableRecv |
1432 IbmVethMcastDisableFiltering,
1434 if (lpar_rc != H_SUCCESS) {
1435 netdev_err(netdev, "h_multicast_ctrl rc=%ld when "
1436 "entering promisc mode\n", lpar_rc);
1438 } else {
1439 struct netdev_hw_addr *ha;
1440 /* clear the filter table & disable filtering */
1441 lpar_rc = h_multicast_ctrl(adapter->vdev->unit_address,
1442 IbmVethMcastEnableRecv |
1443 IbmVethMcastDisableFiltering |
1444 IbmVethMcastClearFilterTable,
1446 if (lpar_rc != H_SUCCESS) {
1447 netdev_err(netdev, "h_multicast_ctrl rc=%ld when "
1448 "attempting to clear filter table\n",
1449 lpar_rc);
1451 /* add the addresses to the filter table */
1452 netdev_for_each_mc_addr(ha, netdev) {
1453 /* add the multicast address to the filter table */
1454 u64 mcast_addr;
1455 mcast_addr = ibmveth_encode_mac_addr(ha->addr);
1456 lpar_rc = h_multicast_ctrl(adapter->vdev->unit_address,
1457 IbmVethMcastAddFilter,
1458 mcast_addr);
1459 if (lpar_rc != H_SUCCESS) {
1460 netdev_err(netdev, "h_multicast_ctrl rc=%ld "
1461 "when adding an entry to the filter "
1462 "table\n", lpar_rc);
1466 /* re-enable filtering */
1467 lpar_rc = h_multicast_ctrl(adapter->vdev->unit_address,
1468 IbmVethMcastEnableFiltering,
1470 if (lpar_rc != H_SUCCESS) {
1471 netdev_err(netdev, "h_multicast_ctrl rc=%ld when "
1472 "enabling filtering\n", lpar_rc);
1477 static int ibmveth_change_mtu(struct net_device *dev, int new_mtu)
1479 struct ibmveth_adapter *adapter = netdev_priv(dev);
1480 struct vio_dev *viodev = adapter->vdev;
1481 int new_mtu_oh = new_mtu + IBMVETH_BUFF_OH;
1482 int i, rc;
1483 int need_restart = 0;
1485 for (i = 0; i < IBMVETH_NUM_BUFF_POOLS; i++)
1486 if (new_mtu_oh <= adapter->rx_buff_pool[i].buff_size)
1487 break;
1489 if (i == IBMVETH_NUM_BUFF_POOLS)
1490 return -EINVAL;
1492 /* Deactivate all the buffer pools so that the next loop can activate
1493 only the buffer pools necessary to hold the new MTU */
1494 if (netif_running(adapter->netdev)) {
1495 need_restart = 1;
1496 adapter->pool_config = 1;
1497 ibmveth_close(adapter->netdev);
1498 adapter->pool_config = 0;
1501 /* Look for an active buffer pool that can hold the new MTU */
1502 for (i = 0; i < IBMVETH_NUM_BUFF_POOLS; i++) {
1503 adapter->rx_buff_pool[i].active = 1;
1505 if (new_mtu_oh <= adapter->rx_buff_pool[i].buff_size) {
1506 dev->mtu = new_mtu;
1507 vio_cmo_set_dev_desired(viodev,
1508 ibmveth_get_desired_dma
1509 (viodev));
1510 if (need_restart) {
1511 return ibmveth_open(adapter->netdev);
1513 return 0;
1517 if (need_restart && (rc = ibmveth_open(adapter->netdev)))
1518 return rc;
1520 return -EINVAL;
1523 #ifdef CONFIG_NET_POLL_CONTROLLER
1524 static void ibmveth_poll_controller(struct net_device *dev)
1526 ibmveth_replenish_task(netdev_priv(dev));
1527 ibmveth_interrupt(dev->irq, dev);
1529 #endif
1532 * ibmveth_get_desired_dma - Calculate IO memory desired by the driver
1534 * @vdev: struct vio_dev for the device whose desired IO mem is to be returned
1536 * Return value:
1537 * Number of bytes of IO data the driver will need to perform well.
