1 // SPDX-License-Identifier: GPL-2.0
2 /* ldmvsw.c: Sun4v LDOM Virtual Switch Driver.
3 *
4 * Copyright (C) 2016-2017 Oracle. All rights reserved.
5 */
6
7 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
8
9 #include <linux/delay.h>
10 #include <linux/etherdevice.h>
11 #include <linux/ethtool.h>
12 #include <linux/highmem.h>
13 #include <linux/if_vlan.h>
14 #include <linux/init.h>
15 #include <linux/kernel.h>
16 #include <linux/module.h>
17 #include <linux/mutex.h>
18 #include <linux/netdevice.h>
19 #include <linux/slab.h>
20 #include <linux/types.h>
21
22 #if defined(CONFIG_IPV6)
23 #include <linux/icmpv6.h>
24 #endif
25
26 #include <net/ip.h>
27 #include <net/icmp.h>
28 #include <net/route.h>
29
30 #include <asm/vio.h>
31 #include <asm/ldc.h>
32
33 /* This driver makes use of the common code in sunvnet_common.c */
34 #include "sunvnet_common.h"
35
36 /* Length of time before we decide the hardware is hung,
37 * and dev->tx_timeout() should be called to fix the problem.
38 */
39 #define VSW_TX_TIMEOUT (10 * HZ)
40
41 /* Static HW Addr used for the network interfaces representing vsw ports */
42 static u8 vsw_port_hwaddr[ETH_ALEN] = {0xFE, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF};
43
44 #define DRV_MODULE_NAME "ldmvsw"
45 #define DRV_MODULE_VERSION "1.2"
46 #define DRV_MODULE_RELDATE "March 4, 2017"
47
48 static char version[] =
49 DRV_MODULE_NAME " " DRV_MODULE_VERSION " (" DRV_MODULE_RELDATE ")";
50 MODULE_AUTHOR("Oracle");
51 MODULE_DESCRIPTION("Sun4v LDOM Virtual Switch Driver");
52 MODULE_LICENSE("GPL");
53 MODULE_VERSION(DRV_MODULE_VERSION);
54
55 /* Ordered from largest major to lowest */
56 static struct vio_version vsw_versions[] = {
57 { .major = 1, .minor = 8 },
58 { .major = 1, .minor = 7 },
59 { .major = 1, .minor = 6 },
60 { .major = 1, .minor = 0 },
61 };
62
vsw_get_drvinfo(struct net_device * dev,struct ethtool_drvinfo * info)63 static void vsw_get_drvinfo(struct net_device *dev,
64 struct ethtool_drvinfo *info)
65 {
66 strlcpy(info->driver, DRV_MODULE_NAME, sizeof(info->driver));
67 strlcpy(info->version, DRV_MODULE_VERSION, sizeof(info->version));
68 }
69
vsw_get_msglevel(struct net_device * dev)70 static u32 vsw_get_msglevel(struct net_device *dev)
71 {
72 struct vnet_port *port = netdev_priv(dev);
73
74 return port->vp->msg_enable;
75 }
76
vsw_set_msglevel(struct net_device * dev,u32 value)77 static void vsw_set_msglevel(struct net_device *dev, u32 value)
78 {
79 struct vnet_port *port = netdev_priv(dev);
80
81 port->vp->msg_enable = value;
82 }
83
84 static const struct ethtool_ops vsw_ethtool_ops = {
85 .get_drvinfo = vsw_get_drvinfo,
86 .get_msglevel = vsw_get_msglevel,
87 .set_msglevel = vsw_set_msglevel,
88 .get_link = ethtool_op_get_link,
89 };
90
91 static LIST_HEAD(vnet_list);
92 static DEFINE_MUTEX(vnet_list_mutex);
93
94 /* func arg to vnet_start_xmit_common() to get the proper tx port */
vsw_tx_port_find(struct sk_buff * skb,struct net_device * dev)95 static struct vnet_port *vsw_tx_port_find(struct sk_buff *skb,
96 struct net_device *dev)
97 {
98 struct vnet_port *port = netdev_priv(dev);
99
100 return port;
101 }
102
vsw_select_queue(struct net_device * dev,struct sk_buff * skb,struct net_device * sb_dev)103 static u16 vsw_select_queue(struct net_device *dev, struct sk_buff *skb,
104 struct net_device *sb_dev)
105 {
106 struct vnet_port *port = netdev_priv(dev);
