blob: 93394d76587ab749bdd1adb7b22b03fb3e9aaec0 [file] [log] [blame]
Linus Torvalds1da177e2005-04-16 15:20:36 -07001/* File veth.c created by Kyle A. Lucke on Mon Aug 7 2000. */
2/*
3 * IBM eServer iSeries Virtual Ethernet Device Driver
4 * Copyright (C) 2001 Kyle A. Lucke (klucke@us.ibm.com), IBM Corp.
5 * Substantially cleaned up by:
6 * Copyright (C) 2003 David Gibson <dwg@au1.ibm.com>, IBM Corporation.
Michael Ellerman269cd382005-09-06 10:39:38 +10007 * Copyright (C) 2004-2005 Michael Ellerman, IBM Corporation.
Linus Torvalds1da177e2005-04-16 15:20:36 -07008 *
9 * This program is free software; you can redistribute it and/or
10 * modify it under the terms of the GNU General Public License as
11 * published by the Free Software Foundation; either version 2 of the
12 * License, or (at your option) any later version.
13 *
14 * This program is distributed in the hope that it will be useful, but
15 * WITHOUT ANY WARRANTY; without even the implied warranty of
16 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
17 * General Public License for more details.
18 *
19 * You should have received a copy of the GNU General Public License
20 * along with this program; if not, write to the Free Software
21 * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307
22 * USA
23 *
24 *
25 * This module implements the virtual ethernet device for iSeries LPAR
26 * Linux. It uses hypervisor message passing to implement an
27 * ethernet-like network device communicating between partitions on
28 * the iSeries.
29 *
30 * The iSeries LPAR hypervisor currently allows for up to 16 different
31 * virtual ethernets. These are all dynamically configurable on
32 * OS/400 partitions, but dynamic configuration is not supported under
33 * Linux yet. An ethXX network device will be created for each
34 * virtual ethernet this partition is connected to.
35 *
36 * - This driver is responsible for routing packets to and from other
37 * partitions. The MAC addresses used by the virtual ethernets
38 * contains meaning and must not be modified.
39 *
40 * - Having 2 virtual ethernets to the same remote partition DOES NOT
41 * double the available bandwidth. The 2 devices will share the
42 * available hypervisor bandwidth.
43 *
44 * - If you send a packet to your own mac address, it will just be
45 * dropped, you won't get it on the receive side.
46 *
47 * - Multicast is implemented by sending the frame frame to every
48 * other partition. It is the responsibility of the receiving
49 * partition to filter the addresses desired.
50 *
51 * Tunable parameters:
52 *
53 * VETH_NUMBUFFERS: This compile time option defaults to 120. It
54 * controls how much memory Linux will allocate per remote partition
55 * it is communicating with. It can be thought of as the maximum
56 * number of packets outstanding to a remote partition at a time.
57 */
58
59#include <linux/config.h>
60#include <linux/module.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070061#include <linux/types.h>
62#include <linux/errno.h>
63#include <linux/ioport.h>
64#include <linux/kernel.h>
65#include <linux/netdevice.h>
66#include <linux/etherdevice.h>
67#include <linux/skbuff.h>
68#include <linux/init.h>
69#include <linux/delay.h>
70#include <linux/mm.h>
71#include <linux/ethtool.h>
Stephen Rothwell07fb3f42006-05-15 13:39:47 +100072#include <linux/if_ether.h>
Stephen Rothwell426c1a12005-10-14 14:51:42 +100073
74#include <asm/abs_addr.h>
Kelly Dalybbc8b622005-11-02 15:10:38 +110075#include <asm/iseries/mf.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070076#include <asm/uaccess.h>
77
Kelly Daly15b17182005-11-02 11:55:28 +110078#include <asm/iseries/hv_lp_config.h>
Kelly Daly1ec65d72005-11-02 13:46:07 +110079#include <asm/iseries/hv_types.h>
Kelly Dalye45423e2005-11-02 12:08:31 +110080#include <asm/iseries/hv_lp_event.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070081#include <asm/iommu.h>
82#include <asm/vio.h>
83
Michael Ellerman61a3c692005-09-01 11:28:57 +100084#undef DEBUG
85
Linus Torvalds1da177e2005-04-16 15:20:36 -070086MODULE_AUTHOR("Kyle Lucke <klucke@us.ibm.com>");
87MODULE_DESCRIPTION("iSeries Virtual ethernet driver");
88MODULE_LICENSE("GPL");
89
Michael Ellerman59f17ae2005-09-01 11:29:27 +100090#define VETH_EVENT_CAP (0)
91#define VETH_EVENT_FRAMES (1)
92#define VETH_EVENT_MONITOR (2)
93#define VETH_EVENT_FRAMES_ACK (3)
Michael Ellerman642d1a42005-09-01 11:29:25 +100094
95#define VETH_MAX_ACKS_PER_MSG (20)
96#define VETH_MAX_FRAMES_PER_MSG (6)
97
Michael Ellerman59f17ae2005-09-01 11:29:27 +100098struct veth_frames_data {
Michael Ellerman642d1a42005-09-01 11:29:25 +100099 u32 addr[VETH_MAX_FRAMES_PER_MSG];
100 u16 len[VETH_MAX_FRAMES_PER_MSG];
101 u32 eofmask;
102};
103#define VETH_EOF_SHIFT (32-VETH_MAX_FRAMES_PER_MSG)
104
Michael Ellerman59f17ae2005-09-01 11:29:27 +1000105struct veth_frames_ack_data {
Michael Ellerman642d1a42005-09-01 11:29:25 +1000106 u16 token[VETH_MAX_ACKS_PER_MSG];
107};
108
Michael Ellerman59f17ae2005-09-01 11:29:27 +1000109struct veth_cap_data {
Michael Ellerman642d1a42005-09-01 11:29:25 +1000110 u8 caps_version;
111 u8 rsvd1;
112 u16 num_buffers;
113 u16 ack_threshold;
114 u16 rsvd2;
115 u32 ack_timeout;
116 u32 rsvd3;
117 u64 rsvd4[3];
118};
119
Michael Ellerman59f17ae2005-09-01 11:29:27 +1000120struct veth_lpevent {
Michael Ellerman642d1a42005-09-01 11:29:25 +1000121 struct HvLpEvent base_event;
122 union {
Michael Ellerman59f17ae2005-09-01 11:29:27 +1000123 struct veth_cap_data caps_data;
124 struct veth_frames_data frames_data;
125 struct veth_frames_ack_data frames_ack_data;
Michael Ellerman642d1a42005-09-01 11:29:25 +1000126 } u;
127
128};
129
Michael Ellermanee05f032005-09-01 11:29:29 +1000130#define DRV_NAME "iseries_veth"
131#define DRV_VERSION "2.0"
132
Linus Torvalds1da177e2005-04-16 15:20:36 -0700133#define VETH_NUMBUFFERS (120)
134#define VETH_ACKTIMEOUT (1000000) /* microseconds */
135#define VETH_MAX_MCAST (12)
136
137#define VETH_MAX_MTU (9000)
138
139#if VETH_NUMBUFFERS < 10
140#define ACK_THRESHOLD (1)
141#elif VETH_NUMBUFFERS < 20
142#define ACK_THRESHOLD (4)
143#elif VETH_NUMBUFFERS < 40
144#define ACK_THRESHOLD (10)
145#else
146#define ACK_THRESHOLD (20)
147#endif
148
149#define VETH_STATE_SHUTDOWN (0x0001)
150#define VETH_STATE_OPEN (0x0002)
151#define VETH_STATE_RESET (0x0004)
152#define VETH_STATE_SENTMON (0x0008)
153#define VETH_STATE_SENTCAPS (0x0010)
154#define VETH_STATE_GOTCAPACK (0x0020)
155#define VETH_STATE_GOTCAPS (0x0040)
156#define VETH_STATE_SENTCAPACK (0x0080)
157#define VETH_STATE_READY (0x0100)
158
159struct veth_msg {
160 struct veth_msg *next;
Michael Ellerman59f17ae2005-09-01 11:29:27 +1000161 struct veth_frames_data data;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700162 int token;
Michael Ellermanb08bd5c2005-09-01 11:29:06 +1000163 int in_use;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700164 struct sk_buff *skb;
165 struct device *dev;
166};
167
168struct veth_lpar_connection {
169 HvLpIndex remote_lp;
170 struct work_struct statemachine_wq;
171 struct veth_msg *msgs;
172 int num_events;
Michael Ellerman59f17ae2005-09-01 11:29:27 +1000173 struct veth_cap_data local_caps;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700174
Michael Ellermanf0c129c2005-09-01 11:29:09 +1000175 struct kobject kobject;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700176 struct timer_list ack_timer;
177
Michael Ellerman24562ff2005-09-01 11:29:17 +1000178 struct timer_list reset_timer;
179 unsigned int reset_timeout;
180 unsigned long last_contact;
181 int outstanding_tx;
182
Linus Torvalds1da177e2005-04-16 15:20:36 -0700183 spinlock_t lock;
184 unsigned long state;