1539 static unsigned long ibmveth_get_desired_dma(struct vio_dev *vdev)
1541 struct net_device *netdev = dev_get_drvdata(&vdev->dev);
1542 struct ibmveth_adapter *adapter;
1543 struct iommu_table *tbl;
1544 unsigned long ret;
1545 int i;
1546 int rxqentries = 1;
1548 tbl = get_iommu_table_base(&vdev->dev);
1550 /* netdev inits at probe time along with the structures we need below*/
1551 if (netdev == NULL)
1552 return IOMMU_PAGE_ALIGN(IBMVETH_IO_ENTITLEMENT_DEFAULT, tbl);
1554 adapter = netdev_priv(netdev);
1556 ret = IBMVETH_BUFF_LIST_SIZE + IBMVETH_FILT_LIST_SIZE;
1557 ret += IOMMU_PAGE_ALIGN(netdev->mtu, tbl);
1559 for (i = 0; i < IBMVETH_NUM_BUFF_POOLS; i++) {
1560 /* add the size of the active receive buffers */
1561 if (adapter->rx_buff_pool[i].active)
1562 ret +=
1563 adapter->rx_buff_pool[i].size *
1564 IOMMU_PAGE_ALIGN(adapter->rx_buff_pool[i].
1565 buff_size, tbl);
1566 rxqentries += adapter->rx_buff_pool[i].size;
1568 /* add the size of the receive queue entries */
1569 ret += IOMMU_PAGE_ALIGN(
1570 rxqentries * sizeof(struct ibmveth_rx_q_entry), tbl);
1572 return ret;
1575 static int ibmveth_set_mac_addr(struct net_device *dev, void *p)
1577 struct ibmveth_adapter *adapter = netdev_priv(dev);
1578 struct sockaddr *addr = p;
1579 u64 mac_address;
1580 int rc;
1582 if (!is_valid_ether_addr(addr->sa_data))
1583 return -EADDRNOTAVAIL;
1585 mac_address = ibmveth_encode_mac_addr(addr->sa_data);
1586 rc = h_change_logical_lan_mac(adapter->vdev->unit_address, mac_address);
1587 if (rc) {
1588 netdev_err(adapter->netdev, "h_change_logical_lan_mac failed with rc=%d\n", rc);
1589 return rc;
1592 ether_addr_copy(dev->dev_addr, addr->sa_data);
1594 return 0;
1597 static const struct net_device_ops ibmveth_netdev_ops = {
1598 .ndo_open = ibmveth_open,
1599 .ndo_stop = ibmveth_close,
1600 .ndo_start_xmit = ibmveth_start_xmit,
1601 .ndo_set_rx_mode = ibmveth_set_multicast_list,
1602 .ndo_do_ioctl = ibmveth_ioctl,
1603 .ndo_change_mtu = ibmveth_change_mtu,
1604 .ndo_fix_features = ibmveth_fix_features,
1605 .ndo_set_features = ibmveth_set_features,
1606 .ndo_validate_addr = eth_validate_addr,
1607 .ndo_set_mac_address = ibmveth_set_mac_addr,
1608 #ifdef CONFIG_NET_POLL_CONTROLLER
1609 .ndo_poll_controller = ibmveth_poll_controller,
1610 #endif
1613 static int ibmveth_probe(struct vio_dev *dev, const struct vio_device_id *id)
1615 int rc, i, mac_len;
1616 struct net_device *netdev;
1617 struct ibmveth_adapter *adapter;
1618 unsigned char *mac_addr_p;
1619 unsigned int *mcastFilterSize_p;
1620 long ret;
1621 unsigned long ret_attr;
1623 dev_dbg(&dev->dev, "entering ibmveth_probe for UA 0x%x\n",
1624 dev->unit_address);
1626 mac_addr_p = (unsigned char *)vio_get_attribute(dev, VETH_MAC_ADDR,
1627 &mac_len);
1628 if (!mac_addr_p) {
1629 dev_err(&dev->dev, "Can't find VETH_MAC_ADDR attribute\n");
1630 return -EINVAL;
1632 /* Workaround for old/broken pHyp */
1633 if (mac_len == 8)
1634 mac_addr_p += 2;
1635 else if (mac_len != 6) {
1636 dev_err(&dev->dev, "VETH_MAC_ADDR attribute wrong len %d\n",
1637 mac_len);
1638 return -EINVAL;
1641 mcastFilterSize_p = (unsigned int *)vio_get_attribute(dev,
1642 VETH_MCAST_FILTER_SIZE, NULL);
1643 if (!mcastFilterSize_p) {
1644 dev_err(&dev->dev, "Can't find VETH_MCAST_FILTER_SIZE "
1645 "attribute\n");
1646 return -EINVAL;
1649 netdev = alloc_etherdev(sizeof(struct ibmveth_adapter));
1651 if (!netdev)
1652 return -ENOMEM;
1654 adapter = netdev_priv(netdev);
1655 dev_set_drvdata(&dev->dev, netdev);
1657 adapter->vdev = dev;
1658 adapter->netdev = netdev;
1659 adapter->mcastFilterSize = *mcastFilterSize_p;
1660 adapter->pool_config = 0;
1662 netif_napi_add(netdev, &adapter->napi, ibmveth_poll, 16);