107
108 if (!port)
109 return 0;
110
111 return port->q_index;
112 }
113
114 /* Wrappers to common functions */
vsw_start_xmit(struct sk_buff * skb,struct net_device * dev)115 static netdev_tx_t vsw_start_xmit(struct sk_buff *skb, struct net_device *dev)
116 {
117 return sunvnet_start_xmit_common(skb, dev, vsw_tx_port_find);
118 }
119
vsw_set_rx_mode(struct net_device * dev)120 static void vsw_set_rx_mode(struct net_device *dev)
121 {
122 struct vnet_port *port = netdev_priv(dev);
123
124 return sunvnet_set_rx_mode_common(dev, port->vp);
125 }
126
ldmvsw_open(struct net_device * dev)127 int ldmvsw_open(struct net_device *dev)
128 {
129 struct vnet_port *port = netdev_priv(dev);
130 struct vio_driver_state *vio = &port->vio;
131
132 /* reset the channel */
133 vio_link_state_change(vio, LDC_EVENT_RESET);
134 vnet_port_reset(port);
135 vio_port_up(vio);
136
137 return 0;
138 }
139 EXPORT_SYMBOL_GPL(ldmvsw_open);
140
141 #ifdef CONFIG_NET_POLL_CONTROLLER
vsw_poll_controller(struct net_device * dev)142 static void vsw_poll_controller(struct net_device *dev)
143 {
144 struct vnet_port *port = netdev_priv(dev);
145
146 return sunvnet_poll_controller_common(dev, port->vp);
147 }
148 #endif
149
150 static const struct net_device_ops vsw_ops = {
151 .ndo_open = ldmvsw_open,
152 .ndo_stop = sunvnet_close_common,
153 .ndo_set_rx_mode = vsw_set_rx_mode,
154 .ndo_set_mac_address = sunvnet_set_mac_addr_common,
155 .ndo_validate_addr = eth_validate_addr,
156 .ndo_tx_timeout = sunvnet_tx_timeout_common,
157 .ndo_start_xmit = vsw_start_xmit,
158 .ndo_select_queue = vsw_select_queue,
159 #ifdef CONFIG_NET_POLL_CONTROLLER
160 .ndo_poll_controller = vsw_poll_controller,
161 #endif
162 };
163
164 static const char *local_mac_prop = "local-mac-address";
165 static const char *cfg_handle_prop = "cfg-handle";
166
vsw_get_vnet(struct mdesc_handle * hp,u64 port_node,u64 * handle)167 static struct vnet *vsw_get_vnet(struct mdesc_handle *hp,
168 u64 port_node,
169 u64 *handle)
170 {
171 struct vnet *vp;
172 struct vnet *iter;
173 const u64 *local_mac = NULL;
174 const u64 *cfghandle = NULL;
175 u64 a;
176
177 /* Get the parent virtual-network-switch macaddr and cfghandle */
178 mdesc_for_each_arc(a, hp, port_node, MDESC_ARC_TYPE_BACK) {
179 u64 target = mdesc_arc_target(hp, a);
180 const char *name;
181
182 name = mdesc_get_property(hp, target, "name", NULL);
183 if (!name || strcmp(name, "virtual-network-switch"))
184 continue;
185
186 local_mac = mdesc_get_property(hp, target,
187 local_mac_prop, NULL);
188 cfghandle = mdesc_get_property(hp, target,
189 cfg_handle_prop, NULL);
190 break;
191 }
192 if (!local_mac || !cfghandle)
193 return ERR_PTR(-ENODEV);
194
195 /* find or create associated vnet */
196 vp = NULL;
197 mutex_lock(&vnet_list_mutex);
198 list_for_each_entry(iter, &vnet_list, list) {
199 if (iter->local_mac == *local_mac) {
200 vp = iter;
201 break;
202 }
203 }
204
205 if (!vp) {
206 vp = kzalloc(sizeof(*vp), GFP_KERNEL);
207 if (unlikely(!vp)) {
208 mutex_unlock(&vnet_list_mutex);
209 return ERR_PTR(-ENOMEM);
210 }
211
212 spin_lock_init(&vp->lock);
213 INIT_LIST_HEAD(&vp->port_list);
214 INIT_LIST_HEAD(&vp->list);
215 vp->local_mac = *local_mac;
216 list_add(&vp->list, &vnet_list);
217 }
218
219 mutex_unlock(&vnet_list_mutex);
220
221 *handle = (u64)*cfghandle;
222
223 return vp;
224 }
225
