2 * Copyright (c) 2009, Microsoft Corporation.
4 * This program is free software; you can redistribute it and/or modify it
5 * under the terms and conditions of the GNU General Public License,
6 * version 2, as published by the Free Software Foundation.
8 * This program is distributed in the hope it will be useful, but WITHOUT
9 * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
10 * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License for
13 * You should have received a copy of the GNU General Public License along with
14 * this program; if not, write to the Free Software Foundation, Inc., 59 Temple
15 * Place - Suite 330, Boston, MA 02111-1307 USA.
18 * Haiyang Zhang <haiyangz@microsoft.com>
19 * Hank Janssen <hjanssen@microsoft.com>
21 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
23 #include <linux/init.h>
24 #include <linux/atomic.h>
25 #include <linux/module.h>
26 #include <linux/highmem.h>
27 #include <linux/device.h>
29 #include <linux/delay.h>
30 #include <linux/netdevice.h>
31 #include <linux/inetdevice.h>
32 #include <linux/etherdevice.h>
33 #include <linux/skbuff.h>
35 #include <linux/slab.h>
37 #include <net/route.h>
39 #include <net/pkt_sched.h>
41 #include "hyperv_net.h"
43 struct net_device_context {
44 /* point back to our device context */
45 struct hv_device *device_ctx;
47 struct delayed_work dwork;
51 #define PACKET_PAGES_LOWATER 8
52 /* Need this many pages to handle worst case fragmented packet */
53 #define PACKET_PAGES_HIWATER (MAX_SKB_FRAGS + 2)
55 static int ring_size = 128;
56 module_param(ring_size, int, S_IRUGO);
57 MODULE_PARM_DESC(ring_size, "Ring buffer size (# of pages)");
59 /* no-op so the netdev core doesn't return -EINVAL when modifying the the
60 * multicast address list in SIOCADDMULTI. hv is setup to get all multicast
61 * when it calls RndisFilterOnOpen() */
62 static void netvsc_set_multicast_list(struct net_device *net)
66 static int netvsc_open(struct net_device *net)
68 struct net_device_context *net_device_ctx = netdev_priv(net);
69 struct hv_device *device_obj = net_device_ctx->device_ctx;
72 if (netif_carrier_ok(net)) {
73 /* Open up the device */
74 ret = rndis_filter_open(device_obj);
76 netdev_err(net, "unable to open device (ret %d).\n",
81 netif_start_queue(net);
83 netdev_err(net, "unable to open device...link is down.\n");
89 static int netvsc_close(struct net_device *net)
91 struct net_device_context *net_device_ctx = netdev_priv(net);
92 struct hv_device *device_obj = net_device_ctx->device_ctx;
95 netif_stop_queue(net);
97 ret = rndis_filter_close(device_obj);
99 netdev_err(net, "unable to close device (ret %d).\n", ret);
104 static void netvsc_xmit_completion(void *context)
106 struct hv_netvsc_packet *packet = (struct hv_netvsc_packet *)context;
107 struct sk_buff *skb = (struct sk_buff *)
108 (unsigned long)packet->completion.send.send_completion_tid;
113 struct net_device *net = skb->dev;
114 struct net_device_context *net_device_ctx = netdev_priv(net);
115 unsigned int num_pages = skb_shinfo(skb)->nr_frags + 2;
117 dev_kfree_skb_any(skb);
119 atomic_add(num_pages, &net_device_ctx->avail);
120 if (atomic_read(&net_device_ctx->avail) >=
121 PACKET_PAGES_HIWATER)
122 netif_wake_queue(net);
126 static int netvsc_start_xmit(struct sk_buff *skb, struct net_device *net)
128 struct net_device_context *net_device_ctx = netdev_priv(net);
129 struct hv_netvsc_packet *packet;
131 unsigned int i, num_pages;
133 /* Add 1 for skb->data and additional one for RNDIS */
134 num_pages = skb_shinfo(skb)->nr_frags + 1 + 1;
135 if (num_pages > atomic_read(&net_device_ctx->avail))
136 return NETDEV_TX_BUSY;
138 /* Allocate a netvsc packet based on # of frags. */
139 packet = kzalloc(sizeof(struct hv_netvsc_packet) +
140 (num_pages * sizeof(struct hv_page_buffer)) +
141 sizeof(struct rndis_filter_packet), GFP_ATOMIC);
143 /* out of memory, silently drop packet */
144 netdev_err(net, "unable to allocate hv_netvsc_packet\n");
147 net->stats.tx_dropped++;
151 packet->extension = (void *)(unsigned long)packet +
