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>
39 #include <linux/ethtool.h>
42 #include <linux/ipv6.h>
43 #include <linux/slab.h>
44 #include <asm/hvcall.h>
45 #include <linux/atomic.h>
47 #include <asm/iommu.h>
48 #include <asm/firmware.h>
52 static irqreturn_t ibmveth_interrupt(int irq, void *dev_instance);
53 static void ibmveth_rxq_harvest_buffer(struct ibmveth_adapter *adapter);
54 static unsigned long ibmveth_get_desired_dma(struct vio_dev *vdev);
56 static struct kobj_type ktype_veth_pool;
59 static const char ibmveth_driver_name[] = "ibmveth";
60 static const char ibmveth_driver_string[] = "IBM Power Virtual Ethernet Driver";
61 #define ibmveth_driver_version "1.06"
63 MODULE_AUTHOR("Santiago Leon <santil@linux.vnet.ibm.com>");
64 MODULE_DESCRIPTION("IBM Power Virtual Ethernet Driver");
65 MODULE_LICENSE("GPL");
66 MODULE_VERSION(ibmveth_driver_version);
68 static unsigned int tx_copybreak __read_mostly = 128;
69 module_param(tx_copybreak, uint, 0644);
70 MODULE_PARM_DESC(tx_copybreak,
71 "Maximum size of packet that is copied to a new buffer on transmit");
73 static unsigned int rx_copybreak __read_mostly = 128;
74 module_param(rx_copybreak, uint, 0644);
75 MODULE_PARM_DESC(rx_copybreak,
76 "Maximum size of packet that is copied to a new buffer on receive");
78 static unsigned int rx_flush __read_mostly = 0;
79 module_param(rx_flush, uint, 0644);
80 MODULE_PARM_DESC(rx_flush, "Flush receive buffers before use");
82 static bool old_large_send __read_mostly;
83 module_param(old_large_send, bool, S_IRUGO);
84 MODULE_PARM_DESC(old_large_send,
85 "Use old large send method on firmware that supports the new method");
88 char name[ETH_GSTRING_LEN];
92 #define IBMVETH_STAT_OFF(stat) offsetof(struct ibmveth_adapter, stat)
93 #define IBMVETH_GET_STAT(a, off) *((u64 *)(((unsigned long)(a)) + off))
95 struct ibmveth_stat ibmveth_stats[] = {
96 { "replenish_task_cycles", IBMVETH_STAT_OFF(replenish_task_cycles) },
97 { "replenish_no_mem", IBMVETH_STAT_OFF(replenish_no_mem) },
98 { "replenish_add_buff_failure",
99 IBMVETH_STAT_OFF(replenish_add_buff_failure) },
100 { "replenish_add_buff_success",
101 IBMVETH_STAT_OFF(replenish_add_buff_success) },
102 { "rx_invalid_buffer", IBMVETH_STAT_OFF(rx_invalid_buffer) },
103 { "rx_no_buffer", IBMVETH_STAT_OFF(rx_no_buffer) },
104 { "tx_map_failed", IBMVETH_STAT_OFF(tx_map_failed) },
105 { "tx_send_failed", IBMVETH_STAT_OFF(tx_send_failed) },
106 { "fw_enabled_ipv4_csum", IBMVETH_STAT_OFF(fw_ipv4_csum_support) },
107 { "fw_enabled_ipv6_csum", IBMVETH_STAT_OFF(fw_ipv6_csum_support) },
108 { "tx_large_packets", IBMVETH_STAT_OFF(tx_large_packets) },
109 { "rx_large_packets", IBMVETH_STAT_OFF(rx_large_packets) },
110 { "fw_enabled_large_send", IBMVETH_STAT_OFF(fw_large_send_support) }
113 /* simple methods of getting data from the current rxq entry */
114 static inline u32 ibmveth_rxq_flags(struct ibmveth_adapter *adapter)
116 return be32_to_cpu(adapter->rx_queue.queue_addr[adapter->rx_queue.index].flags_off);
119 static inline int ibmveth_rxq_toggle(struct ibmveth_adapter *adapter)
121 return (ibmveth_rxq_flags(adapter) & IBMVETH_RXQ_TOGGLE) >>
122 IBMVETH_RXQ_TOGGLE_SHIFT;
125 static inline int ibmveth_rxq_pending_buffer(struct ibmveth_adapter *adapter)
127 return ibmveth_rxq_toggle(adapter) == adapter->rx_queue.toggle;
130 static inline int ibmveth_rxq_buffer_valid(struct ibmveth_adapter *adapter)
132 return ibmveth_rxq_flags(adapter) & IBMVETH_RXQ_VALID;
135 static inline int ibmveth_rxq_frame_offset(struct ibmveth_adapter *adapter)
137 return ibmveth_rxq_flags(adapter) & IBMVETH_RXQ_OFF_MASK;
140 static inline int ibmveth_rxq_large_packet(struct ibmveth_adapter *adapter)
142 return ibmveth_rxq_flags(adapter) & IBMVETH_RXQ_LRG_PKT;
145 static inline int ibmveth_rxq_frame_length(struct ibmveth_adapter *adapter)
147 return be32_to_cpu(adapter->rx_queue.queue_addr[adapter->rx_queue.index].length);
150 static inline int ibmveth_rxq_csum_good(struct ibmveth_adapter *adapter)
152 return ibmveth_rxq_flags(adapter) & IBMVETH_RXQ_CSUM_GOOD;
155 /* setup the initial settings for a buffer pool */
156 static void ibmveth_init_buffer_pool(struct ibmveth_buff_pool *pool,
157 u32 pool_index, u32 pool_size,
158 u32 buff_size, u32 pool_active)
160 pool->size = pool_size;
161 pool->index = pool_index;
162 pool->buff_size = buff_size;
163 pool->threshold = pool_size * 7 / 8;
164 pool->active = pool_active;
167 /* allocate and setup an buffer pool - called during open */
168 static int ibmveth_alloc_buffer_pool(struct ibmveth_buff_pool *pool)
172 pool->free_map = kmalloc(sizeof(u16) * pool->size, GFP_KERNEL);
177 pool->dma_addr = kcalloc(pool->size, sizeof(dma_addr_t), GFP_KERNEL);
178 if (!pool->dma_addr) {
179 kfree(pool->free_map);
180 pool->free_map = NULL;
184 pool->skbuff = kcalloc(pool->size, sizeof(void *), GFP_KERNEL);
187 kfree(pool->dma_addr);
188 pool->dma_addr = NULL;
190 kfree(pool->free_map);
191 pool->free_map = NULL;
195 for (i = 0; i < pool->size; ++i)
196 pool->free_map[i] = i;
198 atomic_set(&pool->available, 0);
199 pool->producer_index = 0;
200 pool->consumer_index = 0;
205 static inline void ibmveth_flush_buffer(void *addr, unsigned long length)
207 unsigned long offset;
209 for (offset = 0; offset < length; offset += SMP_CACHE_BYTES)
210 asm("dcbfl %0,%1" :: "b" (addr), "r" (offset));
213 /* replenish the buffers for a pool. note that we don't need to
214 * skb_reserve these since they are used for incoming...
