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/module.h>
25 #include <linux/highmem.h>
26 #include <linux/device.h>
28 #include <linux/delay.h>
29 #include <linux/netdevice.h>
30 #include <linux/inetdevice.h>
31 #include <linux/etherdevice.h>
32 #include <linux/skbuff.h>
34 #include <linux/slab.h>
35 #include <linux/dmi.h>
36 #include <linux/pci.h>
38 #include <net/route.h>
40 #include <net/pkt_sched.h>
43 #include "hyperv_net.h"
45 struct net_device_context {
46 /* point back to our device context */
47 struct hv_device *device_ctx;
49 struct work_struct work;
53 #define PACKET_PAGES_LOWATER 8
54 /* Need this many pages to handle worst case fragmented packet */
55 #define PACKET_PAGES_HIWATER (MAX_SKB_FRAGS + 2)
57 static int ring_size = 128;
58 module_param(ring_size, int, S_IRUGO);
59 MODULE_PARM_DESC(ring_size, "Ring buffer size (# of pages)");
61 /* no-op so the netdev core doesn't return -EINVAL when modifying the the
62 * multicast address list in SIOCADDMULTI. hv is setup to get all multicast
63 * when it calls RndisFilterOnOpen() */
64 static void netvsc_set_multicast_list(struct net_device *net)
68 static int netvsc_open(struct net_device *net)
70 struct net_device_context *net_device_ctx = netdev_priv(net);
71 struct hv_device *device_obj = net_device_ctx->device_ctx;
74 if (netif_carrier_ok(net)) {
75 /* Open up the device */
76 ret = rndis_filter_open(device_obj);
78 netdev_err(net, "unable to open device (ret %d).\n",
83 netif_start_queue(net);
85 netdev_err(net, "unable to open device...link is down.\n");
91 static int netvsc_close(struct net_device *net)
93 struct net_device_context *net_device_ctx = netdev_priv(net);
94 struct hv_device *device_obj = net_device_ctx->device_ctx;
97 netif_stop_queue(net);
99 ret = rndis_filter_close(device_obj);
101 netdev_err(net, "unable to close device (ret %d).\n", ret);
106 static void netvsc_xmit_completion(void *context)
108 struct hv_netvsc_packet *packet = (struct hv_netvsc_packet *)context;
109 struct sk_buff *skb = (struct sk_buff *)
110 (unsigned long)packet->completion.send.send_completion_tid;
115 struct net_device *net = skb->dev;
116 struct net_device_context *net_device_ctx = netdev_priv(net);
117 unsigned int num_pages = skb_shinfo(skb)->nr_frags + 2;
119 dev_kfree_skb_any(skb);
121 net_device_ctx->avail += num_pages;
122 if (net_device_ctx->avail >= PACKET_PAGES_HIWATER)
123 netif_wake_queue(net);
127 static int netvsc_start_xmit(struct sk_buff *skb, struct net_device *net)
129 struct net_device_context *net_device_ctx = netdev_priv(net);
130 struct netvsc_driver *net_drv_obj =
131 drv_to_netvscdrv(net_device_ctx->device_ctx->device.driver);
132 struct hv_netvsc_packet *packet;
134 unsigned int i, num_pages;
136 /* Add 1 for skb->data and additional one for RNDIS */
137 num_pages = skb_shinfo(skb)->nr_frags + 1 + 1;
138 if (num_pages > net_device_ctx->avail)
139 return NETDEV_TX_BUSY;
141 /* Allocate a netvsc packet based on # of frags. */
142 packet = kzalloc(sizeof(struct hv_netvsc_packet) +
143 (num_pages * sizeof(struct hv_page_buffer)) +
144 net_drv_obj->req_ext_size, GFP_ATOMIC);
146 /* out of memory, silently drop packet */
147 netdev_err(net, "unable to allocate hv_netvsc_packet\n");
150 net->stats.tx_dropped++;
154 packet->extension = (void *)(unsigned long)packet +
155 sizeof(struct hv_netvsc_packet) +
156 (num_pages * sizeof(struct hv_page_buffer));
158 /* Setup the rndis header */
159 packet->page_buf_cnt = num_pages;
161 /* TODO: Flush all write buffers/ memory fence ??? */
164 /* Initialize it from the skb */
165 packet->total_data_buflen = skb->len;
167 /* Start filling in the page buffers starting after RNDIS buffer. */
168 packet->page_buf[1].pfn = virt_to_phys(skb->data) >> PAGE_SHIFT;
169 packet->page_buf[1].offset
170 = (unsigned long)skb->data & (PAGE_SIZE - 1);
