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);
189 #define veth_info(fmt, args...) \
190 printk(KERN_INFO "iseries_veth: " fmt, ## args)
192 #define veth_error(fmt, args...) \
193 printk(KERN_ERR "iseries_veth: Error: " fmt, ## args)
196 #define veth_debug(fmt, args...) \
197 printk(KERN_DEBUG "iseries_veth: " fmt, ## args)
199 #define veth_debug(fmt, args...) do {} while (0)
202 /* You must hold the connection's lock when you call this function. */
203 static inline void veth_stack_push(struct veth_lpar_connection *cnx,
204 struct veth_msg *msg)
206 msg->next = cnx->msg_stack_head;
207 cnx->msg_stack_head = msg;
210 /* You must hold the connection's lock when you call this function. */
211 static inline struct veth_msg *veth_stack_pop(struct veth_lpar_connection *cnx)
213 struct veth_msg *msg;
215 msg = cnx->msg_stack_head;
217 cnx->msg_stack_head = cnx->msg_stack_head->next;
222 /* You must hold the connection's lock when you call this function. */
223 static inline int veth_stack_is_empty(struct veth_lpar_connection *cnx)
225 return cnx->msg_stack_head == NULL;
228 static inline HvLpEvent_Rc
229 veth_signalevent(struct veth_lpar_connection *cnx, u16 subtype,
230 HvLpEvent_AckInd ackind, HvLpEvent_AckType acktype,
232 u64 data1, u64 data2, u64 data3, u64 data4, u64 data5)
234 return HvCallEvent_signalLpEventFast(cnx->remote_lp,
235 HvLpEvent_Type_VirtualLan,
236 subtype, ackind, acktype,
239 token, data1, data2, data3,
243 static inline HvLpEvent_Rc veth_signaldata(struct veth_lpar_connection *cnx,
244 u16 subtype, u64 token, void *data)
246 u64 *p = (u64 *) data;
248 return veth_signalevent(cnx, subtype, HvLpEvent_AckInd_NoAck,
249 HvLpEvent_AckType_ImmediateAck,
250 token, p[0], p[1], p[2], p[3], p[4]);
253 struct veth_allocation {
258 static void veth_complete_allocation(void *parm, int number)
260 struct veth_allocation *vc = (struct veth_allocation *)parm;
266 static int veth_allocate_events(HvLpIndex rlp, int number)
268 struct veth_allocation vc = { COMPLETION_INITIALIZER(vc.c), 0 };
270 mf_allocate_lp_events(rlp, HvLpEvent_Type_VirtualLan,
271 sizeof(struct VethLpEvent), number,
272 &veth_complete_allocation, &vc);
273 wait_for_completion(&vc.c);
282 struct veth_cnx_attribute {
283 struct attribute attr;
284 ssize_t (*show)(struct veth_lpar_connection *, char *buf);
285 ssize_t (*store)(struct veth_lpar_connection *, const char *buf);
288 static ssize_t veth_cnx_attribute_show(struct kobject *kobj,
289 struct attribute *attr, char *buf)
291 struct veth_cnx_attribute *cnx_attr;
292 struct veth_lpar_connection *cnx;
294 cnx_attr = container_of(attr, struct veth_cnx_attribute, attr);
295 cnx = container_of(kobj, struct veth_lpar_connection, kobject);
300 return cnx_attr->show(cnx, buf);
303 #define CUSTOM_CNX_ATTR(_name, _format, _expression) \
304 static ssize_t _name##_show(struct veth_lpar_connection *cnx, char *buf)\
306 return sprintf(buf, _format, _expression); \
308 struct veth_cnx_attribute veth_cnx_attr_##_name = __ATTR_RO(_name)
