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.
8 * This program is free software; you can redistribute it and/or
9 * modify it under the terms of the GNU General Public License as
10 * published by the Free Software Foundation; either version 2 of the
11 * License, or (at your option) any later version.
13 * This program is distributed in the hope that it will be useful, but
14 * WITHOUT ANY WARRANTY; without even the implied warranty of
15 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
16 * General Public License for more details.
18 * You should have received a copy of the GNU General Public License
19 * along with this program; if not, write to the Free Software
20 * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307
24 * This module implements the virtual ethernet device for iSeries LPAR
25 * Linux. It uses hypervisor message passing to implement an
26 * ethernet-like network device communicating between partitions on
29 * The iSeries LPAR hypervisor currently allows for up to 16 different
30 * virtual ethernets. These are all dynamically configurable on
31 * OS/400 partitions, but dynamic configuration is not supported under
32 * Linux yet. An ethXX network device will be created for each
33 * virtual ethernet this partition is connected to.
35 * - This driver is responsible for routing packets to and from other
36 * partitions. The MAC addresses used by the virtual ethernets
37 * contains meaning and must not be modified.
39 * - Having 2 virtual ethernets to the same remote partition DOES NOT
40 * double the available bandwidth. The 2 devices will share the
41 * available hypervisor bandwidth.
43 * - If you send a packet to your own mac address, it will just be
44 * dropped, you won't get it on the receive side.
46 * - Multicast is implemented by sending the frame frame to every
47 * other partition. It is the responsibility of the receiving
48 * partition to filter the addresses desired.
52 * VETH_NUMBUFFERS: This compile time option defaults to 120. It
53 * controls how much memory Linux will allocate per remote partition
54 * it is communicating with. It can be thought of as the maximum
55 * number of packets outstanding to a remote partition at a time.
58 #include <linux/config.h>
59 #include <linux/module.h>
60 #include <linux/version.h>
61 #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>
71 #include <linux/ethtool.h>
72 #include <asm/iSeries/mf.h>
73 #include <asm/iSeries/iSeries_pci.h>
74 #include <asm/uaccess.h>
76 #include <asm/iSeries/HvLpConfig.h>
77 #include <asm/iSeries/HvTypes.h>
78 #include <asm/iSeries/HvLpEvent.h>
79 #include <asm/iommu.h>
84 #include "iseries_veth.h"
86 MODULE_AUTHOR("Kyle Lucke <klucke@us.ibm.com>");
87 MODULE_DESCRIPTION("iSeries Virtual ethernet driver");
88 MODULE_LICENSE("GPL");
90 #define VETH_NUMBUFFERS (120)
91 #define VETH_ACKTIMEOUT (1000000) /* microseconds */
92 #define VETH_MAX_MCAST (12)
94 #define VETH_MAX_MTU (9000)
96 #if VETH_NUMBUFFERS < 10
97 #define ACK_THRESHOLD (1)
98 #elif VETH_NUMBUFFERS < 20
99 #define ACK_THRESHOLD (4)
100 #elif VETH_NUMBUFFERS < 40
101 #define ACK_THRESHOLD (10)
103 #define ACK_THRESHOLD (20)
106 #define VETH_STATE_SHUTDOWN (0x0001)
107 #define VETH_STATE_OPEN (0x0002)
108 #define VETH_STATE_RESET (0x0004)
109 #define VETH_STATE_SENTMON (0x0008)
110 #define VETH_STATE_SENTCAPS (0x0010)
111 #define VETH_STATE_GOTCAPACK (0x0020)
112 #define VETH_STATE_GOTCAPS (0x0040)
113 #define VETH_STATE_SENTCAPACK (0x0080)
114 #define VETH_STATE_READY (0x0100)
117 struct veth_msg *next;
118 struct VethFramesData data;
125 struct veth_lpar_connection {
127 struct work_struct statemachine_wq;
128 struct veth_msg *msgs;
130 struct VethCapData local_caps;
132 struct kobject kobject;
133 struct timer_list ack_timer;
135 struct timer_list reset_timer;
136 unsigned int reset_timeout;
137 unsigned long last_contact;
142 HvLpInstanceId src_inst;
143 HvLpInstanceId dst_inst;
144 struct VethLpEvent cap_event, cap_ack_event;
145 u16 pending_acks[VETH_MAX_ACKS_PER_MSG];
146 u32 num_pending_acks;
149 struct VethCapData remote_caps;
152 struct veth_msg *msg_stack_head;
157 struct net_device_stats stats;
159 HvLpIndexMap lpar_map;
161 /* queue_lock protects the stopped_map and dev's queue. */
162 spinlock_t queue_lock;
163 HvLpIndexMap stopped_map;
165 /* mcast_gate protects promiscuous, num_mcast & mcast_addr. */
169 u64 mcast_addr[VETH_MAX_MCAST];
172 static HvLpIndex this_lp;
173 static struct veth_lpar_connection *veth_cnx[HVMAXARCHITECTEDLPS]; /* = 0 */