185 HvLpInstanceId src_inst;
186 HvLpInstanceId dst_inst;
Michael Ellerman59f17ae2005-09-01 11:29:27 +1000187 struct veth_lpevent cap_event, cap_ack_event;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700188 u16 pending_acks[VETH_MAX_ACKS_PER_MSG];
189 u32 num_pending_acks;
190
191 int num_ack_events;
Michael Ellerman59f17ae2005-09-01 11:29:27 +1000192 struct veth_cap_data remote_caps;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700193 u32 ack_timeout;
194
Linus Torvalds1da177e2005-04-16 15:20:36 -0700195 struct veth_msg *msg_stack_head;
196};
197
198struct veth_port {
199 struct device *dev;
200 struct net_device_stats stats;
201 u64 mac_addr;
202 HvLpIndexMap lpar_map;
203
Michael Ellermane0808492005-09-01 11:29:18 +1000204 /* queue_lock protects the stopped_map and dev's queue. */
205 spinlock_t queue_lock;
206 HvLpIndexMap stopped_map;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700207
Michael Ellermane0808492005-09-01 11:29:18 +1000208 /* mcast_gate protects promiscuous, num_mcast & mcast_addr. */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700209 rwlock_t mcast_gate;
210 int promiscuous;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700211 int num_mcast;
212 u64 mcast_addr[VETH_MAX_MCAST];
Michael Ellerman07a5c172005-09-01 11:29:21 +1000213
214 struct kobject kobject;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700215};
216
217static HvLpIndex this_lp;
218static struct veth_lpar_connection *veth_cnx[HVMAXARCHITECTEDLPS]; /* = 0 */
219static struct net_device *veth_dev[HVMAXARCHITECTEDVIRTUALLANS]; /* = 0 */
220
221static int veth_start_xmit(struct sk_buff *skb, struct net_device *dev);
222static void veth_recycle_msg(struct veth_lpar_connection *, struct veth_msg *);
Michael Ellermane0808492005-09-01 11:29:18 +1000223static void veth_wake_queues(struct veth_lpar_connection *cnx);
224static void veth_stop_queues(struct veth_lpar_connection *cnx);
Michael Ellerman59f17ae2005-09-01 11:29:27 +1000225static void veth_receive(struct veth_lpar_connection *, struct veth_lpevent *);
Michael Ellermanf0c129c2005-09-01 11:29:09 +1000226static void veth_release_connection(struct kobject *kobject);
Michael Ellerman24562ff2005-09-01 11:29:17 +1000227static void veth_timed_ack(unsigned long ptr);
228static void veth_timed_reset(unsigned long ptr);
Michael Ellermanf0c129c2005-09-01 11:29:09 +1000229
Linus Torvalds1da177e2005-04-16 15:20:36 -0700230/*
231 * Utility functions
232 */
233
Michael Ellerman61a3c692005-09-01 11:28:57 +1000234#define veth_info(fmt, args...) \
Michael Ellermanee05f032005-09-01 11:29:29 +1000235 printk(KERN_INFO DRV_NAME ": " fmt, ## args)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700236
237#define veth_error(fmt, args...) \
Michael Ellermanee05f032005-09-01 11:29:29 +1000238 printk(KERN_ERR DRV_NAME ": Error: " fmt, ## args)
Michael Ellerman61a3c692005-09-01 11:28:57 +1000239
240#ifdef DEBUG
241#define veth_debug(fmt, args...) \
Michael Ellermanee05f032005-09-01 11:29:29 +1000242 printk(KERN_DEBUG DRV_NAME ": " fmt, ## args)
Michael Ellerman61a3c692005-09-01 11:28:57 +1000243#else
244#define veth_debug(fmt, args...) do {} while (0)
245#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -0700246
Michael Ellermand7893dd2005-09-01 11:29:05 +1000247/* You must hold the connection's lock when you call this function. */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700248static inline void veth_stack_push(struct veth_lpar_connection *cnx,
249 struct veth_msg *msg)
250{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700251 msg->next = cnx->msg_stack_head;
252 cnx->msg_stack_head = msg;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700253}
254
Michael Ellermand7893dd2005-09-01 11:29:05 +1000255/* You must hold the connection's lock when you call this function. */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700256static inline struct veth_msg *veth_stack_pop(struct veth_lpar_connection *cnx)
257{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700258 struct veth_msg *msg;
259
Linus Torvalds1da177e2005-04-16 15:20:36 -0700260 msg = cnx->msg_stack_head;
261 if (msg)
262 cnx->msg_stack_head = cnx->msg_stack_head->next;
Michael Ellermand7893dd2005-09-01 11:29:05 +1000263
Linus Torvalds1da177e2005-04-16 15:20:36 -0700264 return msg;
265}
266
Michael Ellermane0808492005-09-01 11:29:18 +1000267/* You must hold the connection's lock when you call this function. */
268static inline int veth_stack_is_empty(struct veth_lpar_connection *cnx)
269{
270 return cnx->msg_stack_head == NULL;
271}
272
Linus Torvalds1da177e2005-04-16 15:20:36 -0700273static inline HvLpEvent_Rc
274veth_signalevent(struct veth_lpar_connection *cnx, u16 subtype,
275 HvLpEvent_AckInd ackind, HvLpEvent_AckType acktype,
276 u64 token,
277 u64 data1, u64 data2, u64 data3, u64 data4, u64 data5)
278{
279 return HvCallEvent_signalLpEventFast(cnx->remote_lp,
280 HvLpEvent_Type_VirtualLan,
281 subtype, ackind, acktype,
282 cnx->src_inst,
283 cnx->dst_inst,
284 token, data1, data2, data3,
285 data4, data5);
286}
287
288static inline HvLpEvent_Rc veth_signaldata(struct veth_lpar_connection *cnx,
289 u16 subtype, u64 token, void *data)
290{
291 u64 *p = (u64 *) data;
292
293 return veth_signalevent(cnx, subtype, HvLpEvent_AckInd_NoAck,
294 HvLpEvent_AckType_ImmediateAck,
295 token, p[0], p[1], p[2], p[3], p[4]);
296}
297
298struct veth_allocation {
299 struct completion c;
300 int num;
301};
302
303static void veth_complete_allocation(void *parm, int number)
304{
305 struct veth_allocation *vc = (struct veth_allocation *)parm;
306
307 vc->num = number;
308 complete(&vc->c);
309}
310
311static int veth_allocate_events(HvLpIndex rlp, int number)
312{
313 struct veth_allocation vc = { COMPLETION_INITIALIZER(vc.c), 0 };
314
315 mf_allocate_lp_events(rlp, HvLpEvent_Type_VirtualLan,
Michael Ellerman59f17ae2005-09-01 11:29:27 +1000316 sizeof(struct veth_lpevent), number,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700317 &veth_complete_allocation, &vc);
318 wait_for_completion(&vc.c);
319
320 return vc.num;
321}
322
323/*
Michael Ellerman76812d8122005-09-01 11:29:20 +1000324 * sysfs support
325 */
326
327struct veth_cnx_attribute {
328 struct attribute attr;
329 ssize_t (*show)(struct veth_lpar_connection *, char *buf);
330 ssize_t (*store)(struct veth_lpar_connection *, const char *buf);
331};
332
333static ssize_t veth_cnx_attribute_show(struct kobject *kobj,
334 struct attribute *attr, char *buf)
335{
336 struct veth_cnx_attribute *cnx_attr;
337 struct veth_lpar_connection *cnx;
338
339 cnx_attr = container_of(attr, struct veth_cnx_attribute, attr);
340 cnx = container_of(kobj, struct veth_lpar_connection, kobject);
341
342 if (!cnx_attr->show)
343 return -EIO;
344
345 return cnx_attr->show(cnx, buf);
346}
347
348#define CUSTOM_CNX_ATTR(_name, _format, _expression) \
349static ssize_t _name##_show(struct veth_lpar_connection *cnx, char *buf)\
350{ \
351 return sprintf(buf, _format, _expression); \
352} \
353struct veth_cnx_attribute veth_cnx_attr_##_name = __ATTR_RO(_name)
354
355#define SIMPLE_CNX_ATTR(_name) \
356 CUSTOM_CNX_ATTR(_name, "%lu\n", (unsigned long)cnx->_name)
357
358SIMPLE_CNX_ATTR(outstanding_tx);
359SIMPLE_CNX_ATTR(remote_lp);
360SIMPLE_CNX_ATTR(num_events);
361SIMPLE_CNX_ATTR(src_inst);
362SIMPLE_CNX_ATTR(dst_inst);
363SIMPLE_CNX_ATTR(num_pending_acks);
364SIMPLE_CNX_ATTR(num_ack_events);
365CUSTOM_CNX_ATTR(ack_timeout, "%d\n", jiffies_to_msecs(cnx->ack_timeout));
366CUSTOM_CNX_ATTR(reset_timeout, "%d\n", jiffies_to_msecs(cnx->reset_timeout));
367CUSTOM_CNX_ATTR(state, "0x%.4lX\n", cnx->state);
368CUSTOM_CNX_ATTR(last_contact, "%d\n", cnx->last_contact ?