310 #define SIMPLE_CNX_ATTR(_name) \
311 CUSTOM_CNX_ATTR(_name, "%lu\n", (unsigned long)cnx->_name)
313 SIMPLE_CNX_ATTR(outstanding_tx);
314 SIMPLE_CNX_ATTR(remote_lp);
315 SIMPLE_CNX_ATTR(num_events);
316 SIMPLE_CNX_ATTR(src_inst);
317 SIMPLE_CNX_ATTR(dst_inst);
318 SIMPLE_CNX_ATTR(num_pending_acks);
319 SIMPLE_CNX_ATTR(num_ack_events);
320 CUSTOM_CNX_ATTR(ack_timeout, "%d\n", jiffies_to_msecs(cnx->ack_timeout));
321 CUSTOM_CNX_ATTR(reset_timeout, "%d\n", jiffies_to_msecs(cnx->reset_timeout));
322 CUSTOM_CNX_ATTR(state, "0x%.4lX\n", cnx->state);
323 CUSTOM_CNX_ATTR(last_contact, "%d\n", cnx->last_contact ?
324 jiffies_to_msecs(jiffies - cnx->last_contact) : 0);
326 #define GET_CNX_ATTR(_name) (&veth_cnx_attr_##_name.attr)
328 static struct attribute *veth_cnx_default_attrs[] = {
329 GET_CNX_ATTR(outstanding_tx),
330 GET_CNX_ATTR(remote_lp),
331 GET_CNX_ATTR(num_events),
332 GET_CNX_ATTR(reset_timeout),
333 GET_CNX_ATTR(last_contact),
335 GET_CNX_ATTR(src_inst),
336 GET_CNX_ATTR(dst_inst),
337 GET_CNX_ATTR(num_pending_acks),
338 GET_CNX_ATTR(num_ack_events),
339 GET_CNX_ATTR(ack_timeout),
343 static struct sysfs_ops veth_cnx_sysfs_ops = {
344 .show = veth_cnx_attribute_show
347 static struct kobj_type veth_lpar_connection_ktype = {
348 .release = veth_release_connection,
349 .sysfs_ops = &veth_cnx_sysfs_ops,
350 .default_attrs = veth_cnx_default_attrs
354 * LPAR connection code
357 static inline void veth_kick_statemachine(struct veth_lpar_connection *cnx)
359 schedule_work(&cnx->statemachine_wq);
362 static void veth_take_cap(struct veth_lpar_connection *cnx,
363 struct VethLpEvent *event)
367 spin_lock_irqsave(&cnx->lock, flags);
368 /* Receiving caps may mean the other end has just come up, so
369 * we need to reload the instance ID of the far end */
371 HvCallEvent_getTargetLpInstanceId(cnx->remote_lp,
372 HvLpEvent_Type_VirtualLan);
374 if (cnx->state & VETH_STATE_GOTCAPS) {
375 veth_error("Received a second capabilities from LPAR %d.\n",
377 event->base_event.xRc = HvLpEvent_Rc_BufferNotAvailable;
378 HvCallEvent_ackLpEvent((struct HvLpEvent *) event);
380 memcpy(&cnx->cap_event, event, sizeof(cnx->cap_event));
381 cnx->state |= VETH_STATE_GOTCAPS;
382 veth_kick_statemachine(cnx);
384 spin_unlock_irqrestore(&cnx->lock, flags);
387 static void veth_take_cap_ack(struct veth_lpar_connection *cnx,
388 struct VethLpEvent *event)
392 spin_lock_irqsave(&cnx->lock, flags);
393 if (cnx->state & VETH_STATE_GOTCAPACK) {
394 veth_error("Received a second capabilities ack from LPAR %d.\n",
397 memcpy(&cnx->cap_ack_event, event,
398 sizeof(&cnx->cap_ack_event));
399 cnx->state |= VETH_STATE_GOTCAPACK;
400 veth_kick_statemachine(cnx);
402 spin_unlock_irqrestore(&cnx->lock, flags);
405 static void veth_take_monitor_ack(struct veth_lpar_connection *cnx,
406 struct VethLpEvent *event)
410 spin_lock_irqsave(&cnx->lock, flags);
411 veth_debug("cnx %d: lost connection.\n", cnx->remote_lp);
413 /* Avoid kicking the statemachine once we're shutdown.