174 static struct net_device *veth_dev[HVMAXARCHITECTEDVIRTUALLANS]; /* = 0 */
176 static int veth_start_xmit(struct sk_buff *skb, struct net_device *dev);
177 static void veth_recycle_msg(struct veth_lpar_connection *, struct veth_msg *);
178 static void veth_wake_queues(struct veth_lpar_connection *cnx);
179 static void veth_stop_queues(struct veth_lpar_connection *cnx);
180 static void veth_receive(struct veth_lpar_connection *, struct VethLpEvent *);
181 static void veth_release_connection(struct kobject *kobject);
182 static void veth_timed_ack(unsigned long ptr);
183 static void veth_timed_reset(unsigned long ptr);
185 static struct kobj_type veth_lpar_connection_ktype = {
186 .release = veth_release_connection
193 #define veth_info(fmt, args...) \
194 printk(KERN_INFO "iseries_veth: " fmt, ## args)
196 #define veth_error(fmt, args...) \
197 printk(KERN_ERR "iseries_veth: Error: " fmt, ## args)
200 #define veth_debug(fmt, args...) \
201 printk(KERN_DEBUG "iseries_veth: " fmt, ## args)
203 #define veth_debug(fmt, args...) do {} while (0)
206 /* You must hold the connection's lock when you call this function. */
207 static inline void veth_stack_push(struct veth_lpar_connection *cnx,
208 struct veth_msg *msg)
210 msg->next = cnx->msg_stack_head;
211 cnx->msg_stack_head = msg;
214 /* You must hold the connection's lock when you call this function. */
215 static inline struct veth_msg *veth_stack_pop(struct veth_lpar_connection *cnx)
217 struct veth_msg *msg;
219 msg = cnx->msg_stack_head;
221 cnx->msg_stack_head = cnx->msg_stack_head->next;
226 /* You must hold the connection's lock when you call this function. */
227 static inline int veth_stack_is_empty(struct veth_lpar_connection *cnx)
229 return cnx->msg_stack_head == NULL;
232 static inline HvLpEvent_Rc
233 veth_signalevent(struct veth_lpar_connection *cnx, u16 subtype,
234 HvLpEvent_AckInd ackind, HvLpEvent_AckType acktype,
236 u64 data1, u64 data2, u64 data3, u64 data4, u64 data5)
238 return HvCallEvent_signalLpEventFast(cnx->remote_lp,
239 HvLpEvent_Type_VirtualLan,
240 subtype, ackind, acktype,
243 token, data1, data2, data3,
247 static inline HvLpEvent_Rc veth_signaldata(struct veth_lpar_connection *cnx,
248 u16 subtype, u64 token, void *data)
250 u64 *p = (u64 *) data;
252 return veth_signalevent(cnx, subtype, HvLpEvent_AckInd_NoAck,
253 HvLpEvent_AckType_ImmediateAck,
254 token, p[0], p[1], p[2], p[3], p[4]);
257 struct veth_allocation {
262 static void veth_complete_allocation(void *parm, int number)
264 struct veth_allocation *vc = (struct veth_allocation *)parm;
270 static int veth_allocate_events(HvLpIndex rlp, int number)
272 struct veth_allocation vc = { COMPLETION_INITIALIZER(vc.c), 0 };
274 mf_allocate_lp_events(rlp, HvLpEvent_Type_VirtualLan,
275 sizeof(struct VethLpEvent), number,
276 &veth_complete_allocation, &vc);
277 wait_for_completion(&vc.c);
283 * LPAR connection code
286 static inline void veth_kick_statemachine(struct veth_lpar_connection *cnx)
288 schedule_work(&cnx->statemachine_wq);
291 static void veth_take_cap(struct veth_lpar_connection *cnx,
292 struct VethLpEvent *event)
296 spin_lock_irqsave(&cnx->lock, flags);
297 /* Receiving caps may mean the other end has just come up, so
298 * we need to reload the instance ID of the far end */
300 HvCallEvent_getTargetLpInstanceId(cnx->remote_lp,
301 HvLpEvent_Type_VirtualLan);
303 if (cnx->state & VETH_STATE_GOTCAPS) {
304 veth_error("Received a second capabilities from LPAR %d.\n",
306 event->base_event.xRc = HvLpEvent_Rc_BufferNotAvailable;
307 HvCallEvent_ackLpEvent((struct HvLpEvent *) event);
309 memcpy(&cnx->cap_event, event, sizeof(cnx->cap_event));
310 cnx->state |= VETH_STATE_GOTCAPS;
311 veth_kick_statemachine(cnx);
313 spin_unlock_irqrestore(&cnx->lock, flags);
316 static void veth_take_cap_ack(struct veth_lpar_connection *cnx,
317 struct VethLpEvent *event)
321 spin_lock_irqsave(&cnx->lock, flags);
322 if (cnx->state & VETH_STATE_GOTCAPACK) {
323 veth_error("Received a second capabilities ack from LPAR %d.\n",
326 memcpy(&cnx->cap_ack_event, event,
327 sizeof(&cnx->cap_ack_event));
328 cnx->state |= VETH_STATE_GOTCAPACK;
329 veth_kick_statemachine(cnx);
331 spin_unlock_irqrestore(&cnx->lock, flags);
334 static void veth_take_monitor_ack(struct veth_lpar_connection *cnx,
335 struct VethLpEvent *event)
339 spin_lock_irqsave(&cnx->lock, flags);
340 veth_debug("cnx %d: lost connection.\n", cnx->remote_lp);
342 /* Avoid kicking the statemachine once we're shutdown.