216 static void ibmveth_replenish_buffer_pool(struct ibmveth_adapter *adapter,
217 struct ibmveth_buff_pool *pool)
220 u32 count = pool->size - atomic_read(&pool->available);
221 u32 buffers_added = 0;
223 unsigned int free_index, index;
225 unsigned long lpar_rc;
230 for (i = 0; i < count; ++i) {
231 union ibmveth_buf_desc desc;
233 skb = netdev_alloc_skb(adapter->netdev, pool->buff_size);
236 netdev_dbg(adapter->netdev,
237 "replenish: unable to allocate skb\n");
238 adapter->replenish_no_mem++;
242 free_index = pool->consumer_index;
243 pool->consumer_index++;
244 if (pool->consumer_index >= pool->size)
245 pool->consumer_index = 0;
246 index = pool->free_map[free_index];
248 BUG_ON(index == IBM_VETH_INVALID_MAP);
249 BUG_ON(pool->skbuff[index] != NULL);
251 dma_addr = dma_map_single(&adapter->vdev->dev, skb->data,
252 pool->buff_size, DMA_FROM_DEVICE);
254 if (dma_mapping_error(&adapter->vdev->dev, dma_addr))
257 pool->free_map[free_index] = IBM_VETH_INVALID_MAP;
258 pool->dma_addr[index] = dma_addr;
259 pool->skbuff[index] = skb;
261 correlator = ((u64)pool->index << 32) | index;
262 *(u64 *)skb->data = correlator;
264 desc.fields.flags_len = IBMVETH_BUF_VALID | pool->buff_size;
265 desc.fields.address = dma_addr;
268 unsigned int len = min(pool->buff_size,
269 adapter->netdev->mtu +
271 ibmveth_flush_buffer(skb->data, len);
273 lpar_rc = h_add_logical_lan_buffer(adapter->vdev->unit_address,
276 if (lpar_rc != H_SUCCESS) {
280 adapter->replenish_add_buff_success++;
285 atomic_add(buffers_added, &(pool->available));
289 pool->free_map[free_index] = index;
290 pool->skbuff[index] = NULL;
291 if (pool->consumer_index == 0)
292 pool->consumer_index = pool->size - 1;
294 pool->consumer_index--;
295 if (!dma_mapping_error(&adapter->vdev->dev, dma_addr))
296 dma_unmap_single(&adapter->vdev->dev,
297 pool->dma_addr[index], pool->buff_size,
299 dev_kfree_skb_any(skb);
300 adapter->replenish_add_buff_failure++;
303 atomic_add(buffers_added, &(pool->available));
307 * The final 8 bytes of the buffer list is a counter of frames dropped
308 * because there was not a buffer in the buffer list capable of holding
311 static void ibmveth_update_rx_no_buffer(struct ibmveth_adapter *adapter)
313 __be64 *p = adapter->buffer_list_addr + 4096 - 8;
315 adapter->rx_no_buffer = be64_to_cpup(p);
318 /* replenish routine */
319 static void ibmveth_replenish_task(struct ibmveth_adapter *adapter)
323 adapter->replenish_task_cycles++;
325 for (i = (IBMVETH_NUM_BUFF_POOLS - 1); i >= 0; i--) {
326 struct ibmveth_buff_pool *pool = &adapter->rx_buff_pool[i];
329 (atomic_read(&pool->available) < pool->threshold))
330 ibmveth_replenish_buffer_pool(adapter, pool);
333 ibmveth_update_rx_no_buffer(adapter);
336 /* empty and free ana buffer pool - also used to do cleanup in error paths */
337 static void ibmveth_free_buffer_pool(struct ibmveth_adapter *adapter,
338 struct ibmveth_buff_pool *pool)
342 kfree(pool->free_map);
343 pool->free_map = NULL;
345 if (pool->skbuff && pool->dma_addr) {
346 for (i = 0; i < pool->size; ++i) {
347 struct sk_buff *skb = pool->skbuff[i];
349 dma_unmap_single(&adapter->vdev->dev,
353 dev_kfree_skb_any(skb);
354 pool->skbuff[i] = NULL;
359 if (pool->dma_addr) {
360 kfree(pool->dma_addr);
361 pool->dma_addr = NULL;
370 /* remove a buffer from a pool */
371 static void ibmveth_remove_buffer_from_pool(struct ibmveth_adapter *adapter,
374 unsigned int pool = correlator >> 32;
375 unsigned int index = correlator & 0xffffffffUL;
376 unsigned int free_index;
379 BUG_ON(pool >= IBMVETH_NUM_BUFF_POOLS);
380 BUG_ON(index >= adapter->rx_buff_pool[pool].size);
382 skb = adapter->rx_buff_pool[pool].skbuff[index];
386 adapter->rx_buff_pool[pool].skbuff[index] = NULL;
388 dma_unmap_single(&adapter->vdev->dev,
389 adapter->rx_buff_pool[pool].dma_addr[index],
390 adapter->rx_buff_pool[pool].buff_size,
393 free_index = adapter->rx_buff_pool[pool].producer_index;
394 adapter->rx_buff_pool[pool].producer_index++;
395 if (adapter->rx_buff_pool[pool].producer_index >=
396 adapter->rx_buff_pool[pool].size)
397 adapter->rx_buff_pool[pool].producer_index = 0;
398 adapter->rx_buff_pool[pool].free_map[free_index] = index;
402 atomic_dec(&(adapter->rx_buff_pool[pool].available));
405 /* get the current buffer on the rx queue */
406 static inline struct sk_buff *ibmveth_rxq_get_buffer(struct ibmveth_adapter *adapter)
408 u64 correlator = adapter->rx_queue.queue_addr[adapter->rx_queue.index].correlator;
409 unsigned int pool = correlator >> 32;
410 unsigned int index = correlator & 0xffffffffUL;
412 BUG_ON(pool >= IBMVETH_NUM_BUFF_POOLS);
413 BUG_ON(index >= adapter->rx_buff_pool[pool].size);
415 return adapter->rx_buff_pool[pool].skbuff[index];
418 /* recycle the current buffer on the rx queue */
419 static int ibmveth_rxq_recycle_buffer(struct ibmveth_adapter *adapter)
421 u32 q_index = adapter->rx_queue.index;
422 u64 correlator = adapter->rx_queue.queue_addr[q_index].correlator;
423 unsigned int pool = correlator >> 32;
424 unsigned int index = correlator & 0xffffffffUL;
425 union ibmveth_buf_desc desc;
426 unsigned long lpar_rc;
429 BUG_ON(pool >= IBMVETH_NUM_BUFF_POOLS);
