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;
120 unsigned long in_use;
125 struct veth_lpar_connection {
127 struct work_struct statemachine_wq;
128 struct veth_msg *msgs;
130 struct VethCapData local_caps;
132 struct timer_list ack_timer;
136 HvLpInstanceId src_inst;
137 HvLpInstanceId dst_inst;
138 struct VethLpEvent cap_event, cap_ack_event;
139 u16 pending_acks[VETH_MAX_ACKS_PER_MSG];
140 u32 num_pending_acks;
143 struct VethCapData remote_caps;
146 spinlock_t msg_stack_lock;
147 struct veth_msg *msg_stack_head;
152 struct net_device_stats stats;
154 HvLpIndexMap lpar_map;
156 spinlock_t pending_gate;
157 struct sk_buff *pending_skb;
158 HvLpIndexMap pending_lpmask;
164 u64 mcast_addr[VETH_MAX_MCAST];
167 static HvLpIndex this_lp;
168 static struct veth_lpar_connection *veth_cnx[HVMAXARCHITECTEDLPS]; /* = 0 */
169 static struct net_device *veth_dev[HVMAXARCHITECTEDVIRTUALLANS]; /* = 0 */
171 static int veth_start_xmit(struct sk_buff *skb, struct net_device *dev);
172 static void veth_recycle_msg(struct veth_lpar_connection *, struct veth_msg *);
173 static void veth_flush_pending(struct veth_lpar_connection *cnx);
174 static void veth_receive(struct veth_lpar_connection *, struct VethLpEvent *);
175 static void veth_timed_ack(unsigned long connectionPtr);
181 #define veth_info(fmt, args...) \
182 printk(KERN_INFO "iseries_veth: " fmt, ## args)
184 #define veth_error(fmt, args...) \
185 printk(KERN_ERR "iseries_veth: Error: " fmt, ## args)
188 #define veth_debug(fmt, args...) \
189 printk(KERN_DEBUG "iseries_veth: " fmt, ## args)
191 #define veth_debug(fmt, args...) do {} while (0)
194 static inline void veth_stack_push(struct veth_lpar_connection *cnx,
195 struct veth_msg *msg)
199 spin_lock_irqsave(&cnx->msg_stack_lock, flags);
200 msg->next = cnx->msg_stack_head;
201 cnx->msg_stack_head = msg;
202 spin_unlock_irqrestore(&cnx->msg_stack_lock, flags);
205 static inline struct veth_msg *veth_stack_pop(struct veth_lpar_connection *cnx)
208 struct veth_msg *msg;
210 spin_lock_irqsave(&cnx->msg_stack_lock, flags);
211 msg = cnx->msg_stack_head;
213 cnx->msg_stack_head = cnx->msg_stack_head->next;
214 spin_unlock_irqrestore(&cnx->msg_stack_lock, flags);
218 static inline HvLpEvent_Rc
219 veth_signalevent(struct veth_lpar_connection *cnx, u16 subtype,
220 HvLpEvent_AckInd ackind, HvLpEvent_AckType acktype,
222 u64 data1, u64 data2, u64 data3, u64 data4, u64 data5)
224 return HvCallEvent_signalLpEventFast(cnx->remote_lp,
225 HvLpEvent_Type_VirtualLan,
226 subtype, ackind, acktype,
229 token, data1, data2, data3,
233 static inline HvLpEvent_Rc veth_signaldata(struct veth_lpar_connection *cnx,
234 u16 subtype, u64 token, void *data)
236 u64 *p = (u64 *) data;
238 return veth_signalevent(cnx, subtype, HvLpEvent_AckInd_NoAck,
239 HvLpEvent_AckType_ImmediateAck,
240 token, p[0], p[1], p[2], p[3], p[4]);
243 struct veth_allocation {
248 static void veth_complete_allocation(void *parm, int number)
250 struct veth_allocation *vc = (struct veth_allocation *)parm;
256 static int veth_allocate_events(HvLpIndex rlp, int number)
258 struct veth_allocation vc = { COMPLETION_INITIALIZER(vc.c), 0 };
260 mf_allocate_lp_events(rlp, HvLpEvent_Type_VirtualLan,
261 sizeof(struct VethLpEvent), number,
262 &veth_complete_allocation, &vc);
263 wait_for_completion(&vc.c);
269 * LPAR connection code
272 static inline void veth_kick_statemachine(struct veth_lpar_connection *cnx)
274 schedule_work(&cnx->statemachine_wq);
277 static void veth_take_cap(struct veth_lpar_connection *cnx,
278 struct VethLpEvent *event)
282 spin_lock_irqsave(&cnx->lock, flags);
283 /* Receiving caps may mean the other end has just come up, so
284 * we need to reload the instance ID of the far end */
286 HvCallEvent_getTargetLpInstanceId(cnx->remote_lp,
287 HvLpEvent_Type_VirtualLan);
289 if (cnx->state & VETH_STATE_GOTCAPS) {
290 veth_error("Received a second capabilities from LPAR %d.\n",
292 event->base_event.xRc = HvLpEvent_Rc_BufferNotAvailable;
293 HvCallEvent_ackLpEvent((struct HvLpEvent *) event);
295 memcpy(&cnx->cap_event, event, sizeof(cnx->cap_event));
