1 /* File veth.c created by Kyle A. Lucke on Mon Aug 7 2000. */
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.
7 * Copyright (C) 2004-2005 Michael Ellerman, IBM Corporation.
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.
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.
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
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
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.
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.
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.
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.
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.
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.
59 #include <linux/module.h>
60 #include <linux/types.h>
61 #include <linux/errno.h>
62 #include <linux/ioport.h>
63 #include <linux/kernel.h>
64 #include <linux/netdevice.h>
65 #include <linux/etherdevice.h>
66 #include <linux/skbuff.h>
67 #include <linux/init.h>
68 #include <linux/delay.h>
70 #include <linux/ethtool.h>
71 #include <linux/if_ether.h>
72 #include <linux/slab.h>
74 #include <asm/abs_addr.h>
75 #include <asm/iseries/mf.h>
76 #include <asm/uaccess.h>
77 #include <asm/firmware.h>
78 #include <asm/iseries/hv_lp_config.h>
79 #include <asm/iseries/hv_types.h>
80 #include <asm/iseries/hv_lp_event.h>
81 #include <asm/iommu.h>
86 MODULE_AUTHOR("Kyle Lucke <klucke@us.ibm.com>");
87 MODULE_DESCRIPTION("iSeries Virtual ethernet driver");
88 MODULE_LICENSE("GPL");
90 #define VETH_EVENT_CAP (0)
91 #define VETH_EVENT_FRAMES (1)
92 #define VETH_EVENT_MONITOR (2)
93 #define VETH_EVENT_FRAMES_ACK (3)
95 #define VETH_MAX_ACKS_PER_MSG (20)
96 #define VETH_MAX_FRAMES_PER_MSG (6)
98 struct veth_frames_data {
99 u32 addr[VETH_MAX_FRAMES_PER_MSG];
100 u16 len[VETH_MAX_FRAMES_PER_MSG];
103 #define VETH_EOF_SHIFT (32-VETH_MAX_FRAMES_PER_MSG)
105 struct veth_frames_ack_data {
106 u16 token[VETH_MAX_ACKS_PER_MSG];
109 struct veth_cap_data {
120 struct veth_lpevent {
121 struct HvLpEvent base_event;
123 struct veth_cap_data caps_data;
124 struct veth_frames_data frames_data;
125 struct veth_frames_ack_data frames_ack_data;
130 #define DRV_NAME "iseries_veth"
131 #define DRV_VERSION "2.0"
133 #define VETH_NUMBUFFERS (120)
134 #define VETH_ACKTIMEOUT (1000000) /* microseconds */
135 #define VETH_MAX_MCAST (12)
137 #define VETH_MAX_MTU (9000)
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)
146 #define ACK_THRESHOLD (20)
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)
160 struct veth_msg *next;
161 struct veth_frames_data data;
168 struct veth_lpar_connection {
170 struct delayed_work statemachine_wq;
171 struct veth_msg *msgs;
173 struct veth_cap_data local_caps;
175 struct kobject kobject;
176 struct timer_list ack_timer;
178 struct timer_list reset_timer;
179 unsigned int reset_timeout;
180 unsigned long last_contact;
185 HvLpInstanceId src_inst;
186 HvLpInstanceId dst_inst;
187 struct veth_lpevent cap_event, cap_ack_event;
188 u16 pending_acks[VETH_MAX_ACKS_PER_MSG];
189 u32 num_pending_acks;
192 struct veth_cap_data remote_caps;
195 struct veth_msg *msg_stack_head;
201 HvLpIndexMap lpar_map;
203 /* queue_lock protects the stopped_map and dev's queue. */
204 spinlock_t queue_lock;
205 HvLpIndexMap stopped_map;
207 /* mcast_gate protects promiscuous, num_mcast & mcast_addr. */
211 u64 mcast_addr[VETH_MAX_MCAST];
213 struct kobject kobject;
216 static HvLpIndex this_lp;
217 static struct veth_lpar_connection *veth_cnx[HVMAXARCHITECTEDLPS]; /* = 0 */
218 static struct net_device *veth_dev[HVMAXARCHITECTEDVIRTUALLANS]; /* = 0 */
220 static int veth_start_xmit(struct sk_buff *skb, struct net_device *dev);
221 static void veth_recycle_msg(struct veth_lpar_connection *, struct veth_msg *);
222 static void veth_wake_queues(struct veth_lpar_connection *cnx);
223 static void veth_stop_queues(struct veth_lpar_connection *cnx);
224 static void veth_receive(struct veth_lpar_connection *, struct veth_lpevent *);
225 static void veth_release_connection(struct kobject *kobject);
226 static void veth_timed_ack(unsigned long ptr);
227 static void veth_timed_reset(unsigned long ptr);
233 #define veth_info(fmt, args...) \
234 printk(KERN_INFO DRV_NAME ": " fmt, ## args)
236 #define veth_error(fmt, args...) \
237 printk(KERN_ERR DRV_NAME ": Error: " fmt, ## args)
240 #define veth_debug(fmt, args...) \
241 printk(KERN_DEBUG DRV_NAME ": " fmt, ## args)
243 #define veth_debug(fmt, args...) do {} while (0)
246 /* You must hold the connection's lock when you call this function. */
247 static inline void veth_stack_push(struct veth_lpar_connection *cnx,
248 struct veth_msg *msg)
250 msg->next = cnx->msg_stack_head;
251 cnx->msg_stack_head = msg;
254 /* You must hold the connection's lock when you call this function. */
255 static inline struct veth_msg *veth_stack_pop(struct veth_lpar_connection *cnx)
257 struct veth_msg *msg;
259 msg = cnx->msg_stack_head;
261 cnx->msg_stack_head = cnx->msg_stack_head->next;
266 /* You must hold the connection's lock when you call this function. */
267 static inline int veth_stack_is_empty(struct veth_lpar_connection *cnx)
269 return cnx->msg_stack_head == NULL;
272 static inline HvLpEvent_Rc
273 veth_signalevent(struct veth_lpar_connection *cnx, u16 subtype,
274 HvLpEvent_AckInd ackind, HvLpEvent_AckType acktype,
276 u64 data1, u64 data2, u64 data3, u64 data4, u64 data5)
278 return HvCallEvent_signalLpEventFast(cnx->remote_lp,