vsw_alloc_netdev(u8 hwaddr[],struct vio_dev * vdev,u64 handle,u64 port_id)226 static struct net_device *vsw_alloc_netdev(u8 hwaddr[],
227 struct vio_dev *vdev,
228 u64 handle,
229 u64 port_id)
230 {
231 struct net_device *dev;
232 struct vnet_port *port;
233 int i;
234
235 dev = alloc_etherdev_mqs(sizeof(*port), VNET_MAX_TXQS, 1);
236 if (!dev)
237 return ERR_PTR(-ENOMEM);
238 dev->needed_headroom = VNET_PACKET_SKIP + 8;
239 dev->needed_tailroom = 8;
240
241 for (i = 0; i < ETH_ALEN; i++) {
242 dev->dev_addr[i] = hwaddr[i];
243 dev->perm_addr[i] = dev->dev_addr[i];
244 }
245
246 sprintf(dev->name, "vif%d.%d", (int)handle, (int)port_id);
247
248 dev->netdev_ops = &vsw_ops;
249 dev->ethtool_ops = &vsw_ethtool_ops;
250 dev->watchdog_timeo = VSW_TX_TIMEOUT;
251
252 dev->hw_features = NETIF_F_HW_CSUM | NETIF_F_SG;
253 dev->features = dev->hw_features;
254
255 /* MTU range: 68 - 65535 */
256 dev->min_mtu = ETH_MIN_MTU;
257 dev->max_mtu = VNET_MAX_MTU;
258
259 SET_NETDEV_DEV(dev, &vdev->dev);
260
261 return dev;
262 }
263
264 static struct ldc_channel_config vsw_ldc_cfg = {
265 .event = sunvnet_event_common,
266 .mtu = 64,
267 .mode = LDC_MODE_UNRELIABLE,
268 };
269
270 static struct vio_driver_ops vsw_vio_ops = {
271 .send_attr = sunvnet_send_attr_common,
272 .handle_attr = sunvnet_handle_attr_common,
273 .handshake_complete = sunvnet_handshake_complete_common,
274 };
275
276 static const char *remote_macaddr_prop = "remote-mac-address";
277 static const char *id_prop = "id";
278
vsw_port_probe(struct vio_dev * vdev,const struct vio_device_id * id)279 static int vsw_port_probe(struct vio_dev *vdev, const struct vio_device_id *id)
280 {
281 struct mdesc_handle *hp;
282 struct vnet_port *port;
283 unsigned long flags;
284 struct vnet *vp;
285 struct net_device *dev;
286 const u64 *rmac;
287 int len, i, err;
288 const u64 *port_id;
289 u64 handle;
290
291 hp = mdesc_grab();
292
293 rmac = mdesc_get_property(hp, vdev->mp, remote_macaddr_prop, &len);
294 err = -ENODEV;
295 if (!rmac) {
296 pr_err("Port lacks %s property\n", remote_macaddr_prop);
297 mdesc_release(hp);
298 return err;
299 }
300
301 port_id = mdesc_get_property(hp, vdev->mp, id_prop, NULL);
302 err = -ENODEV;
303 if (!port_id) {
304 pr_err("Port lacks %s property\n", id_prop);
305 mdesc_release(hp);
306 return err;
307 }
308
309 /* Get (or create) the vnet associated with this port */
310 vp = vsw_get_vnet(hp, vdev->mp, &handle);
311 if (IS_ERR(vp)) {
312 err = PTR_ERR(vp);
313 pr_err("Failed to get vnet for vsw-port\n");
314 mdesc_release(hp);
315 return err;
316 }
317
318 mdesc_release(hp);
319
320 dev = vsw_alloc_netdev(vsw_port_hwaddr, vdev, handle, *port_id);
321 if (IS_ERR(dev)) {
322 err = PTR_ERR(dev);
323 pr_err("Failed to alloc netdev for vsw-port\n");
324 return err;
325 }
326
327 port = netdev_priv(dev);
328
329 INIT_LIST_HEAD(&port->list);
330
331 for (i = 0; i < ETH_ALEN; i++)
332 port->raddr[i] = (*rmac >> (5 - i) * 8) & 0xff;
333
334 port->vp = vp;
335 port->dev = dev;
336 port->switch_port = 1;
337 port->tso = false; /* no tso in vsw, misbehaves in bridge */
338 port->tsolen = 0;
339
340 /* Mark the port as belonging to ldmvsw which directs the
341 * the common code to use the net_device in the vnet_port
342 * rather than the net_device in the vnet (which is used
343 * by sunvnet). This bit is used by the VNET_PORT_TO_NET_DEVICE
344 * macro.