343 * It's unnecessary and it could break veth_stop_connection(). */
345 if (! (cnx->state & VETH_STATE_SHUTDOWN)) {
346 cnx->state |= VETH_STATE_RESET;
347 veth_kick_statemachine(cnx);
349 spin_unlock_irqrestore(&cnx->lock, flags);
352 static void veth_handle_ack(struct VethLpEvent *event)
354 HvLpIndex rlp = event->base_event.xTargetLp;
355 struct veth_lpar_connection *cnx = veth_cnx[rlp];
359 switch (event->base_event.xSubtype) {
360 case VethEventTypeCap:
361 veth_take_cap_ack(cnx, event);
363 case VethEventTypeMonitor:
364 veth_take_monitor_ack(cnx, event);
367 veth_error("Unknown ack type %d from LPAR %d.\n",
368 event->base_event.xSubtype, rlp);
372 static void veth_handle_int(struct VethLpEvent *event)
374 HvLpIndex rlp = event->base_event.xSourceLp;
375 struct veth_lpar_connection *cnx = veth_cnx[rlp];
381 switch (event->base_event.xSubtype) {
382 case VethEventTypeCap:
383 veth_take_cap(cnx, event);
385 case VethEventTypeMonitor:
386 /* do nothing... this'll hang out here til we're dead,
387 * and the hypervisor will return it for us. */
389 case VethEventTypeFramesAck:
390 spin_lock_irqsave(&cnx->lock, flags);
392 for (i = 0; i < VETH_MAX_ACKS_PER_MSG; ++i) {
393 u16 msgnum = event->u.frames_ack_data.token[i];
395 if (msgnum < VETH_NUMBUFFERS) {
396 veth_recycle_msg(cnx, cnx->msgs + msgnum);
397 cnx->outstanding_tx--;
403 cnx->last_contact = jiffies;
404 veth_wake_queues(cnx);
407 spin_unlock_irqrestore(&cnx->lock, flags);
409 case VethEventTypeFrames:
410 veth_receive(cnx, event);
413 veth_error("Unknown interrupt type %d from LPAR %d.\n",
414 event->base_event.xSubtype, rlp);
418 static void veth_handle_event(struct HvLpEvent *event, struct pt_regs *regs)
420 struct VethLpEvent *veth_event = (struct VethLpEvent *)event;
422 if (event->xFlags.xFunction == HvLpEvent_Function_Ack)
423 veth_handle_ack(veth_event);
424 else if (event->xFlags.xFunction == HvLpEvent_Function_Int)
425 veth_handle_int(veth_event);
428 static int veth_process_caps(struct veth_lpar_connection *cnx)
430 struct VethCapData *remote_caps = &cnx->remote_caps;
433 /* Convert timer to jiffies */
434 cnx->ack_timeout = remote_caps->ack_timeout * HZ / 1000000;
436 if ( (remote_caps->num_buffers == 0)
437 || (remote_caps->ack_threshold > VETH_MAX_ACKS_PER_MSG)
438 || (remote_caps->ack_threshold == 0)
439 || (cnx->ack_timeout == 0) ) {
440 veth_error("Received incompatible capabilities from LPAR %d.\n",
442 return HvLpEvent_Rc_InvalidSubtypeData;
445 num_acks_needed = (remote_caps->num_buffers
446 / remote_caps->ack_threshold) + 1;
448 /* FIXME: locking on num_ack_events? */
449 if (cnx->num_ack_events < num_acks_needed) {
452 num = veth_allocate_events(cnx->remote_lp,
453 num_acks_needed-cnx->num_ack_events);
455 cnx->num_ack_events += num;
457 if (cnx->num_ack_events < num_acks_needed) {
458 veth_error("Couldn't allocate enough ack events "
459 "for LPAR %d.\n", cnx->remote_lp);
461 return HvLpEvent_Rc_BufferNotAvailable;
466 return HvLpEvent_Rc_Good;
469 /* FIXME: The gotos here are a bit dubious */
470 static void veth_statemachine(void *p)
472 struct veth_lpar_connection *cnx = (struct veth_lpar_connection *)p;
473 int rlp = cnx->remote_lp;
476 spin_lock_irq(&cnx->lock);
479 if (cnx->state & VETH_STATE_RESET) {
480 if (cnx->state & VETH_STATE_OPEN)
481 HvCallEvent_closeLpEventPath(cnx->remote_lp,
482 HvLpEvent_Type_VirtualLan);
485 * Reset ack data. This prevents the ack_timer actually
486 * doing anything, even if it runs one more time when
487 * we drop the lock below.
489 memset(&cnx->pending_acks, 0xff, sizeof (cnx->pending_acks));
490 cnx->num_pending_acks = 0;
492 cnx->state &= ~(VETH_STATE_RESET | VETH_STATE_SENTMON
493 | VETH_STATE_OPEN | VETH_STATE_SENTCAPS
494 | VETH_STATE_GOTCAPACK | VETH_STATE_GOTCAPS
495 | VETH_STATE_SENTCAPACK | VETH_STATE_READY);
497 /* Clean up any leftover messages */
500 for (i = 0; i < VETH_NUMBUFFERS; ++i)
501 veth_recycle_msg(cnx, cnx->msgs + i);
504 cnx->outstanding_tx = 0;
505 veth_wake_queues(cnx);
507 /* Drop the lock so we can do stuff that might sleep or
508 * take other locks. */
509 spin_unlock_irq(&cnx->lock);
511 del_timer_sync(&cnx->ack_timer);
512 del_timer_sync(&cnx->reset_timer);
514 spin_lock_irq(&cnx->lock);
516 if (cnx->state & VETH_STATE_RESET)
519 /* Hack, wait for the other end to reset itself. */
520 if (! (cnx->state & VETH_STATE_SHUTDOWN)) {
521 schedule_delayed_work(&cnx->statemachine_wq, 5 * HZ);
526 if (cnx->state & VETH_STATE_SHUTDOWN)
527 /* It's all over, do nothing */
530 if ( !(cnx->state & VETH_STATE_OPEN) ) {
531 if (! cnx->msgs || (cnx->num_events < (2 + VETH_NUMBUFFERS)) )
