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>
82 #include "iseries_veth.h"
84 extern struct vio_dev *iSeries_veth_dev;
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;
120 unsigned long in_use;
124 struct veth_lpar_connection {
126 struct work_struct statemachine_wq;
127 struct veth_msg *msgs;
129 struct VethCapData local_caps;
131 struct timer_list ack_timer;
135 HvLpInstanceId src_inst;
136 HvLpInstanceId dst_inst;
137 struct VethLpEvent cap_event, cap_ack_event;
138 u16 pending_acks[VETH_MAX_ACKS_PER_MSG];
139 u32 num_pending_acks;
142 struct VethCapData remote_caps;
145 spinlock_t msg_stack_lock;
146 struct veth_msg *msg_stack_head;
150 struct net_device_stats stats;
152 HvLpIndexMap lpar_map;
154 spinlock_t pending_gate;
155 struct sk_buff *pending_skb;
156 HvLpIndexMap pending_lpmask;
162 u64 mcast_addr[VETH_MAX_MCAST];
165 static HvLpIndex this_lp;
166 static struct veth_lpar_connection *veth_cnx[HVMAXARCHITECTEDLPS]; /* = 0 */
167 static struct net_device *veth_dev[HVMAXARCHITECTEDVIRTUALLANS]; /* = 0 */
169 static int veth_start_xmit(struct sk_buff *skb, struct net_device *dev);
170 static void veth_recycle_msg(struct veth_lpar_connection *, struct veth_msg *);
171 static void veth_flush_pending(struct veth_lpar_connection *cnx);
172 static void veth_receive(struct veth_lpar_connection *, struct VethLpEvent *);
173 static void veth_timed_ack(unsigned long connectionPtr);
179 #define veth_printk(prio, fmt, args...) \
180 printk(prio "%s: " fmt, __FILE__, ## args)
182 #define veth_error(fmt, args...) \
183 printk(KERN_ERR "(%s:%3.3d) ERROR: " fmt, __FILE__, __LINE__ , ## args)
185 static inline void veth_stack_push(struct veth_lpar_connection *cnx,
186 struct veth_msg *msg)
190 spin_lock_irqsave(&cnx->msg_stack_lock, flags);
191 msg->next = cnx->msg_stack_head;
192 cnx->msg_stack_head = msg;
193 spin_unlock_irqrestore(&cnx->msg_stack_lock, flags);
196 static inline struct veth_msg *veth_stack_pop(struct veth_lpar_connection *cnx)
199 struct veth_msg *msg;
201 spin_lock_irqsave(&cnx->msg_stack_lock, flags);
202 msg = cnx->msg_stack_head;
204 cnx->msg_stack_head = cnx->msg_stack_head->next;
205 spin_unlock_irqrestore(&cnx->msg_stack_lock, flags);
209 static inline HvLpEvent_Rc
210 veth_signalevent(struct veth_lpar_connection *cnx, u16 subtype,
211 HvLpEvent_AckInd ackind, HvLpEvent_AckType acktype,
213 u64 data1, u64 data2, u64 data3, u64 data4, u64 data5)
215 return HvCallEvent_signalLpEventFast(cnx->remote_lp,
216 HvLpEvent_Type_VirtualLan,
217 subtype, ackind, acktype,
220 token, data1, data2, data3,
224 static inline HvLpEvent_Rc veth_signaldata(struct veth_lpar_connection *cnx,
225 u16 subtype, u64 token, void *data)
227 u64 *p = (u64 *) data;
229 return veth_signalevent(cnx, subtype, HvLpEvent_AckInd_NoAck,
230 HvLpEvent_AckType_ImmediateAck,
231 token, p[0], p[1], p[2], p[3], p[4]);
234 struct veth_allocation {
239 static void veth_complete_allocation(void *parm, int number)
241 struct veth_allocation *vc = (struct veth_allocation *)parm;
247 static int veth_allocate_events(HvLpIndex rlp, int number)
249 struct veth_allocation vc = { COMPLETION_INITIALIZER(vc.c), 0 };
251 mf_allocateLpEvents(rlp, HvLpEvent_Type_VirtualLan,
252 sizeof(struct VethLpEvent), number,
253 &veth_complete_allocation, &vc);
254 wait_for_completion(&vc.c);
260 * LPAR connection code
263 static inline void veth_kick_statemachine(struct veth_lpar_connection *cnx)
265 schedule_work(&cnx->statemachine_wq);
268 static void veth_take_cap(struct veth_lpar_connection *cnx,