296 cnx->state |= VETH_STATE_GOTCAPS;
297 veth_kick_statemachine(cnx);
299 spin_unlock_irqrestore(&cnx->lock, flags);
302 static void veth_take_cap_ack(struct veth_lpar_connection *cnx,
303 struct VethLpEvent *event)
307 spin_lock_irqsave(&cnx->lock, flags);
308 if (cnx->state & VETH_STATE_GOTCAPACK) {
309 veth_error("Received a second capabilities ack from LPAR %d.\n",
312 memcpy(&cnx->cap_ack_event, event,
313 sizeof(&cnx->cap_ack_event));
314 cnx->state |= VETH_STATE_GOTCAPACK;
315 veth_kick_statemachine(cnx);
317 spin_unlock_irqrestore(&cnx->lock, flags);
320 static void veth_take_monitor_ack(struct veth_lpar_connection *cnx,
321 struct VethLpEvent *event)
325 spin_lock_irqsave(&cnx->lock, flags);
326 veth_debug("cnx %d: lost connection.\n", cnx->remote_lp);
327 cnx->state |= VETH_STATE_RESET;
328 veth_kick_statemachine(cnx);
329 spin_unlock_irqrestore(&cnx->lock, flags);
332 static void veth_handle_ack(struct VethLpEvent *event)
334 HvLpIndex rlp = event->base_event.xTargetLp;
335 struct veth_lpar_connection *cnx = veth_cnx[rlp];
339 switch (event->base_event.xSubtype) {
340 case VethEventTypeCap:
341 veth_take_cap_ack(cnx, event);
343 case VethEventTypeMonitor:
344 veth_take_monitor_ack(cnx, event);
347 veth_error("Unknown ack type %d from LPAR %d.\n",
348 event->base_event.xSubtype, rlp);
352 static void veth_handle_int(struct VethLpEvent *event)
354 HvLpIndex rlp = event->base_event.xSourceLp;
355 struct veth_lpar_connection *cnx = veth_cnx[rlp];
361 switch (event->base_event.xSubtype) {
362 case VethEventTypeCap:
363 veth_take_cap(cnx, event);
365 case VethEventTypeMonitor:
366 /* do nothing... this'll hang out here til we're dead,
367 * and the hypervisor will return it for us. */
369 case VethEventTypeFramesAck:
370 spin_lock_irqsave(&cnx->lock, flags);
371 for (i = 0; i < VETH_MAX_ACKS_PER_MSG; ++i) {
372 u16 msgnum = event->u.frames_ack_data.token[i];
374 if (msgnum < VETH_NUMBUFFERS)
375 veth_recycle_msg(cnx, cnx->msgs + msgnum);
377 spin_unlock_irqrestore(&cnx->lock, flags);
378 veth_flush_pending(cnx);
380 case VethEventTypeFrames:
381 veth_receive(cnx, event);
384 veth_error("Unknown interrupt type %d from LPAR %d.\n",
385 event->base_event.xSubtype, rlp);
389 static void veth_handle_event(struct HvLpEvent *event, struct pt_regs *regs)
391 struct VethLpEvent *veth_event = (struct VethLpEvent *)event;
393 if (event->xFlags.xFunction == HvLpEvent_Function_Ack)
394 veth_handle_ack(veth_event);
395 else if (event->xFlags.xFunction == HvLpEvent_Function_Int)
396 veth_handle_int(veth_event);
399 static int veth_process_caps(struct veth_lpar_connection *cnx)
401 struct VethCapData *remote_caps = &cnx->remote_caps;
404 /* Convert timer to jiffies */
405 cnx->ack_timeout = remote_caps->ack_timeout * HZ / 1000000;
407 if ( (remote_caps->num_buffers == 0)
408 || (remote_caps->ack_threshold > VETH_MAX_ACKS_PER_MSG)
409 || (remote_caps->ack_threshold == 0)
410 || (cnx->ack_timeout == 0) ) {
411 veth_error("Received incompatible capabilities from LPAR %d.\n",
413 return HvLpEvent_Rc_InvalidSubtypeData;
416 num_acks_needed = (remote_caps->num_buffers
417 / remote_caps->ack_threshold) + 1;
419 /* FIXME: locking on num_ack_events? */
420 if (cnx->num_ack_events < num_acks_needed) {
423 num = veth_allocate_events(cnx->remote_lp,
424 num_acks_needed-cnx->num_ack_events);
426 cnx->num_ack_events += num;
428 if (cnx->num_ack_events < num_acks_needed) {
429 veth_error("Couldn't allocate enough ack events "
430 "for LPAR %d.\n", cnx->remote_lp);
432 return HvLpEvent_Rc_BufferNotAvailable;
437 return HvLpEvent_Rc_Good;
440 /* FIXME: The gotos here are a bit dubious */
441 static void veth_statemachine(void *p)
443 struct veth_lpar_connection *cnx = (struct veth_lpar_connection *)p;
444 int rlp = cnx->remote_lp;
447 spin_lock_irq(&cnx->lock);
450 if (cnx->state & VETH_STATE_RESET) {
453 if (cnx->state & VETH_STATE_OPEN)
454 HvCallEvent_closeLpEventPath(cnx->remote_lp,
455 HvLpEvent_Type_VirtualLan);
458 * Reset ack data. This prevents the ack_timer actually
459 * doing anything, even if it runs one more time when
460 * we drop the lock below.
