perf tools: Don't clone maps from parent when synthesizing forks
[linux/fpc-iii.git] / drivers / net / ethernet / ibm / ibmveth.c
bloba4681780a55d2e415e755b271585a9983c6ac5dc
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/types.h>
28 #include <linux/errno.h>
29 #include <linux/dma-mapping.h>
30 #include <linux/kernel.h>
31 #include <linux/netdevice.h>
32 #include <linux/etherdevice.h>
33 #include <linux/skbuff.h>
34 #include <linux/init.h>
35 #include <linux/interrupt.h>
36 #include <linux/mm.h>
37 #include <linux/pm.h>
38 #include <linux/ethtool.h>
39 #include <linux/in.h>
40 #include <linux/ip.h>
41 #include <linux/ipv6.h>
42 #include <linux/slab.h>
43 #include <asm/hvcall.h>
44 #include <linux/atomic.h>
45 #include <asm/vio.h>
46 #include <asm/iommu.h>
47 #include <asm/firmware.h>
48 #include <net/tcp.h>
49 #include <net/ip6_checksum.h>
51 #include "ibmveth.h"
53 static irqreturn_t ibmveth_interrupt(int irq, void *dev_instance);
54 static void ibmveth_rxq_harvest_buffer(struct ibmveth_adapter *adapter);
55 static unsigned long ibmveth_get_desired_dma(struct vio_dev *vdev);
57 static struct kobj_type ktype_veth_pool;
60 static const char ibmveth_driver_name[] = "ibmveth";
61 static const char ibmveth_driver_string[] = "IBM Power Virtual Ethernet Driver";
62 #define ibmveth_driver_version "1.06"
64 MODULE_AUTHOR("Santiago Leon <santil@linux.vnet.ibm.com>");
65 MODULE_DESCRIPTION("IBM Power Virtual Ethernet Driver");
66 MODULE_LICENSE("GPL");
67 MODULE_VERSION(ibmveth_driver_version);
69 static unsigned int tx_copybreak __read_mostly = 128;
70 module_param(tx_copybreak, uint, 0644);
71 MODULE_PARM_DESC(tx_copybreak,
72 "Maximum size of packet that is copied to a new buffer on transmit");
74 static unsigned int rx_copybreak __read_mostly = 128;
75 module_param(rx_copybreak, uint, 0644);
76 MODULE_PARM_DESC(rx_copybreak,
77 "Maximum size of packet that is copied to a new buffer on receive");
79 static unsigned int rx_flush __read_mostly = 0;
80 module_param(rx_flush, uint, 0644);
81 MODULE_PARM_DESC(rx_flush, "Flush receive buffers before use");
83 static bool old_large_send __read_mostly;
84 module_param(old_large_send, bool, 0444);
85 MODULE_PARM_DESC(old_large_send,
86 "Use old large send method on firmware that supports the new method");
88 struct ibmveth_stat {
89 char name[ETH_GSTRING_LEN];
90 int offset;
93 #define IBMVETH_STAT_OFF(stat) offsetof(struct ibmveth_adapter, stat)
94 #define IBMVETH_GET_STAT(a, off) *((u64 *)(((unsigned long)(a)) + off))
96 struct ibmveth_stat ibmveth_stats[] = {
97 { "replenish_task_cycles", IBMVETH_STAT_OFF(replenish_task_cycles) },
98 { "replenish_no_mem", IBMVETH_STAT_OFF(replenish_no_mem) },
99 { "replenish_add_buff_failure",
100 IBMVETH_STAT_OFF(replenish_add_buff_failure) },
101 { "replenish_add_buff_success",
102 IBMVETH_STAT_OFF(replenish_add_buff_success) },
103 { "rx_invalid_buffer", IBMVETH_STAT_OFF(rx_invalid_buffer) },
104 { "rx_no_buffer", IBMVETH_STAT_OFF(rx_no_buffer) },
105 { "tx_map_failed", IBMVETH_STAT_OFF(tx_map_failed) },
106 { "tx_send_failed", IBMVETH_STAT_OFF(tx_send_failed) },
107 { "fw_enabled_ipv4_csum", IBMVETH_STAT_OFF(fw_ipv4_csum_support) },
108 { "fw_enabled_ipv6_csum", IBMVETH_STAT_OFF(fw_ipv6_csum_support) },
109 { "tx_large_packets", IBMVETH_STAT_OFF(tx_large_packets) },
110 { "rx_large_packets", IBMVETH_STAT_OFF(rx_large_packets) },
111 { "fw_enabled_large_send", IBMVETH_STAT_OFF(fw_large_send_support) }
114 /* simple methods of getting data from the current rxq entry */
115 static inline u32 ibmveth_rxq_flags(struct ibmveth_adapter *adapter)
117 return be32_to_cpu(adapter->rx_queue.queue_addr[adapter->rx_queue.index].flags_off);
120 static inline int ibmveth_rxq_toggle(struct ibmveth_adapter *adapter)
122 return (ibmveth_rxq_flags(adapter) & IBMVETH_RXQ_TOGGLE) >>
123 IBMVETH_RXQ_TOGGLE_SHIFT;
126 static inline int ibmveth_rxq_pending_buffer(struct ibmveth_adapter *adapter)
128 return ibmveth_rxq_toggle(adapter) == adapter->rx_queue.toggle;
131 static inline int ibmveth_rxq_buffer_valid(struct ibmveth_adapter *adapter)
133 return ibmveth_rxq_flags(adapter) & IBMVETH_RXQ_VALID;
136 static inline int ibmveth_rxq_frame_offset(struct ibmveth_adapter *adapter)
138 return ibmveth_rxq_flags(adapter) & IBMVETH_RXQ_OFF_MASK;
141 static inline int ibmveth_rxq_large_packet(struct ibmveth_adapter *adapter)
143 return ibmveth_rxq_flags(adapter) & IBMVETH_RXQ_LRG_PKT;
146 static inline int ibmveth_rxq_frame_length(struct ibmveth_adapter *adapter)
148 return be32_to_cpu(adapter->rx_queue.queue_addr[adapter->rx_queue.index].length);
151 static inline int ibmveth_rxq_csum_good(struct ibmveth_adapter *adapter)
153 return ibmveth_rxq_flags(adapter) & IBMVETH_RXQ_CSUM_GOOD;
156 /* setup the initial settings for a buffer pool */
157 static void ibmveth_init_buffer_pool(struct ibmveth_buff_pool *pool,
158 u32 pool_index, u32 pool_size,
159 u32 buff_size, u32 pool_active)
161 pool->size = pool_size;
162 pool->index = pool_index;
163 pool->buff_size = buff_size;
164 pool->threshold = pool_size * 7 / 8;
165 pool->active = pool_active;
168 /* allocate and setup an buffer pool - called during open */
169 static int ibmveth_alloc_buffer_pool(struct ibmveth_buff_pool *pool)
171 int i;
173 pool->free_map = kmalloc_array(pool->size, sizeof(u16), GFP_KERNEL);
175 if (!pool->free_map)
176 return -1;
178 pool->dma_addr = kcalloc(pool->size, sizeof(dma_addr_t), GFP_KERNEL);
179 if (!pool->dma_addr) {
180 kfree(pool->free_map);
181 pool->free_map = NULL;
182 return -1;
185 pool->skbuff = kcalloc(pool->size, sizeof(void *), GFP_KERNEL);
187 if (!pool->skbuff) {
188 kfree(pool->dma_addr);
189 pool->dma_addr = NULL;
191 kfree(pool->free_map);
192 pool->free_map = NULL;
193 return -1;
196 for (i = 0; i < pool->size; ++i)
197 pool->free_map[i] = i;
199 atomic_set(&pool->available, 0);
200 pool->producer_index = 0;
201 pool->consumer_index = 0;
203 return 0;
206 static inline void ibmveth_flush_buffer(void *addr, unsigned long length)
208 unsigned long offset;
210 for (offset = 0; offset < length; offset += SMP_CACHE_BYTES)
211 asm("dcbfl %0,%1" :: "b" (addr), "r" (offset));
214 /* replenish the buffers for a pool. note that we don't need to
215 * skb_reserve these since they are used for incoming...