269 struct VethLpEvent *event)
273 spin_lock_irqsave(&cnx->lock, flags);
274 /* Receiving caps may mean the other end has just come up, so
275 * we need to reload the instance ID of the far end */
277 HvCallEvent_getTargetLpInstanceId(cnx->remote_lp,
278 HvLpEvent_Type_VirtualLan);
280 if (cnx->state & VETH_STATE_GOTCAPS) {
281 veth_error("Received a second capabilities from lpar %d\n",
283 event->base_event.xRc = HvLpEvent_Rc_BufferNotAvailable;
284 HvCallEvent_ackLpEvent((struct HvLpEvent *) event);
286 memcpy(&cnx->cap_event, event, sizeof(cnx->cap_event));
287 cnx->state |= VETH_STATE_GOTCAPS;
288 veth_kick_statemachine(cnx);
290 spin_unlock_irqrestore(&cnx->lock, flags);
293 static void veth_take_cap_ack(struct veth_lpar_connection *cnx,
294 struct VethLpEvent *event)
298 spin_lock_irqsave(&cnx->lock, flags);
299 if (cnx->state & VETH_STATE_GOTCAPACK) {
300 veth_error("Received a second capabilities ack from lpar %d\n",
303 memcpy(&cnx->cap_ack_event, event,
304 sizeof(&cnx->cap_ack_event));
305 cnx->state |= VETH_STATE_GOTCAPACK;
306 veth_kick_statemachine(cnx);
308 spin_unlock_irqrestore(&cnx->lock, flags);
311 static void veth_take_monitor_ack(struct veth_lpar_connection *cnx,
312 struct VethLpEvent *event)
316 spin_lock_irqsave(&cnx->lock, flags);
317 veth_printk(KERN_DEBUG, "Monitor ack returned for lpar %d\n",
319 cnx->state |= VETH_STATE_RESET;
320 veth_kick_statemachine(cnx);
321 spin_unlock_irqrestore(&cnx->lock, flags);
324 static void veth_handle_ack(struct VethLpEvent *event)
326 HvLpIndex rlp = event->base_event.xTargetLp;
327 struct veth_lpar_connection *cnx = veth_cnx[rlp];
331 switch (event->base_event.xSubtype) {
332 case VethEventTypeCap:
333 veth_take_cap_ack(cnx, event);
335 case VethEventTypeMonitor:
336 veth_take_monitor_ack(cnx, event);
339 veth_error("Unknown ack type %d from lpar %d\n",
340 event->base_event.xSubtype, rlp);
344 static void veth_handle_int(struct VethLpEvent *event)
346 HvLpIndex rlp = event->base_event.xSourceLp;
347 struct veth_lpar_connection *cnx = veth_cnx[rlp];
353 switch (event->base_event.xSubtype) {
354 case VethEventTypeCap:
355 veth_take_cap(cnx, event);
357 case VethEventTypeMonitor:
358 /* do nothing... this'll hang out here til we're dead,
359 * and the hypervisor will return it for us. */
361 case VethEventTypeFramesAck:
362 spin_lock_irqsave(&cnx->lock, flags);
363 for (i = 0; i < VETH_MAX_ACKS_PER_MSG; ++i) {
364 u16 msgnum = event->u.frames_ack_data.token[i];
366 if (msgnum < VETH_NUMBUFFERS)
367 veth_recycle_msg(cnx, cnx->msgs + msgnum);
369 spin_unlock_irqrestore(&cnx->lock, flags);
370 veth_flush_pending(cnx);
372 case VethEventTypeFrames:
373 veth_receive(cnx, event);
376 veth_error("Unknown interrupt type %d from lpar %d\n",
377 event->base_event.xSubtype, rlp);
381 static void veth_handle_event(struct HvLpEvent *event, struct pt_regs *regs)
383 struct VethLpEvent *veth_event = (struct VethLpEvent *)event;
385 if (event->xFlags.xFunction == HvLpEvent_Function_Ack)
386 veth_handle_ack(veth_event);
387 else if (event->xFlags.xFunction == HvLpEvent_Function_Int)
388 veth_handle_int(veth_event);
391 static int veth_process_caps(struct veth_lpar_connection *cnx)
393 struct VethCapData *remote_caps = &cnx->remote_caps;
396 /* Convert timer to jiffies */
397 cnx->ack_timeout = remote_caps->ack_timeout * HZ / 1000000;
399 if ( (remote_caps->num_buffers == 0)
400 || (remote_caps->ack_threshold > VETH_MAX_ACKS_PER_MSG)
401 || (remote_caps->ack_threshold == 0)
402 || (cnx->ack_timeout == 0) ) {