462 memset(&cnx->pending_acks, 0xff, sizeof (cnx->pending_acks));
463 cnx->num_pending_acks = 0;
465 cnx->state &= ~(VETH_STATE_RESET | VETH_STATE_SENTMON
466 | VETH_STATE_OPEN | VETH_STATE_SENTCAPS
467 | VETH_STATE_GOTCAPACK | VETH_STATE_GOTCAPS
468 | VETH_STATE_SENTCAPACK | VETH_STATE_READY);
470 /* Clean up any leftover messages */
472 for (i = 0; i < VETH_NUMBUFFERS; ++i)
473 veth_recycle_msg(cnx, cnx->msgs + i);
475 /* Drop the lock so we can do stuff that might sleep or
476 * take other locks. */
477 spin_unlock_irq(&cnx->lock);
479 del_timer_sync(&cnx->ack_timer);
480 veth_flush_pending(cnx);
482 spin_lock_irq(&cnx->lock);
484 if (cnx->state & VETH_STATE_RESET)
488 if (cnx->state & VETH_STATE_SHUTDOWN)
489 /* It's all over, do nothing */
492 if ( !(cnx->state & VETH_STATE_OPEN) ) {
493 if (! cnx->msgs || (cnx->num_events < (2 + VETH_NUMBUFFERS)) )
496 HvCallEvent_openLpEventPath(rlp, HvLpEvent_Type_VirtualLan);
498 HvCallEvent_getSourceLpInstanceId(rlp,
499 HvLpEvent_Type_VirtualLan);
501 HvCallEvent_getTargetLpInstanceId(rlp,
502 HvLpEvent_Type_VirtualLan);
503 cnx->state |= VETH_STATE_OPEN;
506 if ( (cnx->state & VETH_STATE_OPEN)
507 && !(cnx->state & VETH_STATE_SENTMON) ) {
508 rc = veth_signalevent(cnx, VethEventTypeMonitor,
509 HvLpEvent_AckInd_DoAck,
510 HvLpEvent_AckType_DeferredAck,
513 if (rc == HvLpEvent_Rc_Good) {
514 cnx->state |= VETH_STATE_SENTMON;
516 if ( (rc != HvLpEvent_Rc_PartitionDead)
517 && (rc != HvLpEvent_Rc_PathClosed) )
518 veth_error("Error sending monitor to LPAR %d, "
519 "rc = %d\n", rlp, rc);
521 /* Oh well, hope we get a cap from the other
522 * end and do better when that kicks us */
527 if ( (cnx->state & VETH_STATE_OPEN)
528 && !(cnx->state & VETH_STATE_SENTCAPS)) {
529 u64 *rawcap = (u64 *)&cnx->local_caps;
531 rc = veth_signalevent(cnx, VethEventTypeCap,
532 HvLpEvent_AckInd_DoAck,
533 HvLpEvent_AckType_ImmediateAck,
534 0, rawcap[0], rawcap[1], rawcap[2],
535 rawcap[3], rawcap[4]);
537 if (rc == HvLpEvent_Rc_Good) {
538 cnx->state |= VETH_STATE_SENTCAPS;
540 if ( (rc != HvLpEvent_Rc_PartitionDead)
541 && (rc != HvLpEvent_Rc_PathClosed) )
542 veth_error("Error sending caps to LPAR %d, "
543 "rc = %d\n", rlp, rc);
545 /* Oh well, hope we get a cap from the other
546 * end and do better when that kicks us */
551 if ((cnx->state & VETH_STATE_GOTCAPS)
552 && !(cnx->state & VETH_STATE_SENTCAPACK)) {
553 struct VethCapData *remote_caps = &cnx->remote_caps;
555 memcpy(remote_caps, &cnx->cap_event.u.caps_data,
556 sizeof(*remote_caps));
558 spin_unlock_irq(&cnx->lock);
559 rc = veth_process_caps(cnx);
560 spin_lock_irq(&cnx->lock);
562 /* We dropped the lock, so recheck for anything which
563 * might mess us up */
564 if (cnx->state & (VETH_STATE_RESET|VETH_STATE_SHUTDOWN))
567 cnx->cap_event.base_event.xRc = rc;
568 HvCallEvent_ackLpEvent((struct HvLpEvent *)&cnx->cap_event);
569 if (rc == HvLpEvent_Rc_Good)
570 cnx->state |= VETH_STATE_SENTCAPACK;
575 if ((cnx->state & VETH_STATE_GOTCAPACK)
576 && (cnx->state & VETH_STATE_GOTCAPS)
577 && !(cnx->state & VETH_STATE_READY)) {
578 if (cnx->cap_ack_event.base_event.xRc == HvLpEvent_Rc_Good) {
579 /* Start the ACK timer */
580 cnx->ack_timer.expires = jiffies + cnx->ack_timeout;
581 add_timer(&cnx->ack_timer);
582 cnx->state |= VETH_STATE_READY;
584 veth_error("Caps rejected by LPAR %d, rc = %d\n",
585 rlp, cnx->cap_ack_event.base_event.xRc);
591 spin_unlock_irq(&cnx->lock);
595 /* FIXME: we get here if something happens we really can't
596 * cope with. The link will never work once we get here, and
597 * all we can do is not lock the rest of the system up */