217 static void ibmveth_replenish_buffer_pool(struct ibmveth_adapter *adapter,
218 struct ibmveth_buff_pool *pool)
220 u32 i;
221 u32 count = pool->size - atomic_read(&pool->available);
222 u32 buffers_added = 0;
223 struct sk_buff *skb;
224 unsigned int free_index, index;
225 u64 correlator;
226 unsigned long lpar_rc;
227 dma_addr_t dma_addr;
229 mb();
231 for (i = 0; i < count; ++i) {
232 union ibmveth_buf_desc desc;
234 skb = netdev_alloc_skb(adapter->netdev, pool->buff_size);
236 if (!skb) {
237 netdev_dbg(adapter->netdev,
238 "replenish: unable to allocate skb\n");
239 adapter->replenish_no_mem++;
240 break;
243 free_index = pool->consumer_index;
244 pool->consumer_index++;
245 if (pool->consumer_index >= pool->size)
246 pool->consumer_index = 0;
247 index = pool->free_map[free_index];
249 BUG_ON(index == IBM_VETH_INVALID_MAP);
250 BUG_ON(pool->skbuff[index] != NULL);
252 dma_addr = dma_map_single(&adapter->vdev->dev, skb->data,
253 pool->buff_size, DMA_FROM_DEVICE);
255 if (dma_mapping_error(&adapter->vdev->dev, dma_addr))
256 goto failure;
258 pool->free_map[free_index] = IBM_VETH_INVALID_MAP;
259 pool->dma_addr[index] = dma_addr;
260 pool->skbuff[index] = skb;
262 correlator = ((u64)pool->index << 32) | index;
263 *(u64 *)skb->data = correlator;
265 desc.fields.flags_len = IBMVETH_BUF_VALID | pool->buff_size;
266 desc.fields.address = dma_addr;
268 if (rx_flush) {
269 unsigned int len = min(pool->buff_size,
270 adapter->netdev->mtu +
271 IBMVETH_BUFF_OH);
272 ibmveth_flush_buffer(skb->data, len);
274 lpar_rc = h_add_logical_lan_buffer(adapter->vdev->unit_address,
275 desc.desc);
277 if (lpar_rc != H_SUCCESS) {
278 goto failure;
279 } else {
280 buffers_added++;
281 adapter->replenish_add_buff_success++;
285 mb();
286 atomic_add(buffers_added, &(pool->available));
287 return;
289 failure:
290 pool->free_map[free_index] = index;
291 pool->skbuff[index] = NULL;
292 if (pool->consumer_index == 0)
293 pool->consumer_index = pool->size - 1;
294 else
295 pool->consumer_index--;
296 if (!dma_mapping_error(&adapter->vdev->dev, dma_addr))
297 dma_unmap_single(&adapter->vdev->dev,
298 pool->dma_addr[index], pool->buff_size,
299 DMA_FROM_DEVICE);
300 dev_kfree_skb_any(skb);
301 adapter->replenish_add_buff_failure++;
303 mb();
304 atomic_add(buffers_added, &(pool->available));
308 * The final 8 bytes of the buffer list is a counter of frames dropped
309 * because there was not a buffer in the buffer list capable of holding
310 * the frame.
312 static void ibmveth_update_rx_no_buffer(struct ibmveth_adapter *adapter)
314 __be64 *p = adapter->buffer_list_addr + 4096 - 8;
316 adapter->rx_no_buffer = be64_to_cpup(p);
319 /* replenish routine */
320 static void ibmveth_replenish_task(struct ibmveth_adapter *adapter)
322 int i;
324 adapter->replenish_task_cycles++;
326 for (i = (IBMVETH_NUM_BUFF_POOLS - 1); i >= 0; i--) {
327 struct ibmveth_buff_pool *pool = &adapter->rx_buff_pool[i];
329 if (pool->active &&
330 (atomic_read(&pool->available) < pool->threshold))
331 ibmveth_replenish_buffer_pool(adapter, pool);
334 ibmveth_update_rx_no_buffer(adapter);
337 /* empty and free ana buffer pool - also used to do cleanup in error paths */
338 static void ibmveth_free_buffer_pool(struct ibmveth_adapter *adapter,
339 struct ibmveth_buff_pool *pool)
341 int i;
343 kfree(pool->free_map);
344 pool->free_map = NULL;
346 if (pool->skbuff && pool->dma_addr) {
347 for (i = 0; i < pool->size; ++i) {
348 struct sk_buff *skb = pool->skbuff[i];
349 if (skb) {
350 dma_unmap_single(&adapter->vdev->dev,
351 pool->dma_addr[i],
352 pool->buff_size,
353 DMA_FROM_DEVICE);
354 dev_kfree_skb_any(skb);
355 pool->skbuff[i] = NULL;
360 if (pool->dma_addr) {
361 kfree(pool->dma_addr);
362 pool->dma_addr = NULL;
365 if (pool->skbuff) {
366 kfree(pool->skbuff);
367 pool->skbuff = NULL;
371 /* remove a buffer from a pool */
372 static void ibmveth_remove_buffer_from_pool(struct ibmveth_adapter *adapter,
373 u64 correlator)
375 unsigned int pool = correlator >> 32;
376 unsigned int index = correlator & 0xffffffffUL;
377 unsigned int free_index;
378 struct sk_buff *skb;
380 BUG_ON(pool >= IBMVETH_NUM_BUFF_POOLS);
381 BUG_ON(index >= adapter->rx_buff_pool[pool].size);
383 skb = adapter->rx_buff_pool[pool].skbuff[index];
385 BUG_ON(skb == NULL);
387 adapter->rx_buff_pool[pool].skbuff[index] = NULL;
389 dma_unmap_single(&adapter->vdev->dev,
390 adapter->rx_buff_pool[pool].dma_addr[index],
391 adapter->rx_buff_pool[pool].buff_size,
392 DMA_FROM_DEVICE);
394 free_index = adapter->rx_buff_pool[pool].producer_index;
395 adapter->rx_buff_pool[pool].producer_index++;
396 if (adapter->rx_buff_pool[pool].producer_index >=
397 adapter->rx_buff_pool[pool].size)
398 adapter->rx_buff_pool[pool].producer_index = 0;
399 adapter->rx_buff_pool[pool].free_map[free_index] = index;
401 mb();
403 atomic_dec(&(adapter->rx_buff_pool[pool].available));
406 /* get the current buffer on the rx queue */
407 static inline struct sk_buff *ibmveth_rxq_get_buffer(struct ibmveth_adapter *adapter)
409 u64 correlator = adapter->rx_queue.queue_addr[adapter->rx_queue.index].correlator;
410 unsigned int pool = correlator >> 32;
411 unsigned int index = correlator & 0xffffffffUL;
413 BUG_ON(pool >= IBMVETH_NUM_BUFF_POOLS);
414 BUG_ON(index >= adapter->rx_buff_pool[pool].size);
416 return adapter->rx_buff_pool[pool].skbuff[index];
419 /* recycle the current buffer on the rx queue */
420 static int ibmveth_rxq_recycle_buffer(struct ibmveth_adapter *adapter)
422 u32 q_index = adapter->rx_queue.index;
423 u64 correlator = adapter->rx_queue.queue_addr[q_index].correlator;
424 unsigned int pool = correlator >> 32;
425 unsigned int index = correlator & 0xffffffffUL;
426 union ibmveth_buf_desc desc;
427 unsigned long lpar_rc;
428 int ret = 1;
430 BUG_ON(pool >= IBMVETH_NUM_BUFF_POOLS);
431 BUG_ON(index >= adapter->rx_buff_pool[pool].size);
433 if (!adapter->rx_buff_pool[pool].active) {
434 ibmveth_rxq_harvest_buffer(adapter);
435 ibmveth_free_buffer_pool(adapter, &adapter->rx_buff_pool[pool]);
436 goto out;
439 desc.fields.flags_len = IBMVETH_BUF_VALID |
440 adapter->rx_buff_pool[pool].buff_size;
441 desc.fields.address = adapter->rx_buff_pool[pool].dma_addr[index];
443 lpar_rc = h_add_logical_lan_buffer(adapter->vdev->unit_address, desc.desc);
445 if (lpar_rc != H_SUCCESS) {
446 netdev_dbg(adapter->netdev, "h_add_logical_lan_buffer failed "
447 "during recycle rc=%ld", lpar_rc);
448 ibmveth_remove_buffer_from_pool(adapter, adapter->rx_queue.queue_addr[adapter->rx_queue.index].correlator);
449 ret = 0;
452 if (++adapter->rx_queue.index == adapter->rx_queue.num_slots) {
453 adapter->rx_queue.index = 0;
454 adapter->rx_queue.toggle = !adapter->rx_queue.toggle;
457 out:
458 return ret;
461 static void ibmveth_rxq_harvest_buffer(struct ibmveth_adapter *adapter)
463 ibmveth_remove_buffer_from_pool(adapter, adapter->rx_queue.queue_addr[adapter->rx_queue.index].correlator);
465 if (++adapter->rx_queue.index == adapter->rx_queue.num_slots) {
466 adapter->rx_queue.index = 0;
467 adapter->rx_queue.toggle = !adapter->rx_queue.toggle;
471 static int ibmveth_register_logical_lan(struct ibmveth_adapter *adapter,
472 union ibmveth_buf_desc rxq_desc, u64 mac_address)
474 int rc, try_again = 1;
477 * After a kexec the adapter will still be open, so our attempt to
478 * open it will fail. So if we get a failure we free the adapter and
479 * try again, but only once.