369 jiffies_to_msecs(jiffies - cnx->last_contact) : 0);
370
371#define GET_CNX_ATTR(_name) (&veth_cnx_attr_##_name.attr)
372
373static struct attribute *veth_cnx_default_attrs[] = {
374 GET_CNX_ATTR(outstanding_tx),
375 GET_CNX_ATTR(remote_lp),
376 GET_CNX_ATTR(num_events),
377 GET_CNX_ATTR(reset_timeout),
378 GET_CNX_ATTR(last_contact),
379 GET_CNX_ATTR(state),
380 GET_CNX_ATTR(src_inst),
381 GET_CNX_ATTR(dst_inst),
382 GET_CNX_ATTR(num_pending_acks),
383 GET_CNX_ATTR(num_ack_events),
384 GET_CNX_ATTR(ack_timeout),
385 NULL
386};
387
388static struct sysfs_ops veth_cnx_sysfs_ops = {
389 .show = veth_cnx_attribute_show
390};
391
392static struct kobj_type veth_lpar_connection_ktype = {
393 .release = veth_release_connection,
394 .sysfs_ops = &veth_cnx_sysfs_ops,
395 .default_attrs = veth_cnx_default_attrs
396};
397
Michael Ellerman07a5c172005-09-01 11:29:21 +1000398struct veth_port_attribute {
399 struct attribute attr;
400 ssize_t (*show)(struct veth_port *, char *buf);
401 ssize_t (*store)(struct veth_port *, const char *buf);
402};
403
404static ssize_t veth_port_attribute_show(struct kobject *kobj,
405 struct attribute *attr, char *buf)
406{
407 struct veth_port_attribute *port_attr;
408 struct veth_port *port;
409
410 port_attr = container_of(attr, struct veth_port_attribute, attr);
411 port = container_of(kobj, struct veth_port, kobject);
412
413 if (!port_attr->show)
414 return -EIO;
415
416 return port_attr->show(port, buf);
417}
418
419#define CUSTOM_PORT_ATTR(_name, _format, _expression) \
420static ssize_t _name##_show(struct veth_port *port, char *buf) \
421{ \
422 return sprintf(buf, _format, _expression); \
423} \
424struct veth_port_attribute veth_port_attr_##_name = __ATTR_RO(_name)
425
426#define SIMPLE_PORT_ATTR(_name) \
427 CUSTOM_PORT_ATTR(_name, "%lu\n", (unsigned long)port->_name)
428
429SIMPLE_PORT_ATTR(promiscuous);
430SIMPLE_PORT_ATTR(num_mcast);
431CUSTOM_PORT_ATTR(lpar_map, "0x%X\n", port->lpar_map);
432CUSTOM_PORT_ATTR(stopped_map, "0x%X\n", port->stopped_map);
433CUSTOM_PORT_ATTR(mac_addr, "0x%lX\n", port->mac_addr);
434
435#define GET_PORT_ATTR(_name) (&veth_port_attr_##_name.attr)
436static struct attribute *veth_port_default_attrs[] = {
437 GET_PORT_ATTR(mac_addr),
438 GET_PORT_ATTR(lpar_map),
439 GET_PORT_ATTR(stopped_map),
440 GET_PORT_ATTR(promiscuous),
441 GET_PORT_ATTR(num_mcast),
442 NULL
443};
444
445static struct sysfs_ops veth_port_sysfs_ops = {
446 .show = veth_port_attribute_show
447};
448
449static struct kobj_type veth_port_ktype = {
450 .sysfs_ops = &veth_port_sysfs_ops,
451 .default_attrs = veth_port_default_attrs
452};
453
Michael Ellerman76812d8122005-09-01 11:29:20 +1000454/*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700455 * LPAR connection code
456 */
457
458static inline void veth_kick_statemachine(struct veth_lpar_connection *cnx)
459{
460 schedule_work(&cnx->statemachine_wq);
461}
462
463static void veth_take_cap(struct veth_lpar_connection *cnx,
Michael Ellerman59f17ae2005-09-01 11:29:27 +1000464 struct veth_lpevent *event)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700465{
466 unsigned long flags;
467
468 spin_lock_irqsave(&cnx->lock, flags);
469 /* Receiving caps may mean the other end has just come up, so
470 * we need to reload the instance ID of the far end */
471 cnx->dst_inst =
472 HvCallEvent_getTargetLpInstanceId(cnx->remote_lp,
473 HvLpEvent_Type_VirtualLan);
474
475 if (cnx->state & VETH_STATE_GOTCAPS) {
Michael Ellerman61a3c692005-09-01 11:28:57 +1000476 veth_error("Received a second capabilities from LPAR %d.\n",
Linus Torvalds1da177e2005-04-16 15:20:36 -0700477 cnx->remote_lp);
478 event->base_event.xRc = HvLpEvent_Rc_BufferNotAvailable;
479 HvCallEvent_ackLpEvent((struct HvLpEvent *) event);
480 } else {
481 memcpy(&cnx->cap_event, event, sizeof(cnx->cap_event));
482 cnx->state |= VETH_STATE_GOTCAPS;
483 veth_kick_statemachine(cnx);
484 }
485 spin_unlock_irqrestore(&cnx->lock, flags);
486}
487
488static void veth_take_cap_ack(struct veth_lpar_connection *cnx,
Michael Ellerman59f17ae2005-09-01 11:29:27 +1000489 struct veth_lpevent *event)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700490{
491 unsigned long flags;
492
493 spin_lock_irqsave(&cnx->lock, flags);
494 if (cnx->state & VETH_STATE_GOTCAPACK) {
Michael Ellerman61a3c692005-09-01 11:28:57 +1000495 veth_error("Received a second capabilities ack from LPAR %d.\n",
Linus Torvalds1da177e2005-04-16 15:20:36 -0700496 cnx->remote_lp);
497 } else {
498 memcpy(&cnx->cap_ack_event, event,
499 sizeof(&cnx->cap_ack_event));
500 cnx->state |= VETH_STATE_GOTCAPACK;
501 veth_kick_statemachine(cnx);
502 }
503 spin_unlock_irqrestore(&cnx->lock, flags);
504}
505
506static void veth_take_monitor_ack(struct veth_lpar_connection *cnx,
Michael Ellerman59f17ae2005-09-01 11:29:27 +1000507 struct veth_lpevent *event)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700508{
509 unsigned long flags;
510
511 spin_lock_irqsave(&cnx->lock, flags);
Michael Ellerman61a3c692005-09-01 11:28:57 +1000512 veth_debug("cnx %d: lost connection.\n", cnx->remote_lp);
Michael Ellerman58c59002005-09-01 11:29:00 +1000513
514 /* Avoid kicking the statemachine once we're shutdown.
515 * It's unnecessary and it could break veth_stop_connection(). */
516
517 if (! (cnx->state & VETH_STATE_SHUTDOWN)) {
518 cnx->state |= VETH_STATE_RESET;
519 veth_kick_statemachine(cnx);
520 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700521 spin_unlock_irqrestore(&cnx->lock, flags);
522}
523
Michael Ellerman59f17ae2005-09-01 11:29:27 +1000524static void veth_handle_ack(struct veth_lpevent *event)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700525{
526 HvLpIndex rlp = event->base_event.xTargetLp;
527 struct veth_lpar_connection *cnx = veth_cnx[rlp];
528
529 BUG_ON(! cnx);
530
531 switch (event->base_event.xSubtype) {
Michael Ellerman59f17ae2005-09-01 11:29:27 +1000532 case VETH_EVENT_CAP:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700533 veth_take_cap_ack(cnx, event);
534 break;
Michael Ellerman59f17ae2005-09-01 11:29:27 +1000535 case VETH_EVENT_MONITOR:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700536 veth_take_monitor_ack(cnx, event);
537 break;
538 default:
Michael Ellerman61a3c692005-09-01 11:28:57 +1000539 veth_error("Unknown ack type %d from LPAR %d.\n",
540 event->base_event.xSubtype, rlp);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700541 };
542}
543
Michael Ellerman59f17ae2005-09-01 11:29:27 +1000544static void veth_handle_int(struct veth_lpevent *event)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700545{
546 HvLpIndex rlp = event->base_event.xSourceLp;
547 struct veth_lpar_connection *cnx = veth_cnx[rlp];
548 unsigned long flags;
Michael Ellerman24562ff2005-09-01 11:29:17 +1000549 int i, acked = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700550
551 BUG_ON(! cnx);
552
553 switch (event->base_event.xSubtype) {
Michael Ellerman59f17ae2005-09-01 11:29:27 +1000554 case VETH_EVENT_CAP:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700555 veth_take_cap(cnx, event);
556 break;
Michael Ellerman59f17ae2005-09-01 11:29:27 +1000557 case VETH_EVENT_MONITOR:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700558 /* do nothing... this'll hang out here til we're dead,
559 * and the hypervisor will return it for us. */
560 break;
Michael Ellerman59f17ae2005-09-01 11:29:27 +1000561 case VETH_EVENT_FRAMES_ACK:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700562 spin_lock_irqsave(&cnx->lock, flags);
Michael Ellerman24562ff2005-09-01 11:29:17 +1000563
Linus Torvalds1da177e2005-04-16 15:20:36 -0700564 for (i = 0; i < VETH_MAX_ACKS_PER_MSG; ++i) {
565 u16 msgnum = event->u.frames_ack_data.token[i];
566
Michael Ellerman24562ff2005-09-01 11:29:17 +1000567 if (msgnum < VETH_NUMBUFFERS) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700568 veth_recycle_msg(cnx, cnx->msgs + msgnum);
Michael Ellerman24562ff2005-09-01 11:29:17 +1000569 cnx->outstanding_tx--;
570 acked++;
571 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700572 }
Michael Ellerman24562ff2005-09-01 11:29:17 +1000573
Michael Ellermane0808492005-09-01 11:29:18 +1000574 if (acked > 0) {
Michael Ellerman24562ff2005-09-01 11:29:17 +1000575 cnx->last_contact = jiffies;
Michael Ellermane0808492005-09-01 11:29:18 +1000576 veth_wake_queues(cnx);
577 }
Michael Ellerman24562ff2005-09-01 11:29:17 +1000578
Linus Torvalds1da177e2005-04-16 15:20:36 -0700579 spin_unlock_irqrestore(&cnx->lock, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700580 break;
Michael Ellerman59f17ae2005-09-01 11:29:27 +1000581 case VETH_EVENT_FRAMES:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700582 veth_receive(cnx, event);
583 break;
584 default:
Michael Ellerman61a3c692005-09-01 11:28:57 +1000585 veth_error("Unknown interrupt type %d from LPAR %d.\n",
586 event->base_event.xSubtype, rlp);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700587 };
588}
589
590static void veth_handle_event(struct HvLpEvent *event, struct pt_regs *regs)
591{
Michael Ellerman59f17ae2005-09-01 11:29:27 +1000592 struct veth_lpevent *veth_event = (struct veth_lpevent *)event;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700593
Stephen Rothwell677f8c02006-01-12 13:47:43 +1100594 if (hvlpevent_is_ack(event))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700595 veth_handle_ack(veth_event);
Stephen Rothwell677f8c02006-01-12 13:47:43 +1100596 else
Linus Torvalds1da177e2005-04-16 15:20:36 -0700597 veth_handle_int(veth_event);
598}
599
600static int veth_process_caps(struct veth_lpar_connection *cnx)
601{
Michael Ellerman59f17ae2005-09-01 11:29:27 +1000602 struct veth_cap_data *remote_caps = &cnx->remote_caps;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700603 int num_acks_needed;
604
605 /* Convert timer to jiffies */
606 cnx->ack_timeout = remote_caps->ack_timeout * HZ / 1000000;
607
608 if ( (remote_caps->num_buffers == 0)
609 || (remote_caps->ack_threshold > VETH_MAX_ACKS_PER_MSG)
610 || (remote_caps->ack_threshold == 0)
611 || (cnx->ack_timeout == 0) ) {
Michael Ellerman61a3c692005-09-01 11:28:57 +1000612 veth_error("Received incompatible capabilities from LPAR %d.\n",
613 cnx->remote_lp);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700614 return HvLpEvent_Rc_InvalidSubtypeData;
615 }
616
617 num_acks_needed = (remote_caps->num_buffers
618 / remote_caps->ack_threshold) + 1;
619
620 /* FIXME: locking on num_ack_events? */
621 if (cnx->num_ack_events < num_acks_needed) {
622 int num;
623
624 num = veth_allocate_events(cnx->remote_lp,
625 num_acks_needed-cnx->num_ack_events);
626 if (num > 0)
627 cnx->num_ack_events += num;
628
629 if (cnx->num_ack_events < num_acks_needed) {
Michael Ellerman61a3c692005-09-01 11:28:57 +1000630 veth_error("Couldn't allocate enough ack events "
631 "for LPAR %d.\n", cnx->remote_lp);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700632
633 return HvLpEvent_Rc_BufferNotAvailable;
634 }
635 }
636
637
638 return HvLpEvent_Rc_Good;
639}
640
641/* FIXME: The gotos here are a bit dubious */
642static void veth_statemachine(void *p)
643{
644 struct veth_lpar_connection *cnx = (struct veth_lpar_connection *)p;
645 int rlp = cnx->remote_lp;
646 int rc;
647
648 spin_lock_irq(&cnx->lock);
649
650 restart:
651 if (cnx->state & VETH_STATE_RESET) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700652 if (cnx->state & VETH_STATE_OPEN)
653 HvCallEvent_closeLpEventPath(cnx->remote_lp,
654 HvLpEvent_Type_VirtualLan);
655
Michael Ellermanabfda472005-09-01 11:28:59 +1000656 /*
657 * Reset ack data. This prevents the ack_timer actually
658 * doing anything, even if it runs one more time when
659 * we drop the lock below.