598 veth_error("Unrecoverable error on connection to LPAR %d, shutting down"
599 " (state = 0x%04lx)\n", rlp, cnx->state);
600 cnx->state |= VETH_STATE_SHUTDOWN;
601 spin_unlock_irq(&cnx->lock);
604 static int veth_init_connection(u8 rlp)
606 struct veth_lpar_connection *cnx;
607 struct veth_msg *msgs;
610 if ( (rlp == this_lp)
611 || ! HvLpConfig_doLpsCommunicateOnVirtualLan(this_lp, rlp) )
614 cnx = kmalloc(sizeof(*cnx), GFP_KERNEL);
617 memset(cnx, 0, sizeof(*cnx));
619 cnx->remote_lp = rlp;
620 spin_lock_init(&cnx->lock);
621 INIT_WORK(&cnx->statemachine_wq, veth_statemachine, cnx);
622 init_timer(&cnx->ack_timer);
623 cnx->ack_timer.function = veth_timed_ack;
624 cnx->ack_timer.data = (unsigned long) cnx;
625 memset(&cnx->pending_acks, 0xff, sizeof (cnx->pending_acks));
629 msgs = kmalloc(VETH_NUMBUFFERS * sizeof(struct veth_msg), GFP_KERNEL);
631 veth_error("Can't allocate buffers for LPAR %d.\n", rlp);
636 memset(msgs, 0, VETH_NUMBUFFERS * sizeof(struct veth_msg));
637 spin_lock_init(&cnx->msg_stack_lock);
639 for (i = 0; i < VETH_NUMBUFFERS; i++) {
641 veth_stack_push(cnx, msgs + i);
644 cnx->num_events = veth_allocate_events(rlp, 2 + VETH_NUMBUFFERS);
646 if (cnx->num_events < (2 + VETH_NUMBUFFERS)) {
647 veth_error("Can't allocate enough events for LPAR %d.\n", rlp);
651 cnx->local_caps.num_buffers = VETH_NUMBUFFERS;
652 cnx->local_caps.ack_threshold = ACK_THRESHOLD;
653 cnx->local_caps.ack_timeout = VETH_ACKTIMEOUT;
658 static void veth_stop_connection(u8 rlp)
660 struct veth_lpar_connection *cnx = veth_cnx[rlp];
665 spin_lock_irq(&cnx->lock);
666 cnx->state |= VETH_STATE_RESET | VETH_STATE_SHUTDOWN;
667 veth_kick_statemachine(cnx);
668 spin_unlock_irq(&cnx->lock);
670 /* Wait for the state machine to run. */
671 flush_scheduled_work();
673 if (cnx->num_events > 0)
674 mf_deallocate_lp_events(cnx->remote_lp,
675 HvLpEvent_Type_VirtualLan,
678 if (cnx->num_ack_events > 0)
679 mf_deallocate_lp_events(cnx->remote_lp,
680 HvLpEvent_Type_VirtualLan,
685 static void veth_destroy_connection(u8 rlp)
687 struct veth_lpar_connection *cnx = veth_cnx[rlp];
694 veth_cnx[rlp] = NULL;
701 static int veth_open(struct net_device *dev)
703 struct veth_port *port = (struct veth_port *) dev->priv;
705 memset(&port->stats, 0, sizeof (port->stats));
706 netif_start_queue(dev);
710 static int veth_close(struct net_device *dev)
712 netif_stop_queue(dev);
716 static struct net_device_stats *veth_get_stats(struct net_device *dev)
718 struct veth_port *port = (struct veth_port *) dev->priv;
723 static int veth_change_mtu(struct net_device *dev, int new_mtu)
725 if ((new_mtu < 68) || (new_mtu > VETH_MAX_MTU))
731 static void veth_set_multicast_list(struct net_device *dev)
733 struct veth_port *port = (struct veth_port *) dev->priv;
736 write_lock_irqsave(&port->mcast_gate, flags);
738 if (dev->flags & IFF_PROMISC) { /* set promiscuous mode */
739 printk(KERN_INFO "%s: Promiscuous mode enabled.\n",
741 port->promiscuous = 1;
742 } else if ( (dev->flags & IFF_ALLMULTI)
743 || (dev->mc_count > VETH_MAX_MCAST) ) {
746 struct dev_mc_list *dmi = dev->mc_list;
752 for (i = 0; i < dev->mc_count; i++) {
753 u8 *addr = dmi->dmi_addr;
756 if (addr[0] & 0x01) {/* multicast address? */
757 memcpy(&xaddr, addr, ETH_ALEN);
758 port->mcast_addr[port->num_mcast] = xaddr;
765 write_unlock_irqrestore(&port->mcast_gate, flags);
768 static void veth_get_drvinfo(struct net_device *dev, struct ethtool_drvinfo *info)
770 strncpy(info->driver, "veth", sizeof(info->driver) - 1);
771 info->driver[sizeof(info->driver) - 1] = '\0';
772 strncpy(info->version, "1.0", sizeof(info->version) - 1);