171 packet->page_buf[1].len = skb_headlen(skb);
173 /* Additional fragments are after SKB data */
174 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
175 skb_frag_t *f = &skb_shinfo(skb)->frags[i];
177 packet->page_buf[i+2].pfn = page_to_pfn(f->page);
178 packet->page_buf[i+2].offset = f->page_offset;
179 packet->page_buf[i+2].len = f->size;
182 /* Set the completion routine */
183 packet->completion.send.send_completion = netvsc_xmit_completion;
184 packet->completion.send.send_completion_ctx = packet;
185 packet->completion.send.send_completion_tid = (unsigned long)skb;
187 ret = net_drv_obj->send(net_device_ctx->device_ctx,
190 net->stats.tx_bytes += skb->len;
191 net->stats.tx_packets++;
193 net_device_ctx->avail -= num_pages;
194 if (net_device_ctx->avail < PACKET_PAGES_LOWATER)
195 netif_stop_queue(net);
197 /* we are shutting down or bus overloaded, just drop packet */
198 net->stats.tx_dropped++;
199 netvsc_xmit_completion(packet);
206 * netvsc_linkstatus_callback - Link up/down notification
208 void netvsc_linkstatus_callback(struct hv_device *device_obj,
211 struct net_device *net = dev_get_drvdata(&device_obj->device);
212 struct net_device_context *ndev_ctx;
215 netdev_err(net, "got link status but net device "
216 "not initialized yet\n");
221 netif_carrier_on(net);
222 netif_wake_queue(net);
223 netif_notify_peers(net);
224 ndev_ctx = netdev_priv(net);
225 schedule_work(&ndev_ctx->work);
227 netif_carrier_off(net);
228 netif_stop_queue(net);
233 * netvsc_recv_callback - Callback when we receive a packet from the
234 * "wire" on the specified device.
236 static int netvsc_recv_callback(struct hv_device *device_obj,
237 struct hv_netvsc_packet *packet)
239 struct net_device *net = dev_get_drvdata(&device_obj->device);
246 netdev_err(net, "got receive callback but net device"
247 " not initialized yet\n");
251 /* Allocate a skb - TODO direct I/O to pages? */
252 skb = netdev_alloc_skb_ip_align(net, packet->total_data_buflen);
253 if (unlikely(!skb)) {
254 ++net->stats.rx_dropped;
258 /* for kmap_atomic */
259 local_irq_save(flags);
262 * Copy to skb. This copy is needed here since the memory pointed by
263 * hv_netvsc_packet cannot be deallocated
265 for (i = 0; i < packet->page_buf_cnt; i++) {
266 data = kmap_atomic(pfn_to_page(packet->page_buf[i].pfn),
268 data = (void *)(unsigned long)data +
269 packet->page_buf[i].offset;
271 memcpy(skb_put(skb, packet->page_buf[i].len), data,
272 packet->page_buf[i].len);
274 kunmap_atomic((void *)((unsigned long)data -
275 packet->page_buf[i].offset), KM_IRQ1);
278 local_irq_restore(flags);
280 skb->protocol = eth_type_trans(skb, net);
281 skb->ip_summed = CHECKSUM_NONE;
283 net->stats.rx_packets++;
284 net->stats.rx_bytes += skb->len;
287 * Pass the skb back up. Network stack will deallocate the skb when it
296 static void netvsc_get_drvinfo(struct net_device *net,
297 struct ethtool_drvinfo *info)
299 strcpy(info->driver, "hv_netvsc");
300 strcpy(info->version, HV_DRV_VERSION);
301 strcpy(info->fw_version, "N/A");
304 static const struct ethtool_ops ethtool_ops = {
305 .get_drvinfo = netvsc_get_drvinfo,
306 .get_link = ethtool_op_get_link,
309 static const struct net_device_ops device_ops = {
310 .ndo_open = netvsc_open,
311 .ndo_stop = netvsc_close,
312 .ndo_start_xmit = netvsc_start_xmit,
313 .ndo_set_multicast_list = netvsc_set_multicast_list,
314 .ndo_change_mtu = eth_change_mtu,
315 .ndo_validate_addr = eth_validate_addr,
316 .ndo_set_mac_address = eth_mac_addr,
320 * Send GARP packet to network peers after migrations.
321 * After Quick Migration, the network is not immediately operational in the
322 * current context when receiving RNDIS_STATUS_MEDIA_CONNECT event. So, add
323 * another netif_notify_peers() into a scheduled work, otherwise GARP packet
324 * will not be sent after quick migration, and cause network disconnection.