534 HvCallEvent_openLpEventPath(rlp, HvLpEvent_Type_VirtualLan);
536 HvCallEvent_getSourceLpInstanceId(rlp,
537 HvLpEvent_Type_VirtualLan);
539 HvCallEvent_getTargetLpInstanceId(rlp,
540 HvLpEvent_Type_VirtualLan);
541 cnx->state |= VETH_STATE_OPEN;
544 if ( (cnx->state & VETH_STATE_OPEN)
545 && !(cnx->state & VETH_STATE_SENTMON) ) {
546 rc = veth_signalevent(cnx, VethEventTypeMonitor,
547 HvLpEvent_AckInd_DoAck,
548 HvLpEvent_AckType_DeferredAck,
551 if (rc == HvLpEvent_Rc_Good) {
552 cnx->state |= VETH_STATE_SENTMON;
554 if ( (rc != HvLpEvent_Rc_PartitionDead)
555 && (rc != HvLpEvent_Rc_PathClosed) )
556 veth_error("Error sending monitor to LPAR %d, "
557 "rc = %d\n", rlp, rc);
559 /* Oh well, hope we get a cap from the other
560 * end and do better when that kicks us */
565 if ( (cnx->state & VETH_STATE_OPEN)
566 && !(cnx->state & VETH_STATE_SENTCAPS)) {
567 u64 *rawcap = (u64 *)&cnx->local_caps;
569 rc = veth_signalevent(cnx, VethEventTypeCap,
570 HvLpEvent_AckInd_DoAck,
571 HvLpEvent_AckType_ImmediateAck,
572 0, rawcap[0], rawcap[1], rawcap[2],
573 rawcap[3], rawcap[4]);
575 if (rc == HvLpEvent_Rc_Good) {
576 cnx->state |= VETH_STATE_SENTCAPS;
578 if ( (rc != HvLpEvent_Rc_PartitionDead)
579 && (rc != HvLpEvent_Rc_PathClosed) )
580 veth_error("Error sending caps to LPAR %d, "
581 "rc = %d\n", rlp, rc);
583 /* Oh well, hope we get a cap from the other
584 * end and do better when that kicks us */
589 if ((cnx->state & VETH_STATE_GOTCAPS)
590 && !(cnx->state & VETH_STATE_SENTCAPACK)) {
591 struct VethCapData *remote_caps = &cnx->remote_caps;
593 memcpy(remote_caps, &cnx->cap_event.u.caps_data,
594 sizeof(*remote_caps));
596 spin_unlock_irq(&cnx->lock);
597 rc = veth_process_caps(cnx);
598 spin_lock_irq(&cnx->lock);
600 /* We dropped the lock, so recheck for anything which
601 * might mess us up */
602 if (cnx->state & (VETH_STATE_RESET|VETH_STATE_SHUTDOWN))
605 cnx->cap_event.base_event.xRc = rc;
606 HvCallEvent_ackLpEvent((struct HvLpEvent *)&cnx->cap_event);
607 if (rc == HvLpEvent_Rc_Good)
608 cnx->state |= VETH_STATE_SENTCAPACK;
613 if ((cnx->state & VETH_STATE_GOTCAPACK)
614 && (cnx->state & VETH_STATE_GOTCAPS)
615 && !(cnx->state & VETH_STATE_READY)) {
616 if (cnx->cap_ack_event.base_event.xRc == HvLpEvent_Rc_Good) {
617 /* Start the ACK timer */
618 cnx->ack_timer.expires = jiffies + cnx->ack_timeout;
619 add_timer(&cnx->ack_timer);
620 cnx->state |= VETH_STATE_READY;
622 veth_error("Caps rejected by LPAR %d, rc = %d\n",
623 rlp, cnx->cap_ack_event.base_event.xRc);
629 spin_unlock_irq(&cnx->lock);
633 /* FIXME: we get here if something happens we really can't
634 * cope with. The link will never work once we get here, and
635 * all we can do is not lock the rest of the system up */
636 veth_error("Unrecoverable error on connection to LPAR %d, shutting down"
637 " (state = 0x%04lx)\n", rlp, cnx->state);
638 cnx->state |= VETH_STATE_SHUTDOWN;
639 spin_unlock_irq(&cnx->lock);
642 static int veth_init_connection(u8 rlp)
644 struct veth_lpar_connection *cnx;
645 struct veth_msg *msgs;
648 if ( (rlp == this_lp)
649 || ! HvLpConfig_doLpsCommunicateOnVirtualLan(this_lp, rlp) )
652 cnx = kmalloc(sizeof(*cnx), GFP_KERNEL);
655 memset(cnx, 0, sizeof(*cnx));
657 cnx->remote_lp = rlp;
658 spin_lock_init(&cnx->lock);
659 INIT_WORK(&cnx->statemachine_wq, veth_statemachine, cnx);
661 init_timer(&cnx->ack_timer);
662 cnx->ack_timer.function = veth_timed_ack;
663 cnx->ack_timer.data = (unsigned long) cnx;
665 init_timer(&cnx->reset_timer);
666 cnx->reset_timer.function = veth_timed_reset;
667 cnx->reset_timer.data = (unsigned long) cnx;
668 cnx->reset_timeout = 5 * HZ * (VETH_ACKTIMEOUT / 1000000);
670 memset(&cnx->pending_acks, 0xff, sizeof (cnx->pending_acks));
674 /* This gets us 1 reference, which is held on behalf of the driver
675 * infrastructure. It's released at module unload. */
676 kobject_init(&cnx->kobject);
677 cnx->kobject.ktype = &veth_lpar_connection_ktype;
678 rc = kobject_set_name(&cnx->kobject, "cnx%.2d", rlp);
682 msgs = kmalloc(VETH_NUMBUFFERS * sizeof(struct veth_msg), GFP_KERNEL);
684 veth_error("Can't allocate buffers for LPAR %d.\n", rlp);
689 memset(msgs, 0, VETH_NUMBUFFERS * sizeof(struct veth_msg));
691 for (i = 0; i < VETH_NUMBUFFERS; i++) {
693 veth_stack_push(cnx, msgs + i);
696 cnx->num_events = veth_allocate_events(rlp, 2 + VETH_NUMBUFFERS);
698 if (cnx->num_events < (2 + VETH_NUMBUFFERS)) {
699 veth_error("Can't allocate enough events for LPAR %d.\n", rlp);
703 cnx->local_caps.num_buffers = VETH_NUMBUFFERS;
704 cnx->local_caps.ack_threshold = ACK_THRESHOLD;