775 static int veth_get_settings(struct net_device *dev, struct ethtool_cmd *ecmd)
777 ecmd->supported = (SUPPORTED_1000baseT_Full
778 | SUPPORTED_Autoneg | SUPPORTED_FIBRE);
779 ecmd->advertising = (SUPPORTED_1000baseT_Full
780 | SUPPORTED_Autoneg | SUPPORTED_FIBRE);
781 ecmd->port = PORT_FIBRE;
782 ecmd->transceiver = XCVR_INTERNAL;
783 ecmd->phy_address = 0;
784 ecmd->speed = SPEED_1000;
785 ecmd->duplex = DUPLEX_FULL;
786 ecmd->autoneg = AUTONEG_ENABLE;
787 ecmd->maxtxpkt = 120;
788 ecmd->maxrxpkt = 120;
792 static u32 veth_get_link(struct net_device *dev)
797 static struct ethtool_ops ops = {
798 .get_drvinfo = veth_get_drvinfo,
799 .get_settings = veth_get_settings,
800 .get_link = veth_get_link,
803 static void veth_tx_timeout(struct net_device *dev)
805 struct veth_port *port = (struct veth_port *)dev->priv;
806 struct net_device_stats *stats = &port->stats;
812 spin_lock_irqsave(&port->pending_gate, flags);
814 if (!port->pending_lpmask) {
815 spin_unlock_irqrestore(&port->pending_gate, flags);
819 printk(KERN_WARNING "%s: Tx timeout! Resetting lp connections: %08x\n",
820 dev->name, port->pending_lpmask);
822 for (i = 0; i < HVMAXARCHITECTEDLPS; i++) {
823 struct veth_lpar_connection *cnx = veth_cnx[i];
825 if (! (port->pending_lpmask & (1<<i)))
828 /* If we're pending on it, we must be connected to it,
829 * so we should certainly have a structure for it. */
832 /* Theoretically we could be kicking a connection
833 * which doesn't deserve it, but in practice if we've
834 * had a Tx timeout, the pending_lpmask will have
835 * exactly one bit set - the connection causing the
837 spin_lock(&cnx->lock);
838 cnx->state |= VETH_STATE_RESET;
839 veth_kick_statemachine(cnx);
840 spin_unlock(&cnx->lock);
843 spin_unlock_irqrestore(&port->pending_gate, flags);
846 static struct net_device * __init veth_probe_one(int vlan, struct device *vdev)
848 struct net_device *dev;
849 struct veth_port *port;
852 dev = alloc_etherdev(sizeof (struct veth_port));
854 veth_error("Unable to allocate net_device structure!\n");
858 port = (struct veth_port *) dev->priv;
860 spin_lock_init(&port->pending_gate);
861 rwlock_init(&port->mcast_gate);
863 for (i = 0; i < HVMAXARCHITECTEDLPS; i++) {
864 HvLpVirtualLanIndexMap map;
868 map = HvLpConfig_getVirtualLanIndexMapForLp(i);
869 if (map & (0x8000 >> vlan))
870 port->lpar_map |= (1 << i);
874 dev->dev_addr[0] = 0x02;
875 dev->dev_addr[1] = 0x01;
876 dev->dev_addr[2] = 0xff;
877 dev->dev_addr[3] = vlan;
878 dev->dev_addr[4] = 0xff;
879 dev->dev_addr[5] = this_lp;
881 dev->mtu = VETH_MAX_MTU;
883 memcpy(&port->mac_addr, dev->dev_addr, 6);
885 dev->open = veth_open;
886 dev->hard_start_xmit = veth_start_xmit;
887 dev->stop = veth_close;
888 dev->get_stats = veth_get_stats;
889 dev->change_mtu = veth_change_mtu;
890 dev->set_mac_address = NULL;
891 dev->set_multicast_list = veth_set_multicast_list;
892 SET_ETHTOOL_OPS(dev, &ops);
894 dev->watchdog_timeo = 2 * (VETH_ACKTIMEOUT * HZ / 1000000);
895 dev->tx_timeout = veth_tx_timeout;
897 SET_NETDEV_DEV(dev, vdev);
899 rc = register_netdev(dev);
901 veth_error("Failed registering net device for vlan%d.\n", vlan);
906 veth_info("%s attached to iSeries vlan %d (LPAR map = 0x%.4X)\n",
907 dev->name, vlan, port->lpar_map);
916 static int veth_transmit_to_one(struct sk_buff *skb, HvLpIndex rlp,
917 struct net_device *dev)
919 struct veth_lpar_connection *cnx = veth_cnx[rlp];
920 struct veth_port *port = (struct veth_port *) dev->priv;
922 u32 dma_address, dma_length;
923 struct veth_msg *msg = NULL;
928 port->stats.tx_errors++;
933 spin_lock_irqsave(&cnx->lock, flags);
935 if (! (cnx->state & VETH_STATE_READY))