279 HvLpEvent_Type_VirtualLan,
280 subtype, ackind, acktype,
283 token, data1, data2, data3,
287 static inline HvLpEvent_Rc veth_signaldata(struct veth_lpar_connection *cnx,
288 u16 subtype, u64 token, void *data)
290 u64 *p = (u64 *) data;
292 return veth_signalevent(cnx, subtype, HvLpEvent_AckInd_NoAck,
293 HvLpEvent_AckType_ImmediateAck,
294 token, p[0], p[1], p[2], p[3], p[4]);
297 struct veth_allocation {
302 static void veth_complete_allocation(void *parm, int number)
304 struct veth_allocation *vc = (struct veth_allocation *)parm;
310 static int veth_allocate_events(HvLpIndex rlp, int number)
312 struct veth_allocation vc =
313 { COMPLETION_INITIALIZER_ONSTACK(vc.c), 0 };
315 mf_allocate_lp_events(rlp, HvLpEvent_Type_VirtualLan,
316 sizeof(struct veth_lpevent), number,
317 &veth_complete_allocation, &vc);
318 wait_for_completion(&vc.c);
327 struct 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);
333 static ssize_t veth_cnx_attribute_show(struct kobject *kobj,
334 struct attribute *attr, char *buf)
336 struct veth_cnx_attribute *cnx_attr;
337 struct veth_lpar_connection *cnx;
339 cnx_attr = container_of(attr, struct veth_cnx_attribute, attr);
340 cnx = container_of(kobj, struct veth_lpar_connection, kobject);
345 return cnx_attr->show(cnx, buf);
348 #define CUSTOM_CNX_ATTR(_name, _format, _expression) \
349 static ssize_t _name##_show(struct veth_lpar_connection *cnx, char *buf)\
351 return sprintf(buf, _format, _expression); \
353 struct veth_cnx_attribute veth_cnx_attr_##_name = __ATTR_RO(_name)
355 #define SIMPLE_CNX_ATTR(_name) \
356 CUSTOM_CNX_ATTR(_name, "%lu\n", (unsigned long)cnx->_name)
358 SIMPLE_CNX_ATTR(outstanding_tx);
359 SIMPLE_CNX_ATTR(remote_lp);
360 SIMPLE_CNX_ATTR(num_events);
361 SIMPLE_CNX_ATTR(src_inst);
362 SIMPLE_CNX_ATTR(dst_inst);
363 SIMPLE_CNX_ATTR(num_pending_acks);
364 SIMPLE_CNX_ATTR(num_ack_events);
365 CUSTOM_CNX_ATTR(ack_timeout, "%d\n", jiffies_to_msecs(cnx->ack_timeout));
366 CUSTOM_CNX_ATTR(reset_timeout, "%d\n", jiffies_to_msecs(cnx->reset_timeout));
367 CUSTOM_CNX_ATTR(state, "0x%.4lX\n", cnx->state);
368 CUSTOM_CNX_ATTR(last_contact, "%d\n", cnx->last_contact ?
369 jiffies_to_msecs(jiffies - cnx->last_contact) : 0);
371 #define GET_CNX_ATTR(_name) (&veth_cnx_attr_##_name.attr)
373 static 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),
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),
388 static const struct sysfs_ops veth_cnx_sysfs_ops = {
389 .show = veth_cnx_attribute_show
392 static 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
398 struct 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);
404 static ssize_t veth_port_attribute_show(struct kobject *kobj,
405 struct attribute *attr, char *buf)
407 struct veth_port_attribute *port_attr;
408 struct veth_port *port;
410 port_attr = container_of(attr, struct veth_port_attribute, attr);
411 port = container_of(kobj, struct veth_port, kobject);
413 if (!port_attr->show)
416 return port_attr->show(port, buf);
419 #define CUSTOM_PORT_ATTR(_name, _format, _expression) \
420 static ssize_t _name##_show(struct veth_port *port, char *buf) \
422 return sprintf(buf, _format, _expression); \
424 struct veth_port_attribute veth_port_attr_##_name = __ATTR_RO(_name)
426 #define SIMPLE_PORT_ATTR(_name) \
427 CUSTOM_PORT_ATTR(_name, "%lu\n", (unsigned long)port->_name)
429 SIMPLE_PORT_ATTR(promiscuous);
430 SIMPLE_PORT_ATTR(num_mcast);
431 CUSTOM_PORT_ATTR(lpar_map, "0x%X\n", port->lpar_map);
432 CUSTOM_PORT_ATTR(stopped_map, "0x%X\n", port->stopped_map);
433 CUSTOM_PORT_ATTR(mac_addr, "0x%llX\n", port->mac_addr);
435 #define GET_PORT_ATTR(_name) (&veth_port_attr_##_name.attr)
436 static 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),
445 static const struct sysfs_ops veth_port_sysfs_ops = {
446 .show = veth_port_attribute_show
449 static struct kobj_type veth_port_ktype = {
450 .sysfs_ops = &veth_port_sysfs_ops,
451 .default_attrs = veth_port_default_attrs
455 * LPAR connection code
458 static inline void veth_kick_statemachine(struct veth_lpar_connection *cnx)
460 schedule_delayed_work(&cnx->statemachine_wq, 0);
463 static void veth_take_cap(struct veth_lpar_connection *cnx,
464 struct veth_lpevent *event)
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 */
472 HvCallEvent_getTargetLpInstanceId(cnx->remote_lp,
473 HvLpEvent_Type_VirtualLan);
475 if (cnx->state & VETH_STATE_GOTCAPS) {
476 veth_error("Received a second capabilities from LPAR %d.\n",
478 event->base_event.xRc = HvLpEvent_Rc_BufferNotAvailable;
479 HvCallEvent_ackLpEvent((struct HvLpEvent *) event);
481 memcpy(&cnx->cap_event, event, sizeof(cnx->cap_event));
482 cnx->state |= VETH_STATE_GOTCAPS;
483 veth_kick_statemachine(cnx);
485 spin_unlock_irqrestore(&cnx->lock, flags);
488 static void veth_take_cap_ack(struct veth_lpar_connection *cnx,
489 struct veth_lpevent *event)
493 spin_lock_irqsave(&cnx->lock, flags);
494 if (cnx->state & VETH_STATE_GOTCAPACK) {
495 veth_error("Received a second capabilities ack from LPAR %d.\n",
498 memcpy(&cnx->cap_ack_event, event,
499 sizeof(cnx->cap_ack_event));
500 cnx->state |= VETH_STATE_GOTCAPACK;
501 veth_kick_statemachine(cnx);
503 spin_unlock_irqrestore(&cnx->lock, flags);
506 static void veth_take_monitor_ack(struct veth_lpar_connection *cnx,
507 struct veth_lpevent *event)
511 spin_lock_irqsave(&cnx->lock, flags);
512 veth_debug("cnx %d: lost connection.\n", cnx->remote_lp);
514 /* Avoid kicking the statemachine once we're shutdown.