481 retry:
482 rc = h_register_logical_lan(adapter->vdev->unit_address,
483 adapter->buffer_list_dma, rxq_desc.desc,
484 adapter->filter_list_dma, mac_address);
486 if (rc != H_SUCCESS && try_again) {
487 do {
488 rc = h_free_logical_lan(adapter->vdev->unit_address);
489 } while (H_IS_LONG_BUSY(rc) || (rc == H_BUSY));
491 try_again = 0;
492 goto retry;
495 return rc;
498 static u64 ibmveth_encode_mac_addr(u8 *mac)
500 int i;
501 u64 encoded = 0;
503 for (i = 0; i < ETH_ALEN; i++)
504 encoded = (encoded << 8) | mac[i];
506 return encoded;
509 static int ibmveth_open(struct net_device *netdev)
511 struct ibmveth_adapter *adapter = netdev_priv(netdev);
512 u64 mac_address;
513 int rxq_entries = 1;
514 unsigned long lpar_rc;
515 int rc;
516 union ibmveth_buf_desc rxq_desc;
517 int i;
518 struct device *dev;
520 netdev_dbg(netdev, "open starting\n");
522 napi_enable(&adapter->napi);
524 for(i = 0; i < IBMVETH_NUM_BUFF_POOLS; i++)
525 rxq_entries += adapter->rx_buff_pool[i].size;
527 rc = -ENOMEM;
528 adapter->buffer_list_addr = (void*) get_zeroed_page(GFP_KERNEL);
529 if (!adapter->buffer_list_addr) {
530 netdev_err(netdev, "unable to allocate list pages\n");
531 goto out;
534 adapter->filter_list_addr = (void*) get_zeroed_page(GFP_KERNEL);
535 if (!adapter->filter_list_addr) {
536 netdev_err(netdev, "unable to allocate filter pages\n");
537 goto out_free_buffer_list;
540 dev = &adapter->vdev->dev;
542 adapter->rx_queue.queue_len = sizeof(struct ibmveth_rx_q_entry) *
543 rxq_entries;
544 adapter->rx_queue.queue_addr =
545 dma_alloc_coherent(dev, adapter->rx_queue.queue_len,
546 &adapter->rx_queue.queue_dma, GFP_KERNEL);
547 if (!adapter->rx_queue.queue_addr)
548 goto out_free_filter_list;
550 adapter->buffer_list_dma = dma_map_single(dev,
551 adapter->buffer_list_addr, 4096, DMA_BIDIRECTIONAL);
552 if (dma_mapping_error(dev, adapter->buffer_list_dma)) {
553 netdev_err(netdev, "unable to map buffer list pages\n");
554 goto out_free_queue_mem;
557 adapter->filter_list_dma = dma_map_single(dev,
558 adapter->filter_list_addr, 4096, DMA_BIDIRECTIONAL);
559 if (dma_mapping_error(dev, adapter->filter_list_dma)) {
560 netdev_err(netdev, "unable to map filter list pages\n");
561 goto out_unmap_buffer_list;
564 adapter->rx_queue.index = 0;
565 adapter->rx_queue.num_slots = rxq_entries;
566 adapter->rx_queue.toggle = 1;
568 mac_address = ibmveth_encode_mac_addr(netdev->dev_addr);
570 rxq_desc.fields.flags_len = IBMVETH_BUF_VALID |
571 adapter->rx_queue.queue_len;
572 rxq_desc.fields.address = adapter->rx_queue.queue_dma;
574 netdev_dbg(netdev, "buffer list @ 0x%p\n", adapter->buffer_list_addr);
575 netdev_dbg(netdev, "filter list @ 0x%p\n", adapter->filter_list_addr);
576 netdev_dbg(netdev, "receive q @ 0x%p\n", adapter->rx_queue.queue_addr);
578 h_vio_signal(adapter->vdev->unit_address, VIO_IRQ_DISABLE);
580 lpar_rc = ibmveth_register_logical_lan(adapter, rxq_desc, mac_address);
582 if (lpar_rc != H_SUCCESS) {
583 netdev_err(netdev, "h_register_logical_lan failed with %ld\n",
584 lpar_rc);
585 netdev_err(netdev, "buffer TCE:0x%llx filter TCE:0x%llx rxq "
586 "desc:0x%llx MAC:0x%llx\n",
587 adapter->buffer_list_dma,
588 adapter->filter_list_dma,
589 rxq_desc.desc,
590 mac_address);
591 rc = -ENONET;
592 goto out_unmap_filter_list;
595 for (i = 0; i < IBMVETH_NUM_BUFF_POOLS; i++) {
596 if (!adapter->rx_buff_pool[i].active)
597 continue;
598 if (ibmveth_alloc_buffer_pool(&adapter->rx_buff_pool[i])) {
599 netdev_err(netdev, "unable to alloc pool\n");
600 adapter->rx_buff_pool[i].active = 0;
601 rc = -ENOMEM;
602 goto out_free_buffer_pools;
606 netdev_dbg(netdev, "registering irq 0x%x\n", netdev->irq);
607 rc = request_irq(netdev->irq, ibmveth_interrupt, 0, netdev->name,
608 netdev);
609 if (rc != 0) {
610 netdev_err(netdev, "unable to request irq 0x%x, rc %d\n",
611 netdev->irq, rc);
612 do {
613 lpar_rc = h_free_logical_lan(adapter->vdev->unit_address);
614 } while (H_IS_LONG_BUSY(lpar_rc) || (lpar_rc == H_BUSY));
616 goto out_free_buffer_pools;
619 rc = -ENOMEM;
620 adapter->bounce_buffer =
621 kmalloc(netdev->mtu + IBMVETH_BUFF_OH, GFP_KERNEL);
622 if (!adapter->bounce_buffer)
623 goto out_free_irq;
625 adapter->bounce_buffer_dma =
626 dma_map_single(&adapter->vdev->dev, adapter->bounce_buffer,
627 netdev->mtu + IBMVETH_BUFF_OH, DMA_BIDIRECTIONAL);
628 if (dma_mapping_error(dev, adapter->bounce_buffer_dma)) {
629 netdev_err(netdev, "unable to map bounce buffer\n");
630 goto out_free_bounce_buffer;
633 netdev_dbg(netdev, "initial replenish cycle\n");
634 ibmveth_interrupt(netdev->irq, netdev);
636 netif_start_queue(netdev);
638 netdev_dbg(netdev, "open complete\n");
640 return 0;
642 out_free_bounce_buffer:
643 kfree(adapter->bounce_buffer);
644 out_free_irq:
645 free_irq(netdev->irq, netdev);
646 out_free_buffer_pools:
647 while (--i >= 0) {
648 if (adapter->rx_buff_pool[i].active)
649 ibmveth_free_buffer_pool(adapter,
650 &adapter->rx_buff_pool[i]);
652 out_unmap_filter_list:
653 dma_unmap_single(dev, adapter->filter_list_dma, 4096,
654 DMA_BIDIRECTIONAL);
655 out_unmap_buffer_list:
656 dma_unmap_single(dev, adapter->buffer_list_dma, 4096,
657 DMA_BIDIRECTIONAL);
658 out_free_queue_mem:
659 dma_free_coherent(dev, adapter->rx_queue.queue_len,
660 adapter->rx_queue.queue_addr,
661 adapter->rx_queue.queue_dma);
662 out_free_filter_list:
663 free_page((unsigned long)adapter->filter_list_addr);
664 out_free_buffer_list:
665 free_page((unsigned long)adapter->buffer_list_addr);
666 out:
667 napi_disable(&adapter->napi);
668 return rc;
671 static int ibmveth_close(struct net_device *netdev)
673 struct ibmveth_adapter *adapter = netdev_priv(netdev);
674 struct device *dev = &adapter->vdev->dev;
675 long lpar_rc;
676 int i;
678 netdev_dbg(netdev, "close starting\n");
680 napi_disable(&adapter->napi);
682 if (!adapter->pool_config)
683 netif_stop_queue(netdev);
685 h_vio_signal(adapter->vdev->unit_address, VIO_IRQ_DISABLE);
687 do {
688 lpar_rc = h_free_logical_lan(adapter->vdev->unit_address);
689 } while (H_IS_LONG_BUSY(lpar_rc) || (lpar_rc == H_BUSY));
691 if (lpar_rc != H_SUCCESS) {
692 netdev_err(netdev, "h_free_logical_lan failed with %lx, "
693 "continuing with close\n", lpar_rc);
696 free_irq(netdev->irq, netdev);
698 ibmveth_update_rx_no_buffer(adapter);
700 dma_unmap_single(dev, adapter->buffer_list_dma, 4096,
701 DMA_BIDIRECTIONAL);
702 free_page((unsigned long)adapter->buffer_list_addr);
704 dma_unmap_single(dev, adapter->filter_list_dma, 4096,
705 DMA_BIDIRECTIONAL);
706 free_page((unsigned long)adapter->filter_list_addr);
708 dma_free_coherent(dev, adapter->rx_queue.queue_len,
709 adapter->rx_queue.queue_addr,
710 adapter->rx_queue.queue_dma);
712 for (i = 0; i < IBMVETH_NUM_BUFF_POOLS; i++)
713 if (adapter->rx_buff_pool[i].active)
714 ibmveth_free_buffer_pool(adapter,
715 &adapter->rx_buff_pool[i]);
717 dma_unmap_single(&adapter->vdev->dev, adapter->bounce_buffer_dma,
718 adapter->netdev->mtu + IBMVETH_BUFF_OH,
719 DMA_BIDIRECTIONAL);
720 kfree(adapter->bounce_buffer);
722 netdev_dbg(netdev, "close complete\n");
724 return 0;
727 static int netdev_get_link_ksettings(struct net_device *dev,
728 struct ethtool_link_ksettings *cmd)
730 u32 supported, advertising;
732 supported = (SUPPORTED_1000baseT_Full | SUPPORTED_Autoneg |
733 SUPPORTED_FIBRE);
734 advertising = (ADVERTISED_1000baseT_Full | ADVERTISED_Autoneg |
735 ADVERTISED_FIBRE);
736 cmd->base.speed = SPEED_1000;
737 cmd->base.duplex = DUPLEX_FULL;
738 cmd->base.port = PORT_FIBRE;
739 cmd->base.phy_address = 0;
740 cmd->base.autoneg = AUTONEG_ENABLE;
742 ethtool_convert_legacy_u32_to_link_mode(cmd->link_modes.supported,
743 supported);
744 ethtool_convert_legacy_u32_to_link_mode(cmd->link_modes.advertising,
745 advertising);
747 return 0;
750 static void netdev_get_drvinfo(struct net_device *dev,
751 struct ethtool_drvinfo *info)
753 strlcpy(info->driver, ibmveth_driver_name, sizeof(info->driver));
754 strlcpy(info->version, ibmveth_driver_version, sizeof(info->version));
757 static netdev_features_t ibmveth_fix_features(struct net_device *dev,
758 netdev_features_t features)
761 * Since the ibmveth firmware interface does not have the
762 * concept of separate tx/rx checksum offload enable, if rx
763 * checksum is disabled we also have to disable tx checksum
764 * offload. Once we disable rx checksum offload, we are no
765 * longer allowed to send tx buffers that are not properly
766 * checksummed.