1664 netdev->irq = dev->irq;
1665 netdev->netdev_ops = &ibmveth_netdev_ops;
1666 netdev->ethtool_ops = &netdev_ethtool_ops;
1667 SET_NETDEV_DEV(netdev, &dev->dev);
1668 netdev->hw_features = NETIF_F_SG;
1669 if (vio_get_attribute(dev, "ibm,illan-options", NULL) != NULL) {
1670 netdev->hw_features |= NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM |
1671 NETIF_F_RXCSUM;
1674 netdev->features |= netdev->hw_features;
1676 ret = h_illan_attributes(adapter->vdev->unit_address, 0, 0, &ret_attr);
1678 /* If running older firmware, TSO should not be enabled by default */
1679 if (ret == H_SUCCESS && (ret_attr & IBMVETH_ILLAN_LRG_SND_SUPPORT) &&
1680 !old_large_send) {
1681 netdev->hw_features |= NETIF_F_TSO | NETIF_F_TSO6;
1682 netdev->features |= netdev->hw_features;
1683 } else {
1684 netdev->hw_features |= NETIF_F_TSO;
1687 adapter->is_active_trunk = false;
1688 if (ret == H_SUCCESS && (ret_attr & IBMVETH_ILLAN_ACTIVE_TRUNK)) {
1689 adapter->is_active_trunk = true;
1690 netdev->hw_features |= NETIF_F_FRAGLIST;
1691 netdev->features |= NETIF_F_FRAGLIST;
1694 netdev->min_mtu = IBMVETH_MIN_MTU;
1695 netdev->max_mtu = ETH_MAX_MTU;
1697 memcpy(netdev->dev_addr, mac_addr_p, ETH_ALEN);
1699 if (firmware_has_feature(FW_FEATURE_CMO))
1700 memcpy(pool_count, pool_count_cmo, sizeof(pool_count));
1702 for (i = 0; i < IBMVETH_NUM_BUFF_POOLS; i++) {
1703 struct kobject *kobj = &adapter->rx_buff_pool[i].kobj;
1704 int error;
1706 ibmveth_init_buffer_pool(&adapter->rx_buff_pool[i], i,
1707 pool_count[i], pool_size[i],
1708 pool_active[i]);
1709 error = kobject_init_and_add(kobj, &ktype_veth_pool,
1710 &dev->dev.kobj, "pool%d", i);
1711 if (!error)
1712 kobject_uevent(kobj, KOBJ_ADD);
1715 netdev_dbg(netdev, "adapter @ 0x%p\n", adapter);
1716 netdev_dbg(netdev, "registering netdev...\n");
1718 ibmveth_set_features(netdev, netdev->features);
1720 rc = register_netdev(netdev);
1722 if (rc) {
1723 netdev_dbg(netdev, "failed to register netdev rc=%d\n", rc);
1724 free_netdev(netdev);
1725 return rc;
1728 netdev_dbg(netdev, "registered\n");
1730 return 0;
1733 static int ibmveth_remove(struct vio_dev *dev)
1735 struct net_device *netdev = dev_get_drvdata(&dev->dev);
1736 struct ibmveth_adapter *adapter = netdev_priv(netdev);
1737 int i;
1739 for (i = 0; i < IBMVETH_NUM_BUFF_POOLS; i++)
1740 kobject_put(&adapter->rx_buff_pool[i].kobj);
1742 unregister_netdev(netdev);
1744 free_netdev(netdev);
1745 dev_set_drvdata(&dev->dev, NULL);
1747 return 0;
1750 static struct attribute veth_active_attr;
1751 static struct attribute veth_num_attr;
1752 static struct attribute veth_size_attr;
1754 static ssize_t veth_pool_show(struct kobject *kobj,
1755 struct attribute *attr, char *buf)
1757 struct ibmveth_buff_pool *pool = container_of(kobj,
1758 struct ibmveth_buff_pool,
1759 kobj);
1761 if (attr == &veth_active_attr)
1762 return sprintf(buf, "%d\n", pool->active);
1763 else if (attr == &veth_num_attr)
1764 return sprintf(buf, "%d\n", pool->size);
1765 else if (attr == &veth_size_attr)
1766 return sprintf(buf, "%d\n", pool->buff_size);
1767 return 0;
1770 static ssize_t veth_pool_store(struct kobject *kobj, struct attribute *attr,
1771 const char *buf, size_t count)
1773 struct ibmveth_buff_pool *pool = container_of(kobj,
1774 struct ibmveth_buff_pool,
1775 kobj);
1776 struct net_device *netdev = dev_get_drvdata(
1777 container_of(kobj->parent, struct device, kobj));
1778 struct ibmveth_adapter *adapter = netdev_priv(netdev);
1779 long value = simple_strtol(buf, NULL, 10);
1780 long rc;
1782 if (attr == &veth_active_attr) {
1783 if (value && !pool->active) {
1784 if (netif_running(netdev)) {