430 BUG_ON(index >= adapter->rx_buff_pool[pool].size);
432 if (!adapter->rx_buff_pool[pool].active) {
433 ibmveth_rxq_harvest_buffer(adapter);
434 ibmveth_free_buffer_pool(adapter, &adapter->rx_buff_pool[pool]);
438 desc.fields.flags_len = IBMVETH_BUF_VALID |
439 adapter->rx_buff_pool[pool].buff_size;
440 desc.fields.address = adapter->rx_buff_pool[pool].dma_addr[index];
442 lpar_rc = h_add_logical_lan_buffer(adapter->vdev->unit_address, desc.desc);
444 if (lpar_rc != H_SUCCESS) {
445 netdev_dbg(adapter->netdev, "h_add_logical_lan_buffer failed "
446 "during recycle rc=%ld", lpar_rc);
447 ibmveth_remove_buffer_from_pool(adapter, adapter->rx_queue.queue_addr[adapter->rx_queue.index].correlator);
451 if (++adapter->rx_queue.index == adapter->rx_queue.num_slots) {
452 adapter->rx_queue.index = 0;
453 adapter->rx_queue.toggle = !adapter->rx_queue.toggle;
460 static void ibmveth_rxq_harvest_buffer(struct ibmveth_adapter *adapter)
462 ibmveth_remove_buffer_from_pool(adapter, adapter->rx_queue.queue_addr[adapter->rx_queue.index].correlator);
464 if (++adapter->rx_queue.index == adapter->rx_queue.num_slots) {
465 adapter->rx_queue.index = 0;
466 adapter->rx_queue.toggle = !adapter->rx_queue.toggle;
470 static void ibmveth_cleanup(struct ibmveth_adapter *adapter)
473 struct device *dev = &adapter->vdev->dev;
475 if (adapter->buffer_list_addr != NULL) {
476 if (!dma_mapping_error(dev, adapter->buffer_list_dma)) {
477 dma_unmap_single(dev, adapter->buffer_list_dma, 4096,
479 adapter->buffer_list_dma = DMA_ERROR_CODE;
481 free_page((unsigned long)adapter->buffer_list_addr);
482 adapter->buffer_list_addr = NULL;
485 if (adapter->filter_list_addr != NULL) {
486 if (!dma_mapping_error(dev, adapter->filter_list_dma)) {
487 dma_unmap_single(dev, adapter->filter_list_dma, 4096,
489 adapter->filter_list_dma = DMA_ERROR_CODE;
491 free_page((unsigned long)adapter->filter_list_addr);
492 adapter->filter_list_addr = NULL;
495 if (adapter->rx_queue.queue_addr != NULL) {
496 dma_free_coherent(dev, adapter->rx_queue.queue_len,
497 adapter->rx_queue.queue_addr,
498 adapter->rx_queue.queue_dma);
499 adapter->rx_queue.queue_addr = NULL;
502 for (i = 0; i < IBMVETH_NUM_BUFF_POOLS; i++)
503 if (adapter->rx_buff_pool[i].active)
504 ibmveth_free_buffer_pool(adapter,
505 &adapter->rx_buff_pool[i]);
507 if (adapter->bounce_buffer != NULL) {
508 if (!dma_mapping_error(dev, adapter->bounce_buffer_dma)) {
509 dma_unmap_single(&adapter->vdev->dev,
510 adapter->bounce_buffer_dma,
511 adapter->netdev->mtu + IBMVETH_BUFF_OH,
513 adapter->bounce_buffer_dma = DMA_ERROR_CODE;
515 kfree(adapter->bounce_buffer);
516 adapter->bounce_buffer = NULL;
520 static int ibmveth_register_logical_lan(struct ibmveth_adapter *adapter,
521 union ibmveth_buf_desc rxq_desc, u64 mac_address)
523 int rc, try_again = 1;
526 * After a kexec the adapter will still be open, so our attempt to
527 * open it will fail. So if we get a failure we free the adapter and
528 * try again, but only once.
531 rc = h_register_logical_lan(adapter->vdev->unit_address,
532 adapter->buffer_list_dma, rxq_desc.desc,
533 adapter->filter_list_dma, mac_address);
535 if (rc != H_SUCCESS && try_again) {
537 rc = h_free_logical_lan(adapter->vdev->unit_address);
538 } while (H_IS_LONG_BUSY(rc) || (rc == H_BUSY));
547 static u64 ibmveth_encode_mac_addr(u8 *mac)
552 for (i = 0; i < ETH_ALEN; i++)
553 encoded = (encoded << 8) | mac[i];
558 static int ibmveth_open(struct net_device *netdev)
560 struct ibmveth_adapter *adapter = netdev_priv(netdev);
563 unsigned long lpar_rc;
565 union ibmveth_buf_desc rxq_desc;
569 netdev_dbg(netdev, "open starting\n");
571 napi_enable(&adapter->napi);
573 for(i = 0; i < IBMVETH_NUM_BUFF_POOLS; i++)
574 rxq_entries += adapter->rx_buff_pool[i].size;
576 adapter->buffer_list_addr = (void*) get_zeroed_page(GFP_KERNEL);
577 adapter->filter_list_addr = (void*) get_zeroed_page(GFP_KERNEL);
579 if (!adapter->buffer_list_addr || !adapter->filter_list_addr) {
580 netdev_err(netdev, "unable to allocate filter or buffer list "
586 dev = &adapter->vdev->dev;
588 adapter->rx_queue.queue_len = sizeof(struct ibmveth_rx_q_entry) *
590 adapter->rx_queue.queue_addr =
591 dma_alloc_coherent(dev, adapter->rx_queue.queue_len,
592 &adapter->rx_queue.queue_dma, GFP_KERNEL);
593 if (!adapter->rx_queue.queue_addr) {
598 adapter->buffer_list_dma = dma_map_single(dev,
599 adapter->buffer_list_addr, 4096, DMA_BIDIRECTIONAL);
600 adapter->filter_list_dma = dma_map_single(dev,
601 adapter->filter_list_addr, 4096, DMA_BIDIRECTIONAL);
603 if ((dma_mapping_error(dev, adapter->buffer_list_dma)) ||
604 (dma_mapping_error(dev, adapter->filter_list_dma))) {
605 netdev_err(netdev, "unable to map filter or buffer list "
611 adapter->rx_queue.index = 0;
612 adapter->rx_queue.num_slots = rxq_entries;
613 adapter->rx_queue.toggle = 1;
615 mac_address = ibmveth_encode_mac_addr(netdev->dev_addr);
617 rxq_desc.fields.flags_len = IBMVETH_BUF_VALID |
618 adapter->rx_queue.queue_len;
619 rxq_desc.fields.address = adapter->rx_queue.queue_dma;
621 netdev_dbg(netdev, "buffer list @ 0x%p\n", adapter->buffer_list_addr);
622 netdev_dbg(netdev, "filter list @ 0x%p\n", adapter->filter_list_addr);