705 cnx->local_caps.ack_timeout = VETH_ACKTIMEOUT;
710 static void veth_stop_connection(struct veth_lpar_connection *cnx)
715 spin_lock_irq(&cnx->lock);
716 cnx->state |= VETH_STATE_RESET | VETH_STATE_SHUTDOWN;
717 veth_kick_statemachine(cnx);
718 spin_unlock_irq(&cnx->lock);
720 /* There's a slim chance the reset code has just queued the
721 * statemachine to run in five seconds. If so we need to cancel
722 * that and requeue the work to run now. */
723 if (cancel_delayed_work(&cnx->statemachine_wq)) {
724 spin_lock_irq(&cnx->lock);
725 veth_kick_statemachine(cnx);
726 spin_unlock_irq(&cnx->lock);
729 /* Wait for the state machine to run. */
730 flush_scheduled_work();
733 static void veth_destroy_connection(struct veth_lpar_connection *cnx)
738 if (cnx->num_events > 0)
739 mf_deallocate_lp_events(cnx->remote_lp,
740 HvLpEvent_Type_VirtualLan,
743 if (cnx->num_ack_events > 0)
744 mf_deallocate_lp_events(cnx->remote_lp,
745 HvLpEvent_Type_VirtualLan,
750 veth_cnx[cnx->remote_lp] = NULL;
754 static void veth_release_connection(struct kobject *kobj)
756 struct veth_lpar_connection *cnx;
757 cnx = container_of(kobj, struct veth_lpar_connection, kobject);
758 veth_stop_connection(cnx);
759 veth_destroy_connection(cnx);
766 static int veth_open(struct net_device *dev)
768 struct veth_port *port = (struct veth_port *) dev->priv;
770 memset(&port->stats, 0, sizeof (port->stats));
771 netif_start_queue(dev);
775 static int veth_close(struct net_device *dev)
777 netif_stop_queue(dev);
781 static struct net_device_stats *veth_get_stats(struct net_device *dev)
783 struct veth_port *port = (struct veth_port *) dev->priv;
788 static int veth_change_mtu(struct net_device *dev, int new_mtu)
790 if ((new_mtu < 68) || (new_mtu > VETH_MAX_MTU))
796 static void veth_set_multicast_list(struct net_device *dev)
798 struct veth_port *port = (struct veth_port *) dev->priv;
801 write_lock_irqsave(&port->mcast_gate, flags);
803 if ((dev->flags & IFF_PROMISC) || (dev->flags & IFF_ALLMULTI) ||
804 (dev->mc_count > VETH_MAX_MCAST)) {
805 port->promiscuous = 1;
807 struct dev_mc_list *dmi = dev->mc_list;
810 port->promiscuous = 0;
815 for (i = 0; i < dev->mc_count; i++) {
816 u8 *addr = dmi->dmi_addr;
819 if (addr[0] & 0x01) {/* multicast address? */
820 memcpy(&xaddr, addr, ETH_ALEN);
821 port->mcast_addr[port->num_mcast] = xaddr;
828 write_unlock_irqrestore(&port->mcast_gate, flags);
831 static void veth_get_drvinfo(struct net_device *dev, struct ethtool_drvinfo *info)
833 strncpy(info->driver, "veth", sizeof(info->driver) - 1);
834 info->driver[sizeof(info->driver) - 1] = '\0';
835 strncpy(info->version, "1.0", sizeof(info->version) - 1);
838 static int veth_get_settings(struct net_device *dev, struct ethtool_cmd *ecmd)
840 ecmd->supported = (SUPPORTED_1000baseT_Full
841 | SUPPORTED_Autoneg | SUPPORTED_FIBRE);
842 ecmd->advertising = (SUPPORTED_1000baseT_Full
843 | SUPPORTED_Autoneg | SUPPORTED_FIBRE);
844 ecmd->port = PORT_FIBRE;
845 ecmd->transceiver = XCVR_INTERNAL;
846 ecmd->phy_address = 0;
847 ecmd->speed = SPEED_1000;
848 ecmd->duplex = DUPLEX_FULL;
849 ecmd->autoneg = AUTONEG_ENABLE;
850 ecmd->maxtxpkt = 120;
851 ecmd->maxrxpkt = 120;
855 static u32 veth_get_link(struct net_device *dev)
860 static struct ethtool_ops ops = {
861 .get_drvinfo = veth_get_drvinfo,
862 .get_settings = veth_get_settings,
863 .get_link = veth_get_link,
866 static struct net_device * __init veth_probe_one(int vlan, struct device *vdev)
868 struct net_device *dev;
869 struct veth_port *port;
872 dev = alloc_etherdev(sizeof (struct veth_port));
874 veth_error("Unable to allocate net_device structure!\n");
878 port = (struct veth_port *) dev->priv;
880 spin_lock_init(&port->queue_lock);
881 rwlock_init(&port->mcast_gate);
882 port->stopped_map = 0;
884 for (i = 0; i < HVMAXARCHITECTEDLPS; i++) {
885 HvLpVirtualLanIndexMap map;
889 map = HvLpConfig_getVirtualLanIndexMapForLp(i);
890 if (map & (0x8000 >> vlan))
891 port->lpar_map |= (1 << i);
895 dev->dev_addr[0] = 0x02;
896 dev->dev_addr[1] = 0x01;
897 dev->dev_addr[2] = 0xff;
898 dev->dev_addr[3] = vlan;
899 dev->dev_addr[4] = 0xff;
900 dev->dev_addr[5] = this_lp;
902 dev->mtu = VETH_MAX_MTU;
904 memcpy(&port->mac_addr, dev->dev_addr, 6);
906 dev->open = veth_open;
907 dev->hard_start_xmit = veth_start_xmit;
908 dev->stop = veth_close;
909 dev->get_stats = veth_get_stats;
910 dev->change_mtu = veth_change_mtu;
911 dev->set_mac_address = NULL;
912 dev->set_multicast_list = veth_set_multicast_list;
913 SET_ETHTOOL_OPS(dev, &ops);
915 SET_NETDEV_DEV(dev, vdev);
917 rc = register_netdev(dev);