414 * It's unnecessary and it could break veth_stop_connection(). */
416 if (! (cnx->state & VETH_STATE_SHUTDOWN)) {
417 cnx->state |= VETH_STATE_RESET;
418 veth_kick_statemachine(cnx);
420 spin_unlock_irqrestore(&cnx->lock, flags);
423 static void veth_handle_ack(struct VethLpEvent *event)
425 HvLpIndex rlp = event->base_event.xTargetLp;
426 struct veth_lpar_connection *cnx = veth_cnx[rlp];
430 switch (event->base_event.xSubtype) {
431 case VethEventTypeCap:
432 veth_take_cap_ack(cnx, event);
434 case VethEventTypeMonitor:
435 veth_take_monitor_ack(cnx, event);
438 veth_error("Unknown ack type %d from LPAR %d.\n",
439 event->base_event.xSubtype, rlp);
443 static void veth_handle_int(struct VethLpEvent *event)
445 HvLpIndex rlp = event->base_event.xSourceLp;
446 struct veth_lpar_connection *cnx = veth_cnx[rlp];
452 switch (event->base_event.xSubtype) {
453 case VethEventTypeCap:
454 veth_take_cap(cnx, event);
456 case VethEventTypeMonitor:
457 /* do nothing... this'll hang out here til we're dead,
458 * and the hypervisor will return it for us. */
460 case VethEventTypeFramesAck:
461 spin_lock_irqsave(&cnx->lock, flags);
463 for (i = 0; i < VETH_MAX_ACKS_PER_MSG; ++i) {
464 u16 msgnum = event->u.frames_ack_data.token[i];
466 if (msgnum < VETH_NUMBUFFERS) {
467 veth_recycle_msg(cnx, cnx->msgs + msgnum);
468 cnx->outstanding_tx--;
474 cnx->last_contact = jiffies;
475 veth_wake_queues(cnx);
478 spin_unlock_irqrestore(&cnx->lock, flags);
480 case VethEventTypeFrames:
481 veth_receive(cnx, event);
484 veth_error("Unknown interrupt type %d from LPAR %d.\n",
485 event->base_event.xSubtype, rlp);
489 static void veth_handle_event(struct HvLpEvent *event, struct pt_regs *regs)
491 struct VethLpEvent *veth_event = (struct VethLpEvent *)event;
493 if (event->xFlags.xFunction == HvLpEvent_Function_Ack)
494 veth_handle_ack(veth_event);
495 else if (event->xFlags.xFunction == HvLpEvent_Function_Int)
496 veth_handle_int(veth_event);
499 static int veth_process_caps(struct veth_lpar_connection *cnx)
501 struct VethCapData *remote_caps = &cnx->remote_caps;
504 /* Convert timer to jiffies */
505 cnx->ack_timeout = remote_caps->ack_timeout * HZ / 1000000;
507 if ( (remote_caps->num_buffers == 0)
508 || (remote_caps->ack_threshold > VETH_MAX_ACKS_PER_MSG)
509 || (remote_caps->ack_threshold == 0)
510 || (cnx->ack_timeout == 0) ) {
511 veth_error("Received incompatible capabilities from LPAR %d.\n",
513 return HvLpEvent_Rc_InvalidSubtypeData;
516 num_acks_needed = (remote_caps->num_buffers
517 / remote_caps->ack_threshold) + 1;
519 /* FIXME: locking on num_ack_events? */
520 if (cnx->num_ack_events < num_acks_needed) {
523 num = veth_allocate_events(cnx->remote_lp,
524 num_acks_needed-cnx->num_ack_events);
526 cnx->num_ack_events += num;
528 if (cnx->num_ack_events < num_acks_needed) {
529 veth_error("Couldn't allocate enough ack events "
530 "for LPAR %d.\n", cnx->remote_lp);
532 return HvLpEvent_Rc_BufferNotAvailable;
537 return HvLpEvent_Rc_Good;
540 /* FIXME: The gotos here are a bit dubious */
541 static void veth_statemachine(void *p)
543 struct veth_lpar_connection *cnx = (struct veth_lpar_connection *)p;
544 int rlp = cnx->remote_lp;
547 spin_lock_irq(&cnx->lock);
550 if (cnx->state & VETH_STATE_RESET) {
551 if (cnx->state & VETH_STATE_OPEN)
552 HvCallEvent_closeLpEventPath(cnx->remote_lp,
553 HvLpEvent_Type_VirtualLan);
556 * Reset ack data. This prevents the ack_timer actually
557 * doing anything, even if it runs one more time when
558 * we drop the lock below.