623 netdev_dbg(netdev, "receive q @ 0x%p\n", adapter->rx_queue.queue_addr);
625 h_vio_signal(adapter->vdev->unit_address, VIO_IRQ_DISABLE);
627 lpar_rc = ibmveth_register_logical_lan(adapter, rxq_desc, mac_address);
629 if (lpar_rc != H_SUCCESS) {
630 netdev_err(netdev, "h_register_logical_lan failed with %ld\n",
632 netdev_err(netdev, "buffer TCE:0x%llx filter TCE:0x%llx rxq "
633 "desc:0x%llx MAC:0x%llx\n",
634 adapter->buffer_list_dma,
635 adapter->filter_list_dma,
642 for (i = 0; i < IBMVETH_NUM_BUFF_POOLS; i++) {
643 if (!adapter->rx_buff_pool[i].active)
645 if (ibmveth_alloc_buffer_pool(&adapter->rx_buff_pool[i])) {
646 netdev_err(netdev, "unable to alloc pool\n");
647 adapter->rx_buff_pool[i].active = 0;
653 netdev_dbg(netdev, "registering irq 0x%x\n", netdev->irq);
654 rc = request_irq(netdev->irq, ibmveth_interrupt, 0, netdev->name,
657 netdev_err(netdev, "unable to request irq 0x%x, rc %d\n",
660 lpar_rc = h_free_logical_lan(adapter->vdev->unit_address);
661 } while (H_IS_LONG_BUSY(lpar_rc) || (lpar_rc == H_BUSY));
666 adapter->bounce_buffer =
667 kmalloc(netdev->mtu + IBMVETH_BUFF_OH, GFP_KERNEL);
668 if (!adapter->bounce_buffer) {
670 goto err_out_free_irq;
672 adapter->bounce_buffer_dma =
673 dma_map_single(&adapter->vdev->dev, adapter->bounce_buffer,
674 netdev->mtu + IBMVETH_BUFF_OH, DMA_BIDIRECTIONAL);
675 if (dma_mapping_error(dev, adapter->bounce_buffer_dma)) {
676 netdev_err(netdev, "unable to map bounce buffer\n");
678 goto err_out_free_irq;
681 netdev_dbg(netdev, "initial replenish cycle\n");
682 ibmveth_interrupt(netdev->irq, netdev);
684 netif_start_queue(netdev);
686 netdev_dbg(netdev, "open complete\n");
691 free_irq(netdev->irq, netdev);
693 ibmveth_cleanup(adapter);
694 napi_disable(&adapter->napi);
698 static int ibmveth_close(struct net_device *netdev)
700 struct ibmveth_adapter *adapter = netdev_priv(netdev);
703 netdev_dbg(netdev, "close starting\n");
705 napi_disable(&adapter->napi);
707 if (!adapter->pool_config)
708 netif_stop_queue(netdev);
710 h_vio_signal(adapter->vdev->unit_address, VIO_IRQ_DISABLE);
713 lpar_rc = h_free_logical_lan(adapter->vdev->unit_address);
714 } while (H_IS_LONG_BUSY(lpar_rc) || (lpar_rc == H_BUSY));
716 if (lpar_rc != H_SUCCESS) {
717 netdev_err(netdev, "h_free_logical_lan failed with %lx, "
718 "continuing with close\n", lpar_rc);
721 free_irq(netdev->irq, netdev);
723 ibmveth_update_rx_no_buffer(adapter);
725 ibmveth_cleanup(adapter);
727 netdev_dbg(netdev, "close complete\n");
732 static int netdev_get_settings(struct net_device *dev, struct ethtool_cmd *cmd)
734 cmd->supported = (SUPPORTED_1000baseT_Full | SUPPORTED_Autoneg |
736 cmd->advertising = (ADVERTISED_1000baseT_Full | ADVERTISED_Autoneg |
738 ethtool_cmd_speed_set(cmd, SPEED_1000);
739 cmd->duplex = DUPLEX_FULL;
740 cmd->port = PORT_FIBRE;
741 cmd->phy_address = 0;
742 cmd->transceiver = XCVR_INTERNAL;
743 cmd->autoneg = AUTONEG_ENABLE;
749 static void netdev_get_drvinfo(struct net_device *dev,
750 struct ethtool_drvinfo *info)
752 strlcpy(info->driver, ibmveth_driver_name, sizeof(info->driver));
753 strlcpy(info->version, ibmveth_driver_version, sizeof(info->version));
756 static netdev_features_t ibmveth_fix_features(struct net_device *dev,
757 netdev_features_t features)
760 * Since the ibmveth firmware interface does not have the
761 * concept of separate tx/rx checksum offload enable, if rx
762 * checksum is disabled we also have to disable tx checksum
763 * offload. Once we disable rx checksum offload, we are no
764 * longer allowed to send tx buffers that are not properly
768 if (!(features & NETIF_F_RXCSUM))
769 features &= ~NETIF_F_CSUM_MASK;
774 static int ibmveth_set_csum_offload(struct net_device *dev, u32 data)
776 struct ibmveth_adapter *adapter = netdev_priv(dev);
777 unsigned long set_attr, clr_attr, ret_attr;
778 unsigned long set_attr6, clr_attr6;
779 long ret, ret4, ret6;
780 int rc1 = 0, rc2 = 0;
783 if (netif_running(dev)) {
785 adapter->pool_config = 1;
787 adapter->pool_config = 0;
796 set_attr = IBMVETH_ILLAN_IPV4_TCP_CSUM;
797 set_attr6 = IBMVETH_ILLAN_IPV6_TCP_CSUM;
799 clr_attr = IBMVETH_ILLAN_IPV4_TCP_CSUM;
800 clr_attr6 = IBMVETH_ILLAN_IPV6_TCP_CSUM;
803 ret = h_illan_attributes(adapter->vdev->unit_address, 0, 0, &ret_attr);
805 if (ret == H_SUCCESS && !(ret_attr & IBMVETH_ILLAN_ACTIVE_TRUNK) &&
806 !(ret_attr & IBMVETH_ILLAN_TRUNK_PRI_MASK) &&
807 (ret_attr & IBMVETH_ILLAN_PADDED_PKT_CSUM)) {
808 ret4 = h_illan_attributes(adapter->vdev->unit_address, clr_attr,
809 set_attr, &ret_attr);
811 if (ret4 != H_SUCCESS) {
812 netdev_err(dev, "unable to change IPv4 checksum "
813 "offload settings. %d rc=%ld\n",
816 h_illan_attributes(adapter->vdev->unit_address,
817 set_attr, clr_attr, &ret_attr);
820 dev->features &= ~NETIF_F_IP_CSUM;
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",
834 h_illan_attributes(adapter->vdev->unit_address,
835 set_attr6, clr_attr6, &ret_attr);
838 dev->features &= ~NETIF_F_IPV6_CSUM;
841 adapter->fw_ipv6_csum_support = data;
843 if (ret4 == H_SUCCESS || ret6 == H_SUCCESS)
844 adapter->rx_csum = data;
849 netdev_err(dev, "unable to change checksum offload settings."