938 if ((skb->len - 14) > VETH_MAX_MTU)
941 msg = veth_stack_pop(cnx);
948 dma_length = skb->len;
949 dma_address = dma_map_single(port->dev, skb->data,
950 dma_length, DMA_TO_DEVICE);
952 if (dma_mapping_error(dma_address))
953 goto recycle_and_drop;
955 /* Is it really necessary to check the length and address
956 * fields of the first entry here? */
958 msg->dev = port->dev;
959 msg->data.addr[0] = dma_address;
960 msg->data.len[0] = dma_length;
961 msg->data.eofmask = 1 << VETH_EOF_SHIFT;
962 set_bit(0, &(msg->in_use));
963 rc = veth_signaldata(cnx, VethEventTypeFrames, msg->token, &msg->data);
965 if (rc != HvLpEvent_Rc_Good)
966 goto recycle_and_drop;
968 spin_unlock_irqrestore(&cnx->lock, flags);
973 /* need to set in use to make veth_recycle_msg in case this
974 * was a mapping failure */
975 set_bit(0, &msg->in_use);
976 veth_recycle_msg(cnx, msg);
978 port->stats.tx_errors++;
980 spin_unlock_irqrestore(&cnx->lock, flags);
984 static HvLpIndexMap veth_transmit_to_many(struct sk_buff *skb,
986 struct net_device *dev)
988 struct veth_port *port = (struct veth_port *) dev->priv;
992 for (i = 0; i < HVMAXARCHITECTEDLPS; i++) {
993 if ((lpmask & (1 << i)) == 0)
996 rc = veth_transmit_to_one(skb_get(skb), i, dev);
1002 port->stats.tx_packets++;
1003 port->stats.tx_bytes += skb->len;
1009 static int veth_start_xmit(struct sk_buff *skb, struct net_device *dev)
1011 unsigned char *frame = skb->data;
1012 struct veth_port *port = (struct veth_port *) dev->priv;
1013 unsigned long flags;
1014 HvLpIndexMap lpmask;
1016 if (! (frame[0] & 0x01)) {
1017 /* unicast packet */
1018 HvLpIndex rlp = frame[5];
1020 if ( ! ((1 << rlp) & port->lpar_map) ) {
1027 lpmask = port->lpar_map;
1030 spin_lock_irqsave(&port->pending_gate, flags);
1032 lpmask = veth_transmit_to_many(skb, lpmask, dev);
1034 dev->trans_start = jiffies;
1039 if (port->pending_skb) {
1040 veth_error("%s: TX while skb was pending!\n",
1043 spin_unlock_irqrestore(&port->pending_gate, flags);
1047 port->pending_skb = skb;
1048 port->pending_lpmask = lpmask;
1049 netif_stop_queue(dev);
1052 spin_unlock_irqrestore(&port->pending_gate, flags);
1057 static void veth_recycle_msg(struct veth_lpar_connection *cnx,
1058 struct veth_msg *msg)
1060 u32 dma_address, dma_length;
1062 if (test_and_clear_bit(0, &msg->in_use)) {
1063 dma_address = msg->data.addr[0];
1064 dma_length = msg->data.len[0];
1066 dma_unmap_single(msg->dev, dma_address, dma_length,
1070 dev_kfree_skb_any(msg->skb);
1074 memset(&msg->data, 0, sizeof(msg->data));
1075 veth_stack_push(cnx, msg);
1076 } else if (cnx->state & VETH_STATE_OPEN) {
1077 veth_error("Non-pending frame (# %d) acked by LPAR %d.\n",
1078 cnx->remote_lp, msg->token);
1082 static void veth_flush_pending(struct veth_lpar_connection *cnx)
1085 for (i = 0; i < HVMAXARCHITECTEDVIRTUALLANS; i++) {
1086 struct net_device *dev = veth_dev[i];
1087 struct veth_port *port;
1088 unsigned long flags;
1093 port = (struct veth_port *)dev->priv;
1095 if (! (port->lpar_map & (1<<cnx->remote_lp)))
1098 spin_lock_irqsave(&port->pending_gate, flags);
1099 if (port->pending_skb) {
1100 port->pending_lpmask =
1101 veth_transmit_to_many(port->pending_skb,
1102 port->pending_lpmask,
1104 if (! port->pending_lpmask) {
1105 dev_kfree_skb_any(port->pending_skb);
1106 port->pending_skb = NULL;
1107 netif_wake_queue(dev);
1110 spin_unlock_irqrestore(&port->pending_gate, flags);
1118 static inline int veth_frame_wanted(struct veth_port *port, u64 mac_addr)
1122 unsigned long flags;
1124 if ( (mac_addr == port->mac_addr) || (mac_addr == 0xffffffffffff0000) )
1127 if (! (((char *) &mac_addr)[0] & 0x01))