769 if (!(features & NETIF_F_RXCSUM))
770 features &= ~NETIF_F_CSUM_MASK;
772 return features;
775 static int ibmveth_set_csum_offload(struct net_device *dev, u32 data)
777 struct ibmveth_adapter *adapter = netdev_priv(dev);
778 unsigned long set_attr, clr_attr, ret_attr;
779 unsigned long set_attr6, clr_attr6;
780 long ret, ret4, ret6;
781 int rc1 = 0, rc2 = 0;
782 int restart = 0;
784 if (netif_running(dev)) {
785 restart = 1;
786 adapter->pool_config = 1;
787 ibmveth_close(dev);
788 adapter->pool_config = 0;
791 set_attr = 0;
792 clr_attr = 0;
793 set_attr6 = 0;
794 clr_attr6 = 0;
796 if (data) {
797 set_attr = IBMVETH_ILLAN_IPV4_TCP_CSUM;
798 set_attr6 = IBMVETH_ILLAN_IPV6_TCP_CSUM;
799 } else {
800 clr_attr = IBMVETH_ILLAN_IPV4_TCP_CSUM;
801 clr_attr6 = IBMVETH_ILLAN_IPV6_TCP_CSUM;
804 ret = h_illan_attributes(adapter->vdev->unit_address, 0, 0, &ret_attr);
806 if (ret == H_SUCCESS &&
807 (ret_attr & IBMVETH_ILLAN_PADDED_PKT_CSUM)) {
808 ret4 = h_illan_attributes(adapter->vdev->unit_address, clr_attr,
809 set_attr, &ret_attr);
811 if (ret4 != H_SUCCESS) {
812 netdev_err(dev, "unable to change IPv4 checksum "
813 "offload settings. %d rc=%ld\n",
814 data, ret4);
816 h_illan_attributes(adapter->vdev->unit_address,
817 set_attr, clr_attr, &ret_attr);
819 if (data == 1)
820 dev->features &= ~NETIF_F_IP_CSUM;
822 } else {
823 adapter->fw_ipv4_csum_support = data;
826 ret6 = h_illan_attributes(adapter->vdev->unit_address,
827 clr_attr6, set_attr6, &ret_attr);
829 if (ret6 != H_SUCCESS) {
830 netdev_err(dev, "unable to change IPv6 checksum "
831 "offload settings. %d rc=%ld\n",
832 data, ret6);
834 h_illan_attributes(adapter->vdev->unit_address,
835 set_attr6, clr_attr6, &ret_attr);
837 if (data == 1)
838 dev->features &= ~NETIF_F_IPV6_CSUM;
840 } else
841 adapter->fw_ipv6_csum_support = data;
843 if (ret4 == H_SUCCESS || ret6 == H_SUCCESS)
844 adapter->rx_csum = data;
845 else
846 rc1 = -EIO;
847 } else {
848 rc1 = -EIO;
849 netdev_err(dev, "unable to change checksum offload settings."
850 " %d rc=%ld ret_attr=%lx\n", data, ret,
851 ret_attr);
854 if (restart)
855 rc2 = ibmveth_open(dev);
857 return rc1 ? rc1 : rc2;
860 static int ibmveth_set_tso(struct net_device *dev, u32 data)
862 struct ibmveth_adapter *adapter = netdev_priv(dev);
863 unsigned long set_attr, clr_attr, ret_attr;
864 long ret1, ret2;
865 int rc1 = 0, rc2 = 0;
866 int restart = 0;
868 if (netif_running(dev)) {
869 restart = 1;
870 adapter->pool_config = 1;
871 ibmveth_close(dev);
872 adapter->pool_config = 0;
875 set_attr = 0;
876 clr_attr = 0;
878 if (data)
879 set_attr = IBMVETH_ILLAN_LRG_SR_ENABLED;
880 else
881 clr_attr = IBMVETH_ILLAN_LRG_SR_ENABLED;
883 ret1 = h_illan_attributes(adapter->vdev->unit_address, 0, 0, &ret_attr);
885 if (ret1 == H_SUCCESS && (ret_attr & IBMVETH_ILLAN_LRG_SND_SUPPORT) &&
886 !old_large_send) {
887 ret2 = h_illan_attributes(adapter->vdev->unit_address, clr_attr,
888 set_attr, &ret_attr);
890 if (ret2 != H_SUCCESS) {
891 netdev_err(dev, "unable to change tso settings. %d rc=%ld\n",
892 data, ret2);
894 h_illan_attributes(adapter->vdev->unit_address,
895 set_attr, clr_attr, &ret_attr);
897 if (data == 1)
898 dev->features &= ~(NETIF_F_TSO | NETIF_F_TSO6);
899 rc1 = -EIO;
901 } else {
902 adapter->fw_large_send_support = data;
903 adapter->large_send = data;
905 } else {
906 /* Older firmware version of large send offload does not
907 * support tcp6/ipv6
909 if (data == 1) {
910 dev->features &= ~NETIF_F_TSO6;
911 netdev_info(dev, "TSO feature requires all partitions to have updated driver");
913 adapter->large_send = data;
916 if (restart)
917 rc2 = ibmveth_open(dev);
919 return rc1 ? rc1 : rc2;
922 static int ibmveth_set_features(struct net_device *dev,
923 netdev_features_t features)
925 struct ibmveth_adapter *adapter = netdev_priv(dev);
926 int rx_csum = !!(features & NETIF_F_RXCSUM);
927 int large_send = !!(features & (NETIF_F_TSO | NETIF_F_TSO6));
928 int rc1 = 0, rc2 = 0;
930 if (rx_csum != adapter->rx_csum) {
931 rc1 = ibmveth_set_csum_offload(dev, rx_csum);
932 if (rc1 && !adapter->rx_csum)
933 dev->features =
934 features & ~(NETIF_F_CSUM_MASK |
935 NETIF_F_RXCSUM);
938 if (large_send != adapter->large_send) {
939 rc2 = ibmveth_set_tso(dev, large_send);
940 if (rc2 && !adapter->large_send)
941 dev->features =
942 features & ~(NETIF_F_TSO | NETIF_F_TSO6);
945 return rc1 ? rc1 : rc2;
948 static void ibmveth_get_strings(struct net_device *dev, u32 stringset, u8 *data)
950 int i;
952 if (stringset != ETH_SS_STATS)
953 return;
955 for (i = 0; i < ARRAY_SIZE(ibmveth_stats); i++, data += ETH_GSTRING_LEN)
956 memcpy(data, ibmveth_stats[i].name, ETH_GSTRING_LEN);
959 static int ibmveth_get_sset_count(struct net_device *dev, int sset)
961 switch (sset) {
962 case ETH_SS_STATS:
963 return ARRAY_SIZE(ibmveth_stats);
964 default:
965 return -EOPNOTSUPP;
969 static void ibmveth_get_ethtool_stats(struct net_device *dev,
970 struct ethtool_stats *stats, u64 *data)
972 int i;
973 struct ibmveth_adapter *adapter = netdev_priv(dev);
975 for (i = 0; i < ARRAY_SIZE(ibmveth_stats); i++)
976 data[i] = IBMVETH_GET_STAT(adapter, ibmveth_stats[i].offset);
979 static const struct ethtool_ops netdev_ethtool_ops = {
980 .get_drvinfo = netdev_get_drvinfo,
981 .get_link = ethtool_op_get_link,
982 .get_strings = ibmveth_get_strings,
983 .get_sset_count = ibmveth_get_sset_count,
984 .get_ethtool_stats = ibmveth_get_ethtool_stats,
985 .get_link_ksettings = netdev_get_link_ksettings,
988 static int ibmveth_ioctl(struct net_device *dev, struct ifreq *ifr, int cmd)
990 return -EOPNOTSUPP;
993 #define page_offset(v) ((unsigned long)(v) & ((1 << 12) - 1))
995 static int ibmveth_send(struct ibmveth_adapter *adapter,
996 union ibmveth_buf_desc *descs, unsigned long mss)
998 unsigned long correlator;
999 unsigned int retry_count;
1000 unsigned long ret;
1003 * The retry count sets a maximum for the number of broadcast and
1004 * multicast destinations within the system.