850 " %d rc=%ld ret_attr=%lx\n", data, ret,
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;
865 int rc1 = 0, rc2 = 0;
868 if (netif_running(dev)) {
870 adapter->pool_config = 1;
872 adapter->pool_config = 0;
879 set_attr = IBMVETH_ILLAN_LRG_SR_ENABLED;
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) &&
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",
894 h_illan_attributes(adapter->vdev->unit_address,
895 set_attr, clr_attr, &ret_attr);
898 dev->features &= ~(NETIF_F_TSO | NETIF_F_TSO6);
902 adapter->fw_large_send_support = data;
903 adapter->large_send = data;
906 /* Older firmware version of large send offload does not
910 dev->features &= ~NETIF_F_TSO6;
911 netdev_info(dev, "TSO feature requires all partitions to have updated driver");
913 adapter->large_send = data;
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)
934 features & ~(NETIF_F_CSUM_MASK |
938 if (large_send != adapter->large_send) {
939 rc2 = ibmveth_set_tso(dev, large_send);
940 if (rc2 && !adapter->large_send)
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)
952 if (stringset != ETH_SS_STATS)
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)
963 return ARRAY_SIZE(ibmveth_stats);
969 static void ibmveth_get_ethtool_stats(struct net_device *dev,
970 struct ethtool_stats *stats, u64 *data)
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_settings = netdev_get_settings,
982 .get_link = ethtool_op_get_link,
983 .get_strings = ibmveth_get_strings,
984 .get_sset_count = ibmveth_get_sset_count,
985 .get_ethtool_stats = ibmveth_get_ethtool_stats,
988 static int ibmveth_ioctl(struct net_device *dev, struct ifreq *ifr, int cmd)
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;
1003 * The retry count sets a maximum for the number of broadcast and
1004 * multicast destinations within the system.
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);
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];
1033 int force_bounce = 0;
1034 dma_addr_t dma_addr;
1035 unsigned long mss = 0;
1038 * veth handles a maximum of 6 segments including the header, so
1039 * we have to linearize the skb if there are more than this.
1041 if (skb_shinfo(skb)->nr_frags > 5 && __skb_linearize(skb)) {
1042 netdev->stats.tx_dropped++;
1046 /* veth can't checksum offload UDP */
1047 if (skb->ip_summed == CHECKSUM_PARTIAL &&
1048 ((skb->protocol == htons(ETH_P_IP) &&
1049 ip_hdr(skb)->protocol != IPPROTO_TCP) ||
1050 (skb->protocol == htons(ETH_P_IPV6) &&
1051 ipv6_hdr(skb)->nexthdr != IPPROTO_TCP)) &&
1052 skb_checksum_help(skb)) {
1054 netdev_err(netdev, "tx: failed to checksum packet\n");
1055 netdev->stats.tx_dropped++;
1059 desc_flags = IBMVETH_BUF_VALID;
1061 if (skb_is_gso(skb) && adapter->fw_large_send_support)
1062 desc_flags |= IBMVETH_BUF_LRG_SND;
1064 if (skb->ip_summed == CHECKSUM_PARTIAL) {
1065 unsigned char *buf = skb_transport_header(skb) +
1068 desc_flags |= (IBMVETH_BUF_NO_CSUM | IBMVETH_BUF_CSUM_GOOD);
1070 /* Need to zero out the checksum */
1076 memset(descs, 0, sizeof(descs));
1079 * If a linear packet is below the rx threshold then
1080 * copy it into the static bounce buffer. This avoids the
1081 * cost of a TCE insert and remove.
1083 if (force_bounce || (!skb_is_nonlinear(skb) &&
1084 (skb->len < tx_copybreak))) {
1085 skb_copy_from_linear_data(skb, adapter->bounce_buffer,
1088 descs[0].fields.flags_len = desc_flags | skb->len;
1089 descs[0].fields.address = adapter->bounce_buffer_dma;
1091 if (ibmveth_send(adapter, descs, 0)) {
1092 adapter->tx_send_failed++;
1093 netdev->stats.tx_dropped++;
1095 netdev->stats.tx_packets++;
1096 netdev->stats.tx_bytes += skb->len;
1102 /* Map the header */
1103 dma_addr = dma_map_single(&adapter->vdev->dev, skb->data,
1104 skb_headlen(skb), DMA_TO_DEVICE);
1105 if (dma_mapping_error(&adapter->vdev->dev, dma_addr))
1108 descs[0].fields.flags_len = desc_flags | skb_headlen(skb);
1109 descs[0].fields.address = dma_addr;
1112 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
1113 const skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
1115 dma_addr = skb_frag_dma_map(&adapter->vdev->dev, frag, 0,
1116 skb_frag_size(frag), DMA_TO_DEVICE);
1118 if (dma_mapping_error(&adapter->vdev->dev, dma_addr))
1119 goto map_failed_frags;
1121 descs[i+1].fields.flags_len = desc_flags | skb_frag_size(frag);
1122 descs[i+1].fields.address = dma_addr;
1125 if (skb_is_gso(skb)) {
1126 if (adapter->fw_large_send_support) {
1127 mss = (unsigned long)skb_shinfo(skb)->gso_size;
1128 adapter->tx_large_packets++;
1129 } else if (!skb_is_gso_v6(skb)) {
1130 /* Put -1 in the IP checksum to tell phyp it
1131 * is a largesend packet. Put the mss in
1134 ip_hdr(skb)->check = 0xffff;
1135 tcp_hdr(skb)->check =
1136 cpu_to_be16(skb_shinfo(skb)->gso_size);
1137 adapter->tx_large_packets++;
1141 if (ibmveth_send(adapter, descs, mss)) {
1142 adapter->tx_send_failed++;
1143 netdev->stats.tx_dropped++;
1145 netdev->stats.tx_packets++;
1146 netdev->stats.tx_bytes += skb->len;
1149 dma_unmap_single(&adapter->vdev->dev,
1150 descs[0].fields.address,
1151 descs[0].fields.flags_len & IBMVETH_BUF_LEN_MASK,
1154 for (i = 1; i < skb_shinfo(skb)->nr_frags + 1; i++)
1155 dma_unmap_page(&adapter->vdev->dev, descs[i].fields.address,
1156 descs[i].fields.flags_len & IBMVETH_BUF_LEN_MASK,
1160 dev_consume_skb_any(skb);
1161 return NETDEV_TX_OK;
1165 for (i = 0; i < last; i++)
1166 dma_unmap_page(&adapter->vdev->dev, descs[i].fields.address,
1167 descs[i].fields.flags_len & IBMVETH_BUF_LEN_MASK,
1171 if (!firmware_has_feature(FW_FEATURE_CMO))
1172 netdev_err(netdev, "tx: unable to map xmit buffer\n");
1173 adapter->tx_map_failed++;
1174 if (skb_linearize(skb)) {
1175 netdev->stats.tx_dropped++;
1182 static void ibmveth_rx_mss_helper(struct sk_buff *skb, u16 mss, int lrg_pkt)
1184 struct tcphdr *tcph;
1188 /* only TCP packets will be aggregated */
1189 if (skb->protocol == htons(ETH_P_IP)) {
1190 struct iphdr *iph = (struct iphdr *)skb->data;
1192 if (iph->protocol == IPPROTO_TCP) {
1193 offset = iph->ihl * 4;
1194 skb_shinfo(skb)->gso_type = SKB_GSO_TCPV4;
1198 } else if (skb->protocol == htons(ETH_P_IPV6)) {
1199 struct ipv6hdr *iph6 = (struct ipv6hdr *)skb->data;
1201 if (iph6->nexthdr == IPPROTO_TCP) {
1202 offset = sizeof(struct ipv6hdr);
1203 skb_shinfo(skb)->gso_type = SKB_GSO_TCPV6;
1210 /* if mss is not set through Large Packet bit/mss in rx buffer,
1211 * expect that the mss will be written to the tcp header checksum.