403 veth_error("Received incompatible capabilities from lpar %d\n",
405 return HvLpEvent_Rc_InvalidSubtypeData;
408 num_acks_needed = (remote_caps->num_buffers
409 / remote_caps->ack_threshold) + 1;
411 /* FIXME: locking on num_ack_events? */
412 if (cnx->num_ack_events < num_acks_needed) {
415 num = veth_allocate_events(cnx->remote_lp,
416 num_acks_needed-cnx->num_ack_events);
418 cnx->num_ack_events += num;
420 if (cnx->num_ack_events < num_acks_needed) {
421 veth_error("Couldn't allocate enough ack events for lpar %d\n",
424 return HvLpEvent_Rc_BufferNotAvailable;
429 return HvLpEvent_Rc_Good;
432 /* FIXME: The gotos here are a bit dubious */
433 static void veth_statemachine(void *p)
435 struct veth_lpar_connection *cnx = (struct veth_lpar_connection *)p;
436 int rlp = cnx->remote_lp;
439 spin_lock_irq(&cnx->lock);
442 if (cnx->state & VETH_STATE_RESET) {
445 del_timer(&cnx->ack_timer);
447 if (cnx->state & VETH_STATE_OPEN)
448 HvCallEvent_closeLpEventPath(cnx->remote_lp,
449 HvLpEvent_Type_VirtualLan);
452 memset(&cnx->pending_acks, 0xff, sizeof (cnx->pending_acks));
453 cnx->num_pending_acks = 0;
455 cnx->state &= ~(VETH_STATE_RESET | VETH_STATE_SENTMON
456 | VETH_STATE_OPEN | VETH_STATE_SENTCAPS
457 | VETH_STATE_GOTCAPACK | VETH_STATE_GOTCAPS
458 | VETH_STATE_SENTCAPACK | VETH_STATE_READY);
460 /* Clean up any leftover messages */
462 for (i = 0; i < VETH_NUMBUFFERS; ++i)
463 veth_recycle_msg(cnx, cnx->msgs + i);
466 if (cnx->state & VETH_STATE_SHUTDOWN)
467 /* It's all over, do nothing */
470 if ( !(cnx->state & VETH_STATE_OPEN) ) {
471 if (! cnx->msgs || (cnx->num_events < (2 + VETH_NUMBUFFERS)) )
474 HvCallEvent_openLpEventPath(rlp, HvLpEvent_Type_VirtualLan);
476 HvCallEvent_getSourceLpInstanceId(rlp,
477 HvLpEvent_Type_VirtualLan);
479 HvCallEvent_getTargetLpInstanceId(rlp,
480 HvLpEvent_Type_VirtualLan);
481 cnx->state |= VETH_STATE_OPEN;
484 if ( (cnx->state & VETH_STATE_OPEN)
485 && !(cnx->state & VETH_STATE_SENTMON) ) {
486 rc = veth_signalevent(cnx, VethEventTypeMonitor,
487 HvLpEvent_AckInd_DoAck,
488 HvLpEvent_AckType_DeferredAck,
491 if (rc == HvLpEvent_Rc_Good) {
492 cnx->state |= VETH_STATE_SENTMON;
494 if ( (rc != HvLpEvent_Rc_PartitionDead)
495 && (rc != HvLpEvent_Rc_PathClosed) )
496 veth_error("Error sending monitor to "
500 /* Oh well, hope we get a cap from the other
501 * end and do better when that kicks us */
506 if ( (cnx->state & VETH_STATE_OPEN)
507 && !(cnx->state & VETH_STATE_SENTCAPS)) {
508 u64 *rawcap = (u64 *)&cnx->local_caps;
510 rc = veth_signalevent(cnx, VethEventTypeCap,
511 HvLpEvent_AckInd_DoAck,
512 HvLpEvent_AckType_ImmediateAck,
513 0, rawcap[0], rawcap[1], rawcap[2],
514 rawcap[3], rawcap[4]);
516 if (rc == HvLpEvent_Rc_Good) {
517 cnx->state |= VETH_STATE_SENTCAPS;
519 if ( (rc != HvLpEvent_Rc_PartitionDead)
520 && (rc != HvLpEvent_Rc_PathClosed) )
521 veth_error("Error sending caps to "
524 /* Oh well, hope we get a cap from the other
525 * end and do better when that kicks us */
530 if ((cnx->state & VETH_STATE_GOTCAPS)
531 && !(cnx->state & VETH_STATE_SENTCAPACK)) {
532 struct VethCapData *remote_caps = &cnx->remote_caps;
534 memcpy(remote_caps, &cnx->cap_event.u.caps_data,
535 sizeof(*remote_caps));
537 spin_unlock_irq(&cnx->lock);
538 rc = veth_process_caps(cnx);
539 spin_lock_irq(&cnx->lock);
541 /* We dropped the lock, so recheck for anything which
542 * might mess us up */
543 if (cnx->state & (VETH_STATE_RESET|VETH_STATE_SHUTDOWN))
546 cnx->cap_event.base_event.xRc = rc;