515 * It's unnecessary and it could break veth_stop_connection(). */
517 if (! (cnx->state & VETH_STATE_SHUTDOWN)) {
518 cnx->state |= VETH_STATE_RESET;
519 veth_kick_statemachine(cnx);
521 spin_unlock_irqrestore(&cnx->lock, flags);
524 static void veth_handle_ack(struct veth_lpevent *event)
526 HvLpIndex rlp = event->base_event.xTargetLp;
527 struct veth_lpar_connection *cnx = veth_cnx[rlp];
531 switch (event->base_event.xSubtype) {
533 veth_take_cap_ack(cnx, event);
535 case VETH_EVENT_MONITOR:
536 veth_take_monitor_ack(cnx, event);
539 veth_error("Unknown ack type %d from LPAR %d.\n",
540 event->base_event.xSubtype, rlp);
544 static void veth_handle_int(struct veth_lpevent *event)
546 HvLpIndex rlp = event->base_event.xSourceLp;
547 struct veth_lpar_connection *cnx = veth_cnx[rlp];
553 switch (event->base_event.xSubtype) {
555 veth_take_cap(cnx, event);
557 case VETH_EVENT_MONITOR:
558 /* do nothing... this'll hang out here til we're dead,
559 * and the hypervisor will return it for us. */
561 case VETH_EVENT_FRAMES_ACK:
562 spin_lock_irqsave(&cnx->lock, flags);
564 for (i = 0; i < VETH_MAX_ACKS_PER_MSG; ++i) {
565 u16 msgnum = event->u.frames_ack_data.token[i];
567 if (msgnum < VETH_NUMBUFFERS) {
568 veth_recycle_msg(cnx, cnx->msgs + msgnum);
569 cnx->outstanding_tx--;
575 cnx->last_contact = jiffies;
576 veth_wake_queues(cnx);
579 spin_unlock_irqrestore(&cnx->lock, flags);
581 case VETH_EVENT_FRAMES:
582 veth_receive(cnx, event);
585 veth_error("Unknown interrupt type %d from LPAR %d.\n",
586 event->base_event.xSubtype, rlp);
590 static void veth_handle_event(struct HvLpEvent *event)
592 struct veth_lpevent *veth_event = (struct veth_lpevent *)event;
594 if (hvlpevent_is_ack(event))
595 veth_handle_ack(veth_event);
597 veth_handle_int(veth_event);
600 static int veth_process_caps(struct veth_lpar_connection *cnx)
602 struct veth_cap_data *remote_caps = &cnx->remote_caps;
605 /* Convert timer to jiffies */
606 cnx->ack_timeout = remote_caps->ack_timeout * HZ / 1000000;
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) ) {
612 veth_error("Received incompatible capabilities from LPAR %d.\n",
614 return HvLpEvent_Rc_InvalidSubtypeData;
617 num_acks_needed = (remote_caps->num_buffers
618 / remote_caps->ack_threshold) + 1;
620 /* FIXME: locking on num_ack_events? */
621 if (cnx->num_ack_events < num_acks_needed) {
624 num = veth_allocate_events(cnx->remote_lp,
625 num_acks_needed-cnx->num_ack_events);
627 cnx->num_ack_events += num;
629 if (cnx->num_ack_events < num_acks_needed) {
630 veth_error("Couldn't allocate enough ack events "
631 "for LPAR %d.\n", cnx->remote_lp);
633 return HvLpEvent_Rc_BufferNotAvailable;
638 return HvLpEvent_Rc_Good;
641 /* FIXME: The gotos here are a bit dubious */
642 static void veth_statemachine(struct work_struct *work)
644 struct veth_lpar_connection *cnx =
645 container_of(work, struct veth_lpar_connection,
646 statemachine_wq.work);
647 int rlp = cnx->remote_lp;
650 spin_lock_irq(&cnx->lock);
653 if (cnx->state & VETH_STATE_RESET) {
654 if (cnx->state & VETH_STATE_OPEN)
655 HvCallEvent_closeLpEventPath(cnx->remote_lp,
656 HvLpEvent_Type_VirtualLan);
659 * Reset ack data. This prevents the ack_timer actually
660 * doing anything, even if it runs one more time when
661 * we drop the lock below.