1213 tcph = (struct tcphdr *)(skb->data + offset);
1215 skb_shinfo(skb)->gso_size = mss;
1216 } else if (offset) {
1217 skb_shinfo(skb)->gso_size = ntohs(tcph->check);
1221 if (skb_shinfo(skb)->gso_size) {
1222 hdr_len = offset + tcph->doff * 4;
1223 skb_shinfo(skb)->gso_segs =
1224 DIV_ROUND_UP(skb->len - hdr_len,
1225 skb_shinfo(skb)->gso_size);
1229 static int ibmveth_poll(struct napi_struct *napi, int budget)
1231 struct ibmveth_adapter *adapter =
1232 container_of(napi, struct ibmveth_adapter, napi);
1233 struct net_device *netdev = adapter->netdev;
1234 int frames_processed = 0;
1235 unsigned long lpar_rc;
1240 while (frames_processed < budget) {
1241 if (!ibmveth_rxq_pending_buffer(adapter))
1245 if (!ibmveth_rxq_buffer_valid(adapter)) {
1246 wmb(); /* suggested by larson1 */
1247 adapter->rx_invalid_buffer++;
1248 netdev_dbg(netdev, "recycling invalid buffer\n");
1249 ibmveth_rxq_recycle_buffer(adapter);
1251 struct sk_buff *skb, *new_skb;
1252 int length = ibmveth_rxq_frame_length(adapter);
1253 int offset = ibmveth_rxq_frame_offset(adapter);
1254 int csum_good = ibmveth_rxq_csum_good(adapter);
1255 int lrg_pkt = ibmveth_rxq_large_packet(adapter);
1257 skb = ibmveth_rxq_get_buffer(adapter);
1259 /* if the large packet bit is set in the rx queue
1260 * descriptor, the mss will be written by PHYP eight
1261 * bytes from the start of the rx buffer, which is
1262 * skb->data at this stage
1265 __be64 *rxmss = (__be64 *)(skb->data + 8);
1267 mss = (u16)be64_to_cpu(*rxmss);
1271 if (length < rx_copybreak)
1272 new_skb = netdev_alloc_skb(netdev, length);
1275 skb_copy_to_linear_data(new_skb,
1279 ibmveth_flush_buffer(skb->data,
1281 if (!ibmveth_rxq_recycle_buffer(adapter))
1285 ibmveth_rxq_harvest_buffer(adapter);
1286 skb_reserve(skb, offset);
1289 skb_put(skb, length);
1290 skb->protocol = eth_type_trans(skb, netdev);
1293 skb->ip_summed = CHECKSUM_UNNECESSARY;
1294 if (be16_to_cpu(skb->protocol) == ETH_P_IP) {
1295 iph = (struct iphdr *)skb->data;
1297 /* If the IP checksum is not offloaded and if the packet
1298 * is large send, the checksum must be rebuilt.
1300 if (iph->check == 0xffff) {
1302 iph->check = ip_fast_csum((unsigned char *)iph, iph->ihl);
1307 if (length > netdev->mtu + ETH_HLEN) {
1308 ibmveth_rx_mss_helper(skb, mss, lrg_pkt);
1309 adapter->rx_large_packets++;
1312 napi_gro_receive(napi, skb); /* send it up */
1314 netdev->stats.rx_packets++;
1315 netdev->stats.rx_bytes += length;
1320 ibmveth_replenish_task(adapter);
1322 if (frames_processed < budget) {
1323 napi_complete(napi);
1325 /* We think we are done - reenable interrupts,
1326 * then check once more to make sure we are done.
1328 lpar_rc = h_vio_signal(adapter->vdev->unit_address,
1331 BUG_ON(lpar_rc != H_SUCCESS);
1333 if (ibmveth_rxq_pending_buffer(adapter) &&
1334 napi_reschedule(napi)) {
1335 lpar_rc = h_vio_signal(adapter->vdev->unit_address,
1341 return frames_processed;
1344 static irqreturn_t ibmveth_interrupt(int irq, void *dev_instance)
1346 struct net_device *netdev = dev_instance;
1347 struct ibmveth_adapter *adapter = netdev_priv(netdev);
1348 unsigned long lpar_rc;
1350 if (napi_schedule_prep(&adapter->napi)) {
1351 lpar_rc = h_vio_signal(adapter->vdev->unit_address,
1353 BUG_ON(lpar_rc != H_SUCCESS);
1354 __napi_schedule(&adapter->napi);
1359 static void ibmveth_set_multicast_list(struct net_device *netdev)
1361 struct ibmveth_adapter *adapter = netdev_priv(netdev);
1362 unsigned long lpar_rc;
1364 if ((netdev->flags & IFF_PROMISC) ||
1365 (netdev_mc_count(netdev) > adapter->mcastFilterSize)) {
1366 lpar_rc = h_multicast_ctrl(adapter->vdev->unit_address,
1367 IbmVethMcastEnableRecv |
1368 IbmVethMcastDisableFiltering,
1370 if (lpar_rc != H_SUCCESS) {
1371 netdev_err(netdev, "h_multicast_ctrl rc=%ld when "
1372 "entering promisc mode\n", lpar_rc);
1375 struct netdev_hw_addr *ha;
1376 /* clear the filter table & disable filtering */
1377 lpar_rc = h_multicast_ctrl(adapter->vdev->unit_address,
1378 IbmVethMcastEnableRecv |
1379 IbmVethMcastDisableFiltering |
1380 IbmVethMcastClearFilterTable,
1382 if (lpar_rc != H_SUCCESS) {
1383 netdev_err(netdev, "h_multicast_ctrl rc=%ld when "
1384 "attempting to clear filter table\n",
1387 /* add the addresses to the filter table */
1388 netdev_for_each_mc_addr(ha, netdev) {
1389 /* add the multicast address to the filter table */
1391 mcast_addr = ibmveth_encode_mac_addr(ha->addr);
1392 lpar_rc = h_multicast_ctrl(adapter->vdev->unit_address,
1393 IbmVethMcastAddFilter,
1395 if (lpar_rc != H_SUCCESS) {
1396 netdev_err(netdev, "h_multicast_ctrl rc=%ld "
1397 "when adding an entry to the filter "
1398 "table\n", lpar_rc);
1402 /* re-enable filtering */
1403 lpar_rc = h_multicast_ctrl(adapter->vdev->unit_address,
1404 IbmVethMcastEnableFiltering,
1406 if (lpar_rc != H_SUCCESS) {
1407 netdev_err(netdev, "h_multicast_ctrl rc=%ld when "
1408 "enabling filtering\n", lpar_rc);
1413 static int ibmveth_change_mtu(struct net_device *dev, int new_mtu)
1415 struct ibmveth_adapter *adapter = netdev_priv(dev);
1416 struct vio_dev *viodev = adapter->vdev;
1417 int new_mtu_oh = new_mtu + IBMVETH_BUFF_OH;
1419 int need_restart = 0;
1421 for (i = 0; i < IBMVETH_NUM_BUFF_POOLS; i++)
1422 if (new_mtu_oh <= adapter->rx_buff_pool[i].buff_size)
1425 if (i == IBMVETH_NUM_BUFF_POOLS)
1428 /* Deactivate all the buffer pools so that the next loop can activate
1429 only the buffer pools necessary to hold the new MTU */
1430 if (netif_running(adapter->netdev)) {
1432 adapter->pool_config = 1;
1433 ibmveth_close(adapter->netdev);
1434 adapter->pool_config = 0;
1437 /* Look for an active buffer pool that can hold the new MTU */
1438 for (i = 0; i < IBMVETH_NUM_BUFF_POOLS; i++) {
1439 adapter->rx_buff_pool[i].active = 1;
1441 if (new_mtu_oh <= adapter->rx_buff_pool[i].buff_size) {
1443 vio_cmo_set_dev_desired(viodev,
1444 ibmveth_get_desired_dma
1447 return ibmveth_open(adapter->netdev);
1453 if (need_restart && (rc = ibmveth_open(adapter->netdev)))
1459 #ifdef CONFIG_NET_POLL_CONTROLLER
1460 static void ibmveth_poll_controller(struct net_device *dev)
1462 ibmveth_replenish_task(netdev_priv(dev));
1463 ibmveth_interrupt(dev->irq, dev);
1468 * ibmveth_get_desired_dma - Calculate IO memory desired by the driver
1470 * @vdev: struct vio_dev for the device whose desired IO mem is to be returned
1473 * Number of bytes of IO data the driver will need to perform well.