1785 if (ibmveth_alloc_buffer_pool(pool)) {
1786 netdev_err(netdev,
1787 "unable to alloc pool\n");
1788 return -ENOMEM;
1790 pool->active = 1;
1791 adapter->pool_config = 1;
1792 ibmveth_close(netdev);
1793 adapter->pool_config = 0;
1794 if ((rc = ibmveth_open(netdev)))
1795 return rc;
1796 } else {
1797 pool->active = 1;
1799 } else if (!value && pool->active) {
1800 int mtu = netdev->mtu + IBMVETH_BUFF_OH;
1801 int i;
1802 /* Make sure there is a buffer pool with buffers that
1803 can hold a packet of the size of the MTU */
1804 for (i = 0; i < IBMVETH_NUM_BUFF_POOLS; i++) {
1805 if (pool == &adapter->rx_buff_pool[i])
1806 continue;
1807 if (!adapter->rx_buff_pool[i].active)
1808 continue;
1809 if (mtu <= adapter->rx_buff_pool[i].buff_size)
1810 break;
1813 if (i == IBMVETH_NUM_BUFF_POOLS) {
1814 netdev_err(netdev, "no active pool >= MTU\n");
1815 return -EPERM;
1818 if (netif_running(netdev)) {
1819 adapter->pool_config = 1;
1820 ibmveth_close(netdev);
1821 pool->active = 0;
1822 adapter->pool_config = 0;
1823 if ((rc = ibmveth_open(netdev)))
1824 return rc;
1826 pool->active = 0;
1828 } else if (attr == &veth_num_attr) {
1829 if (value <= 0 || value > IBMVETH_MAX_POOL_COUNT) {
1830 return -EINVAL;
1831 } else {
1832 if (netif_running(netdev)) {
1833 adapter->pool_config = 1;
1834 ibmveth_close(netdev);
1835 adapter->pool_config = 0;
1836 pool->size = value;
1837 if ((rc = ibmveth_open(netdev)))
1838 return rc;
1839 } else {
1840 pool->size = value;
1843 } else if (attr == &veth_size_attr) {
1844 if (value <= IBMVETH_BUFF_OH || value > IBMVETH_MAX_BUF_SIZE) {
1845 return -EINVAL;
1846 } else {
1847 if (netif_running(netdev)) {
1848 adapter->pool_config = 1;
1849 ibmveth_close(netdev);
1850 adapter->pool_config = 0;
1851 pool->buff_size = value;
1852 if ((rc = ibmveth_open(netdev)))
1853 return rc;
1854 } else {
1855 pool->buff_size = value;
1860 /* kick the interrupt handler to allocate/deallocate pools */
1861 ibmveth_interrupt(netdev->irq, netdev);
1862 return count;
1866 #define ATTR(_name, _mode) \
1867 struct attribute veth_##_name##_attr = { \
1868 .name = __stringify(_name), .mode = _mode, \
1871 static ATTR(active, 0644);
1872 static ATTR(num, 0644);
1873 static ATTR(size, 0644);
1875 static struct attribute *veth_pool_attrs[] = {
1876 &veth_active_attr,
1877 &veth_num_attr,
1878 &veth_size_attr,
1879 NULL,
1882 static const struct sysfs_ops veth_pool_ops = {
1883 .show = veth_pool_show,
1884 .store = veth_pool_store,
1887 static struct kobj_type ktype_veth_pool = {
1888 .release = NULL,
1889 .sysfs_ops = &veth_pool_ops,
1890 .default_attrs = veth_pool_attrs,
1893 static int ibmveth_resume(struct device *dev)
1895 struct net_device *netdev = dev_get_drvdata(dev);
1896 ibmveth_interrupt(netdev->irq, netdev);
1897 return 0;
1900 static const struct vio_device_id ibmveth_device_table[] = {
1901 { "network", "IBM,l-lan"},
1902 { "", "" }
1904 MODULE_DEVICE_TABLE(vio, ibmveth_device_table);
1906 static const struct dev_pm_ops ibmveth_pm_ops = {
1907 .resume = ibmveth_resume
1910 static struct vio_driver ibmveth_driver = {
1911 .id_table = ibmveth_device_table,
1912 .probe = ibmveth_probe,
1913 .remove = ibmveth_remove,
1914 .get_desired_dma = ibmveth_get_desired_dma,
1915 .name = ibmveth_driver_name,
1916 .pm = &ibmveth_pm_ops,
1919 static int __init ibmveth_module_init(void)
1921 printk(KERN_DEBUG "%s: %s %s\n", ibmveth_driver_name,
1922 ibmveth_driver_string, ibmveth_driver_version);
1924 return vio_register_driver(&ibmveth_driver);
1927 static void __exit ibmveth_module_exit(void)
1929 vio_unregister_driver(&ibmveth_driver);
1932 module_init(ibmveth_module_init);
1933 module_exit(ibmveth_module_exit);