663 memset(&cnx->pending_acks, 0xff, sizeof (cnx->pending_acks));
664 cnx->num_pending_acks = 0;
666 cnx->state &= ~(VETH_STATE_RESET | VETH_STATE_SENTMON
667 | VETH_STATE_OPEN | VETH_STATE_SENTCAPS
668 | VETH_STATE_GOTCAPACK | VETH_STATE_GOTCAPS
669 | VETH_STATE_SENTCAPACK | VETH_STATE_READY);
671 /* Clean up any leftover messages */
674 for (i = 0; i < VETH_NUMBUFFERS; ++i)
675 veth_recycle_msg(cnx, cnx->msgs + i);
678 cnx->outstanding_tx = 0;
679 veth_wake_queues(cnx);
681 /* Drop the lock so we can do stuff that might sleep or
682 * take other locks. */
683 spin_unlock_irq(&cnx->lock);
685 del_timer_sync(&cnx->ack_timer);
686 del_timer_sync(&cnx->reset_timer);
688 spin_lock_irq(&cnx->lock);
690 if (cnx->state & VETH_STATE_RESET)
693 /* Hack, wait for the other end to reset itself. */
694 if (! (cnx->state & VETH_STATE_SHUTDOWN)) {
695 schedule_delayed_work(&cnx->statemachine_wq, 5 * HZ);
700 if (cnx->state & VETH_STATE_SHUTDOWN)
701 /* It's all over, do nothing */
704 if ( !(cnx->state & VETH_STATE_OPEN) ) {
705 if (! cnx->msgs || (cnx->num_events < (2 + VETH_NUMBUFFERS)) )
708 HvCallEvent_openLpEventPath(rlp, HvLpEvent_Type_VirtualLan);
710 HvCallEvent_getSourceLpInstanceId(rlp,
711 HvLpEvent_Type_VirtualLan);
713 HvCallEvent_getTargetLpInstanceId(rlp,
714 HvLpEvent_Type_VirtualLan);
715 cnx->state |= VETH_STATE_OPEN;
718 if ( (cnx->state & VETH_STATE_OPEN) &&
719 !(cnx->state & VETH_STATE_SENTMON) ) {
720 rc = veth_signalevent(cnx, VETH_EVENT_MONITOR,
721 HvLpEvent_AckInd_DoAck,
722 HvLpEvent_AckType_DeferredAck,
725 if (rc == HvLpEvent_Rc_Good) {
726 cnx->state |= VETH_STATE_SENTMON;
728 if ( (rc != HvLpEvent_Rc_PartitionDead) &&
729 (rc != HvLpEvent_Rc_PathClosed) )
730 veth_error("Error sending monitor to LPAR %d, "
731 "rc = %d\n", rlp, rc);
733 /* Oh well, hope we get a cap from the other
734 * end and do better when that kicks us */
739 if ( (cnx->state & VETH_STATE_OPEN) &&
740 !(cnx->state & VETH_STATE_SENTCAPS)) {
741 u64 *rawcap = (u64 *)&cnx->local_caps;
743 rc = veth_signalevent(cnx, VETH_EVENT_CAP,
744 HvLpEvent_AckInd_DoAck,
745 HvLpEvent_AckType_ImmediateAck,
746 0, rawcap[0], rawcap[1], rawcap[2],
747 rawcap[3], rawcap[4]);
749 if (rc == HvLpEvent_Rc_Good) {
750 cnx->state |= VETH_STATE_SENTCAPS;
752 if ( (rc != HvLpEvent_Rc_PartitionDead) &&
753 (rc != HvLpEvent_Rc_PathClosed) )
754 veth_error("Error sending caps to LPAR %d, "
755 "rc = %d\n", rlp, rc);
757 /* Oh well, hope we get a cap from the other
758 * end and do better when that kicks us */
763 if ((cnx->state & VETH_STATE_GOTCAPS) &&
764 !(cnx->state & VETH_STATE_SENTCAPACK)) {
765 struct veth_cap_data *remote_caps = &cnx->remote_caps;
767 memcpy(remote_caps, &cnx->cap_event.u.caps_data,
768 sizeof(*remote_caps));
770 spin_unlock_irq(&cnx->lock);
771 rc = veth_process_caps(cnx);
772 spin_lock_irq(&cnx->lock);
774 /* We dropped the lock, so recheck for anything which
775 * might mess us up */
776 if (cnx->state & (VETH_STATE_RESET|VETH_STATE_SHUTDOWN))
779 cnx->cap_event.base_event.xRc = rc;
780 HvCallEvent_ackLpEvent((struct HvLpEvent *)&cnx->cap_event);
781 if (rc == HvLpEvent_Rc_Good)
782 cnx->state |= VETH_STATE_SENTCAPACK;
787 if ((cnx->state & VETH_STATE_GOTCAPACK) &&
788 (cnx->state & VETH_STATE_GOTCAPS) &&
789 !(cnx->state & VETH_STATE_READY)) {
790 if (cnx->cap_ack_event.base_event.xRc == HvLpEvent_Rc_Good) {
791 /* Start the ACK timer */
792 cnx->ack_timer.expires = jiffies + cnx->ack_timeout;
793 add_timer(&cnx->ack_timer);
794 cnx->state |= VETH_STATE_READY;
796 veth_error("Caps rejected by LPAR %d, rc = %d\n",
797 rlp, cnx->cap_ack_event.base_event.xRc);
803 spin_unlock_irq(&cnx->lock);
807 /* FIXME: we get here if something happens we really can't
808 * cope with. The link will never work once we get here, and
809 * all we can do is not lock the rest of the system up */
810 veth_error("Unrecoverable error on connection to LPAR %d, shutting down"
811 " (state = 0x%04lx)\n", rlp, cnx->state);
812 cnx->state |= VETH_STATE_SHUTDOWN;
813 spin_unlock_irq(&cnx->lock);
816 static int veth_init_connection(u8 rlp)
818 struct veth_lpar_connection *cnx;
819 struct veth_msg *msgs;
822 if ( (rlp == this_lp) ||
823 ! HvLpConfig_doLpsCommunicateOnVirtualLan(this_lp, rlp) )
826 cnx = kzalloc(sizeof(*cnx), GFP_KERNEL);
830 cnx->remote_lp = rlp;
831 spin_lock_init(&cnx->lock);
832 INIT_DELAYED_WORK(&cnx->statemachine_wq, veth_statemachine);
834 init_timer(&cnx->ack_timer);
835 cnx->ack_timer.function = veth_timed_ack;
836 cnx->ack_timer.data = (unsigned long) cnx;
838 init_timer(&cnx->reset_timer);
839 cnx->reset_timer.function = veth_timed_reset;
840 cnx->reset_timer.data = (unsigned long) cnx;
841 cnx->reset_timeout = 5 * HZ * (VETH_ACKTIMEOUT / 1000000);
843 memset(&cnx->pending_acks, 0xff, sizeof (cnx->pending_acks));
847 /* This gets us 1 reference, which is held on behalf of the driver
848 * infrastructure. It's released at module unload. */