1475 static unsigned long ibmveth_get_desired_dma(struct vio_dev *vdev)
1477 struct net_device *netdev = dev_get_drvdata(&vdev->dev);
1478 struct ibmveth_adapter *adapter;
1479 struct iommu_table *tbl;
1484 tbl = get_iommu_table_base(&vdev->dev);
1486 /* netdev inits at probe time along with the structures we need below*/
1488 return IOMMU_PAGE_ALIGN(IBMVETH_IO_ENTITLEMENT_DEFAULT, tbl);
1490 adapter = netdev_priv(netdev);
1492 ret = IBMVETH_BUFF_LIST_SIZE + IBMVETH_FILT_LIST_SIZE;
1493 ret += IOMMU_PAGE_ALIGN(netdev->mtu, tbl);
1495 for (i = 0; i < IBMVETH_NUM_BUFF_POOLS; i++) {
1496 /* add the size of the active receive buffers */
1497 if (adapter->rx_buff_pool[i].active)
1499 adapter->rx_buff_pool[i].size *
1500 IOMMU_PAGE_ALIGN(adapter->rx_buff_pool[i].
1502 rxqentries += adapter->rx_buff_pool[i].size;
1504 /* add the size of the receive queue entries */
1505 ret += IOMMU_PAGE_ALIGN(
1506 rxqentries * sizeof(struct ibmveth_rx_q_entry), tbl);
1511 static int ibmveth_set_mac_addr(struct net_device *dev, void *p)
1513 struct ibmveth_adapter *adapter = netdev_priv(dev);
1514 struct sockaddr *addr = p;
1518 if (!is_valid_ether_addr(addr->sa_data))
1519 return -EADDRNOTAVAIL;
1521 mac_address = ibmveth_encode_mac_addr(addr->sa_data);
1522 rc = h_change_logical_lan_mac(adapter->vdev->unit_address, mac_address);
1524 netdev_err(adapter->netdev, "h_change_logical_lan_mac failed with rc=%d\n", rc);
1528 ether_addr_copy(dev->dev_addr, addr->sa_data);
1533 static const struct net_device_ops ibmveth_netdev_ops = {
1534 .ndo_open = ibmveth_open,
1535 .ndo_stop = ibmveth_close,
1536 .ndo_start_xmit = ibmveth_start_xmit,
1537 .ndo_set_rx_mode = ibmveth_set_multicast_list,
1538 .ndo_do_ioctl = ibmveth_ioctl,
1539 .ndo_change_mtu = ibmveth_change_mtu,
1540 .ndo_fix_features = ibmveth_fix_features,
1541 .ndo_set_features = ibmveth_set_features,
1542 .ndo_validate_addr = eth_validate_addr,
1543 .ndo_set_mac_address = ibmveth_set_mac_addr,
1544 #ifdef CONFIG_NET_POLL_CONTROLLER
1545 .ndo_poll_controller = ibmveth_poll_controller,
1549 static int ibmveth_probe(struct vio_dev *dev, const struct vio_device_id *id)
1552 struct net_device *netdev;
1553 struct ibmveth_adapter *adapter;
1554 unsigned char *mac_addr_p;
1555 unsigned int *mcastFilterSize_p;
1557 unsigned long ret_attr;
1559 dev_dbg(&dev->dev, "entering ibmveth_probe for UA 0x%x\n",
1562 mac_addr_p = (unsigned char *)vio_get_attribute(dev, VETH_MAC_ADDR,
1565 dev_err(&dev->dev, "Can't find VETH_MAC_ADDR attribute\n");
1568 /* Workaround for old/broken pHyp */
1571 else if (mac_len != 6) {
1572 dev_err(&dev->dev, "VETH_MAC_ADDR attribute wrong len %d\n",
1577 mcastFilterSize_p = (unsigned int *)vio_get_attribute(dev,
1578 VETH_MCAST_FILTER_SIZE, NULL);
1579 if (!mcastFilterSize_p) {
1580 dev_err(&dev->dev, "Can't find VETH_MCAST_FILTER_SIZE "
1585 netdev = alloc_etherdev(sizeof(struct ibmveth_adapter));
1590 adapter = netdev_priv(netdev);
1591 dev_set_drvdata(&dev->dev, netdev);
1593 adapter->vdev = dev;
1594 adapter->netdev = netdev;
1595 adapter->mcastFilterSize = *mcastFilterSize_p;
1596 adapter->pool_config = 0;
1598 netif_napi_add(netdev, &adapter->napi, ibmveth_poll, 16);
1600 netdev->irq = dev->irq;
1601 netdev->netdev_ops = &ibmveth_netdev_ops;
1602 netdev->ethtool_ops = &netdev_ethtool_ops;
1603 SET_NETDEV_DEV(netdev, &dev->dev);
1604 netdev->hw_features = NETIF_F_SG;
1605 if (vio_get_attribute(dev, "ibm,illan-options", NULL) != NULL) {
1606 netdev->hw_features |= NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM |
1610 netdev->features |= netdev->hw_features;
1612 ret = h_illan_attributes(adapter->vdev->unit_address, 0, 0, &ret_attr);
1614 /* If running older firmware, TSO should not be enabled by default */
1615 if (ret == H_SUCCESS && (ret_attr & IBMVETH_ILLAN_LRG_SND_SUPPORT) &&
1617 netdev->hw_features |= NETIF_F_TSO | NETIF_F_TSO6;
1618 netdev->features |= netdev->hw_features;
1620 netdev->hw_features |= NETIF_F_TSO;
1623 netdev->min_mtu = IBMVETH_MIN_MTU;
1624 netdev->max_mtu = ETH_MAX_MTU;
1626 memcpy(netdev->dev_addr, mac_addr_p, ETH_ALEN);
1628 if (firmware_has_feature(FW_FEATURE_CMO))
1629 memcpy(pool_count, pool_count_cmo, sizeof(pool_count));
1631 for (i = 0; i < IBMVETH_NUM_BUFF_POOLS; i++) {
1632 struct kobject *kobj = &adapter->rx_buff_pool[i].kobj;
1635 ibmveth_init_buffer_pool(&adapter->rx_buff_pool[i], i,
1636 pool_count[i], pool_size[i],
1638 error = kobject_init_and_add(kobj, &ktype_veth_pool,
1639 &dev->dev.kobj, "pool%d", i);
1641 kobject_uevent(kobj, KOBJ_ADD);
1644 netdev_dbg(netdev, "adapter @ 0x%p\n", adapter);
1646 adapter->buffer_list_dma = DMA_ERROR_CODE;
1647 adapter->filter_list_dma = DMA_ERROR_CODE;
1648 adapter->rx_queue.queue_dma = DMA_ERROR_CODE;
1650 netdev_dbg(netdev, "registering netdev...\n");