547 HvCallEvent_ackLpEvent((struct HvLpEvent *)&cnx->cap_event);
548 if (rc == HvLpEvent_Rc_Good)
549 cnx->state |= VETH_STATE_SENTCAPACK;
554 if ((cnx->state & VETH_STATE_GOTCAPACK)
555 && (cnx->state & VETH_STATE_GOTCAPS)
556 && !(cnx->state & VETH_STATE_READY)) {
557 if (cnx->cap_ack_event.base_event.xRc == HvLpEvent_Rc_Good) {
558 /* Start the ACK timer */
559 cnx->ack_timer.expires = jiffies + cnx->ack_timeout;
560 add_timer(&cnx->ack_timer);
561 cnx->state |= VETH_STATE_READY;
563 veth_printk(KERN_ERR, "Caps rejected (rc=%d) by "
565 cnx->cap_ack_event.base_event.xRc,
572 spin_unlock_irq(&cnx->lock);
576 /* FIXME: we get here if something happens we really can't
577 * cope with. The link will never work once we get here, and
578 * all we can do is not lock the rest of the system up */
579 veth_error("Badness on connection to lpar %d (state=%04lx) "
580 " - shutting down\n", rlp, cnx->state);
581 cnx->state |= VETH_STATE_SHUTDOWN;
582 spin_unlock_irq(&cnx->lock);
585 static int veth_init_connection(u8 rlp)
587 struct veth_lpar_connection *cnx;
588 struct veth_msg *msgs;
591 if ( (rlp == this_lp)
592 || ! HvLpConfig_doLpsCommunicateOnVirtualLan(this_lp, rlp) )
595 cnx = kmalloc(sizeof(*cnx), GFP_KERNEL);
598 memset(cnx, 0, sizeof(*cnx));
600 cnx->remote_lp = rlp;
601 spin_lock_init(&cnx->lock);
602 INIT_WORK(&cnx->statemachine_wq, veth_statemachine, cnx);
603 init_timer(&cnx->ack_timer);
604 cnx->ack_timer.function = veth_timed_ack;
605 cnx->ack_timer.data = (unsigned long) cnx;
606 memset(&cnx->pending_acks, 0xff, sizeof (cnx->pending_acks));
610 msgs = kmalloc(VETH_NUMBUFFERS * sizeof(struct veth_msg), GFP_KERNEL);
612 veth_error("Can't allocate buffers for lpar %d\n", rlp);
617 memset(msgs, 0, VETH_NUMBUFFERS * sizeof(struct veth_msg));
618 spin_lock_init(&cnx->msg_stack_lock);
620 for (i = 0; i < VETH_NUMBUFFERS; i++) {
622 veth_stack_push(cnx, msgs + i);
625 cnx->num_events = veth_allocate_events(rlp, 2 + VETH_NUMBUFFERS);
627 if (cnx->num_events < (2 + VETH_NUMBUFFERS)) {
628 veth_error("Can't allocate events for lpar %d, only got %d\n",
629 rlp, cnx->num_events);
633 cnx->local_caps.num_buffers = VETH_NUMBUFFERS;
634 cnx->local_caps.ack_threshold = ACK_THRESHOLD;
635 cnx->local_caps.ack_timeout = VETH_ACKTIMEOUT;
640 static void veth_destroy_connection(u8 rlp)
642 struct veth_lpar_connection *cnx = veth_cnx[rlp];
647 spin_lock_irq(&cnx->lock);
648 cnx->state |= VETH_STATE_RESET | VETH_STATE_SHUTDOWN;
649 veth_kick_statemachine(cnx);
650 spin_unlock_irq(&cnx->lock);
652 flush_scheduled_work();
654 /* FIXME: not sure if this is necessary - will already have
655 * been deleted by the state machine, just want to make sure
656 * its not running any more */
657 del_timer_sync(&cnx->ack_timer);
659 if (cnx->num_events > 0)
660 mf_deallocateLpEvents(cnx->remote_lp,
661 HvLpEvent_Type_VirtualLan,
664 if (cnx->num_ack_events > 0)
665 mf_deallocateLpEvents(cnx->remote_lp,
666 HvLpEvent_Type_VirtualLan,
678 static int veth_open(struct net_device *dev)
680 struct veth_port *port = (struct veth_port *) dev->priv;
682 memset(&port->stats, 0, sizeof (port->stats));
683 netif_start_queue(dev);
687 static int veth_close(struct net_device *dev)
689 netif_stop_queue(dev);
693 static struct net_device_stats *veth_get_stats(struct net_device *dev)
695 struct veth_port *port = (struct veth_port *) dev->priv;
700 static int veth_change_mtu(struct net_device *dev, int new_mtu)
702 if ((new_mtu < 68) || (new_mtu > VETH_MAX_MTU))
708 static void veth_set_multicast_list(struct net_device *dev)