1130 read_lock_irqsave(&port->mcast_gate, flags);
1132 if (port->promiscuous || port->all_mcast) {
1137 for (i = 0; i < port->num_mcast; ++i) {
1138 if (port->mcast_addr[i] == mac_addr) {
1145 read_unlock_irqrestore(&port->mcast_gate, flags);
1155 #define VETH_MAX_PAGES_PER_FRAME ( (VETH_MAX_MTU+PAGE_SIZE-2)/PAGE_SIZE + 1 )
1157 static inline void veth_build_dma_list(struct dma_chunk *list,
1158 unsigned char *p, unsigned long length)
1163 /* FIXME: skbs are continguous in real addresses. Do we
1164 * really need to break it into PAGE_SIZE chunks, or can we do
1165 * it just at the granularity of iSeries real->absolute
1166 * mapping? Indeed, given the way the allocator works, can we
1167 * count on them being absolutely contiguous? */
1168 list[0].addr = ISERIES_HV_ADDR(p);
1169 list[0].size = min(length,
1170 PAGE_SIZE - ((unsigned long)p & ~PAGE_MASK));
1172 done = list[0].size;
1173 while (done < length) {
1174 list[i].addr = ISERIES_HV_ADDR(p + done);
1175 list[i].size = min(length-done, PAGE_SIZE);
1176 done += list[i].size;
1181 static void veth_flush_acks(struct veth_lpar_connection *cnx)
1185 rc = veth_signaldata(cnx, VethEventTypeFramesAck,
1186 0, &cnx->pending_acks);
1188 if (rc != HvLpEvent_Rc_Good)
1189 veth_error("Failed acking frames from LPAR %d, rc = %d\n",
1190 cnx->remote_lp, (int)rc);
1192 cnx->num_pending_acks = 0;
1193 memset(&cnx->pending_acks, 0xff, sizeof(cnx->pending_acks));
1196 static void veth_receive(struct veth_lpar_connection *cnx,
1197 struct VethLpEvent *event)
1199 struct VethFramesData *senddata = &event->u.frames_data;
1202 unsigned long flags;
1207 struct sk_buff *skb;
1208 struct dma_chunk local_list[VETH_MAX_PAGES_PER_FRAME];
1209 struct dma_chunk remote_list[VETH_MAX_FRAMES_PER_MSG];
1211 HvLpVirtualLanIndex vlan;
1212 struct net_device *dev;
1213 struct veth_port *port;
1215 /* FIXME: do we need this? */
1216 memset(local_list, 0, sizeof(local_list));
1217 memset(remote_list, 0, sizeof(VETH_MAX_FRAMES_PER_MSG));
1219 /* a 0 address marks the end of the valid entries */
1220 if (senddata->addr[startchunk] == 0)
1223 /* make sure that we have at least 1 EOF entry in the
1224 * remaining entries */
1225 if (! (senddata->eofmask >> (startchunk + VETH_EOF_SHIFT))) {
1226 veth_error("Missing EOF fragment in event "
1227 "eofmask = 0x%x startchunk = %d\n",
1228 (unsigned)senddata->eofmask,
1233 /* build list of chunks in this frame */
1236 remote_list[nchunks].addr =
1237 (u64) senddata->addr[startchunk+nchunks] << 32;
1238 remote_list[nchunks].size =
1239 senddata->len[startchunk+nchunks];
1240 length += remote_list[nchunks].size;
1241 } while (! (senddata->eofmask &
1242 (1 << (VETH_EOF_SHIFT + startchunk + nchunks++))));
1244 /* length == total length of all chunks */
1245 /* nchunks == # of chunks in this frame */
1247 if ((length - ETH_HLEN) > VETH_MAX_MTU) {
1248 veth_error("Received oversize frame from LPAR %d "
1250 cnx->remote_lp, length);
1254 skb = alloc_skb(length, GFP_ATOMIC);
1258 veth_build_dma_list(local_list, skb->data, length);
1260 rc = HvCallEvent_dmaBufList(HvLpEvent_Type_VirtualLan,
1261 event->base_event.xSourceLp,
1262 HvLpDma_Direction_RemoteToLocal,
1265 HvLpDma_AddressType_RealAddress,
1266 HvLpDma_AddressType_TceIndex,
1267 ISERIES_HV_ADDR(&local_list),
1268 ISERIES_HV_ADDR(&remote_list),
1270 if (rc != HvLpDma_Rc_Good) {
1271 dev_kfree_skb_irq(skb);
1275 vlan = skb->data[9];
1276 dev = veth_dev[vlan];
1279 * Some earlier versions of the driver sent
1280 * broadcasts down all connections, even to lpars
1281 * that weren't on the relevant vlan. So ignore
1282 * packets belonging to a vlan we're not on.