326 static void netvsc_send_garp(struct work_struct *w)
328 struct net_device_context *ndev_ctx;
329 struct net_device *net;
332 ndev_ctx = container_of(w, struct net_device_context, work);
333 net = dev_get_drvdata(&ndev_ctx->device_ctx->device);
334 netif_notify_peers(net);
338 static int netvsc_probe(struct hv_device *dev)
340 struct net_device *net = NULL;
341 struct net_device_context *net_device_ctx;
342 struct netvsc_device_info device_info;
345 net = alloc_etherdev(sizeof(struct net_device_context));
349 /* Set initial state */
350 netif_carrier_off(net);
352 net_device_ctx = netdev_priv(net);
353 net_device_ctx->device_ctx = dev;
354 net_device_ctx->avail = ring_size;
355 dev_set_drvdata(&dev->device, net);
356 INIT_WORK(&net_device_ctx->work, netvsc_send_garp);
358 /* Notify the netvsc driver of the new device */
359 ret = rndis_filte_device_add(dev, &device_info);
362 dev_set_drvdata(&dev->device, NULL);
364 netdev_err(net, "unable to add netvsc device (ret %d)\n", ret);
369 * If carrier is still off ie we did not get a link status callback,
370 * update it if necessary
373 * FIXME: We should use a atomic or test/set instead to avoid getting
374 * out of sync with the device's link status
376 if (!netif_carrier_ok(net))
377 if (!device_info.link_state)
378 netif_carrier_on(net);
380 memcpy(net->dev_addr, device_info.mac_adr, ETH_ALEN);
382 net->netdev_ops = &device_ops;
384 /* TODO: Add GSO and Checksum offload */
385 net->hw_features = NETIF_F_SG;
386 net->features = NETIF_F_SG;
388 SET_ETHTOOL_OPS(net, ðtool_ops);
389 SET_NETDEV_DEV(net, &dev->device);
391 ret = register_netdev(net);
393 /* Remove the device and release the resource */
394 rndis_filter_device_remove(dev);
401 static int netvsc_remove(struct hv_device *dev)
403 struct net_device *net = dev_get_drvdata(&dev->device);
407 dev_err(&dev->device, "No net device to remove\n");
411 /* Stop outbound asap */
412 netif_stop_queue(net);
413 /* netif_carrier_off(net); */
415 unregister_netdev(net);
418 * Call to the vsc driver to let it know that the device is being
421 ret = rndis_filter_device_remove(dev);
424 netdev_err(net, "unable to remove vsc device (ret %d)\n", ret);
431 /* The one and only one */
432 static struct netvsc_driver netvsc_drv = {
433 .base.probe = netvsc_probe,
434 .base.remove = netvsc_remove,
437 static void netvsc_drv_exit(void)
439 vmbus_child_driver_unregister(&netvsc_drv.base.driver);
442 static int netvsc_drv_init(int (*drv_init)(struct hv_driver *drv))
444 struct netvsc_driver *net_drv_obj = &netvsc_drv;
445 struct hv_driver *drv = &netvsc_drv.base;
448 net_drv_obj->ring_buf_size = ring_size * PAGE_SIZE;
449 net_drv_obj->recv_cb = netvsc_recv_callback;
450 net_drv_obj->link_status_change = netvsc_linkstatus_callback;
452 /* Callback to client driver to complete the initialization */
453 drv_init(&net_drv_obj->base);
455 drv->driver.name = net_drv_obj->base.name;
457 /* The driver belongs to vmbus */
458 ret = vmbus_child_driver_register(&drv->driver);
463 static const struct dmi_system_id __initconst
464 hv_netvsc_dmi_table[] __maybe_unused = {
468 DMI_MATCH(DMI_SYS_VENDOR, "Microsoft Corporation"),
469 DMI_MATCH(DMI_PRODUCT_NAME, "Virtual Machine"),
470 DMI_MATCH(DMI_BOARD_NAME, "Virtual Machine"),
475 MODULE_DEVICE_TABLE(dmi, hv_netvsc_dmi_table);
477 static int __init netvsc_init(void)
479 pr_info("initializing....");
481 if (!dmi_check_system(hv_netvsc_dmi_table))
484 return netvsc_drv_init(netvsc_initialize);
487 static void __exit netvsc_exit(void)
492 static const struct pci_device_id __initconst
493 hv_netvsc_pci_table[] __maybe_unused = {
494 { PCI_DEVICE(0x1414, 0x5353) }, /* VGA compatible controller */
497 MODULE_DEVICE_TABLE(pci, hv_netvsc_pci_table);
499 MODULE_LICENSE("GPL");
500 MODULE_VERSION(HV_DRV_VERSION);
501 MODULE_DESCRIPTION("Microsoft Hyper-V network driver");
503 module_init(netvsc_init);
504 module_exit(netvsc_exit);