849 kobject_init(&cnx->kobject, &veth_lpar_connection_ktype);
851 msgs = kcalloc(VETH_NUMBUFFERS, sizeof(struct veth_msg), GFP_KERNEL);
853 veth_error("Can't allocate buffers for LPAR %d.\n", rlp);
859 for (i = 0; i < VETH_NUMBUFFERS; i++) {
861 veth_stack_push(cnx, msgs + i);
864 cnx->num_events = veth_allocate_events(rlp, 2 + VETH_NUMBUFFERS);
866 if (cnx->num_events < (2 + VETH_NUMBUFFERS)) {
867 veth_error("Can't allocate enough events for LPAR %d.\n", rlp);
871 cnx->local_caps.num_buffers = VETH_NUMBUFFERS;
872 cnx->local_caps.ack_threshold = ACK_THRESHOLD;
873 cnx->local_caps.ack_timeout = VETH_ACKTIMEOUT;
878 static void veth_stop_connection(struct veth_lpar_connection *cnx)
883 spin_lock_irq(&cnx->lock);
884 cnx->state |= VETH_STATE_RESET | VETH_STATE_SHUTDOWN;
885 veth_kick_statemachine(cnx);
886 spin_unlock_irq(&cnx->lock);
888 /* There's a slim chance the reset code has just queued the
889 * statemachine to run in five seconds. If so we need to cancel
890 * that and requeue the work to run now. */
891 if (cancel_delayed_work(&cnx->statemachine_wq)) {
892 spin_lock_irq(&cnx->lock);
893 veth_kick_statemachine(cnx);
894 spin_unlock_irq(&cnx->lock);
897 /* Wait for the state machine to run. */
898 flush_scheduled_work();
901 static void veth_destroy_connection(struct veth_lpar_connection *cnx)
906 if (cnx->num_events > 0)
907 mf_deallocate_lp_events(cnx->remote_lp,
908 HvLpEvent_Type_VirtualLan,
911 if (cnx->num_ack_events > 0)
912 mf_deallocate_lp_events(cnx->remote_lp,
913 HvLpEvent_Type_VirtualLan,
918 veth_cnx[cnx->remote_lp] = NULL;
922 static void veth_release_connection(struct kobject *kobj)
924 struct veth_lpar_connection *cnx;
925 cnx = container_of(kobj, struct veth_lpar_connection, kobject);
926 veth_stop_connection(cnx);
927 veth_destroy_connection(cnx);
934 static int veth_open(struct net_device *dev)
936 netif_start_queue(dev);
940 static int veth_close(struct net_device *dev)
942 netif_stop_queue(dev);
946 static int veth_change_mtu(struct net_device *dev, int new_mtu)
948 if ((new_mtu < 68) || (new_mtu > VETH_MAX_MTU))
954 static void veth_set_multicast_list(struct net_device *dev)
956 struct veth_port *port = netdev_priv(dev);
959 write_lock_irqsave(&port->mcast_gate, flags);
961 if ((dev->flags & IFF_PROMISC) || (dev->flags & IFF_ALLMULTI) ||
962 (netdev_mc_count(dev) > VETH_MAX_MCAST)) {
963 port->promiscuous = 1;
965 struct netdev_hw_addr *ha;
967 port->promiscuous = 0;
972 netdev_for_each_mc_addr(ha, dev) {
976 if (addr[0] & 0x01) {/* multicast address? */
977 memcpy(&xaddr, addr, ETH_ALEN);
978 port->mcast_addr[port->num_mcast] = xaddr;
984 write_unlock_irqrestore(&port->mcast_gate, flags);
987 static void veth_get_drvinfo(struct net_device *dev, struct ethtool_drvinfo *info)
989 strncpy(info->driver, DRV_NAME, sizeof(info->driver) - 1);
990 info->driver[sizeof(info->driver) - 1] = '\0';
991 strncpy(info->version, DRV_VERSION, sizeof(info->version) - 1);
992 info->version[sizeof(info->version) - 1] = '\0';
995 static int veth_get_settings(struct net_device *dev, struct ethtool_cmd *ecmd)
997 ecmd->supported = (SUPPORTED_1000baseT_Full
998 | SUPPORTED_Autoneg | SUPPORTED_FIBRE);
999 ecmd->advertising = (SUPPORTED_1000baseT_Full
1000 | SUPPORTED_Autoneg | SUPPORTED_FIBRE);
1001 ecmd->port = PORT_FIBRE;
1002 ecmd->transceiver = XCVR_INTERNAL;
1003 ecmd->phy_address = 0;
1004 ecmd->speed = SPEED_1000;
1005 ecmd->duplex = DUPLEX_FULL;
1006 ecmd->autoneg = AUTONEG_ENABLE;
1007 ecmd->maxtxpkt = 120;
1008 ecmd->maxrxpkt = 120;
1012 static const struct ethtool_ops ops = {
1013 .get_drvinfo = veth_get_drvinfo,
1014 .get_settings = veth_get_settings,
1015 .get_link = ethtool_op_get_link,
1018 static const struct net_device_ops veth_netdev_ops = {
1019 .ndo_open = veth_open,
1020 .ndo_stop = veth_close,
1021 .ndo_start_xmit = veth_start_xmit,
1022 .ndo_change_mtu = veth_change_mtu,
1023 .ndo_set_multicast_list = veth_set_multicast_list,
1024 .ndo_set_mac_address = NULL,
1025 .ndo_validate_addr = eth_validate_addr,
1028 static struct net_device *veth_probe_one(int vlan,
1029 struct vio_dev *vio_dev)
1031 struct net_device *dev;
1032 struct veth_port *port;
1033 struct device *vdev = &vio_dev->dev;
1035 const unsigned char *mac_addr;
1037 mac_addr = vio_get_attribute(vio_dev, "local-mac-address", NULL);
1038 if (mac_addr == NULL)
1039 mac_addr = vio_get_attribute(vio_dev, "mac-address", NULL);
1040 if (mac_addr == NULL) {
1041 veth_error("Unable to fetch MAC address from device tree.\n");
1045 dev = alloc_etherdev(sizeof (struct veth_port));
1047 veth_error("Unable to allocate net_device structure!\n");
1051 port = netdev_priv(dev);
1053 spin_lock_init(&port->queue_lock);
1054 rwlock_init(&port->mcast_gate);
1055 port->stopped_map = 0;
1057 for (i = 0; i < HVMAXARCHITECTEDLPS; i++) {
1058 HvLpVirtualLanIndexMap map;
1062 map = HvLpConfig_getVirtualLanIndexMapForLp(i);
1063 if (map & (0x8000 >> vlan))
1064 port->lpar_map |= (1 << i);