660 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700661 memset(&cnx->pending_acks, 0xff, sizeof (cnx->pending_acks));
662 cnx->num_pending_acks = 0;
663
664 cnx->state &= ~(VETH_STATE_RESET | VETH_STATE_SENTMON
665 | VETH_STATE_OPEN | VETH_STATE_SENTCAPS
666 | VETH_STATE_GOTCAPACK | VETH_STATE_GOTCAPS
667 | VETH_STATE_SENTCAPACK | VETH_STATE_READY);
668
669 /* Clean up any leftover messages */
Michael Ellerman24562ff2005-09-01 11:29:17 +1000670 if (cnx->msgs) {
671 int i;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700672 for (i = 0; i < VETH_NUMBUFFERS; ++i)
673 veth_recycle_msg(cnx, cnx->msgs + i);
Michael Ellerman24562ff2005-09-01 11:29:17 +1000674 }
Michael Ellermane0808492005-09-01 11:29:18 +1000675
Michael Ellerman24562ff2005-09-01 11:29:17 +1000676 cnx->outstanding_tx = 0;
Michael Ellermane0808492005-09-01 11:29:18 +1000677 veth_wake_queues(cnx);
Michael Ellermanabfda472005-09-01 11:28:59 +1000678
679 /* Drop the lock so we can do stuff that might sleep or
680 * take other locks. */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700681 spin_unlock_irq(&cnx->lock);
Michael Ellermanabfda472005-09-01 11:28:59 +1000682
683 del_timer_sync(&cnx->ack_timer);
Michael Ellerman24562ff2005-09-01 11:29:17 +1000684 del_timer_sync(&cnx->reset_timer);
685
Linus Torvalds1da177e2005-04-16 15:20:36 -0700686 spin_lock_irq(&cnx->lock);
Michael Ellermanabfda472005-09-01 11:28:59 +1000687
Linus Torvalds1da177e2005-04-16 15:20:36 -0700688 if (cnx->state & VETH_STATE_RESET)
689 goto restart;
Michael Ellerman58c59002005-09-01 11:29:00 +1000690
691 /* Hack, wait for the other end to reset itself. */
692 if (! (cnx->state & VETH_STATE_SHUTDOWN)) {
693 schedule_delayed_work(&cnx->statemachine_wq, 5 * HZ);
694 goto out;
695 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700696 }
697
698 if (cnx->state & VETH_STATE_SHUTDOWN)
699 /* It's all over, do nothing */
700 goto out;
701
702 if ( !(cnx->state & VETH_STATE_OPEN) ) {
703 if (! cnx->msgs || (cnx->num_events < (2 + VETH_NUMBUFFERS)) )
704 goto cant_cope;
705
706 HvCallEvent_openLpEventPath(rlp, HvLpEvent_Type_VirtualLan);
707 cnx->src_inst =
708 HvCallEvent_getSourceLpInstanceId(rlp,
709 HvLpEvent_Type_VirtualLan);
710 cnx->dst_inst =
711 HvCallEvent_getTargetLpInstanceId(rlp,
712 HvLpEvent_Type_VirtualLan);
713 cnx->state |= VETH_STATE_OPEN;
714 }
715
716 if ( (cnx->state & VETH_STATE_OPEN)
717 && !(cnx->state & VETH_STATE_SENTMON) ) {
Michael Ellerman59f17ae2005-09-01 11:29:27 +1000718 rc = veth_signalevent(cnx, VETH_EVENT_MONITOR,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700719 HvLpEvent_AckInd_DoAck,
720 HvLpEvent_AckType_DeferredAck,
721 0, 0, 0, 0, 0, 0);
722
723 if (rc == HvLpEvent_Rc_Good) {
724 cnx->state |= VETH_STATE_SENTMON;
725 } else {
726 if ( (rc != HvLpEvent_Rc_PartitionDead)
727 && (rc != HvLpEvent_Rc_PathClosed) )
Michael Ellerman61a3c692005-09-01 11:28:57 +1000728 veth_error("Error sending monitor to LPAR %d, "
729 "rc = %d\n", rlp, rc);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700730
731 /* Oh well, hope we get a cap from the other
732 * end and do better when that kicks us */
733 goto out;
734 }
735 }
736
737 if ( (cnx->state & VETH_STATE_OPEN)
738 && !(cnx->state & VETH_STATE_SENTCAPS)) {
739 u64 *rawcap = (u64 *)&cnx->local_caps;
740
Michael Ellerman59f17ae2005-09-01 11:29:27 +1000741 rc = veth_signalevent(cnx, VETH_EVENT_CAP,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700742 HvLpEvent_AckInd_DoAck,
743 HvLpEvent_AckType_ImmediateAck,
744 0, rawcap[0], rawcap[1], rawcap[2],
745 rawcap[3], rawcap[4]);
746
747 if (rc == HvLpEvent_Rc_Good) {
748 cnx->state |= VETH_STATE_SENTCAPS;
749 } else {
750 if ( (rc != HvLpEvent_Rc_PartitionDead)
751 && (rc != HvLpEvent_Rc_PathClosed) )
Michael Ellerman61a3c692005-09-01 11:28:57 +1000752 veth_error("Error sending caps to LPAR %d, "
753 "rc = %d\n", rlp, rc);
754
Linus Torvalds1da177e2005-04-16 15:20:36 -0700755 /* Oh well, hope we get a cap from the other
756 * end and do better when that kicks us */
757 goto out;
758 }
759 }
760
761 if ((cnx->state & VETH_STATE_GOTCAPS)
762 && !(cnx->state & VETH_STATE_SENTCAPACK)) {
Michael Ellerman59f17ae2005-09-01 11:29:27 +1000763 struct veth_cap_data *remote_caps = &cnx->remote_caps;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700764
765 memcpy(remote_caps, &cnx->cap_event.u.caps_data,
766 sizeof(*remote_caps));
767
768 spin_unlock_irq(&cnx->lock);
769 rc = veth_process_caps(cnx);
770 spin_lock_irq(&cnx->lock);
771
772 /* We dropped the lock, so recheck for anything which
773 * might mess us up */
774 if (cnx->state & (VETH_STATE_RESET|VETH_STATE_SHUTDOWN))
775 goto restart;
776
777 cnx->cap_event.base_event.xRc = rc;
778 HvCallEvent_ackLpEvent((struct HvLpEvent *)&cnx->cap_event);
779 if (rc == HvLpEvent_Rc_Good)
780 cnx->state |= VETH_STATE_SENTCAPACK;
781 else
782 goto cant_cope;
783 }
784
785 if ((cnx->state & VETH_STATE_GOTCAPACK)
786 && (cnx->state & VETH_STATE_GOTCAPS)
787 && !(cnx->state & VETH_STATE_READY)) {
788 if (cnx->cap_ack_event.base_event.xRc == HvLpEvent_Rc_Good) {
789 /* Start the ACK timer */
790 cnx->ack_timer.expires = jiffies + cnx->ack_timeout;
791 add_timer(&cnx->ack_timer);
792 cnx->state |= VETH_STATE_READY;
793 } else {
Michael Ellerman61a3c692005-09-01 11:28:57 +1000794 veth_error("Caps rejected by LPAR %d, rc = %d\n",
795 rlp, cnx->cap_ack_event.base_event.xRc);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700796 goto cant_cope;
797 }
798 }
799
800 out:
801 spin_unlock_irq(&cnx->lock);
802 return;
803
804 cant_cope:
805 /* FIXME: we get here if something happens we really can't
806 * cope with. The link will never work once we get here, and
807 * all we can do is not lock the rest of the system up */
Michael Ellerman61a3c692005-09-01 11:28:57 +1000808 veth_error("Unrecoverable error on connection to LPAR %d, shutting down"
809 " (state = 0x%04lx)\n", rlp, cnx->state);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700810 cnx->state |= VETH_STATE_SHUTDOWN;
811 spin_unlock_irq(&cnx->lock);
812}
813
814static int veth_init_connection(u8 rlp)
815{
816 struct veth_lpar_connection *cnx;
817 struct veth_msg *msgs;
Michael Ellermanf0c129c2005-09-01 11:29:09 +1000818 int i, rc;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700819
820 if ( (rlp == this_lp)
821 || ! HvLpConfig_doLpsCommunicateOnVirtualLan(this_lp, rlp) )
822 return 0;
823
824 cnx = kmalloc(sizeof(*cnx), GFP_KERNEL);
825 if (! cnx)
826 return -ENOMEM;
827 memset(cnx, 0, sizeof(*cnx));
828
829 cnx->remote_lp = rlp;
830 spin_lock_init(&cnx->lock);
831 INIT_WORK(&cnx->statemachine_wq, veth_statemachine, cnx);
Michael Ellerman24562ff2005-09-01 11:29:17 +1000832
Linus Torvalds1da177e2005-04-16 15:20:36 -0700833 init_timer(&cnx->ack_timer);
834 cnx->ack_timer.function = veth_timed_ack;
835 cnx->ack_timer.data = (unsigned long) cnx;
Michael Ellerman24562ff2005-09-01 11:29:17 +1000836
837 init_timer(&cnx->reset_timer);
838 cnx->reset_timer.function = veth_timed_reset;
839 cnx->reset_timer.data = (unsigned long) cnx;
840 cnx->reset_timeout = 5 * HZ * (VETH_ACKTIMEOUT / 1000000);
841
Linus Torvalds1da177e2005-04-16 15:20:36 -0700842 memset(&cnx->pending_acks, 0xff, sizeof (cnx->pending_acks));
843
844 veth_cnx[rlp] = cnx;
845
Michael Ellermanf0c129c2005-09-01 11:29:09 +1000846 /* This gets us 1 reference, which is held on behalf of the driver
847 * infrastructure. It's released at module unload. */
848 kobject_init(&cnx->kobject);
849 cnx->kobject.ktype = &veth_lpar_connection_ktype;
850 rc = kobject_set_name(&cnx->kobject, "cnx%.2d", rlp);
851 if (rc != 0)
852 return rc;
853
Linus Torvalds1da177e2005-04-16 15:20:36 -0700854 msgs = kmalloc(VETH_NUMBUFFERS * sizeof(struct veth_msg), GFP_KERNEL);
855 if (! msgs) {
Michael Ellerman61a3c692005-09-01 11:28:57 +1000856 veth_error("Can't allocate buffers for LPAR %d.\n", rlp);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700857 return -ENOMEM;
858 }
859
860 cnx->msgs = msgs;
861 memset(msgs, 0, VETH_NUMBUFFERS * sizeof(struct veth_msg));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700862
863 for (i = 0; i < VETH_NUMBUFFERS; i++) {
864 msgs[i].token = i;
865 veth_stack_push(cnx, msgs + i);
866 }
867
868 cnx->num_events = veth_allocate_events(rlp, 2 + VETH_NUMBUFFERS);
869
870 if (cnx->num_events < (2 + VETH_NUMBUFFERS)) {
Michael Ellerman61a3c692005-09-01 11:28:57 +1000871 veth_error("Can't allocate enough events for LPAR %d.\n", rlp);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700872 return -ENOMEM;
873 }
874
875 cnx->local_caps.num_buffers = VETH_NUMBUFFERS;
876 cnx->local_caps.ack_threshold = ACK_THRESHOLD;
877 cnx->local_caps.ack_timeout = VETH_ACKTIMEOUT;
878
879 return 0;
880}
881