1283 * We can also be here if we receive packets while
1284 * the driver is going down, because then dev is NULL.
1286 dev_kfree_skb_irq(skb);
1290 port = (struct veth_port *)dev->priv;
1291 dest = *((u64 *) skb->data) & 0xFFFFFFFFFFFF0000;
1293 if ((vlan > HVMAXARCHITECTEDVIRTUALLANS) || !port) {
1294 dev_kfree_skb_irq(skb);
1297 if (! veth_frame_wanted(port, dest)) {
1298 dev_kfree_skb_irq(skb);
1302 skb_put(skb, length);
1304 skb->protocol = eth_type_trans(skb, dev);
1305 skb->ip_summed = CHECKSUM_NONE;
1306 netif_rx(skb); /* send it up */
1307 port->stats.rx_packets++;
1308 port->stats.rx_bytes += length;
1309 } while (startchunk += nchunks, startchunk < VETH_MAX_FRAMES_PER_MSG);
1312 spin_lock_irqsave(&cnx->lock, flags);
1313 BUG_ON(cnx->num_pending_acks > VETH_MAX_ACKS_PER_MSG);
1315 cnx->pending_acks[cnx->num_pending_acks++] =
1316 event->base_event.xCorrelationToken;
1318 if ( (cnx->num_pending_acks >= cnx->remote_caps.ack_threshold)
1319 || (cnx->num_pending_acks >= VETH_MAX_ACKS_PER_MSG) )
1320 veth_flush_acks(cnx);
1322 spin_unlock_irqrestore(&cnx->lock, flags);
1325 static void veth_timed_ack(unsigned long ptr)
1327 struct veth_lpar_connection *cnx = (struct veth_lpar_connection *) ptr;
1328 unsigned long flags;
1330 /* Ack all the events */
1331 spin_lock_irqsave(&cnx->lock, flags);
1332 if (cnx->num_pending_acks > 0)
1333 veth_flush_acks(cnx);
1335 /* Reschedule the timer */
1336 cnx->ack_timer.expires = jiffies + cnx->ack_timeout;
1337 add_timer(&cnx->ack_timer);
1338 spin_unlock_irqrestore(&cnx->lock, flags);
1341 static int veth_remove(struct vio_dev *vdev)
1343 int i = vdev->unit_address;
1344 struct net_device *dev;
1349 unregister_netdev(dev);
1355 static int veth_probe(struct vio_dev *vdev, const struct vio_device_id *id)
1357 int i = vdev->unit_address;
1358 struct net_device *dev;
1360 dev = veth_probe_one(i, &vdev->dev);
1367 /* Start the state machine on each connection, to commence
1368 * link negotiation */
1369 for (i = 0; i < HVMAXARCHITECTEDLPS; i++)
1371 veth_kick_statemachine(veth_cnx[i]);
1377 * veth_device_table: Used by vio.c to match devices that we
1380 static struct vio_device_id veth_device_table[] __devinitdata = {
1384 MODULE_DEVICE_TABLE(vio, veth_device_table);
1386 static struct vio_driver veth_driver = {
1387 .name = "iseries_veth",
1388 .id_table = veth_device_table,
1389 .probe = veth_probe,
1390 .remove = veth_remove
1394 * Module initialization/cleanup
1397 void __exit veth_module_cleanup(void)
1401 /* Stop the queues first to stop any new packets being sent. */
1402 for (i = 0; i < HVMAXARCHITECTEDVIRTUALLANS; i++)
1404 netif_stop_queue(veth_dev[i]);
1406 /* Stop the connections before we unregister the driver. This
1407 * ensures there's no skbs lying around holding the device open. */
1408 for (i = 0; i < HVMAXARCHITECTEDLPS; ++i)
1409 veth_stop_connection(i);
1411 HvLpEvent_unregisterHandler(HvLpEvent_Type_VirtualLan);
1413 /* Hypervisor callbacks may have scheduled more work while we
1414 * were stoping connections. Now that we've disconnected from
1415 * the hypervisor make sure everything's finished. */
1416 flush_scheduled_work();
1418 vio_unregister_driver(&veth_driver);
1420 for (i = 0; i < HVMAXARCHITECTEDLPS; ++i)
1421 veth_destroy_connection(i);
1424 module_exit(veth_module_cleanup);
1426 int __init veth_module_init(void)
1431 this_lp = HvLpConfig_getLpIndex_outline();
1433 for (i = 0; i < HVMAXARCHITECTEDLPS; ++i) {
1434 rc = veth_init_connection(i);
1436 veth_module_cleanup();
1441 HvLpEvent_registerHandler(HvLpEvent_Type_VirtualLan,
1442 &veth_handle_event);
1444 return vio_register_driver(&veth_driver);
1446 module_init(veth_module_init);