919 veth_error("Failed registering net device for vlan%d.\n", vlan);
924 veth_info("%s attached to iSeries vlan %d (LPAR map = 0x%.4X)\n",
925 dev->name, vlan, port->lpar_map);
934 static int veth_transmit_to_one(struct sk_buff *skb, HvLpIndex rlp,
935 struct net_device *dev)
937 struct veth_lpar_connection *cnx = veth_cnx[rlp];
938 struct veth_port *port = (struct veth_port *) dev->priv;
940 struct veth_msg *msg = NULL;
945 port->stats.tx_errors++;
950 spin_lock_irqsave(&cnx->lock, flags);
952 if (! (cnx->state & VETH_STATE_READY))
955 if ((skb->len - 14) > VETH_MAX_MTU)
958 msg = veth_stack_pop(cnx);
967 msg->data.addr[0] = dma_map_single(port->dev, skb->data,
968 skb->len, DMA_TO_DEVICE);
970 if (dma_mapping_error(msg->data.addr[0]))
971 goto recycle_and_drop;
973 /* Is it really necessary to check the length and address
974 * fields of the first entry here? */
976 msg->dev = port->dev;
977 msg->data.len[0] = skb->len;
978 msg->data.eofmask = 1 << VETH_EOF_SHIFT;
980 rc = veth_signaldata(cnx, VethEventTypeFrames, msg->token, &msg->data);
982 if (rc != HvLpEvent_Rc_Good)
983 goto recycle_and_drop;
985 /* If the timer's not already running, start it now. */
986 if (0 == cnx->outstanding_tx)
987 mod_timer(&cnx->reset_timer, jiffies + cnx->reset_timeout);
989 cnx->last_contact = jiffies;
990 cnx->outstanding_tx++;
992 if (veth_stack_is_empty(cnx))
993 veth_stop_queues(cnx);
995 spin_unlock_irqrestore(&cnx->lock, flags);
999 /* we free the skb below, so tell veth_recycle_msg() not to. */
1001 veth_recycle_msg(cnx, msg);
1003 port->stats.tx_errors++;
1005 spin_unlock_irqrestore(&cnx->lock, flags);
1009 static HvLpIndexMap veth_transmit_to_many(struct sk_buff *skb,
1010 HvLpIndexMap lpmask,
1011 struct net_device *dev)
1013 struct veth_port *port = (struct veth_port *) dev->priv;
1017 for (i = 0; i < HVMAXARCHITECTEDLPS; i++) {
1018 if ((lpmask & (1 << i)) == 0)
1021 rc = veth_transmit_to_one(skb_get(skb), i, dev);
1027 port->stats.tx_packets++;
1028 port->stats.tx_bytes += skb->len;
1034 static int veth_start_xmit(struct sk_buff *skb, struct net_device *dev)
1036 unsigned char *frame = skb->data;
1037 struct veth_port *port = (struct veth_port *) dev->priv;
1038 HvLpIndexMap lpmask;
1040 if (! (frame[0] & 0x01)) {
1041 /* unicast packet */
1042 HvLpIndex rlp = frame[5];
1044 if ( ! ((1 << rlp) & port->lpar_map) ) {
1051 lpmask = port->lpar_map;
1054 veth_transmit_to_many(skb, lpmask, dev);
1061 /* You must hold the connection's lock when you call this function. */
1062 static void veth_recycle_msg(struct veth_lpar_connection *cnx,
1063 struct veth_msg *msg)
1065 u32 dma_address, dma_length;
1069 dma_address = msg->data.addr[0];
1070 dma_length = msg->data.len[0];
1072 if (!dma_mapping_error(dma_address))
1073 dma_unmap_single(msg->dev, dma_address, dma_length,
1077 dev_kfree_skb_any(msg->skb);
1081 memset(&msg->data, 0, sizeof(msg->data));
1082 veth_stack_push(cnx, msg);
1083 } else if (cnx->state & VETH_STATE_OPEN) {
1084 veth_error("Non-pending frame (# %d) acked by LPAR %d.\n",
1085 cnx->remote_lp, msg->token);
1089 static void veth_wake_queues(struct veth_lpar_connection *cnx)
1093 for (i = 0; i < HVMAXARCHITECTEDVIRTUALLANS; i++) {
1094 struct net_device *dev = veth_dev[i];
1095 struct veth_port *port;
1096 unsigned long flags;
1101 port = (struct veth_port *)dev->priv;
1103 if (! (port->lpar_map & (1<<cnx->remote_lp)))
1106 spin_lock_irqsave(&port->queue_lock, flags);
1108 port->stopped_map &= ~(1 << cnx->remote_lp);
1110 if (0 == port->stopped_map && netif_queue_stopped(dev)) {
1111 veth_debug("cnx %d: woke queue for %s.\n",
1112 cnx->remote_lp, dev->name);
1113 netif_wake_queue(dev);
1115 spin_unlock_irqrestore(&port->queue_lock, flags);
1119 static void veth_stop_queues(struct veth_lpar_connection *cnx)
1123 for (i = 0; i < HVMAXARCHITECTEDVIRTUALLANS; i++) {
1124 struct net_device *dev = veth_dev[i];
1125 struct veth_port *port;
1130 port = (struct veth_port *)dev->priv;
1132 /* If this cnx is not on the vlan for this port, continue */
1133 if (! (port->lpar_map & (1 << cnx->remote_lp)))
1136 spin_lock(&port->queue_lock);
1138 netif_stop_queue(dev);
1139 port->stopped_map |= (1 << cnx->remote_lp);
1141 veth_debug("cnx %d: stopped queue for %s, map = 0x%x.\n",
1142 cnx->remote_lp, dev->name, port->stopped_map);
1144 spin_unlock(&port->queue_lock);
1148 static void veth_timed_reset(unsigned long ptr)
1150 struct veth_lpar_connection *cnx = (struct veth_lpar_connection *)ptr;
1151 unsigned long trigger_time, flags;
1153 /* FIXME is it possible this fires after veth_stop_connection()?