1006 retry_count = 1024;
1007 correlator = 0;
1008 do {
1009 ret = h_send_logical_lan(adapter->vdev->unit_address,
1010 descs[0].desc, descs[1].desc,
1011 descs[2].desc, descs[3].desc,
1012 descs[4].desc, descs[5].desc,
1013 correlator, &correlator, mss,
1014 adapter->fw_large_send_support);
1015 } while ((ret == H_BUSY) && (retry_count--));
1017 if (ret != H_SUCCESS && ret != H_DROPPED) {
1018 netdev_err(adapter->netdev, "tx: h_send_logical_lan failed "
1019 "with rc=%ld\n", ret);
1020 return 1;
1023 return 0;
1026 static netdev_tx_t ibmveth_start_xmit(struct sk_buff *skb,
1027 struct net_device *netdev)
1029 struct ibmveth_adapter *adapter = netdev_priv(netdev);
1030 unsigned int desc_flags;
1031 union ibmveth_buf_desc descs[6];
1032 int last, i;
1033 int force_bounce = 0;
1034 dma_addr_t dma_addr;
1035 unsigned long mss = 0;
1037 /* veth doesn't handle frag_list, so linearize the skb.
1038 * When GRO is enabled SKB's can have frag_list.
1040 if (adapter->is_active_trunk &&
1041 skb_has_frag_list(skb) && __skb_linearize(skb)) {
1042 netdev->stats.tx_dropped++;
1043 goto out;
1047 * veth handles a maximum of 6 segments including the header, so
1048 * we have to linearize the skb if there are more than this.
1050 if (skb_shinfo(skb)->nr_frags > 5 && __skb_linearize(skb)) {
1051 netdev->stats.tx_dropped++;
1052 goto out;
1055 /* veth can't checksum offload UDP */
1056 if (skb->ip_summed == CHECKSUM_PARTIAL &&
1057 ((skb->protocol == htons(ETH_P_IP) &&
1058 ip_hdr(skb)->protocol != IPPROTO_TCP) ||
1059 (skb->protocol == htons(ETH_P_IPV6) &&
1060 ipv6_hdr(skb)->nexthdr != IPPROTO_TCP)) &&
1061 skb_checksum_help(skb)) {
1063 netdev_err(netdev, "tx: failed to checksum packet\n");
1064 netdev->stats.tx_dropped++;
1065 goto out;
1068 desc_flags = IBMVETH_BUF_VALID;
1070 if (skb->ip_summed == CHECKSUM_PARTIAL) {
1071 unsigned char *buf = skb_transport_header(skb) +
1072 skb->csum_offset;
1074 desc_flags |= (IBMVETH_BUF_NO_CSUM | IBMVETH_BUF_CSUM_GOOD);
1076 /* Need to zero out the checksum */
1077 buf[0] = 0;
1078 buf[1] = 0;
1080 if (skb_is_gso(skb) && adapter->fw_large_send_support)
1081 desc_flags |= IBMVETH_BUF_LRG_SND;
1084 retry_bounce:
1085 memset(descs, 0, sizeof(descs));
1088 * If a linear packet is below the rx threshold then
1089 * copy it into the static bounce buffer. This avoids the
1090 * cost of a TCE insert and remove.
1092 if (force_bounce || (!skb_is_nonlinear(skb) &&
1093 (skb->len < tx_copybreak))) {
1094 skb_copy_from_linear_data(skb, adapter->bounce_buffer,
1095 skb->len);
1097 descs[0].fields.flags_len = desc_flags | skb->len;
1098 descs[0].fields.address = adapter->bounce_buffer_dma;
1100 if (ibmveth_send(adapter, descs, 0)) {
1101 adapter->tx_send_failed++;
1102 netdev->stats.tx_dropped++;
1103 } else {
1104 netdev->stats.tx_packets++;
1105 netdev->stats.tx_bytes += skb->len;
1108 goto out;
1111 /* Map the header */
1112 dma_addr = dma_map_single(&adapter->vdev->dev, skb->data,
1113 skb_headlen(skb), DMA_TO_DEVICE);
1114 if (dma_mapping_error(&adapter->vdev->dev, dma_addr))
1115 goto map_failed;
1117 descs[0].fields.flags_len = desc_flags | skb_headlen(skb);
1118 descs[0].fields.address = dma_addr;
1120 /* Map the frags */
1121 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
1122 const skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
1124 dma_addr = skb_frag_dma_map(&adapter->vdev->dev, frag, 0,
1125 skb_frag_size(frag), DMA_TO_DEVICE);
1127 if (dma_mapping_error(&adapter->vdev->dev, dma_addr))
1128 goto map_failed_frags;
1130 descs[i+1].fields.flags_len = desc_flags | skb_frag_size(frag);
1131 descs[i+1].fields.address = dma_addr;
1134 if (skb->ip_summed == CHECKSUM_PARTIAL && skb_is_gso(skb)) {
1135 if (adapter->fw_large_send_support) {
1136 mss = (unsigned long)skb_shinfo(skb)->gso_size;
1137 adapter->tx_large_packets++;
1138 } else if (!skb_is_gso_v6(skb)) {
1139 /* Put -1 in the IP checksum to tell phyp it
1140 * is a largesend packet. Put the mss in
1141 * the TCP checksum.
1143 ip_hdr(skb)->check = 0xffff;
1144 tcp_hdr(skb)->check =
1145 cpu_to_be16(skb_shinfo(skb)->gso_size);
1146 adapter->tx_large_packets++;
1150 if (ibmveth_send(adapter, descs, mss)) {
1151 adapter->tx_send_failed++;
1152 netdev->stats.tx_dropped++;
1153 } else {
1154 netdev->stats.tx_packets++;
1155 netdev->stats.tx_bytes += skb->len;
1158 dma_unmap_single(&adapter->vdev->dev,
1159 descs[0].fields.address,
1160 descs[0].fields.flags_len & IBMVETH_BUF_LEN_MASK,
1161 DMA_TO_DEVICE);
1163 for (i = 1; i < skb_shinfo(skb)->nr_frags + 1; i++)
1164 dma_unmap_page(&adapter->vdev->dev, descs[i].fields.address,
1165 descs[i].fields.flags_len & IBMVETH_BUF_LEN_MASK,
1166 DMA_TO_DEVICE);
1168 out:
1169 dev_consume_skb_any(skb);
1170 return NETDEV_TX_OK;
1172 map_failed_frags:
1173 last = i+1;
1174 for (i = 0; i < last; i++)
1175 dma_unmap_page(&adapter->vdev->dev, descs[i].fields.address,
1176 descs[i].fields.flags_len & IBMVETH_BUF_LEN_MASK,
1177 DMA_TO_DEVICE);
1179 map_failed:
1180 if (!firmware_has_feature(FW_FEATURE_CMO))
1181 netdev_err(netdev, "tx: unable to map xmit buffer\n");
1182 adapter->tx_map_failed++;
1183 if (skb_linearize(skb)) {
1184 netdev->stats.tx_dropped++;
1185 goto out;
1187 force_bounce = 1;
1188 goto retry_bounce;
1191 static void ibmveth_rx_mss_helper(struct sk_buff *skb, u16 mss, int lrg_pkt)
1193 struct tcphdr *tcph;
1194 int offset = 0;
1195 int hdr_len;
1197 /* only TCP packets will be aggregated */
1198 if (skb->protocol == htons(ETH_P_IP)) {
1199 struct iphdr *iph = (struct iphdr *)skb->data;
1201 if (iph->protocol == IPPROTO_TCP) {
1202 offset = iph->ihl * 4;
1203 skb_shinfo(skb)->gso_type = SKB_GSO_TCPV4;
1204 } else {
1205 return;
1207 } else if (skb->protocol == htons(ETH_P_IPV6)) {
1208 struct ipv6hdr *iph6 = (struct ipv6hdr *)skb->data;
1210 if (iph6->nexthdr == IPPROTO_TCP) {
1211 offset = sizeof(struct ipv6hdr);
1212 skb_shinfo(skb)->gso_type = SKB_GSO_TCPV6;
1213 } else {
1214 return;
1216 } else {
1217 return;
1219 /* if mss is not set through Large Packet bit/mss in rx buffer,
1220 * expect that the mss will be written to the tcp header checksum.