345 */
346 port->vsw = 1;
347
348 err = vio_driver_init(&port->vio, vdev, VDEV_NETWORK,
349 vsw_versions, ARRAY_SIZE(vsw_versions),
350 &vsw_vio_ops, dev->name);
351 if (err)
352 goto err_out_free_dev;
353
354 err = vio_ldc_alloc(&port->vio, &vsw_ldc_cfg, port);
355 if (err)
356 goto err_out_free_dev;
357
358 dev_set_drvdata(&vdev->dev, port);
359
360 netif_napi_add(dev, &port->napi, sunvnet_poll_common,
361 NAPI_POLL_WEIGHT);
362
363 spin_lock_irqsave(&vp->lock, flags);
364 list_add_rcu(&port->list, &vp->port_list);
365 spin_unlock_irqrestore(&vp->lock, flags);
366
367 timer_setup(&port->clean_timer, sunvnet_clean_timer_expire_common, 0);
368
369 err = register_netdev(dev);
370 if (err) {
371 pr_err("Cannot register net device, aborting\n");
372 goto err_out_del_timer;
373 }
374
375 spin_lock_irqsave(&vp->lock, flags);
376 sunvnet_port_add_txq_common(port);
377 spin_unlock_irqrestore(&vp->lock, flags);
378
379 napi_enable(&port->napi);
380 vio_port_up(&port->vio);
381
382 /* assure no carrier until we receive an LDC_EVENT_UP,
383 * even if the vsw config script tries to force us up
384 */
385 netif_carrier_off(dev);
386
387 netdev_info(dev, "LDOM vsw-port %pM\n", dev->dev_addr);
388
389 pr_info("%s: PORT ( remote-mac %pM%s )\n", dev->name,
390 port->raddr, " switch-port");
391
392 return 0;
393
394 err_out_del_timer:
395 del_timer_sync(&port->clean_timer);
396 list_del_rcu(&port->list);
397 synchronize_rcu();
398 netif_napi_del(&port->napi);
399 dev_set_drvdata(&vdev->dev, NULL);
400 vio_ldc_free(&port->vio);
401
402 err_out_free_dev:
403 free_netdev(dev);
404 return err;
405 }
406
vsw_port_remove(struct vio_dev * vdev)407 static int vsw_port_remove(struct vio_dev *vdev)
408 {
409 struct vnet_port *port = dev_get_drvdata(&vdev->dev);
410 unsigned long flags;
411
412 if (port) {
413 del_timer_sync(&port->vio.timer);
414 del_timer_sync(&port->clean_timer);
415
416 napi_disable(&port->napi);
417 unregister_netdev(port->dev);
418
419 list_del_rcu(&port->list);
420
421 synchronize_rcu();
422 spin_lock_irqsave(&port->vp->lock, flags);
423 sunvnet_port_rm_txq_common(port);
424 spin_unlock_irqrestore(&port->vp->lock, flags);
425 netif_napi_del(&port->napi);
426 sunvnet_port_free_tx_bufs_common(port);
427 vio_ldc_free(&port->vio);
428
429 dev_set_drvdata(&vdev->dev, NULL);
430
431 free_netdev(port->dev);
432 }
433
434 return 0;
435 }
436
vsw_cleanup(void)437 static void vsw_cleanup(void)
438 {
439 struct vnet *vp;
440
441 /* just need to free up the vnet list */
442 mutex_lock(&vnet_list_mutex);
443 while (!list_empty(&vnet_list)) {
444 vp = list_first_entry(&vnet_list, struct vnet, list);
445 list_del(&vp->list);
446 /* vio_unregister_driver() should have cleaned up port_list */
447 if (!list_empty(&vp->port_list))
448 pr_err("Ports not removed by VIO subsystem!\n");
449 kfree(vp);
450 }
451 mutex_unlock(&vnet_list_mutex);
452 }
453
454 static const struct vio_device_id vsw_port_match[] = {
455 {
456 .type = "vsw-port",
457 },
458 {},
459 };
460 MODULE_DEVICE_TABLE(vio, vsw_port_match);
461
462 static struct vio_driver vsw_port_driver = {
463 .id_table = vsw_port_match,
464 .probe = vsw_port_probe,
465 .remove = vsw_port_remove,
466 .name = "vsw_port",
467 };
468
vsw_init(void)469 static int __init vsw_init(void)
470 {
471 pr_info("%s\n", version);
472 return vio_register_driver(&vsw_port_driver);
473 }
474
vsw_exit(void)475 static void __exit vsw_exit(void)
476 {
477 vio_unregister_driver(&vsw_port_driver);
478 vsw_cleanup();
479 }
480
481 module_init(vsw_init);
482 module_exit(vsw_exit);
483