710 struct veth_port *port = (struct veth_port *) dev->priv;
713 write_lock_irqsave(&port->mcast_gate, flags);
715 if (dev->flags & IFF_PROMISC) { /* set promiscuous mode */
716 printk(KERN_INFO "%s: Promiscuous mode enabled.\n",
718 port->promiscuous = 1;
719 } else if ( (dev->flags & IFF_ALLMULTI)
720 || (dev->mc_count > VETH_MAX_MCAST) ) {
723 struct dev_mc_list *dmi = dev->mc_list;
729 for (i = 0; i < dev->mc_count; i++) {
730 u8 *addr = dmi->dmi_addr;
733 if (addr[0] & 0x01) {/* multicast address? */
734 memcpy(&xaddr, addr, ETH_ALEN);
735 port->mcast_addr[port->num_mcast] = xaddr;
742 write_unlock_irqrestore(&port->mcast_gate, flags);
745 static int veth_ioctl(struct net_device *dev, struct ifreq *ifr, int cmd)
748 struct ethtool_cmd ecmd;
750 if (cmd != SIOCETHTOOL)
752 if (copy_from_user(&ecmd, ifr->ifr_data, sizeof (ecmd)))
756 ecmd.supported = (SUPPORTED_1000baseT_Full
757 | SUPPORTED_Autoneg | SUPPORTED_FIBRE);
758 ecmd.advertising = (SUPPORTED_1000baseT_Full
759 | SUPPORTED_Autoneg | SUPPORTED_FIBRE);
761 ecmd.port = PORT_FIBRE;
762 ecmd.transceiver = XCVR_INTERNAL;
763 ecmd.phy_address = 0;
764 ecmd.speed = SPEED_1000;
765 ecmd.duplex = DUPLEX_FULL;
766 ecmd.autoneg = AUTONEG_ENABLE;
769 if (copy_to_user(ifr->ifr_data, &ecmd, sizeof(ecmd)))
773 case ETHTOOL_GDRVINFO:{
774 struct ethtool_drvinfo info = { ETHTOOL_GDRVINFO };
775 strncpy(info.driver, "veth", sizeof(info.driver) - 1);
776 info.driver[sizeof(info.driver) - 1] = '\0';
777 strncpy(info.version, "1.0", sizeof(info.version) - 1);
778 if (copy_to_user(ifr->ifr_data, &info, sizeof(info)))
782 /* get link status */
784 struct ethtool_value edata = { ETHTOOL_GLINK };
786 if (copy_to_user(ifr->ifr_data, &edata, sizeof(edata)))
799 struct net_device * __init veth_probe_one(int vlan)
801 struct net_device *dev;
802 struct veth_port *port;
805 dev = alloc_etherdev(sizeof (struct veth_port));
807 veth_error("Unable to allocate net_device structure!\n");
811 port = (struct veth_port *) dev->priv;
813 spin_lock_init(&port->pending_gate);
814 rwlock_init(&port->mcast_gate);
816 for (i = 0; i < HVMAXARCHITECTEDLPS; i++) {
817 HvLpVirtualLanIndexMap map;
821 map = HvLpConfig_getVirtualLanIndexMapForLp(i);
822 if (map & (0x8000 >> vlan))
823 port->lpar_map |= (1 << i);
826 dev->dev_addr[0] = 0x02;
827 dev->dev_addr[1] = 0x01;
828 dev->dev_addr[2] = 0xff;
829 dev->dev_addr[3] = vlan;
830 dev->dev_addr[4] = 0xff;
831 dev->dev_addr[5] = this_lp;
833 dev->mtu = VETH_MAX_MTU;
835 memcpy(&port->mac_addr, dev->dev_addr, 6);
837 dev->open = veth_open;
838 dev->hard_start_xmit = veth_start_xmit;
839 dev->stop = veth_close;
840 dev->get_stats = veth_get_stats;
841 dev->change_mtu = veth_change_mtu;
842 dev->set_mac_address = NULL;
843 dev->set_multicast_list = veth_set_multicast_list;
844 dev->do_ioctl = veth_ioctl;
846 rc = register_netdev(dev);
848 veth_printk(KERN_ERR,
849 "Failed to register ethernet device for vlan %d\n",
855 veth_printk(KERN_DEBUG, "%s attached to iSeries vlan %d (lpar_map=0x%04x)\n",
856 dev->name, vlan, port->lpar_map);
865 static int veth_transmit_to_one(struct sk_buff *skb, HvLpIndex rlp,
866 struct net_device *dev)
868 struct veth_lpar_connection *cnx = veth_cnx[rlp];
869 struct veth_port *port = (struct veth_port *) dev->priv;
871 u32 dma_address, dma_length;
872 struct veth_msg *msg = NULL;
877 port->stats.tx_errors++;
882 spin_lock_irqsave(&cnx->lock, flags);
884 if (! cnx->state & VETH_STATE_READY)
887 if ((skb->len - 14) > VETH_MAX_MTU)
890 msg = veth_stack_pop(cnx);