1652 ibmveth_set_features(netdev, netdev->features);
1654 rc = register_netdev(netdev);
1657 netdev_dbg(netdev, "failed to register netdev rc=%d\n", rc);
1658 free_netdev(netdev);
1662 netdev_dbg(netdev, "registered\n");
1667 static int ibmveth_remove(struct vio_dev *dev)
1669 struct net_device *netdev = dev_get_drvdata(&dev->dev);
1670 struct ibmveth_adapter *adapter = netdev_priv(netdev);
1673 for (i = 0; i < IBMVETH_NUM_BUFF_POOLS; i++)
1674 kobject_put(&adapter->rx_buff_pool[i].kobj);
1676 unregister_netdev(netdev);
1678 free_netdev(netdev);
1679 dev_set_drvdata(&dev->dev, NULL);
1684 static struct attribute veth_active_attr;
1685 static struct attribute veth_num_attr;
1686 static struct attribute veth_size_attr;
1688 static ssize_t veth_pool_show(struct kobject *kobj,
1689 struct attribute *attr, char *buf)
1691 struct ibmveth_buff_pool *pool = container_of(kobj,
1692 struct ibmveth_buff_pool,
1695 if (attr == &veth_active_attr)
1696 return sprintf(buf, "%d\n", pool->active);
1697 else if (attr == &veth_num_attr)
1698 return sprintf(buf, "%d\n", pool->size);
1699 else if (attr == &veth_size_attr)
1700 return sprintf(buf, "%d\n", pool->buff_size);
1704 static ssize_t veth_pool_store(struct kobject *kobj, struct attribute *attr,
1705 const char *buf, size_t count)
1707 struct ibmveth_buff_pool *pool = container_of(kobj,
1708 struct ibmveth_buff_pool,
1710 struct net_device *netdev = dev_get_drvdata(
1711 container_of(kobj->parent, struct device, kobj));
1712 struct ibmveth_adapter *adapter = netdev_priv(netdev);
1713 long value = simple_strtol(buf, NULL, 10);
1716 if (attr == &veth_active_attr) {
1717 if (value && !pool->active) {
1718 if (netif_running(netdev)) {
1719 if (ibmveth_alloc_buffer_pool(pool)) {
1721 "unable to alloc pool\n");
1725 adapter->pool_config = 1;
1726 ibmveth_close(netdev);
1727 adapter->pool_config = 0;
1728 if ((rc = ibmveth_open(netdev)))
1733 } else if (!value && pool->active) {
1734 int mtu = netdev->mtu + IBMVETH_BUFF_OH;
1736 /* Make sure there is a buffer pool with buffers that
1737 can hold a packet of the size of the MTU */
1738 for (i = 0; i < IBMVETH_NUM_BUFF_POOLS; i++) {
1739 if (pool == &adapter->rx_buff_pool[i])
1741 if (!adapter->rx_buff_pool[i].active)
1743 if (mtu <= adapter->rx_buff_pool[i].buff_size)
1747 if (i == IBMVETH_NUM_BUFF_POOLS) {
1748 netdev_err(netdev, "no active pool >= MTU\n");
1752 if (netif_running(netdev)) {
1753 adapter->pool_config = 1;
1754 ibmveth_close(netdev);
1756 adapter->pool_config = 0;
1757 if ((rc = ibmveth_open(netdev)))
1762 } else if (attr == &veth_num_attr) {
1763 if (value <= 0 || value > IBMVETH_MAX_POOL_COUNT) {
1766 if (netif_running(netdev)) {
1767 adapter->pool_config = 1;
1768 ibmveth_close(netdev);
1769 adapter->pool_config = 0;
1771 if ((rc = ibmveth_open(netdev)))
1777 } else if (attr == &veth_size_attr) {
1778 if (value <= IBMVETH_BUFF_OH || value > IBMVETH_MAX_BUF_SIZE) {
1781 if (netif_running(netdev)) {
1782 adapter->pool_config = 1;
1783 ibmveth_close(netdev);
1784 adapter->pool_config = 0;
1785 pool->buff_size = value;
1786 if ((rc = ibmveth_open(netdev)))
1789 pool->buff_size = value;
1794 /* kick the interrupt handler to allocate/deallocate pools */
1795 ibmveth_interrupt(netdev->irq, netdev);
1800 #define ATTR(_name, _mode) \
1801 struct attribute veth_##_name##_attr = { \
1802 .name = __stringify(_name), .mode = _mode, \
1805 static ATTR(active, 0644);
1806 static ATTR(num, 0644);
1807 static ATTR(size, 0644);
1809 static struct attribute *veth_pool_attrs[] = {
1816 static const struct sysfs_ops veth_pool_ops = {
1817 .show = veth_pool_show,
1818 .store = veth_pool_store,
1821 static struct kobj_type ktype_veth_pool = {
1823 .sysfs_ops = &veth_pool_ops,
1824 .default_attrs = veth_pool_attrs,
1827 static int ibmveth_resume(struct device *dev)
1829 struct net_device *netdev = dev_get_drvdata(dev);
1830 ibmveth_interrupt(netdev->irq, netdev);
1834 static struct vio_device_id ibmveth_device_table[] = {
1835 { "network", "IBM,l-lan"},
1838 MODULE_DEVICE_TABLE(vio, ibmveth_device_table);
1840 static struct dev_pm_ops ibmveth_pm_ops = {
1841 .resume = ibmveth_resume
1844 static struct vio_driver ibmveth_driver = {
1845 .id_table = ibmveth_device_table,
1846 .probe = ibmveth_probe,
1847 .remove = ibmveth_remove,
1848 .get_desired_dma = ibmveth_get_desired_dma,
1849 .name = ibmveth_driver_name,
1850 .pm = &ibmveth_pm_ops,
1853 static int __init ibmveth_module_init(void)
1855 printk(KERN_DEBUG "%s: %s %s\n", ibmveth_driver_name,
1856 ibmveth_driver_string, ibmveth_driver_version);
1858 return vio_register_driver(&ibmveth_driver);
1861 static void __exit ibmveth_module_exit(void)
1863 vio_unregister_driver(&ibmveth_driver);
1866 module_init(ibmveth_module_init);
1867 module_exit(ibmveth_module_exit);