Michael Ellermanf0c129c2005-09-01 11:29:09 +1000882static void veth_stop_connection(struct veth_lpar_connection *cnx)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700883{
Michael Ellermanf0c129c2005-09-01 11:29:09 +1000884 if (!cnx)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700885 return;
886
887 spin_lock_irq(&cnx->lock);
888 cnx->state |= VETH_STATE_RESET | VETH_STATE_SHUTDOWN;
889 veth_kick_statemachine(cnx);
890 spin_unlock_irq(&cnx->lock);
891
Michael Ellerman58c59002005-09-01 11:29:00 +1000892 /* There's a slim chance the reset code has just queued the
893 * statemachine to run in five seconds. If so we need to cancel
894 * that and requeue the work to run now. */
895 if (cancel_delayed_work(&cnx->statemachine_wq)) {
896 spin_lock_irq(&cnx->lock);
897 veth_kick_statemachine(cnx);
898 spin_unlock_irq(&cnx->lock);
899 }
900
Michael Ellermanabfda472005-09-01 11:28:59 +1000901 /* Wait for the state machine to run. */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700902 flush_scheduled_work();
Michael Ellermanec60bee2005-09-01 11:29:08 +1000903}
904
Michael Ellermanf0c129c2005-09-01 11:29:09 +1000905static void veth_destroy_connection(struct veth_lpar_connection *cnx)
Michael Ellermanec60bee2005-09-01 11:29:08 +1000906{
Michael Ellermanf0c129c2005-09-01 11:29:09 +1000907 if (!cnx)
Michael Ellermanec60bee2005-09-01 11:29:08 +1000908 return;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700909
Linus Torvalds1da177e2005-04-16 15:20:36 -0700910 if (cnx->num_events > 0)
911 mf_deallocate_lp_events(cnx->remote_lp,
912 HvLpEvent_Type_VirtualLan,
913 cnx->num_events,
914 NULL, NULL);
915 if (cnx->num_ack_events > 0)
916 mf_deallocate_lp_events(cnx->remote_lp,
917 HvLpEvent_Type_VirtualLan,
918 cnx->num_ack_events,
919 NULL, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700920
921 kfree(cnx->msgs);
Michael Ellermanf0c129c2005-09-01 11:29:09 +1000922 veth_cnx[cnx->remote_lp] = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700923 kfree(cnx);
Michael Ellermanf0c129c2005-09-01 11:29:09 +1000924}
925
926static void veth_release_connection(struct kobject *kobj)
927{
928 struct veth_lpar_connection *cnx;
929 cnx = container_of(kobj, struct veth_lpar_connection, kobject);
930 veth_stop_connection(cnx);
931 veth_destroy_connection(cnx);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700932}
933
934/*
935 * net_device code
936 */
937
938static int veth_open(struct net_device *dev)
939{
940 struct veth_port *port = (struct veth_port *) dev->priv;
941
942 memset(&port->stats, 0, sizeof (port->stats));
943 netif_start_queue(dev);
944 return 0;
945}
946
947static int veth_close(struct net_device *dev)
948{
949 netif_stop_queue(dev);
950 return 0;
951}
952
953static struct net_device_stats *veth_get_stats(struct net_device *dev)
954{
955 struct veth_port *port = (struct veth_port *) dev->priv;
956
957 return &port->stats;
958}
959
960static int veth_change_mtu(struct net_device *dev, int new_mtu)
961{
962 if ((new_mtu < 68) || (new_mtu > VETH_MAX_MTU))
963 return -EINVAL;
964 dev->mtu = new_mtu;
965 return 0;
966}
967
968static void veth_set_multicast_list(struct net_device *dev)
969{
970 struct veth_port *port = (struct veth_port *) dev->priv;
971 unsigned long flags;
972
973 write_lock_irqsave(&port->mcast_gate, flags);
974
Michael Ellerman2a5391a2005-09-01 11:29:02 +1000975 if ((dev->flags & IFF_PROMISC) || (dev->flags & IFF_ALLMULTI) ||
976 (dev->mc_count > VETH_MAX_MCAST)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700977 port->promiscuous = 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700978 } else {
979 struct dev_mc_list *dmi = dev->mc_list;
980 int i;
981
Michael Ellerman2a5391a2005-09-01 11:29:02 +1000982 port->promiscuous = 0;
983
Linus Torvalds1da177e2005-04-16 15:20:36 -0700984 /* Update table */
985 port->num_mcast = 0;
986
987 for (i = 0; i < dev->mc_count; i++) {
988 u8 *addr = dmi->dmi_addr;
989 u64 xaddr = 0;
990
991 if (addr[0] & 0x01) {/* multicast address? */
992 memcpy(&xaddr, addr, ETH_ALEN);
993 port->mcast_addr[port->num_mcast] = xaddr;
994 port->num_mcast++;
995 }
996 dmi = dmi->next;
997 }
998 }
999
1000 write_unlock_irqrestore(&port->mcast_gate, flags);
1001}
1002
1003static void veth_get_drvinfo(struct net_device *dev, struct ethtool_drvinfo *info)
1004{
Michael Ellermanee05f032005-09-01 11:29:29 +10001005 strncpy(info->driver, DRV_NAME, sizeof(info->driver) - 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001006 info->driver[sizeof(info->driver) - 1] = '\0';
Michael Ellermanee05f032005-09-01 11:29:29 +10001007 strncpy(info->version, DRV_VERSION, sizeof(info->version) - 1);
1008 info->version[sizeof(info->version) - 1] = '\0';
Linus Torvalds1da177e2005-04-16 15:20:36 -07001009}
1010
1011static int veth_get_settings(struct net_device *dev, struct ethtool_cmd *ecmd)
1012{
1013 ecmd->supported = (SUPPORTED_1000baseT_Full
1014 | SUPPORTED_Autoneg | SUPPORTED_FIBRE);
1015 ecmd->advertising = (SUPPORTED_1000baseT_Full
1016 | SUPPORTED_Autoneg | SUPPORTED_FIBRE);
1017 ecmd->port = PORT_FIBRE;
1018 ecmd->transceiver = XCVR_INTERNAL;
1019 ecmd->phy_address = 0;
1020 ecmd->speed = SPEED_1000;
1021 ecmd->duplex = DUPLEX_FULL;
1022 ecmd->autoneg = AUTONEG_ENABLE;
1023 ecmd->maxtxpkt = 120;
1024 ecmd->maxrxpkt = 120;
1025 return 0;
1026}
1027
1028static u32 veth_get_link(struct net_device *dev)
1029{
1030 return 1;
1031}
1032
1033static struct ethtool_ops ops = {
1034 .get_drvinfo = veth_get_drvinfo,
1035 .get_settings = veth_get_settings,
1036 .get_link = veth_get_link,
1037};
1038
Stephen Rothwell07fb3f42006-05-15 13:39:47 +10001039static struct net_device * __init veth_probe_one(int vlan,
1040 struct vio_dev *vio_dev)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001041{
1042 struct net_device *dev;
1043 struct veth_port *port;
Stephen Rothwell07fb3f42006-05-15 13:39:47 +10001044 struct device *vdev = &vio_dev->dev;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001045 int i, rc;
Stephen Rothwell07fb3f42006-05-15 13:39:47 +10001046 const unsigned char *mac_addr;
1047
1048 mac_addr = vio_get_attribute(vio_dev, "local-mac-address", NULL);
1049 if (mac_addr == NULL)
1050 mac_addr = vio_get_attribute(vio_dev, "mac-address", NULL);
1051 if (mac_addr == NULL) {
1052 veth_error("Unable to fetch MAC address from device tree.\n");
1053 return NULL;
1054 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001055
1056 dev = alloc_etherdev(sizeof (struct veth_port));
1057 if (! dev) {
1058 veth_error("Unable to allocate net_device structure!\n");
1059 return NULL;
1060 }
1061
1062 port = (struct veth_port *) dev->priv;
1063
Michael Ellermane0808492005-09-01 11:29:18 +10001064 spin_lock_init(&port->queue_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001065 rwlock_init(&port->mcast_gate);
Michael Ellermane0808492005-09-01 11:29:18 +10001066 port->stopped_map = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001067
1068 for (i = 0; i < HVMAXARCHITECTEDLPS; i++) {
1069 HvLpVirtualLanIndexMap map;
1070
1071 if (i == this_lp)
1072 continue;
1073 map = HvLpConfig_getVirtualLanIndexMapForLp(i);
1074 if (map & (0x8000 >> vlan))
1075 port->lpar_map |= (1 << i);
1076 }
1077 port->dev = vdev;
1078
Stephen Rothwell07fb3f42006-05-15 13:39:47 +10001079 memcpy(dev->dev_addr, mac_addr, ETH_ALEN);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001080
1081 dev->mtu = VETH_MAX_MTU;
1082
Stephen Rothwell07fb3f42006-05-15 13:39:47 +10001083 memcpy(&port->mac_addr, mac_addr, ETH_ALEN);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001084
1085 dev->open = veth_open;
1086 dev->hard_start_xmit = veth_start_xmit;
1087 dev->stop = veth_close;
1088 dev->get_stats = veth_get_stats;
1089 dev->change_mtu = veth_change_mtu;
1090 dev->set_mac_address = NULL;
1091 dev->set_multicast_list = veth_set_multicast_list;
1092 SET_ETHTOOL_OPS(dev, &ops);
1093
Linus Torvalds1da177e2005-04-16 15:20:36 -07001094 SET_NETDEV_DEV(dev, vdev);
1095
1096 rc = register_netdev(dev);
1097 if (rc != 0) {
Michael Ellerman61a3c692005-09-01 11:28:57 +10001098 veth_error("Failed registering net device for vlan%d.\n", vlan);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001099 free_netdev(dev);
1100 return NULL;
1101 }
1102
Michael Ellerman07a5c172005-09-01 11:29:21 +10001103 kobject_init(&port->kobject);
1104 port->kobject.parent = &dev->class_dev.kobj;
1105 port->kobject.ktype = &veth_port_ktype;
1106 kobject_set_name(&port->kobject, "veth_port");
1107 if (0 != kobject_add(&port->kobject))
1108 veth_error("Failed adding port for %s to sysfs.\n", dev->name);
1109
Michael Ellerman61a3c692005-09-01 11:28:57 +10001110 veth_info("%s attached to iSeries vlan %d (LPAR map = 0x%.4X)\n",
1111 dev->name, vlan, port->lpar_map);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001112