152 sizeof(struct hv_netvsc_packet) +
153 (num_pages * sizeof(struct hv_page_buffer));
155 /* Setup the rndis header */
156 packet->page_buf_cnt = num_pages;
158 /* Initialize it from the skb */
159 packet->total_data_buflen = skb->len;
161 /* Start filling in the page buffers starting after RNDIS buffer. */
162 packet->page_buf[1].pfn = virt_to_phys(skb->data) >> PAGE_SHIFT;
163 packet->page_buf[1].offset
164 = (unsigned long)skb->data & (PAGE_SIZE - 1);
165 packet->page_buf[1].len = skb_headlen(skb);
167 /* Additional fragments are after SKB data */
168 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
169 skb_frag_t *f = &skb_shinfo(skb)->frags[i];
171 packet->page_buf[i+2].pfn = page_to_pfn(f->page);
172 packet->page_buf[i+2].offset = f->page_offset;
173 packet->page_buf[i+2].len = f->size;
176 /* Set the completion routine */
177 packet->completion.send.send_completion = netvsc_xmit_completion;
178 packet->completion.send.send_completion_ctx = packet;
179 packet->completion.send.send_completion_tid = (unsigned long)skb;
181 ret = rndis_filter_send(net_device_ctx->device_ctx,
184 net->stats.tx_bytes += skb->len;
185 net->stats.tx_packets++;
187 atomic_sub(num_pages, &net_device_ctx->avail);
188 if (atomic_read(&net_device_ctx->avail) < PACKET_PAGES_LOWATER)
189 netif_stop_queue(net);
191 /* we are shutting down or bus overloaded, just drop packet */
192 net->stats.tx_dropped++;
193 netvsc_xmit_completion(packet);
200 * netvsc_linkstatus_callback - Link up/down notification
202 void netvsc_linkstatus_callback(struct hv_device *device_obj,
205 struct net_device *net = dev_get_drvdata(&device_obj->device);
206 struct net_device_context *ndev_ctx;
209 netdev_err(net, "got link status but net device "
210 "not initialized yet\n");
215 netif_carrier_on(net);
216 netif_wake_queue(net);
217 ndev_ctx = netdev_priv(net);
218 schedule_delayed_work(&ndev_ctx->dwork, 0);
219 schedule_delayed_work(&ndev_ctx->dwork, msecs_to_jiffies(20));
221 netif_carrier_off(net);
222 netif_stop_queue(net);
227 * netvsc_recv_callback - Callback when we receive a packet from the
228 * "wire" on the specified device.
230 int netvsc_recv_callback(struct hv_device *device_obj,
231 struct hv_netvsc_packet *packet)
233 struct net_device *net = dev_get_drvdata(&device_obj->device);
240 netdev_err(net, "got receive callback but net device"
241 " not initialized yet\n");
245 /* Allocate a skb - TODO direct I/O to pages? */
246 skb = netdev_alloc_skb_ip_align(net, packet->total_data_buflen);
247 if (unlikely(!skb)) {
248 ++net->stats.rx_dropped;
252 /* for kmap_atomic */
253 local_irq_save(flags);
256 * Copy to skb. This copy is needed here since the memory pointed by
257 * hv_netvsc_packet cannot be deallocated
259 for (i = 0; i < packet->page_buf_cnt; i++) {
260 data = kmap_atomic(pfn_to_page(packet->page_buf[i].pfn),
262 data = (void *)(unsigned long)data +
263 packet->page_buf[i].offset;
265 memcpy(skb_put(skb, packet->page_buf[i].len), data,
266 packet->page_buf[i].len);
268 kunmap_atomic((void *)((unsigned long)data -
269 packet->page_buf[i].offset), KM_IRQ1);
272 local_irq_restore(flags);
274 skb->protocol = eth_type_trans(skb, net);
275 skb->ip_summed = CHECKSUM_NONE;
277 net->stats.rx_packets++;
278 net->stats.rx_bytes += skb->len;
281 * Pass the skb back up. Network stack will deallocate the skb when it
290 static void netvsc_get_drvinfo(struct net_device *net,
291 struct ethtool_drvinfo *info)
293 strcpy(info->driver, "hv_netvsc");
294 strcpy(info->version, HV_DRV_VERSION);
295 strcpy(info->fw_version, "N/A");
298 static const struct ethtool_ops ethtool_ops = {
299 .get_drvinfo = netvsc_get_drvinfo,
300 .get_link = ethtool_op_get_link,
303 static const struct net_device_ops device_ops = {
304 .ndo_open = netvsc_open,
305 .ndo_stop = netvsc_close,
306 .ndo_start_xmit = netvsc_start_xmit,
307 .ndo_set_multicast_list = netvsc_set_multicast_list,
308 .ndo_change_mtu = eth_change_mtu,
309 .ndo_validate_addr = eth_validate_addr,
310 .ndo_set_mac_address = eth_mac_addr,
314 * Send GARP packet to network peers after migrations.