1154 * That would reschedule the statemachine for 5 seconds and probably
1155 * execute it after the module's been unloaded. Hmm. */
1157 spin_lock_irqsave(&cnx->lock, flags);
1159 if (cnx->outstanding_tx > 0) {
1160 trigger_time = cnx->last_contact + cnx->reset_timeout;
1162 if (trigger_time < jiffies) {
1163 cnx->state |= VETH_STATE_RESET;
1164 veth_kick_statemachine(cnx);
1165 veth_error("%d packets not acked by LPAR %d within %d "
1166 "seconds, resetting.\n",
1167 cnx->outstanding_tx, cnx->remote_lp,
1168 cnx->reset_timeout / HZ);
1170 /* Reschedule the timer */
1171 trigger_time = jiffies + cnx->reset_timeout;
1172 mod_timer(&cnx->reset_timer, trigger_time);
1176 spin_unlock_irqrestore(&cnx->lock, flags);
1183 static inline int veth_frame_wanted(struct veth_port *port, u64 mac_addr)
1187 unsigned long flags;
1189 if ( (mac_addr == port->mac_addr) || (mac_addr == 0xffffffffffff0000) )
1192 read_lock_irqsave(&port->mcast_gate, flags);
1194 if (port->promiscuous) {
1199 for (i = 0; i < port->num_mcast; ++i) {
1200 if (port->mcast_addr[i] == mac_addr) {
1207 read_unlock_irqrestore(&port->mcast_gate, flags);
1217 #define VETH_MAX_PAGES_PER_FRAME ( (VETH_MAX_MTU+PAGE_SIZE-2)/PAGE_SIZE + 1 )
1219 static inline void veth_build_dma_list(struct dma_chunk *list,
1220 unsigned char *p, unsigned long length)
1225 /* FIXME: skbs are continguous in real addresses. Do we
1226 * really need to break it into PAGE_SIZE chunks, or can we do
1227 * it just at the granularity of iSeries real->absolute
1228 * mapping? Indeed, given the way the allocator works, can we
1229 * count on them being absolutely contiguous? */
1230 list[0].addr = ISERIES_HV_ADDR(p);
1231 list[0].size = min(length,
1232 PAGE_SIZE - ((unsigned long)p & ~PAGE_MASK));
1234 done = list[0].size;
1235 while (done < length) {
1236 list[i].addr = ISERIES_HV_ADDR(p + done);
1237 list[i].size = min(length-done, PAGE_SIZE);
1238 done += list[i].size;
1243 static void veth_flush_acks(struct veth_lpar_connection *cnx)
1247 rc = veth_signaldata(cnx, VethEventTypeFramesAck,
1248 0, &cnx->pending_acks);
1250 if (rc != HvLpEvent_Rc_Good)
1251 veth_error("Failed acking frames from LPAR %d, rc = %d\n",
1252 cnx->remote_lp, (int)rc);
1254 cnx->num_pending_acks = 0;
1255 memset(&cnx->pending_acks, 0xff, sizeof(cnx->pending_acks));
1258 static void veth_receive(struct veth_lpar_connection *cnx,
1259 struct VethLpEvent *event)
1261 struct VethFramesData *senddata = &event->u.frames_data;
1264 unsigned long flags;
1269 struct sk_buff *skb;
1270 struct dma_chunk local_list[VETH_MAX_PAGES_PER_FRAME];
1271 struct dma_chunk remote_list[VETH_MAX_FRAMES_PER_MSG];
1273 HvLpVirtualLanIndex vlan;
1274 struct net_device *dev;
1275 struct veth_port *port;
1277 /* FIXME: do we need this? */
1278 memset(local_list, 0, sizeof(local_list));
1279 memset(remote_list, 0, sizeof(VETH_MAX_FRAMES_PER_MSG));
1281 /* a 0 address marks the end of the valid entries */
1282 if (senddata->addr[startchunk] == 0)
1285 /* make sure that we have at least 1 EOF entry in the
1286 * remaining entries */
1287 if (! (senddata->eofmask >> (startchunk + VETH_EOF_SHIFT))) {
1288 veth_error("Missing EOF fragment in event "
1289 "eofmask = 0x%x startchunk = %d\n",
1290 (unsigned)senddata->eofmask,
1295 /* build list of chunks in this frame */
1298 remote_list[nchunks].addr =
1299 (u64) senddata->addr[startchunk+nchunks] << 32;
1300 remote_list[nchunks].size =
1301 senddata->len[startchunk+nchunks];
1302 length += remote_list[nchunks].size;
1303 } while (! (senddata->eofmask &
1304 (1 << (VETH_EOF_SHIFT + startchunk + nchunks++))));
1306 /* length == total length of all chunks */
1307 /* nchunks == # of chunks in this frame */
1309 if ((length - ETH_HLEN) > VETH_MAX_MTU) {
1310 veth_error("Received oversize frame from LPAR %d "
1312 cnx->remote_lp, length);
1316 skb = alloc_skb(length, GFP_ATOMIC);
1320 veth_build_dma_list(local_list, skb->data, length);
1322 rc = HvCallEvent_dmaBufList(HvLpEvent_Type_VirtualLan,
1323 event->base_event.xSourceLp,
1324 HvLpDma_Direction_RemoteToLocal,
1327 HvLpDma_AddressType_RealAddress,
1328 HvLpDma_AddressType_TceIndex,
1329 ISERIES_HV_ADDR(&local_list),
1330 ISERIES_HV_ADDR(&remote_list),
1332 if (rc != HvLpDma_Rc_Good) {
1333 dev_kfree_skb_irq(skb);
1337 vlan = skb->data[9];
1338 dev = veth_dev[vlan];
1341 * Some earlier versions of the driver sent
1342 * broadcasts down all connections, even to lpars
1343 * that weren't on the relevant vlan. So ignore
1344 * packets belonging to a vlan we're not on.