897 dma_length = skb->len;
898 dma_address = vio_map_single(iSeries_veth_dev, skb->data,
899 dma_length, DMA_TO_DEVICE);
901 if (dma_mapping_error(dma_address))
902 goto recycle_and_drop;
904 /* Is it really necessary to check the length and address
905 * fields of the first entry here? */
907 msg->data.addr[0] = dma_address;
908 msg->data.len[0] = dma_length;
909 msg->data.eofmask = 1 << VETH_EOF_SHIFT;
910 set_bit(0, &(msg->in_use));
911 rc = veth_signaldata(cnx, VethEventTypeFrames, msg->token, &msg->data);
913 if (rc != HvLpEvent_Rc_Good)
914 goto recycle_and_drop;
916 spin_unlock_irqrestore(&cnx->lock, flags);
921 /* need to set in use to make veth_recycle_msg in case this
922 * was a mapping failure */
923 set_bit(0, &msg->in_use);
924 veth_recycle_msg(cnx, msg);
926 port->stats.tx_errors++;
928 spin_unlock_irqrestore(&cnx->lock, flags);
932 static HvLpIndexMap veth_transmit_to_many(struct sk_buff *skb,
934 struct net_device *dev)
936 struct veth_port *port = (struct veth_port *) dev->priv;
940 for (i = 0; i < HVMAXARCHITECTEDLPS; i++) {
941 struct sk_buff *clone;
943 if (! lpmask & (1<<i))
946 clone = skb_clone(skb, GFP_ATOMIC);
948 veth_error("%s: skb_clone failed %p\n",
953 rc = veth_transmit_to_one(clone, i, dev);
959 port->stats.tx_packets++;
960 port->stats.tx_bytes += skb->len;
966 static int veth_start_xmit(struct sk_buff *skb, struct net_device *dev)
968 unsigned char *frame = skb->data;
969 struct veth_port *port = (struct veth_port *) dev->priv;
973 if (! (frame[0] & 0x01)) {
975 HvLpIndex rlp = frame[5];
977 if ( ! ((1 << rlp) & port->lpar_map) ) {
984 lpmask = port->lpar_map;
987 lpmask = veth_transmit_to_many(skb, lpmask, dev);
992 spin_lock_irqsave(&port->pending_gate, flags);
993 if (port->pending_skb) {
994 veth_error("%s: Tx while skb was pending!\n",
1000 port->pending_skb = skb;
1001 port->pending_lpmask = lpmask;
1002 netif_stop_queue(dev);
1004 spin_unlock_irqrestore(&port->pending_gate, flags);
1010 static void veth_recycle_msg(struct veth_lpar_connection *cnx,
1011 struct veth_msg *msg)
1013 u32 dma_address, dma_length;
1015 if (test_and_clear_bit(0, &msg->in_use)) {
1016 dma_address = msg->data.addr[0];
1017 dma_length = msg->data.len[0];
1019 vio_unmap_single(iSeries_veth_dev, dma_address, dma_length,
1023 dev_kfree_skb_any(msg->skb);
1027 memset(&msg->data, 0, sizeof(msg->data));
1028 veth_stack_push(cnx, msg);
1030 if (cnx->state & VETH_STATE_OPEN)
1031 veth_error("Bogus frames ack from lpar %d (#%d)\n",
1032 cnx->remote_lp, msg->token);
1035 static void veth_flush_pending(struct veth_lpar_connection *cnx)
1038 for (i = 0; i < HVMAXARCHITECTEDVIRTUALLANS; i++) {
1039 struct net_device *dev = veth_dev[i];
1040 struct veth_port *port;
1041 unsigned long flags;
1046 port = (struct veth_port *)dev->priv;
1048 if (! (port->lpar_map & (1<<cnx->remote_lp)))
1051 spin_lock_irqsave(&port->pending_gate, flags);
1052 if (port->pending_skb) {
1053 port->pending_lpmask =
1054 veth_transmit_to_many(port->pending_skb,
1055 port->pending_lpmask,
1057 if (! port->pending_lpmask) {
1058 dev_kfree_skb_any(port->pending_skb);
1059 port->pending_skb = NULL;
1060 netif_start_queue(dev);
1063 spin_unlock_irqrestore(&port->pending_gate, flags);
1071 static inline int veth_frame_wanted(struct veth_port *port, u64 mac_addr)
1075 unsigned long flags;
1077 if ( (mac_addr == port->mac_addr) || (mac_addr == 0xffffffffffff0000) )
1080 if (! (((char *) &mac_addr)[0] & 0x01))
1083 read_lock_irqsave(&port->mcast_gate, flags);
1085 if (port->promiscuous || port->all_mcast) {