1222 tcph = (struct tcphdr *)(skb->data + offset);
1223 if (lrg_pkt) {
1224 skb_shinfo(skb)->gso_size = mss;
1225 } else if (offset) {
1226 skb_shinfo(skb)->gso_size = ntohs(tcph->check);
1227 tcph->check = 0;
1230 if (skb_shinfo(skb)->gso_size) {
1231 hdr_len = offset + tcph->doff * 4;
1232 skb_shinfo(skb)->gso_segs =
1233 DIV_ROUND_UP(skb->len - hdr_len,
1234 skb_shinfo(skb)->gso_size);
1238 static void ibmveth_rx_csum_helper(struct sk_buff *skb,
1239 struct ibmveth_adapter *adapter)
1241 struct iphdr *iph = NULL;
1242 struct ipv6hdr *iph6 = NULL;
1243 __be16 skb_proto = 0;
1244 u16 iphlen = 0;
1245 u16 iph_proto = 0;
1246 u16 tcphdrlen = 0;
1248 skb_proto = be16_to_cpu(skb->protocol);
1250 if (skb_proto == ETH_P_IP) {
1251 iph = (struct iphdr *)skb->data;
1253 /* If the IP checksum is not offloaded and if the packet
1254 * is large send, the checksum must be rebuilt.
1256 if (iph->check == 0xffff) {
1257 iph->check = 0;
1258 iph->check = ip_fast_csum((unsigned char *)iph,
1259 iph->ihl);
1262 iphlen = iph->ihl * 4;
1263 iph_proto = iph->protocol;
1264 } else if (skb_proto == ETH_P_IPV6) {
1265 iph6 = (struct ipv6hdr *)skb->data;
1266 iphlen = sizeof(struct ipv6hdr);
1267 iph_proto = iph6->nexthdr;
1270 /* In OVS environment, when a flow is not cached, specifically for a
1271 * new TCP connection, the first packet information is passed up
1272 * the user space for finding a flow. During this process, OVS computes
1273 * checksum on the first packet when CHECKSUM_PARTIAL flag is set.
1275 * Given that we zeroed out TCP checksum field in transmit path
1276 * (refer ibmveth_start_xmit routine) as we set "no checksum bit",
1277 * OVS computed checksum will be incorrect w/o TCP pseudo checksum
1278 * in the packet. This leads to OVS dropping the packet and hence
1279 * TCP retransmissions are seen.
1281 * So, re-compute TCP pseudo header checksum.
1283 if (iph_proto == IPPROTO_TCP && adapter->is_active_trunk) {
1284 struct tcphdr *tcph = (struct tcphdr *)(skb->data + iphlen);
1286 tcphdrlen = skb->len - iphlen;
1288 /* Recompute TCP pseudo header checksum */
1289 if (skb_proto == ETH_P_IP)
1290 tcph->check = ~csum_tcpudp_magic(iph->saddr,
1291 iph->daddr, tcphdrlen, iph_proto, 0);
1292 else if (skb_proto == ETH_P_IPV6)
1293 tcph->check = ~csum_ipv6_magic(&iph6->saddr,
1294 &iph6->daddr, tcphdrlen, iph_proto, 0);
1296 /* Setup SKB fields for checksum offload */
1297 skb_partial_csum_set(skb, iphlen,
1298 offsetof(struct tcphdr, check));
1299 skb_reset_network_header(skb);
1303 static int ibmveth_poll(struct napi_struct *napi, int budget)
1305 struct ibmveth_adapter *adapter =
1306 container_of(napi, struct ibmveth_adapter, napi);
1307 struct net_device *netdev = adapter->netdev;
1308 int frames_processed = 0;
1309 unsigned long lpar_rc;
1310 u16 mss = 0;
1312 restart_poll:
1313 while (frames_processed < budget) {
1314 if (!ibmveth_rxq_pending_buffer(adapter))
1315 break;
1317 smp_rmb();
1318 if (!ibmveth_rxq_buffer_valid(adapter)) {
1319 wmb(); /* suggested by larson1 */
1320 adapter->rx_invalid_buffer++;
1321 netdev_dbg(netdev, "recycling invalid buffer\n");
1322 ibmveth_rxq_recycle_buffer(adapter);
1323 } else {
1324 struct sk_buff *skb, *new_skb;
1325 int length = ibmveth_rxq_frame_length(adapter);
1326 int offset = ibmveth_rxq_frame_offset(adapter);
1327 int csum_good = ibmveth_rxq_csum_good(adapter);
1328 int lrg_pkt = ibmveth_rxq_large_packet(adapter);
1330 skb = ibmveth_rxq_get_buffer(adapter);
1332 /* if the large packet bit is set in the rx queue
1333 * descriptor, the mss will be written by PHYP eight
1334 * bytes from the start of the rx buffer, which is
1335 * skb->data at this stage
1337 if (lrg_pkt) {
1338 __be64 *rxmss = (__be64 *)(skb->data + 8);
1340 mss = (u16)be64_to_cpu(*rxmss);
1343 new_skb = NULL;
1344 if (length < rx_copybreak)
1345 new_skb = netdev_alloc_skb(netdev, length);
1347 if (new_skb) {
1348 skb_copy_to_linear_data(new_skb,
1349 skb->data + offset,
1350 length);
1351 if (rx_flush)
1352 ibmveth_flush_buffer(skb->data,
1353 length + offset);
1354 if (!ibmveth_rxq_recycle_buffer(adapter))
1355 kfree_skb(skb);
1356 skb = new_skb;
1357 } else {
1358 ibmveth_rxq_harvest_buffer(adapter);
1359 skb_reserve(skb, offset);
1362 skb_put(skb, length);
1363 skb->protocol = eth_type_trans(skb, netdev);
1365 if (csum_good) {
1366 skb->ip_summed = CHECKSUM_UNNECESSARY;
1367 ibmveth_rx_csum_helper(skb, adapter);
1370 if (length > netdev->mtu + ETH_HLEN) {
1371 ibmveth_rx_mss_helper(skb, mss, lrg_pkt);
1372 adapter->rx_large_packets++;
1375 napi_gro_receive(napi, skb); /* send it up */
1377 netdev->stats.rx_packets++;
1378 netdev->stats.rx_bytes += length;
1379 frames_processed++;
1383 ibmveth_replenish_task(adapter);
1385 if (frames_processed < budget) {
1386 napi_complete_done(napi, frames_processed);
1388 /* We think we are done - reenable interrupts,
1389 * then check once more to make sure we are done.