1068 memcpy(dev->dev_addr, mac_addr, ETH_ALEN);
1070 dev->mtu = VETH_MAX_MTU;
1072 memcpy(&port->mac_addr, mac_addr, ETH_ALEN);
1074 dev->netdev_ops = &veth_netdev_ops;
1075 SET_ETHTOOL_OPS(dev, &ops);
1077 SET_NETDEV_DEV(dev, vdev);
1079 rc = register_netdev(dev);
1081 veth_error("Failed registering net device for vlan%d.\n", vlan);
1086 kobject_init(&port->kobject, &veth_port_ktype);
1087 if (0 != kobject_add(&port->kobject, &dev->dev.kobj, "veth_port"))
1088 veth_error("Failed adding port for %s to sysfs.\n", dev->name);
1090 veth_info("%s attached to iSeries vlan %d (LPAR map = 0x%.4X)\n",
1091 dev->name, vlan, port->lpar_map);
1100 static int veth_transmit_to_one(struct sk_buff *skb, HvLpIndex rlp,
1101 struct net_device *dev)
1103 struct veth_lpar_connection *cnx = veth_cnx[rlp];
1104 struct veth_port *port = netdev_priv(dev);
1106 struct veth_msg *msg = NULL;
1107 unsigned long flags;
1112 spin_lock_irqsave(&cnx->lock, flags);
1114 if (! (cnx->state & VETH_STATE_READY))
1117 if ((skb->len - ETH_HLEN) > VETH_MAX_MTU)
1120 msg = veth_stack_pop(cnx);
1125 msg->skb = skb_get(skb);
1127 msg->data.addr[0] = dma_map_single(port->dev, skb->data,
1128 skb->len, DMA_TO_DEVICE);
1130 if (dma_mapping_error(port->dev, msg->data.addr[0]))
1131 goto recycle_and_drop;
1133 msg->dev = port->dev;
1134 msg->data.len[0] = skb->len;
1135 msg->data.eofmask = 1 << VETH_EOF_SHIFT;
1137 rc = veth_signaldata(cnx, VETH_EVENT_FRAMES, msg->token, &msg->data);
1139 if (rc != HvLpEvent_Rc_Good)
1140 goto recycle_and_drop;
1142 /* If the timer's not already running, start it now. */
1143 if (0 == cnx->outstanding_tx)
1144 mod_timer(&cnx->reset_timer, jiffies + cnx->reset_timeout);
1146 cnx->last_contact = jiffies;
1147 cnx->outstanding_tx++;
1149 if (veth_stack_is_empty(cnx))
1150 veth_stop_queues(cnx);
1153 spin_unlock_irqrestore(&cnx->lock, flags);
1157 veth_recycle_msg(cnx, msg);
1159 spin_unlock_irqrestore(&cnx->lock, flags);
1163 static void veth_transmit_to_many(struct sk_buff *skb,
1164 HvLpIndexMap lpmask,
1165 struct net_device *dev)
1167 int i, success, error;
1169 success = error = 0;
1171 for (i = 0; i < HVMAXARCHITECTEDLPS; i++) {
1172 if ((lpmask & (1 << i)) == 0)
1175 if (veth_transmit_to_one(skb, i, dev))
1182 dev->stats.tx_errors++;
1185 dev->stats.tx_packets++;
1186 dev->stats.tx_bytes += skb->len;
1190 static int veth_start_xmit(struct sk_buff *skb, struct net_device *dev)
1192 unsigned char *frame = skb->data;
1193 struct veth_port *port = netdev_priv(dev);
1194 HvLpIndexMap lpmask;
1196 if (! (frame[0] & 0x01)) {
1197 /* unicast packet */
1198 HvLpIndex rlp = frame[5];
1200 if ( ! ((1 << rlp) & port->lpar_map) ) {
1202 return NETDEV_TX_OK;
1207 lpmask = port->lpar_map;
1210 veth_transmit_to_many(skb, lpmask, dev);
1214 return NETDEV_TX_OK;
1217 /* You must hold the connection's lock when you call this function. */
1218 static void veth_recycle_msg(struct veth_lpar_connection *cnx,
1219 struct veth_msg *msg)
1221 u32 dma_address, dma_length;
1225 dma_address = msg->data.addr[0];
1226 dma_length = msg->data.len[0];
1228 if (!dma_mapping_error(msg->dev, dma_address))
1229 dma_unmap_single(msg->dev, dma_address, dma_length,
1233 dev_kfree_skb_any(msg->skb);
1237 memset(&msg->data, 0, sizeof(msg->data));
1238 veth_stack_push(cnx, msg);
1239 } else if (cnx->state & VETH_STATE_OPEN) {
1240 veth_error("Non-pending frame (# %d) acked by LPAR %d.\n",
1241 cnx->remote_lp, msg->token);
1245 static void veth_wake_queues(struct veth_lpar_connection *cnx)
1249 for (i = 0; i < HVMAXARCHITECTEDVIRTUALLANS; i++) {
1250 struct net_device *dev = veth_dev[i];
1251 struct veth_port *port;
1252 unsigned long flags;
1257 port = netdev_priv(dev);
1259 if (! (port->lpar_map & (1<<cnx->remote_lp)))
1262 spin_lock_irqsave(&port->queue_lock, flags);
1264 port->stopped_map &= ~(1 << cnx->remote_lp);
1266 if (0 == port->stopped_map && netif_queue_stopped(dev)) {
1267 veth_debug("cnx %d: woke queue for %s.\n",
1268 cnx->remote_lp, dev->name);
1269 netif_wake_queue(dev);
1271 spin_unlock_irqrestore(&port->queue_lock, flags);
1275 static void veth_stop_queues(struct veth_lpar_connection *cnx)
1279 for (i = 0; i < HVMAXARCHITECTEDVIRTUALLANS; i++) {
1280 struct net_device *dev = veth_dev[i];
1281 struct veth_port *port;
1286 port = netdev_priv(dev);
1288 /* If this cnx is not on the vlan for this port, continue */
1289 if (! (port->lpar_map & (1 << cnx->remote_lp)))
1292 spin_lock(&port->queue_lock);
1294 netif_stop_queue(dev);
1295 port->stopped_map |= (1 << cnx->remote_lp);
1297 veth_debug("cnx %d: stopped queue for %s, map = 0x%x.\n",
1298 cnx->remote_lp, dev->name, port->stopped_map);
1300 spin_unlock(&port->queue_lock);
1304 static void veth_timed_reset(unsigned long ptr)
1306 struct veth_lpar_connection *cnx = (struct veth_lpar_connection *)ptr;
1307 unsigned long trigger_time, flags;
1309 /* FIXME is it possible this fires after veth_stop_connection()?