1090 for (i = 0; i < port->num_mcast; ++i) {
1091 if (port->mcast_addr[i] == mac_addr) {
1098 read_unlock_irqrestore(&port->mcast_gate, flags);
1108 #define VETH_MAX_PAGES_PER_FRAME ( (VETH_MAX_MTU+PAGE_SIZE-2)/PAGE_SIZE + 1 )
1110 static inline void veth_build_dma_list(struct dma_chunk *list,
1111 unsigned char *p, unsigned long length)
1116 /* FIXME: skbs are continguous in real addresses. Do we
1117 * really need to break it into PAGE_SIZE chunks, or can we do
1118 * it just at the granularity of iSeries real->absolute
1119 * mapping? Indeed, given the way the allocator works, can we
1120 * count on them being absolutely contiguous? */
1121 list[0].addr = ISERIES_HV_ADDR(p);
1122 list[0].size = min(length,
1123 PAGE_SIZE - ((unsigned long)p & ~PAGE_MASK));
1125 done = list[0].size;
1126 while (done < length) {
1127 list[i].addr = ISERIES_HV_ADDR(p + done);
1128 list[i].size = min(length-done, PAGE_SIZE);
1129 done += list[i].size;
1134 static void veth_flush_acks(struct veth_lpar_connection *cnx)
1138 rc = veth_signaldata(cnx, VethEventTypeFramesAck,
1139 0, &cnx->pending_acks);
1141 if (rc != HvLpEvent_Rc_Good)
1142 veth_error("Error 0x%x acking frames from lpar %d!\n",
1143 (unsigned)rc, cnx->remote_lp);
1145 cnx->num_pending_acks = 0;
1146 memset(&cnx->pending_acks, 0xff, sizeof(cnx->pending_acks));
1149 static void veth_receive(struct veth_lpar_connection *cnx,
1150 struct VethLpEvent *event)
1152 struct VethFramesData *senddata = &event->u.frames_data;
1155 unsigned long flags;
1160 struct sk_buff *skb;
1161 struct dma_chunk local_list[VETH_MAX_PAGES_PER_FRAME];
1162 struct dma_chunk remote_list[VETH_MAX_FRAMES_PER_MSG];
1164 HvLpVirtualLanIndex vlan;
1165 struct net_device *dev;
1166 struct veth_port *port;
1168 /* FIXME: do we need this? */
1169 memset(local_list, 0, sizeof(local_list));
1170 memset(remote_list, 0, sizeof(VETH_MAX_FRAMES_PER_MSG));
1172 /* a 0 address marks the end of the valid entries */
1173 if (senddata->addr[startchunk] == 0)
1176 /* make sure that we have at least 1 EOF entry in the
1177 * remaining entries */
1178 if (! (senddata->eofmask >> (startchunk + VETH_EOF_SHIFT))) {
1179 veth_error("missing EOF frag in event "
1180 "eofmask=0x%x startchunk=%d\n",
1181 (unsigned) senddata->eofmask, startchunk);
1185 /* build list of chunks in this frame */
1188 remote_list[nchunks].addr =
1189 (u64) senddata->addr[startchunk+nchunks] << 32;
1190 remote_list[nchunks].size =
1191 senddata->len[startchunk+nchunks];
1192 length += remote_list[nchunks].size;
1193 } while (! (senddata->eofmask &
1194 (1 << (VETH_EOF_SHIFT + startchunk + nchunks++))));
1196 /* length == total length of all chunks */
1197 /* nchunks == # of chunks in this frame */
1199 if ((length - ETH_HLEN) > VETH_MAX_MTU) {
1200 veth_error("Received oversize frame from lpar %d "
1201 "(length=%d)\n", cnx->remote_lp, length);
1205 skb = alloc_skb(length, GFP_ATOMIC);
1209 veth_build_dma_list(local_list, skb->data, length);
1211 rc = HvCallEvent_dmaBufList(HvLpEvent_Type_VirtualLan,
1212 event->base_event.xSourceLp,
1213 HvLpDma_Direction_RemoteToLocal,
1216 HvLpDma_AddressType_RealAddress,
1217 HvLpDma_AddressType_TceIndex,
1218 ISERIES_HV_ADDR(&local_list),
1219 ISERIES_HV_ADDR(&remote_list),
1221 if (rc != HvLpDma_Rc_Good) {
1222 dev_kfree_skb_irq(skb);
1226 vlan = skb->data[9];
1227 dev = veth_dev[vlan];
1229 /* Some earlier versions of the driver sent
1230 broadcasts down all connections, even to
1231 lpars that weren't on the relevant vlan.