1113 return dev;
1114}
1115
1116/*
1117 * Tx path
1118 */
1119
1120static int veth_transmit_to_one(struct sk_buff *skb, HvLpIndex rlp,
1121 struct net_device *dev)
1122{
1123 struct veth_lpar_connection *cnx = veth_cnx[rlp];
1124 struct veth_port *port = (struct veth_port *) dev->priv;
1125 HvLpEvent_Rc rc;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001126 struct veth_msg *msg = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001127 unsigned long flags;
1128
Michael Ellermandb5e8712005-09-01 11:29:19 +10001129 if (! cnx)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001130 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001131
1132 spin_lock_irqsave(&cnx->lock, flags);
1133
Michael Ellermanf27eff12005-05-12 17:47:27 +10001134 if (! (cnx->state & VETH_STATE_READY))
Michael Ellermandb5e8712005-09-01 11:29:19 +10001135 goto no_error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001136
Michael Ellermandb5e8712005-09-01 11:29:19 +10001137 if ((skb->len - ETH_HLEN) > VETH_MAX_MTU)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001138 goto drop;
1139
1140 msg = veth_stack_pop(cnx);
Michael Ellermandb5e8712005-09-01 11:29:19 +10001141 if (! msg)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001142 goto drop;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001143
Michael Ellermanb08bd5c2005-09-01 11:29:06 +10001144 msg->in_use = 1;
Michael Ellermandb5e8712005-09-01 11:29:19 +10001145 msg->skb = skb_get(skb);
Michael Ellermanb08bd5c2005-09-01 11:29:06 +10001146
Michael Ellermancbf90742005-09-01 11:29:07 +10001147 msg->data.addr[0] = dma_map_single(port->dev, skb->data,
1148 skb->len, DMA_TO_DEVICE);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001149
Michael Ellermancbf90742005-09-01 11:29:07 +10001150 if (dma_mapping_error(msg->data.addr[0]))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001151 goto recycle_and_drop;
1152
Linus Torvalds1da177e2005-04-16 15:20:36 -07001153 msg->dev = port->dev;
Michael Ellermancbf90742005-09-01 11:29:07 +10001154 msg->data.len[0] = skb->len;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001155 msg->data.eofmask = 1 << VETH_EOF_SHIFT;
Michael Ellermancbf90742005-09-01 11:29:07 +10001156
Michael Ellerman59f17ae2005-09-01 11:29:27 +10001157 rc = veth_signaldata(cnx, VETH_EVENT_FRAMES, msg->token, &msg->data);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001158
1159 if (rc != HvLpEvent_Rc_Good)
1160 goto recycle_and_drop;
1161
Michael Ellerman24562ff2005-09-01 11:29:17 +10001162 /* If the timer's not already running, start it now. */
1163 if (0 == cnx->outstanding_tx)
1164 mod_timer(&cnx->reset_timer, jiffies + cnx->reset_timeout);
1165
1166 cnx->last_contact = jiffies;
1167 cnx->outstanding_tx++;
1168
Michael Ellermane0808492005-09-01 11:29:18 +10001169 if (veth_stack_is_empty(cnx))
1170 veth_stop_queues(cnx);
1171
Michael Ellermandb5e8712005-09-01 11:29:19 +10001172 no_error:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001173 spin_unlock_irqrestore(&cnx->lock, flags);
1174 return 0;
1175
1176 recycle_and_drop:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001177 veth_recycle_msg(cnx, msg);
1178 drop:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001179 spin_unlock_irqrestore(&cnx->lock, flags);
Michael Ellermandb5e8712005-09-01 11:29:19 +10001180 return 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001181}
1182
Michael Ellermandb5e8712005-09-01 11:29:19 +10001183static void veth_transmit_to_many(struct sk_buff *skb,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001184 HvLpIndexMap lpmask,
1185 struct net_device *dev)
1186{
1187 struct veth_port *port = (struct veth_port *) dev->priv;
Michael Ellermandb5e8712005-09-01 11:29:19 +10001188 int i, success, error;
1189
1190 success = error = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001191
1192 for (i = 0; i < HVMAXARCHITECTEDLPS; i++) {
1193 if ((lpmask & (1 << i)) == 0)
1194 continue;
1195
Michael Ellermandb5e8712005-09-01 11:29:19 +10001196 if (veth_transmit_to_one(skb, i, dev))
1197 error = 1;
1198 else
1199 success = 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001200 }
1201
Michael Ellermandb5e8712005-09-01 11:29:19 +10001202 if (error)
1203 port->stats.tx_errors++;
1204
1205 if (success) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001206 port->stats.tx_packets++;
1207 port->stats.tx_bytes += skb->len;
1208 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001209}
1210
1211static int veth_start_xmit(struct sk_buff *skb, struct net_device *dev)
1212{
1213 unsigned char *frame = skb->data;
1214 struct veth_port *port = (struct veth_port *) dev->priv;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001215 HvLpIndexMap lpmask;
1216
1217 if (! (frame[0] & 0x01)) {
1218 /* unicast packet */
1219 HvLpIndex rlp = frame[5];
1220
1221 if ( ! ((1 << rlp) & port->lpar_map) ) {
1222 dev_kfree_skb(skb);
1223 return 0;
1224 }
1225
1226 lpmask = 1 << rlp;
1227 } else {
1228 lpmask = port->lpar_map;
1229 }
1230
Michael Ellermane0808492005-09-01 11:29:18 +10001231 veth_transmit_to_many(skb, lpmask, dev);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001232
Michael Ellermane0808492005-09-01 11:29:18 +10001233 dev_kfree_skb(skb);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001234
1235 return 0;
1236}
1237
Michael Ellermanb08bd5c2005-09-01 11:29:06 +10001238/* You must hold the connection's lock when you call this function. */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001239static void veth_recycle_msg(struct veth_lpar_connection *cnx,
1240 struct veth_msg *msg)
1241{
1242 u32 dma_address, dma_length;
1243
Michael Ellermanb08bd5c2005-09-01 11:29:06 +10001244 if (msg->in_use) {
1245 msg->in_use = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001246 dma_address = msg->data.addr[0];
1247 dma_length = msg->data.len[0];
1248
Michael Ellermancbf90742005-09-01 11:29:07 +10001249 if (!dma_mapping_error(dma_address))
1250 dma_unmap_single(msg->dev, dma_address, dma_length,
1251 DMA_TO_DEVICE);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001252
1253 if (msg->skb) {
1254 dev_kfree_skb_any(msg->skb);
1255 msg->skb = NULL;
1256 }
1257
1258 memset(&msg->data, 0, sizeof(msg->data));
1259 veth_stack_push(cnx, msg);
Michael Ellerman61a3c692005-09-01 11:28:57 +10001260 } else if (cnx->state & VETH_STATE_OPEN) {
1261 veth_error("Non-pending frame (# %d) acked by LPAR %d.\n",
1262 cnx->remote_lp, msg->token);
1263 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001264}
1265
Michael Ellermane0808492005-09-01 11:29:18 +10001266static void veth_wake_queues(struct veth_lpar_connection *cnx)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001267{
1268 int i;
Michael Ellermane0808492005-09-01 11:29:18 +10001269
Linus Torvalds1da177e2005-04-16 15:20:36 -07001270 for (i = 0; i < HVMAXARCHITECTEDVIRTUALLANS; i++) {
1271 struct net_device *dev = veth_dev[i];
1272 struct veth_port *port;
1273 unsigned long flags;
1274
1275 if (! dev)
1276 continue;
1277
1278 port = (struct veth_port *)dev->priv;
1279
1280 if (! (port->lpar_map & (1<<cnx->remote_lp)))
1281 continue;
1282
Michael Ellermane0808492005-09-01 11:29:18 +10001283 spin_lock_irqsave(&port->queue_lock, flags);
1284
1285 port->stopped_map &= ~(1 << cnx->remote_lp);
1286
1287 if (0 == port->stopped_map && netif_queue_stopped(dev)) {
1288 veth_debug("cnx %d: woke queue for %s.\n",
1289 cnx->remote_lp, dev->name);
1290 netif_wake_queue(dev);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001291 }
Michael Ellermane0808492005-09-01 11:29:18 +10001292 spin_unlock_irqrestore(&port->queue_lock, flags);
1293 }
1294}
1295
1296static void veth_stop_queues(struct veth_lpar_connection *cnx)
1297{
1298 int i;
1299
1300 for (i = 0; i < HVMAXARCHITECTEDVIRTUALLANS; i++) {
1301 struct net_device *dev = veth_dev[i];
1302 struct veth_port *port;
1303
1304 if (! dev)
1305 continue;
1306
1307 port = (struct veth_port *)dev->priv;
1308
1309 /* If this cnx is not on the vlan for this port, continue */
1310 if (! (port->lpar_map & (1 << cnx->remote_lp)))
1311 continue;
1312
1313 spin_lock(&port->queue_lock);
1314
1315 netif_stop_queue(dev);
1316 port->stopped_map |= (1 << cnx->remote_lp);
1317
1318 veth_debug("cnx %d: stopped queue for %s, map = 0x%x.\n",
1319 cnx->remote_lp, dev->name, port->stopped_map);
1320
1321 spin_unlock(&port->queue_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001322 }
1323}
1324
Michael Ellerman24562ff2005-09-01 11:29:17 +10001325static void veth_timed_reset(unsigned long ptr)
1326{
1327 struct veth_lpar_connection *cnx = (struct veth_lpar_connection *)ptr;
1328 unsigned long trigger_time, flags;
1329
1330 /* FIXME is it possible this fires after veth_stop_connection()?