1310 * That would reschedule the statemachine for 5 seconds and probably
1311 * execute it after the module's been unloaded. Hmm. */
1313 spin_lock_irqsave(&cnx->lock, flags);
1315 if (cnx->outstanding_tx > 0) {
1316 trigger_time = cnx->last_contact + cnx->reset_timeout;
1318 if (trigger_time < jiffies) {
1319 cnx->state |= VETH_STATE_RESET;
1320 veth_kick_statemachine(cnx);
1321 veth_error("%d packets not acked by LPAR %d within %d "
1322 "seconds, resetting.\n",
1323 cnx->outstanding_tx, cnx->remote_lp,
1324 cnx->reset_timeout / HZ);
1326 /* Reschedule the timer */
1327 trigger_time = jiffies + cnx->reset_timeout;
1328 mod_timer(&cnx->reset_timer, trigger_time);
1332 spin_unlock_irqrestore(&cnx->lock, flags);
1339 static inline int veth_frame_wanted(struct veth_port *port, u64 mac_addr)
1343 unsigned long flags;
1345 if ( (mac_addr == port->mac_addr) || (mac_addr == 0xffffffffffff0000) )
1348 read_lock_irqsave(&port->mcast_gate, flags);
1350 if (port->promiscuous) {
1355 for (i = 0; i < port->num_mcast; ++i) {
1356 if (port->mcast_addr[i] == mac_addr) {
1363 read_unlock_irqrestore(&port->mcast_gate, flags);
1373 #define VETH_MAX_PAGES_PER_FRAME ( (VETH_MAX_MTU+PAGE_SIZE-2)/PAGE_SIZE + 1 )
1375 static inline void veth_build_dma_list(struct dma_chunk *list,
1376 unsigned char *p, unsigned long length)
1381 /* FIXME: skbs are contiguous in real addresses. Do we
1382 * really need to break it into PAGE_SIZE chunks, or can we do
1383 * it just at the granularity of iSeries real->absolute
1384 * mapping? Indeed, given the way the allocator works, can we
1385 * count on them being absolutely contiguous? */
1386 list[0].addr = iseries_hv_addr(p);
1387 list[0].size = min(length,
1388 PAGE_SIZE - ((unsigned long)p & ~PAGE_MASK));
1390 done = list[0].size;
1391 while (done < length) {
1392 list[i].addr = iseries_hv_addr(p + done);
1393 list[i].size = min(length-done, PAGE_SIZE);
1394 done += list[i].size;
1399 static void veth_flush_acks(struct veth_lpar_connection *cnx)
1403 rc = veth_signaldata(cnx, VETH_EVENT_FRAMES_ACK,
1404 0, &cnx->pending_acks);
1406 if (rc != HvLpEvent_Rc_Good)
1407 veth_error("Failed acking frames from LPAR %d, rc = %d\n",
1408 cnx->remote_lp, (int)rc);
1410 cnx->num_pending_acks = 0;
1411 memset(&cnx->pending_acks, 0xff, sizeof(cnx->pending_acks));
1414 static void veth_receive(struct veth_lpar_connection *cnx,
1415 struct veth_lpevent *event)
1417 struct veth_frames_data *senddata = &event->u.frames_data;
1420 unsigned long flags;
1425 struct sk_buff *skb;
1426 struct dma_chunk local_list[VETH_MAX_PAGES_PER_FRAME];
1427 struct dma_chunk remote_list[VETH_MAX_FRAMES_PER_MSG];
1429 HvLpVirtualLanIndex vlan;
1430 struct net_device *dev;
1431 struct veth_port *port;
1433 /* FIXME: do we need this? */
1434 memset(local_list, 0, sizeof(local_list));
1435 memset(remote_list, 0, sizeof(VETH_MAX_FRAMES_PER_MSG));
1437 /* a 0 address marks the end of the valid entries */
1438 if (senddata->addr[startchunk] == 0)
1441 /* make sure that we have at least 1 EOF entry in the
1442 * remaining entries */
1443 if (! (senddata->eofmask >> (startchunk + VETH_EOF_SHIFT))) {
1444 veth_error("Missing EOF fragment in event "
1445 "eofmask = 0x%x startchunk = %d\n",
1446 (unsigned)senddata->eofmask,
1451 /* build list of chunks in this frame */
1454 remote_list[nchunks].addr =
1455 (u64) senddata->addr[startchunk+nchunks] << 32;
1456 remote_list[nchunks].size =
1457 senddata->len[startchunk+nchunks];
1458 length += remote_list[nchunks].size;
1459 } while (! (senddata->eofmask &
1460 (1 << (VETH_EOF_SHIFT + startchunk + nchunks++))));
1462 /* length == total length of all chunks */
1463 /* nchunks == # of chunks in this frame */
1465 if ((length - ETH_HLEN) > VETH_MAX_MTU) {
1466 veth_error("Received oversize frame from LPAR %d "
1468 cnx->remote_lp, length);
1472 skb = alloc_skb(length, GFP_ATOMIC);
1476 veth_build_dma_list(local_list, skb->data, length);
1478 rc = HvCallEvent_dmaBufList(HvLpEvent_Type_VirtualLan,
1479 event->base_event.xSourceLp,
1480 HvLpDma_Direction_RemoteToLocal,
1483 HvLpDma_AddressType_RealAddress,
1484 HvLpDma_AddressType_TceIndex,
1485 iseries_hv_addr(&local_list),
1486 iseries_hv_addr(&remote_list),
1488 if (rc != HvLpDma_Rc_Good) {
1489 dev_kfree_skb_irq(skb);
1493 vlan = skb->data[9];
1494 dev = veth_dev[vlan];
1497 * Some earlier versions of the driver sent
1498 * broadcasts down all connections, even to lpars
1499 * that weren't on the relevant vlan. So ignore
1500 * packets belonging to a vlan we're not on.