1232 So ignore packets belonging to a vlan we're
1236 port = (struct veth_port *)dev->priv;
1237 dest = *((u64 *) skb->data) & 0xFFFFFFFFFFFF0000;
1239 if ((vlan > HVMAXARCHITECTEDVIRTUALLANS) || !port) {
1240 dev_kfree_skb_irq(skb);
1243 if (! veth_frame_wanted(port, dest)) {
1244 dev_kfree_skb_irq(skb);
1248 skb_put(skb, length);
1250 skb->protocol = eth_type_trans(skb, dev);
1251 skb->ip_summed = CHECKSUM_NONE;
1252 netif_rx(skb); /* send it up */
1253 port->stats.rx_packets++;
1254 port->stats.rx_bytes += length;
1255 } while (startchunk += nchunks, startchunk < VETH_MAX_FRAMES_PER_MSG);
1258 spin_lock_irqsave(&cnx->lock, flags);
1259 BUG_ON(cnx->num_pending_acks > VETH_MAX_ACKS_PER_MSG);
1261 cnx->pending_acks[cnx->num_pending_acks++] =
1262 event->base_event.xCorrelationToken;
1264 if ( (cnx->num_pending_acks >= cnx->remote_caps.ack_threshold)
1265 || (cnx->num_pending_acks >= VETH_MAX_ACKS_PER_MSG) )
1266 veth_flush_acks(cnx);
1268 spin_unlock_irqrestore(&cnx->lock, flags);
1271 static void veth_timed_ack(unsigned long ptr)
1273 struct veth_lpar_connection *cnx = (struct veth_lpar_connection *) ptr;
1274 unsigned long flags;
1276 /* Ack all the events */
1277 spin_lock_irqsave(&cnx->lock, flags);
1278 if (cnx->num_pending_acks > 0)
1279 veth_flush_acks(cnx);
1281 /* Reschedule the timer */
1282 cnx->ack_timer.expires = jiffies + cnx->ack_timeout;
1283 add_timer(&cnx->ack_timer);
1284 spin_unlock_irqrestore(&cnx->lock, flags);
1288 * Module initialization/cleanup
1291 void __exit veth_module_cleanup(void)
1295 for (i = 0; i < HVMAXARCHITECTEDLPS; ++i)
1296 veth_destroy_connection(i);
1298 HvLpEvent_unregisterHandler(HvLpEvent_Type_VirtualLan);
1300 for (i = 0; i < HVMAXARCHITECTEDVIRTUALLANS; ++i) {
1301 struct net_device *dev = veth_dev[i];
1307 unregister_netdev(dev);
1311 module_exit(veth_module_cleanup);
1313 int __init veth_module_init(void)
1315 HvLpIndexMap vlan_map = HvLpConfig_getVirtualLanIndexMap();
1319 this_lp = HvLpConfig_getLpIndex_outline();
1321 for (i = 0; i < HVMAXARCHITECTEDLPS; ++i) {
1322 rc = veth_init_connection(i);
1324 veth_module_cleanup();
1329 for (i = 0; i < HVMAXARCHITECTEDVIRTUALLANS; ++i) {
1330 struct net_device *dev;
1332 if (! (vlan_map & (0x8000 >> i)))
1335 dev = veth_probe_one(i);
1338 veth_module_cleanup();
1345 HvLpEvent_registerHandler(HvLpEvent_Type_VirtualLan,
1346 &veth_handle_event);
1348 /* Start the state machine on each connection, to commence
1349 * link negotiation */
1350 for (i = 0; i < HVMAXARCHITECTEDLPS; i++)
1352 veth_kick_statemachine(veth_cnx[i]);
1356 module_init(veth_module_init);