1345 * We can also be here if we receive packets while
1346 * the driver is going down, because then dev is NULL.
1348 dev_kfree_skb_irq(skb);
1352 port = (struct veth_port *)dev->priv;
1353 dest = *((u64 *) skb->data) & 0xFFFFFFFFFFFF0000;
1355 if ((vlan > HVMAXARCHITECTEDVIRTUALLANS) || !port) {
1356 dev_kfree_skb_irq(skb);
1359 if (! veth_frame_wanted(port, dest)) {
1360 dev_kfree_skb_irq(skb);
1364 skb_put(skb, length);
1366 skb->protocol = eth_type_trans(skb, dev);
1367 skb->ip_summed = CHECKSUM_NONE;
1368 netif_rx(skb); /* send it up */
1369 port->stats.rx_packets++;
1370 port->stats.rx_bytes += length;
1371 } while (startchunk += nchunks, startchunk < VETH_MAX_FRAMES_PER_MSG);
1374 spin_lock_irqsave(&cnx->lock, flags);
1375 BUG_ON(cnx->num_pending_acks > VETH_MAX_ACKS_PER_MSG);
1377 cnx->pending_acks[cnx->num_pending_acks++] =
1378 event->base_event.xCorrelationToken;
1380 if ( (cnx->num_pending_acks >= cnx->remote_caps.ack_threshold)
1381 || (cnx->num_pending_acks >= VETH_MAX_ACKS_PER_MSG) )
1382 veth_flush_acks(cnx);
1384 spin_unlock_irqrestore(&cnx->lock, flags);
1387 static void veth_timed_ack(unsigned long ptr)
1389 struct veth_lpar_connection *cnx = (struct veth_lpar_connection *) ptr;
1390 unsigned long flags;
1392 /* Ack all the events */
1393 spin_lock_irqsave(&cnx->lock, flags);
1394 if (cnx->num_pending_acks > 0)
1395 veth_flush_acks(cnx);
1397 /* Reschedule the timer */
1398 cnx->ack_timer.expires = jiffies + cnx->ack_timeout;
1399 add_timer(&cnx->ack_timer);
1400 spin_unlock_irqrestore(&cnx->lock, flags);
1403 static int veth_remove(struct vio_dev *vdev)
1405 struct veth_lpar_connection *cnx;
1406 struct net_device *dev;
1407 struct veth_port *port;
1410 dev = veth_dev[vdev->unit_address];
1415 port = netdev_priv(dev);
1417 for (i = 0; i < HVMAXARCHITECTEDLPS; i++) {
1420 if (cnx && (port->lpar_map & (1 << i))) {
1421 /* Drop our reference to connections on our VLAN */
1422 kobject_put(&cnx->kobject);
1426 veth_dev[vdev->unit_address] = NULL;
1427 unregister_netdev(dev);
1433 static int veth_probe(struct vio_dev *vdev, const struct vio_device_id *id)
1435 int i = vdev->unit_address;
1436 struct net_device *dev;
1437 struct veth_port *port;
1439 dev = veth_probe_one(i, &vdev->dev);
1446 port = (struct veth_port*)netdev_priv(dev);
1448 /* Start the state machine on each connection on this vlan. If we're
1449 * the first dev to do so this will commence link negotiation */
1450 for (i = 0; i < HVMAXARCHITECTEDLPS; i++) {
1451 struct veth_lpar_connection *cnx;
1453 if (! (port->lpar_map & (1 << i)))
1460 kobject_get(&cnx->kobject);
1461 veth_kick_statemachine(cnx);
1468 * veth_device_table: Used by vio.c to match devices that we
1471 static struct vio_device_id veth_device_table[] __devinitdata = {
1475 MODULE_DEVICE_TABLE(vio, veth_device_table);
1477 static struct vio_driver veth_driver = {
1478 .name = "iseries_veth",
1479 .id_table = veth_device_table,
1480 .probe = veth_probe,
1481 .remove = veth_remove
1485 * Module initialization/cleanup
1488 void __exit veth_module_cleanup(void)
1491 struct veth_lpar_connection *cnx;
1493 /* Disconnect our "irq" to stop events coming from the Hypervisor. */
1494 HvLpEvent_unregisterHandler(HvLpEvent_Type_VirtualLan);
1496 /* Make sure any work queued from Hypervisor callbacks is finished. */
1497 flush_scheduled_work();
1499 for (i = 0; i < HVMAXARCHITECTEDLPS; ++i) {
1505 /* Drop the driver's reference to the connection */
1506 kobject_put(&cnx->kobject);
1509 /* Unregister the driver, which will close all the netdevs and stop
1510 * the connections when they're no longer referenced. */
1511 vio_unregister_driver(&veth_driver);
1513 module_exit(veth_module_cleanup);
1515 int __init veth_module_init(void)
1520 this_lp = HvLpConfig_getLpIndex_outline();
1522 for (i = 0; i < HVMAXARCHITECTEDLPS; ++i) {
1523 rc = veth_init_connection(i);
1528 HvLpEvent_registerHandler(HvLpEvent_Type_VirtualLan,
1529 &veth_handle_event);
1531 rc = vio_register_driver(&veth_driver);
1538 for (i = 0; i < HVMAXARCHITECTEDLPS; ++i) {
1539 veth_destroy_connection(veth_cnx[i]);
1544 module_init(veth_module_init);