1391 lpar_rc = h_vio_signal(adapter->vdev->unit_address,
1392 VIO_IRQ_ENABLE);
1394 BUG_ON(lpar_rc != H_SUCCESS);
1396 if (ibmveth_rxq_pending_buffer(adapter) &&
1397 napi_reschedule(napi)) {
1398 lpar_rc = h_vio_signal(adapter->vdev->unit_address,
1399 VIO_IRQ_DISABLE);
1400 goto restart_poll;
1404 return frames_processed;
1407 static irqreturn_t ibmveth_interrupt(int irq, void *dev_instance)
1409 struct net_device *netdev = dev_instance;
1410 struct ibmveth_adapter *adapter = netdev_priv(netdev);
1411 unsigned long lpar_rc;
1413 if (napi_schedule_prep(&adapter->napi)) {
1414 lpar_rc = h_vio_signal(adapter->vdev->unit_address,
1415 VIO_IRQ_DISABLE);
1416 BUG_ON(lpar_rc != H_SUCCESS);
1417 __napi_schedule(&adapter->napi);
1419 return IRQ_HANDLED;
1422 static void ibmveth_set_multicast_list(struct net_device *netdev)
1424 struct ibmveth_adapter *adapter = netdev_priv(netdev);
1425 unsigned long lpar_rc;
1427 if ((netdev->flags & IFF_PROMISC) ||
1428 (netdev_mc_count(netdev) > adapter->mcastFilterSize)) {
1429 lpar_rc = h_multicast_ctrl(adapter->vdev->unit_address,
1430 IbmVethMcastEnableRecv |
1431 IbmVethMcastDisableFiltering,
1433 if (lpar_rc != H_SUCCESS) {
1434 netdev_err(netdev, "h_multicast_ctrl rc=%ld when "
1435 "entering promisc mode\n", lpar_rc);
1437 } else {
1438 struct netdev_hw_addr *ha;
1439 /* clear the filter table & disable filtering */
1440 lpar_rc = h_multicast_ctrl(adapter->vdev->unit_address,
1441 IbmVethMcastEnableRecv |
1442 IbmVethMcastDisableFiltering |
1443 IbmVethMcastClearFilterTable,
1445 if (lpar_rc != H_SUCCESS) {
1446 netdev_err(netdev, "h_multicast_ctrl rc=%ld when "
1447 "attempting to clear filter table\n",
1448 lpar_rc);
1450 /* add the addresses to the filter table */
1451 netdev_for_each_mc_addr(ha, netdev) {
1452 /* add the multicast address to the filter table */
1453 u64 mcast_addr;
1454 mcast_addr = ibmveth_encode_mac_addr(ha->addr);
1455 lpar_rc = h_multicast_ctrl(adapter->vdev->unit_address,
1456 IbmVethMcastAddFilter,
1457 mcast_addr);
1458 if (lpar_rc != H_SUCCESS) {
1459 netdev_err(netdev, "h_multicast_ctrl rc=%ld "
1460 "when adding an entry to the filter "
1461 "table\n", lpar_rc);
1465 /* re-enable filtering */
1466 lpar_rc = h_multicast_ctrl(adapter->vdev->unit_address,
1467 IbmVethMcastEnableFiltering,
1469 if (lpar_rc != H_SUCCESS) {
1470 netdev_err(netdev, "h_multicast_ctrl rc=%ld when "
1471 "enabling filtering\n", lpar_rc);
1476 static int ibmveth_change_mtu(struct net_device *dev, int new_mtu)
1478 struct ibmveth_adapter *adapter = netdev_priv(dev);
1479 struct vio_dev *viodev = adapter->vdev;
1480 int new_mtu_oh = new_mtu + IBMVETH_BUFF_OH;
1481 int i, rc;
1482 int need_restart = 0;
1484 for (i = 0; i < IBMVETH_NUM_BUFF_POOLS; i++)
1485 if (new_mtu_oh <= adapter->rx_buff_pool[i].buff_size)
1486 break;
1488 if (i == IBMVETH_NUM_BUFF_POOLS)
1489 return -EINVAL;
1491 /* Deactivate all the buffer pools so that the next loop can activate
1492 only the buffer pools necessary to hold the new MTU */
1493 if (netif_running(adapter->netdev)) {
1494 need_restart = 1;
1495 adapter->pool_config = 1;
1496 ibmveth_close(adapter->netdev);
1497 adapter->pool_config = 0;
1500 /* Look for an active buffer pool that can hold the new MTU */
1501 for (i = 0; i < IBMVETH_NUM_BUFF_POOLS; i++) {
1502 adapter->rx_buff_pool[i].active = 1;
1504 if (new_mtu_oh <= adapter->rx_buff_pool[i].buff_size) {
1505 dev->mtu = new_mtu;
1506 vio_cmo_set_dev_desired(viodev,
1507 ibmveth_get_desired_dma
1508 (viodev));
1509 if (need_restart) {
1510 return ibmveth_open(adapter->netdev);
1512 return 0;
1516 if (need_restart && (rc = ibmveth_open(adapter->netdev)))
1517 return rc;
1519 return -EINVAL;
1522 #ifdef CONFIG_NET_POLL_CONTROLLER
1523 static void ibmveth_poll_controller(struct net_device *dev)
1525 ibmveth_replenish_task(netdev_priv(dev));
1526 ibmveth_interrupt(dev->irq, dev);
1528 #endif
1531 * ibmveth_get_desired_dma - Calculate IO memory desired by the driver
1533 * @vdev: struct vio_dev for the device whose desired IO mem is to be returned
1535 * Return value:
1536 * Number of bytes of IO data the driver will need to perform well.
1538 static unsigned long ibmveth_get_desired_dma(struct vio_dev *vdev)
1540 struct net_device *netdev = dev_get_drvdata(&vdev->dev);
1541 struct ibmveth_adapter *adapter;
1542 struct iommu_table *tbl;
1543 unsigned long ret;
1544 int i;
1545 int rxqentries = 1;
1547 tbl = get_iommu_table_base(&vdev->dev);
1549 /* netdev inits at probe time along with the structures we need below*/
1550 if (netdev == NULL)
1551 return IOMMU_PAGE_ALIGN(IBMVETH_IO_ENTITLEMENT_DEFAULT, tbl);
1553 adapter = netdev_priv(netdev);
1555 ret = IBMVETH_BUFF_LIST_SIZE + IBMVETH_FILT_LIST_SIZE;
1556 ret += IOMMU_PAGE_ALIGN(netdev->mtu, tbl);
1558 for (i = 0; i < IBMVETH_NUM_BUFF_POOLS; i++) {
1559 /* add the size of the active receive buffers */
1560 if (adapter->rx_buff_pool[i].active)
1561 ret +=
1562 adapter->rx_buff_pool[i].size *
1563 IOMMU_PAGE_ALIGN(adapter->rx_buff_pool[i].
1564 buff_size, tbl);
1565 rxqentries += adapter->rx_buff_pool[i].size;
1567 /* add the size of the receive queue entries */
1568 ret += IOMMU_PAGE_ALIGN(
1569 rxqentries * sizeof(struct ibmveth_rx_q_entry), tbl);
1571 return ret;
1574 static int ibmveth_set_mac_addr(struct net_device *dev, void *p)
1576 struct ibmveth_adapter *adapter = netdev_priv(dev);
1577 struct sockaddr *addr = p;
1578 u64 mac_address;
1579 int rc;
1581 if (!is_valid_ether_addr(addr->sa_data))
1582 return -EADDRNOTAVAIL;
1584 mac_address = ibmveth_encode_mac_addr(addr->sa_data);
1585 rc = h_change_logical_lan_mac(adapter->vdev->unit_address, mac_address);
1586 if (rc) {
1587 netdev_err(adapter->netdev, "h_change_logical_lan_mac failed with rc=%d\n", rc);
1588 return rc;
1591 ether_addr_copy(dev->dev_addr, addr->sa_data);
1593 return 0;
1596 static const struct net_device_ops ibmveth_netdev_ops = {
1597 .ndo_open = ibmveth_open,
1598 .ndo_stop = ibmveth_close,
1599 .ndo_start_xmit = ibmveth_start_xmit,
1600 .ndo_set_rx_mode = ibmveth_set_multicast_list,
1601 .ndo_do_ioctl = ibmveth_ioctl,
1602 .ndo_change_mtu = ibmveth_change_mtu,
1603 .ndo_fix_features = ibmveth_fix_features,
1604 .ndo_set_features = ibmveth_set_features,
1605 .ndo_validate_addr = eth_validate_addr,
1606 .ndo_set_mac_address = ibmveth_set_mac_addr,
1607 #ifdef CONFIG_NET_POLL_CONTROLLER
1608 .ndo_poll_controller = ibmveth_poll_controller,
1609 #endif
1612 static int ibmveth_probe(struct vio_dev *dev, const struct vio_device_id *id)
1614 int rc, i, mac_len;
1615 struct net_device *netdev;
1616 struct ibmveth_adapter *adapter;
1617 unsigned char *mac_addr_p;
1618 unsigned int *mcastFilterSize_p;
1619 long ret;
1620 unsigned long ret_attr;
1622 dev_dbg(&dev->dev, "entering ibmveth_probe for UA 0x%x\n",
1623 dev->unit_address);
1625 mac_addr_p = (unsigned char *)vio_get_attribute(dev, VETH_MAC_ADDR,
1626 &mac_len);
1627 if (!mac_addr_p) {
1628 dev_err(&dev->dev, "Can't find VETH_MAC_ADDR attribute\n");
1629 return -EINVAL;
1631 /* Workaround for old/broken pHyp */
1632 if (mac_len == 8)
1633 mac_addr_p += 2;
1634 else if (mac_len != 6) {
1635 dev_err(&dev->dev, "VETH_MAC_ADDR attribute wrong len %d\n",
1636 mac_len);
1637 return -EINVAL;
1640 mcastFilterSize_p = (unsigned int *)vio_get_attribute(dev,
1641 VETH_MCAST_FILTER_SIZE, NULL);
1642 if (!mcastFilterSize_p) {
1643 dev_err(&dev->dev, "Can't find VETH_MCAST_FILTER_SIZE "
1644 "attribute\n");
1645 return -EINVAL;
1648 netdev = alloc_etherdev(sizeof(struct ibmveth_adapter));
1650 if (!netdev)
1651 return -ENOMEM;
1653 adapter = netdev_priv(netdev);