1501 * We can also be here if we receive packets while
1502 * the driver is going down, because then dev is NULL.
1504 dev_kfree_skb_irq(skb);
1508 port = netdev_priv(dev);
1509 dest = *((u64 *) skb->data) & 0xFFFFFFFFFFFF0000;
1511 if ((vlan > HVMAXARCHITECTEDVIRTUALLANS) || !port) {
1512 dev_kfree_skb_irq(skb);
1515 if (! veth_frame_wanted(port, dest)) {
1516 dev_kfree_skb_irq(skb);
1520 skb_put(skb, length);
1521 skb->protocol = eth_type_trans(skb, dev);
1522 skb_checksum_none_assert(skb);
1523 netif_rx(skb); /* send it up */
1524 dev->stats.rx_packets++;
1525 dev->stats.rx_bytes += length;
1526 } while (startchunk += nchunks, startchunk < VETH_MAX_FRAMES_PER_MSG);
1529 spin_lock_irqsave(&cnx->lock, flags);
1530 BUG_ON(cnx->num_pending_acks > VETH_MAX_ACKS_PER_MSG);
1532 cnx->pending_acks[cnx->num_pending_acks++] =
1533 event->base_event.xCorrelationToken;
1535 if ( (cnx->num_pending_acks >= cnx->remote_caps.ack_threshold) ||
1536 (cnx->num_pending_acks >= VETH_MAX_ACKS_PER_MSG) )
1537 veth_flush_acks(cnx);
1539 spin_unlock_irqrestore(&cnx->lock, flags);
1542 static void veth_timed_ack(unsigned long ptr)
1544 struct veth_lpar_connection *cnx = (struct veth_lpar_connection *) ptr;
1545 unsigned long flags;
1547 /* Ack all the events */
1548 spin_lock_irqsave(&cnx->lock, flags);
1549 if (cnx->num_pending_acks > 0)
1550 veth_flush_acks(cnx);
1552 /* Reschedule the timer */
1553 cnx->ack_timer.expires = jiffies + cnx->ack_timeout;
1554 add_timer(&cnx->ack_timer);
1555 spin_unlock_irqrestore(&cnx->lock, flags);
1558 static int veth_remove(struct vio_dev *vdev)
1560 struct veth_lpar_connection *cnx;
1561 struct net_device *dev;
1562 struct veth_port *port;
1565 dev = veth_dev[vdev->unit_address];
1570 port = netdev_priv(dev);
1572 for (i = 0; i < HVMAXARCHITECTEDLPS; i++) {
1575 if (cnx && (port->lpar_map & (1 << i))) {
1576 /* Drop our reference to connections on our VLAN */
1577 kobject_put(&cnx->kobject);
1581 veth_dev[vdev->unit_address] = NULL;
1582 kobject_del(&port->kobject);
1583 kobject_put(&port->kobject);
1584 unregister_netdev(dev);
1590 static int veth_probe(struct vio_dev *vdev, const struct vio_device_id *id)
1592 int i = vdev->unit_address;
1593 struct net_device *dev;
1594 struct veth_port *port;
1596 dev = veth_probe_one(i, vdev);
1603 port = netdev_priv(dev);
1605 /* Start the state machine on each connection on this vlan. If we're
1606 * the first dev to do so this will commence link negotiation */
1607 for (i = 0; i < HVMAXARCHITECTEDLPS; i++) {
1608 struct veth_lpar_connection *cnx;
1610 if (! (port->lpar_map & (1 << i)))
1617 kobject_get(&cnx->kobject);
1618 veth_kick_statemachine(cnx);
1625 * veth_device_table: Used by vio.c to match devices that we
1628 static struct vio_device_id veth_device_table[] __devinitdata = {
1629 { "network", "IBM,iSeries-l-lan" },
1632 MODULE_DEVICE_TABLE(vio, veth_device_table);
1634 static struct vio_driver veth_driver = {
1635 .id_table = veth_device_table,
1636 .probe = veth_probe,
1637 .remove = veth_remove,
1640 .owner = THIS_MODULE,
1645 * Module initialization/cleanup
1648 static void __exit veth_module_cleanup(void)
1651 struct veth_lpar_connection *cnx;
1653 /* Disconnect our "irq" to stop events coming from the Hypervisor. */
1654 HvLpEvent_unregisterHandler(HvLpEvent_Type_VirtualLan);
1656 /* Make sure any work queued from Hypervisor callbacks is finished. */
1657 flush_scheduled_work();
1659 for (i = 0; i < HVMAXARCHITECTEDLPS; ++i) {
1665 /* Remove the connection from sysfs */
1666 kobject_del(&cnx->kobject);
1667 /* Drop the driver's reference to the connection */
1668 kobject_put(&cnx->kobject);
1671 /* Unregister the driver, which will close all the netdevs and stop
1672 * the connections when they're no longer referenced. */
1673 vio_unregister_driver(&veth_driver);
1675 module_exit(veth_module_cleanup);
1677 static int __init veth_module_init(void)
1682 if (!firmware_has_feature(FW_FEATURE_ISERIES))
1685 this_lp = HvLpConfig_getLpIndex_outline();
1687 for (i = 0; i < HVMAXARCHITECTEDLPS; ++i) {
1688 rc = veth_init_connection(i);
1693 HvLpEvent_registerHandler(HvLpEvent_Type_VirtualLan,
1694 &veth_handle_event);
1696 rc = vio_register_driver(&veth_driver);
1700 for (i = 0; i < HVMAXARCHITECTEDLPS; ++i) {
1701 struct kobject *kobj;
1706 kobj = &veth_cnx[i]->kobject;
1707 /* If the add failes, complain but otherwise continue */
1708 if (0 != driver_add_kobj(&veth_driver.driver, kobj,
1709 "cnx%.2d", veth_cnx[i]->remote_lp))
1710 veth_error("cnx %d: Failed adding to sysfs.\n", i);
1716 for (i = 0; i < HVMAXARCHITECTEDLPS; ++i) {
1717 veth_destroy_connection(veth_cnx[i]);
1722 module_init(veth_module_init);