315 * After Quick Migration, the network is not immediately operational in the
316 * current context when receiving RNDIS_STATUS_MEDIA_CONNECT event. So, add
317 * another netif_notify_peers() into a delayed work, otherwise GARP packet
318 * will not be sent after quick migration, and cause network disconnection.
320 static void netvsc_send_garp(struct work_struct *w)
322 struct net_device_context *ndev_ctx;
323 struct net_device *net;
325 ndev_ctx = container_of(w, struct net_device_context, dwork.work);
326 net = dev_get_drvdata(&ndev_ctx->device_ctx->device);
327 netif_notify_peers(net);
331 static int netvsc_probe(struct hv_device *dev)
333 struct net_device *net = NULL;
334 struct net_device_context *net_device_ctx;
335 struct netvsc_device_info device_info;
338 net = alloc_etherdev(sizeof(struct net_device_context));
342 /* Set initial state */
343 netif_carrier_off(net);
345 net_device_ctx = netdev_priv(net);
346 net_device_ctx->device_ctx = dev;
347 atomic_set(&net_device_ctx->avail, ring_size);
348 dev_set_drvdata(&dev->device, net);
349 INIT_DELAYED_WORK(&net_device_ctx->dwork, netvsc_send_garp);
351 /* Notify the netvsc driver of the new device */
352 device_info.ring_size = ring_size;
353 ret = rndis_filter_device_add(dev, &device_info);
356 dev_set_drvdata(&dev->device, NULL);
360 netif_carrier_on(net);
362 memcpy(net->dev_addr, device_info.mac_adr, ETH_ALEN);
364 net->netdev_ops = &device_ops;
366 /* TODO: Add GSO and Checksum offload */
367 net->hw_features = NETIF_F_SG;
368 net->features = NETIF_F_SG;
370 SET_ETHTOOL_OPS(net, ðtool_ops);
371 SET_NETDEV_DEV(net, &dev->device);
373 ret = register_netdev(net);
375 /* Remove the device and release the resource */
376 rndis_filter_device_remove(dev);
383 static int netvsc_remove(struct hv_device *dev)
385 struct net_device *net = dev_get_drvdata(&dev->device);
386 struct net_device_context *ndev_ctx;
389 dev_err(&dev->device, "No net device to remove\n");
393 ndev_ctx = netdev_priv(net);
394 cancel_delayed_work_sync(&ndev_ctx->dwork);
396 /* Stop outbound asap */
397 netif_stop_queue(net);
399 unregister_netdev(net);
402 * Call to the vsc driver to let it know that the device is being
405 rndis_filter_device_remove(dev);
411 static const struct hv_vmbus_device_id id_table[] = {
413 { VMBUS_DEVICE(0x63, 0x51, 0x61, 0xF8, 0x3E, 0xDF, 0xc5, 0x46,
414 0x91, 0x3F, 0xF2, 0xD2, 0xF9, 0x65, 0xED, 0x0E) },
418 MODULE_DEVICE_TABLE(vmbus, id_table);
420 /* The one and only one */
421 static struct hv_driver netvsc_drv = {
423 .id_table = id_table,
424 .probe = netvsc_probe,
425 .remove = netvsc_remove,
428 static void __exit netvsc_drv_exit(void)
430 vmbus_driver_unregister(&netvsc_drv);
433 static int __init netvsc_drv_init(void)
435 return vmbus_driver_register(&netvsc_drv);
438 MODULE_LICENSE("GPL");
439 MODULE_VERSION(HV_DRV_VERSION);
440 MODULE_DESCRIPTION("Microsoft Hyper-V network driver");
442 module_init(netvsc_drv_init);
443 module_exit(netvsc_drv_exit);