1654 dev_set_drvdata(&dev->dev, netdev);
1656 adapter->vdev = dev;
1657 adapter->netdev = netdev;
1658 adapter->mcastFilterSize = *mcastFilterSize_p;
1659 adapter->pool_config = 0;
1661 netif_napi_add(netdev, &adapter->napi, ibmveth_poll, 16);
1663 netdev->irq = dev->irq;
1664 netdev->netdev_ops = &ibmveth_netdev_ops;
1665 netdev->ethtool_ops = &netdev_ethtool_ops;
1666 SET_NETDEV_DEV(netdev, &dev->dev);
1667 netdev->hw_features = NETIF_F_SG;
1668 if (vio_get_attribute(dev, "ibm,illan-options", NULL) != NULL) {
1669 netdev->hw_features |= NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM |
1670 NETIF_F_RXCSUM;
1673 netdev->features |= netdev->hw_features;
1675 ret = h_illan_attributes(adapter->vdev->unit_address, 0, 0, &ret_attr);
1677 /* If running older firmware, TSO should not be enabled by default */
1678 if (ret == H_SUCCESS && (ret_attr & IBMVETH_ILLAN_LRG_SND_SUPPORT) &&
1679 !old_large_send) {
1680 netdev->hw_features |= NETIF_F_TSO | NETIF_F_TSO6;
1681 netdev->features |= netdev->hw_features;
1682 } else {
1683 netdev->hw_features |= NETIF_F_TSO;
1686 adapter->is_active_trunk = false;
1687 if (ret == H_SUCCESS && (ret_attr & IBMVETH_ILLAN_ACTIVE_TRUNK)) {
1688 adapter->is_active_trunk = true;
1689 netdev->hw_features |= NETIF_F_FRAGLIST;
1690 netdev->features |= NETIF_F_FRAGLIST;
1693 netdev->min_mtu = IBMVETH_MIN_MTU;
1694 netdev->max_mtu = ETH_MAX_MTU;
1696 memcpy(netdev->dev_addr, mac_addr_p, ETH_ALEN);
1698 if (firmware_has_feature(FW_FEATURE_CMO))
1699 memcpy(pool_count, pool_count_cmo, sizeof(pool_count));
1701 for (i = 0; i < IBMVETH_NUM_BUFF_POOLS; i++) {
1702 struct kobject *kobj = &adapter->rx_buff_pool[i].kobj;
1703 int error;
1705 ibmveth_init_buffer_pool(&adapter->rx_buff_pool[i], i,
1706 pool_count[i], pool_size[i],
1707 pool_active[i]);
1708 error = kobject_init_and_add(kobj, &ktype_veth_pool,
1709 &dev->dev.kobj, "pool%d", i);
1710 if (!error)
1711 kobject_uevent(kobj, KOBJ_ADD);
1714 netdev_dbg(netdev, "adapter @ 0x%p\n", adapter);
1715 netdev_dbg(netdev, "registering netdev...\n");
1717 ibmveth_set_features(netdev, netdev->features);
1719 rc = register_netdev(netdev);
1721 if (rc) {
1722 netdev_dbg(netdev, "failed to register netdev rc=%d\n", rc);
1723 free_netdev(netdev);
1724 return rc;
1727 netdev_dbg(netdev, "registered\n");
1729 return 0;
1732 static int ibmveth_remove(struct vio_dev *dev)
1734 struct net_device *netdev = dev_get_drvdata(&dev->dev);
1735 struct ibmveth_adapter *adapter = netdev_priv(netdev);
1736 int i;
1738 for (i = 0; i < IBMVETH_NUM_BUFF_POOLS; i++)
1739 kobject_put(&adapter->rx_buff_pool[i].kobj);
1741 unregister_netdev(netdev);
1743 free_netdev(netdev);
1744 dev_set_drvdata(&dev->dev, NULL);
1746 return 0;
1749 static struct attribute veth_active_attr;
1750 static struct attribute veth_num_attr;
1751 static struct attribute veth_size_attr;
1753 static ssize_t veth_pool_show(struct kobject *kobj,
1754 struct attribute *attr, char *buf)
1756 struct ibmveth_buff_pool *pool = container_of(kobj,
1757 struct ibmveth_buff_pool,
1758 kobj);
1760 if (attr == &veth_active_attr)
1761 return sprintf(buf, "%d\n", pool->active);
1762 else if (attr == &veth_num_attr)
1763 return sprintf(buf, "%d\n", pool->size);
1764 else if (attr == &veth_size_attr)
1765 return sprintf(buf, "%d\n", pool->buff_size);
1766 return 0;
1769 static ssize_t veth_pool_store(struct kobject *kobj, struct attribute *attr,
1770 const char *buf, size_t count)
1772 struct ibmveth_buff_pool *pool = container_of(kobj,
1773 struct ibmveth_buff_pool,
1774 kobj);
1775 struct net_device *netdev = dev_get_drvdata(
1776 container_of(kobj->parent, struct device, kobj));
1777 struct ibmveth_adapter *adapter = netdev_priv(netdev);
1778 long value = simple_strtol(buf, NULL, 10);
1779 long rc;
1781 if (attr == &veth_active_attr) {
1782 if (value && !pool->active) {
1783 if (netif_running(netdev)) {
1784 if (ibmveth_alloc_buffer_pool(pool)) {
1785 netdev_err(netdev,
1786 "unable to alloc pool\n");
1787 return -ENOMEM;
1789 pool->active = 1;
1790 adapter->pool_config = 1;
1791 ibmveth_close(netdev);
1792 adapter->pool_config = 0;
1793 if ((rc = ibmveth_open(netdev)))
1794 return rc;
1795 } else {
1796 pool->active = 1;
1798 } else if (!value && pool->active) {
1799 int mtu = netdev->mtu + IBMVETH_BUFF_OH;
1800 int i;
1801 /* Make sure there is a buffer pool with buffers that
1802 can hold a packet of the size of the MTU */
1803 for (i = 0; i < IBMVETH_NUM_BUFF_POOLS; i++) {
1804 if (pool == &adapter->rx_buff_pool[i])
1805 continue;
1806 if (!adapter->rx_buff_pool[i].active)
1807 continue;
1808 if (mtu <= adapter->rx_buff_pool[i].buff_size)
1809 break;
1812 if (i == IBMVETH_NUM_BUFF_POOLS) {
1813 netdev_err(netdev, "no active pool >= MTU\n");
1814 return -EPERM;
1817 if (netif_running(netdev)) {
1818 adapter->pool_config = 1;
1819 ibmveth_close(netdev);
1820 pool->active = 0;
1821 adapter->pool_config = 0;
1822 if ((rc = ibmveth_open(netdev)))
1823 return rc;
1825 pool->active = 0;
1827 } else if (attr == &veth_num_attr) {
1828 if (value <= 0 || value > IBMVETH_MAX_POOL_COUNT) {
1829 return -EINVAL;
1830 } else {
1831 if (netif_running(netdev)) {
1832 adapter->pool_config = 1;
1833 ibmveth_close(netdev);
1834 adapter->pool_config = 0;
1835 pool->size = value;
1836 if ((rc = ibmveth_open(netdev)))
1837 return rc;
1838 } else {
1839 pool->size = value;
1842 } else if (attr == &veth_size_attr) {
1843 if (value <= IBMVETH_BUFF_OH || value > IBMVETH_MAX_BUF_SIZE) {
1844 return -EINVAL;
1845 } else {
1846 if (netif_running(netdev)) {
1847 adapter->pool_config = 1;
1848 ibmveth_close(netdev);
1849 adapter->pool_config = 0;
1850 pool->buff_size = value;
1851 if ((rc = ibmveth_open(netdev)))
1852 return rc;
1853 } else {
1854 pool->buff_size = value;
1859 /* kick the interrupt handler to allocate/deallocate pools */
1860 ibmveth_interrupt(netdev->irq, netdev);
1861 return count;
1865 #define ATTR(_name, _mode) \
1866 struct attribute veth_##_name##_attr = { \
1867 .name = __stringify(_name), .mode = _mode, \
1870 static ATTR(active, 0644);
1871 static ATTR(num, 0644);
1872 static ATTR(size, 0644);
1874 static struct attribute *veth_pool_attrs[] = {
1875 &veth_active_attr,
1876 &veth_num_attr,
1877 &veth_size_attr,
1878 NULL,
1881 static const struct sysfs_ops veth_pool_ops = {
1882 .show = veth_pool_show,
1883 .store = veth_pool_store,
1886 static struct kobj_type ktype_veth_pool = {
1887 .release = NULL,
1888 .sysfs_ops = &veth_pool_ops,
1889 .default_attrs = veth_pool_attrs,
1892 static int ibmveth_resume(struct device *dev)
1894 struct net_device *netdev = dev_get_drvdata(dev);
1895 ibmveth_interrupt(netdev->irq, netdev);
1896 return 0;
1899 static const struct vio_device_id ibmveth_device_table[] = {
1900 { "network", "IBM,l-lan"},
1901 { "", "" }
1903 MODULE_DEVICE_TABLE(vio, ibmveth_device_table);
1905 static const struct dev_pm_ops ibmveth_pm_ops = {
1906 .resume = ibmveth_resume
1909 static struct vio_driver ibmveth_driver = {
1910 .id_table = ibmveth_device_table,
1911 .probe = ibmveth_probe,
1912 .remove = ibmveth_remove,
1913 .get_desired_dma = ibmveth_get_desired_dma,
1914 .name = ibmveth_driver_name,
1915 .pm = &ibmveth_pm_ops,
1918 static int __init ibmveth_module_init(void)
1920 printk(KERN_DEBUG "%s: %s %s\n", ibmveth_driver_name,
1921 ibmveth_driver_string, ibmveth_driver_version);
1923 return vio_register_driver(&ibmveth_driver);
1926 static void __exit ibmveth_module_exit(void)
1928 vio_unregister_driver(&ibmveth_driver);
1931 module_init(ibmveth_module_init);
1932 module_exit(ibmveth_module_exit);