1331 * That would reschedule the statemachine for 5 seconds and probably
1332 * execute it after the module's been unloaded. Hmm. */
1333
1334 spin_lock_irqsave(&cnx->lock, flags);
1335
1336 if (cnx->outstanding_tx > 0) {
1337 trigger_time = cnx->last_contact + cnx->reset_timeout;
1338
1339 if (trigger_time < jiffies) {
1340 cnx->state |= VETH_STATE_RESET;
1341 veth_kick_statemachine(cnx);
1342 veth_error("%d packets not acked by LPAR %d within %d "
1343 "seconds, resetting.\n",
1344 cnx->outstanding_tx, cnx->remote_lp,
1345 cnx->reset_timeout / HZ);
1346 } else {
1347 /* Reschedule the timer */
1348 trigger_time = jiffies + cnx->reset_timeout;
1349 mod_timer(&cnx->reset_timer, trigger_time);
1350 }
1351 }
1352
1353 spin_unlock_irqrestore(&cnx->lock, flags);
1354}
1355
Linus Torvalds1da177e2005-04-16 15:20:36 -07001356/*
1357 * Rx path
1358 */
1359
1360static inline int veth_frame_wanted(struct veth_port *port, u64 mac_addr)
1361{
1362 int wanted = 0;
1363 int i;
1364 unsigned long flags;
1365
1366 if ( (mac_addr == port->mac_addr) || (mac_addr == 0xffffffffffff0000) )
1367 return 1;
1368
Linus Torvalds1da177e2005-04-16 15:20:36 -07001369 read_lock_irqsave(&port->mcast_gate, flags);
1370
Michael Ellerman2a5391a2005-09-01 11:29:02 +10001371 if (port->promiscuous) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001372 wanted = 1;
1373 goto out;
1374 }
1375
1376 for (i = 0; i < port->num_mcast; ++i) {
1377 if (port->mcast_addr[i] == mac_addr) {
1378 wanted = 1;
1379 break;
1380 }
1381 }
1382
1383 out:
1384 read_unlock_irqrestore(&port->mcast_gate, flags);
1385
1386 return wanted;
1387}
1388
1389struct dma_chunk {
1390 u64 addr;
1391 u64 size;
1392};
1393
1394#define VETH_MAX_PAGES_PER_FRAME ( (VETH_MAX_MTU+PAGE_SIZE-2)/PAGE_SIZE + 1 )
1395
1396static inline void veth_build_dma_list(struct dma_chunk *list,
1397 unsigned char *p, unsigned long length)
1398{
1399 unsigned long done;
1400 int i = 1;
1401
1402 /* FIXME: skbs are continguous in real addresses. Do we
1403 * really need to break it into PAGE_SIZE chunks, or can we do
1404 * it just at the granularity of iSeries real->absolute
1405 * mapping? Indeed, given the way the allocator works, can we
1406 * count on them being absolutely contiguous? */
Stephen Rothwell426c1a12005-10-14 14:51:42 +10001407 list[0].addr = iseries_hv_addr(p);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001408 list[0].size = min(length,
1409 PAGE_SIZE - ((unsigned long)p & ~PAGE_MASK));
1410
1411 done = list[0].size;
1412 while (done < length) {
Stephen Rothwell426c1a12005-10-14 14:51:42 +10001413 list[i].addr = iseries_hv_addr(p + done);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001414 list[i].size = min(length-done, PAGE_SIZE);
1415 done += list[i].size;
1416 i++;
1417 }
1418}
1419
1420static void veth_flush_acks(struct veth_lpar_connection *cnx)
1421{
1422 HvLpEvent_Rc rc;
1423
Michael Ellerman59f17ae2005-09-01 11:29:27 +10001424 rc = veth_signaldata(cnx, VETH_EVENT_FRAMES_ACK,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001425 0, &cnx->pending_acks);
1426
1427 if (rc != HvLpEvent_Rc_Good)
Michael Ellerman61a3c692005-09-01 11:28:57 +10001428 veth_error("Failed acking frames from LPAR %d, rc = %d\n",
1429 cnx->remote_lp, (int)rc);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001430
1431 cnx->num_pending_acks = 0;
1432 memset(&cnx->pending_acks, 0xff, sizeof(cnx->pending_acks));
1433}
1434
1435static void veth_receive(struct veth_lpar_connection *cnx,
Michael Ellerman59f17ae2005-09-01 11:29:27 +10001436 struct veth_lpevent *event)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001437{
Michael Ellerman59f17ae2005-09-01 11:29:27 +10001438 struct veth_frames_data *senddata = &event->u.frames_data;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001439 int startchunk = 0;
1440 int nchunks;
1441 unsigned long flags;
1442 HvLpDma_Rc rc;
1443
1444 do {
1445 u16 length = 0;
1446 struct sk_buff *skb;
1447 struct dma_chunk local_list[VETH_MAX_PAGES_PER_FRAME];
1448 struct dma_chunk remote_list[VETH_MAX_FRAMES_PER_MSG];
1449 u64 dest;
1450 HvLpVirtualLanIndex vlan;
1451 struct net_device *dev;
1452 struct veth_port *port;
1453
1454 /* FIXME: do we need this? */
1455 memset(local_list, 0, sizeof(local_list));
1456 memset(remote_list, 0, sizeof(VETH_MAX_FRAMES_PER_MSG));
1457
1458 /* a 0 address marks the end of the valid entries */
1459 if (senddata->addr[startchunk] == 0)
1460 break;
1461
1462 /* make sure that we have at least 1 EOF entry in the
1463 * remaining entries */
1464 if (! (senddata->eofmask >> (startchunk + VETH_EOF_SHIFT))) {
Michael Ellerman61a3c692005-09-01 11:28:57 +10001465 veth_error("Missing EOF fragment in event "
1466 "eofmask = 0x%x startchunk = %d\n",
1467 (unsigned)senddata->eofmask,
1468 startchunk);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001469 break;
1470 }
1471
1472 /* build list of chunks in this frame */
1473 nchunks = 0;
1474 do {
1475 remote_list[nchunks].addr =
1476 (u64) senddata->addr[startchunk+nchunks] << 32;
1477 remote_list[nchunks].size =
1478 senddata->len[startchunk+nchunks];
1479 length += remote_list[nchunks].size;
1480 } while (! (senddata->eofmask &
1481 (1 << (VETH_EOF_SHIFT + startchunk + nchunks++))));
1482
1483 /* length == total length of all chunks */
1484 /* nchunks == # of chunks in this frame */
1485
1486 if ((length - ETH_HLEN) > VETH_MAX_MTU) {
Michael Ellerman61a3c692005-09-01 11:28:57 +10001487 veth_error("Received oversize frame from LPAR %d "
1488 "(length = %d)\n",
1489 cnx->remote_lp, length);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001490 continue;
1491 }
1492
1493 skb = alloc_skb(length, GFP_ATOMIC);
1494 if (!skb)
1495 continue;
1496
1497 veth_build_dma_list(local_list, skb->data, length);
1498
1499 rc = HvCallEvent_dmaBufList(HvLpEvent_Type_VirtualLan,
1500 event->base_event.xSourceLp,
1501 HvLpDma_Direction_RemoteToLocal,
1502 cnx->src_inst,
1503 cnx->dst_inst,
1504 HvLpDma_AddressType_RealAddress,
1505 HvLpDma_AddressType_TceIndex,
Stephen Rothwell426c1a12005-10-14 14:51:42 +10001506 iseries_hv_addr(&local_list),
1507 iseries_hv_addr(&remote_list),
Linus Torvalds1da177e2005-04-16 15:20:36 -07001508 length);
1509 if (rc != HvLpDma_Rc_Good) {
1510 dev_kfree_skb_irq(skb);
1511 continue;
1512 }
1513
1514 vlan = skb->data[9];
1515 dev = veth_dev[vlan];
Michael Ellerman41664c02005-05-12 17:55:08 +10001516 if (! dev) {
1517 /*
1518 * Some earlier versions of the driver sent
1519 * broadcasts down all connections, even to lpars
1520 * that weren't on the relevant vlan. So ignore
1521 * packets belonging to a vlan we're not on.
1522 * We can also be here if we receive packets while
1523 * the driver is going down, because then dev is NULL.
1524 */
1525 dev_kfree_skb_irq(skb);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001526 continue;
Michael Ellerman41664c02005-05-12 17:55:08 +10001527 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001528
1529 port = (struct veth_port *)dev->priv;
1530 dest = *((u64 *) skb->data) & 0xFFFFFFFFFFFF0000;
1531
1532 if ((vlan > HVMAXARCHITECTEDVIRTUALLANS) || !port) {
1533 dev_kfree_skb_irq(skb);
1534 continue;
1535 }
1536 if (! veth_frame_wanted(port, dest)) {
1537 dev_kfree_skb_irq(skb);
1538 continue;
1539 }
1540
1541 skb_put(skb, length);
1542 skb->dev = dev;
1543 skb->protocol = eth_type_trans(skb, dev);
1544 skb->ip_summed = CHECKSUM_NONE;
1545 netif_rx(skb); /* send it up */
1546 port->stats.rx_packets++;
1547 port->stats.rx_bytes += length;
1548 } while (startchunk += nchunks, startchunk < VETH_MAX_FRAMES_PER_MSG);
1549
1550 /* Ack it */
1551 spin_lock_irqsave(&cnx->lock, flags);
1552 BUG_ON(cnx->num_pending_acks > VETH_MAX_ACKS_PER_MSG);
1553
1554 cnx->pending_acks[cnx->num_pending_acks++] =
1555 event->base_event.xCorrelationToken;
1556
1557 if ( (cnx->num_pending_acks >= cnx->remote_caps.ack_threshold)
1558 || (cnx->num_pending_acks >= VETH_MAX_ACKS_PER_MSG) )
1559 veth_flush_acks(cnx);
1560
1561 spin_unlock_irqrestore(&cnx->lock, flags);
1562}
1563
1564static void veth_timed_ack(unsigned long ptr)
1565{
1566 struct veth_lpar_connection *cnx = (struct veth_lpar_connection *) ptr;
1567 unsigned long flags;
1568
1569 /* Ack all the events */
1570 spin_lock_irqsave(&cnx->lock, flags);
1571 if (cnx->num_pending_acks > 0)
1572 veth_flush_acks(cnx);
1573
1574 /* Reschedule the timer */
1575 cnx->ack_timer.expires = jiffies + cnx->ack_timeout;
1576 add_timer(&cnx->ack_timer);
1577 spin_unlock_irqrestore(&cnx->lock, flags);
1578}
1579
1580static int veth_remove(struct vio_dev *vdev)
1581{
Michael Ellermanf0c129c2005-09-01 11:29:09 +10001582 struct veth_lpar_connection *cnx;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001583 struct net_device *dev;
Michael Ellermanf0c129c2005-09-01 11:29:09 +10001584 struct veth_port *port;
1585 int i;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001586
Michael Ellermanf0c129c2005-09-01 11:29:09 +10001587 dev = veth_dev[vdev->unit_address];
1588
1589 if (! dev)
1590 return 0;
1591
1592 port = netdev_priv(dev);
1593
1594 for (i = 0; i < HVMAXARCHITECTEDLPS; i++) {
1595 cnx = veth_cnx[i];
1596
1597 if (cnx && (port->lpar_map & (1 << i))) {
1598 /* Drop our reference to connections on our VLAN */
1599 kobject_put(&cnx->kobject);
1600 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001601 }
Michael Ellermanf0c129c2005-09-01 11:29:09 +10001602
1603 veth_dev[vdev->unit_address] = NULL;
Michael Ellerman07a5c172005-09-01 11:29:21 +10001604 kobject_del(&port->kobject);
1605 kobject_put(&port->kobject);
Michael Ellermanf0c129c2005-09-01 11:29:09 +10001606 unregister_netdev(dev);
1607 free_netdev(dev);
1608
Linus Torvalds1da177e2005-04-16 15:20:36 -07001609 return 0;
1610}
1611
1612static int veth_probe(struct vio_dev *vdev, const struct vio_device_id *id)
1613{
1614 int i = vdev->unit_address;
1615 struct net_device *dev;
Michael Ellermanf0c129c2005-09-01 11:29:09 +10001616 struct veth_port *port;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001617
Stephen Rothwell07fb3f42006-05-15 13:39:47 +10001618 dev = veth_probe_one(i, vdev);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001619 if (dev == NULL) {
1620 veth_remove(vdev);
1621 return 1;
1622 }
1623 veth_dev[i] = dev;
1624
Michael Ellermanf0c129c2005-09-01 11:29:09 +10001625 port = (struct veth_port*)netdev_priv(dev);
1626
1627 /* Start the state machine on each connection on this vlan. If we're
1628 * the first dev to do so this will commence link negotiation */
1629 for (i = 0; i < HVMAXARCHITECTEDLPS; i++) {
1630 struct veth_lpar_connection *cnx;
1631
1632 if (! (port->lpar_map & (1 << i)))
1633 continue;
1634
1635 cnx = veth_cnx[i];
1636 if (!cnx)
1637 continue;
1638
1639 kobject_get(&cnx->kobject);
1640 veth_kick_statemachine(cnx);
1641 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001642
1643 return 0;
1644}
1645
1646/**
1647 * veth_device_table: Used by vio.c to match devices that we
1648 * support.
1649 */
1650static struct vio_device_id veth_device_table[] __devinitdata = {
Stephen Rothwell07fb3f42006-05-15 13:39:47 +10001651 { "network", "IBM,iSeries-l-lan" },
Stephen Rothwellfb120da2005-08-17 16:42:59 +10001652 { "", "" }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001653};
1654MODULE_DEVICE_TABLE(vio, veth_device_table);
1655
1656static struct vio_driver veth_driver = {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001657 .id_table = veth_device_table,
1658 .probe = veth_probe,
Stephen Rothwell6fdf5392005-10-24 14:53:21 +10001659 .remove = veth_remove,
1660 .driver = {
1661 .name = DRV_NAME,
Stephen Rothwell915124d2005-10-24 15:12:22 +10001662 .owner = THIS_MODULE,
Stephen Rothwell6fdf5392005-10-24 14:53:21 +10001663 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001664};
1665
1666/*
1667 * Module initialization/cleanup
1668 */
1669
1670void __exit veth_module_cleanup(void)
1671{
1672 int i;
Michael Ellermanf0c129c2005-09-01 11:29:09 +10001673 struct veth_lpar_connection *cnx;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001674
Michael Ellermanf0c129c2005-09-01 11:29:09 +10001675 /* Disconnect our "irq" to stop events coming from the Hypervisor. */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001676 HvLpEvent_unregisterHandler(HvLpEvent_Type_VirtualLan);
1677
Michael Ellermanf0c129c2005-09-01 11:29:09 +10001678 /* Make sure any work queued from Hypervisor callbacks is finished. */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001679 flush_scheduled_work();
1680
Michael Ellermanf0c129c2005-09-01 11:29:09 +10001681 for (i = 0; i < HVMAXARCHITECTEDLPS; ++i) {
1682 cnx = veth_cnx[i];
1683
1684 if (!cnx)
1685 continue;
1686
Michael Ellerman76812d8122005-09-01 11:29:20 +10001687 /* Remove the connection from sysfs */
1688 kobject_del(&cnx->kobject);
Michael Ellermanf0c129c2005-09-01 11:29:09 +10001689 /* Drop the driver's reference to the connection */
1690 kobject_put(&cnx->kobject);
1691 }
1692
1693 /* Unregister the driver, which will close all the netdevs and stop
1694 * the connections when they're no longer referenced. */
Michael Ellermanb2e08522005-05-12 18:09:45 +10001695 vio_unregister_driver(&veth_driver);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001696}
1697module_exit(veth_module_cleanup);
1698
1699int __init veth_module_init(void)
1700{
1701 int i;
1702 int rc;
1703
1704 this_lp = HvLpConfig_getLpIndex_outline();
1705
1706 for (i = 0; i < HVMAXARCHITECTEDLPS; ++i) {
1707 rc = veth_init_connection(i);
Michael Ellermanec60bee2005-09-01 11:29:08 +10001708 if (rc != 0)
1709 goto error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001710 }
1711
1712 HvLpEvent_registerHandler(HvLpEvent_Type_VirtualLan,
1713 &veth_handle_event);
1714
Michael Ellermanec60bee2005-09-01 11:29:08 +10001715 rc = vio_register_driver(&veth_driver);
1716 if (rc != 0)
1717 goto error;
1718
Michael Ellerman76812d8122005-09-01 11:29:20 +10001719 for (i = 0; i < HVMAXARCHITECTEDLPS; ++i) {
1720 struct kobject *kobj;
1721
1722 if (!veth_cnx[i])
1723 continue;
1724
1725 kobj = &veth_cnx[i]->kobject;
1726 kobj->parent = &veth_driver.driver.kobj;
1727 /* If the add failes, complain but otherwise continue */
1728 if (0 != kobject_add(kobj))
1729 veth_error("cnx %d: Failed adding to sysfs.\n", i);
1730 }
1731
Michael Ellermanec60bee2005-09-01 11:29:08 +10001732 return 0;
1733
1734error:
1735 for (i = 0; i < HVMAXARCHITECTEDLPS; ++i) {
Michael Ellermanf0c129c2005-09-01 11:29:09 +10001736 veth_destroy_connection(veth_cnx[i]);
Michael Ellermanec60bee2005-09-01 11:29:08 +10001737 }
1738
1739 return rc;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001740}
1741module_init(veth_module_init);