560 memset(&cnx->pending_acks, 0xff, sizeof (cnx->pending_acks));
561 cnx->num_pending_acks = 0;
563 cnx->state &= ~(VETH_STATE_RESET | VETH_STATE_SENTMON
564 | VETH_STATE_OPEN | VETH_STATE_SENTCAPS
565 | VETH_STATE_GOTCAPACK | VETH_STATE_GOTCAPS
566 | VETH_STATE_SENTCAPACK | VETH_STATE_READY);
568 /* Clean up any leftover messages */
571 for (i = 0; i < VETH_NUMBUFFERS; ++i)
572 veth_recycle_msg(cnx, cnx->msgs + i);
575 cnx->outstanding_tx = 0;
576 veth_wake_queues(cnx);
578 /* Drop the lock so we can do stuff that might sleep or
579 * take other locks. */
580 spin_unlock_irq(&cnx->lock);
582 del_timer_sync(&cnx->ack_timer);
583 del_timer_sync(&cnx->reset_timer);
585 spin_lock_irq(&cnx->lock);
587 if (cnx->state & VETH_STATE_RESET)
590 /* Hack, wait for the other end to reset itself. */
591 if (! (cnx->state & VETH_STATE_SHUTDOWN)) {
592 schedule_delayed_work(&cnx->statemachine_wq, 5 * HZ);
597 if (cnx->state & VETH_STATE_SHUTDOWN)
598 /* It's all over, do nothing */
601 if ( !(cnx->state & VETH_STATE_OPEN) ) {
602 if (! cnx->msgs || (cnx->num_events < (2 + VETH_NUMBUFFERS)) )
605 HvCallEvent_openLpEventPath(rlp, HvLpEvent_Type_VirtualLan);
607 HvCallEvent_getSourceLpInstanceId(rlp,
608 HvLpEvent_Type_VirtualLan);
610 HvCallEvent_getTargetLpInstanceId(rlp,
611 HvLpEvent_Type_VirtualLan);
612 cnx->state |= VETH_STATE_OPEN;
615 if ( (cnx->state & VETH_STATE_OPEN)
616 && !(cnx->state & VETH_STATE_SENTMON) ) {
617 rc = veth_signalevent(cnx, VethEventTypeMonitor,
618 HvLpEvent_AckInd_DoAck,
619 HvLpEvent_AckType_DeferredAck,
622 if (rc == HvLpEvent_Rc_Good) {
623 cnx->state |= VETH_STATE_SENTMON;
625 if ( (rc != HvLpEvent_Rc_PartitionDead)
626 && (rc != HvLpEvent_Rc_PathClosed) )
627 veth_error("Error sending monitor to LPAR %d, "
628 "rc = %d\n", rlp, rc);
630 /* Oh well, hope we get a cap from the other
631 * end and do better when that kicks us */
636 if ( (cnx->state & VETH_STATE_OPEN)
637 && !(cnx->state & VETH_STATE_SENTCAPS)) {
638 u64 *rawcap = (u64 *)&cnx->local_caps;
640 rc = veth_signalevent(cnx, VethEventTypeCap,
641 HvLpEvent_AckInd_DoAck,
642 HvLpEvent_AckType_ImmediateAck,
643 0, rawcap[0], rawcap[1], rawcap[2],
644 rawcap[3], rawcap[4]);
646 if (rc == HvLpEvent_Rc_Good) {
647 cnx->state |= VETH_STATE_SENTCAPS;
649 if ( (rc != HvLpEvent_Rc_PartitionDead)
650 && (rc != HvLpEvent_Rc_PathClosed) )
651 veth_error("Error sending caps to LPAR %d, "
652 "rc = %d\n", rlp, rc);
654 /* Oh well, hope we get a cap from the other
655 * end and do better when that kicks us */
660 if ((cnx->state & VETH_STATE_GOTCAPS)
661 && !(cnx->state & VETH_STATE_SENTCAPACK)) {
662 struct VethCapData *remote_caps = &cnx->remote_caps;
664 memcpy(remote_caps, &cnx->cap_event.u.caps_data,
665 sizeof(*remote_caps));
667 spin_unlock_irq(&cnx->lock);
668 rc = veth_process_caps(cnx);
669 spin_lock_irq(&cnx->lock);
671 /* We dropped the lock, so recheck for anything which
672 * might mess us up */
673 if (cnx->state & (VETH_STATE_RESET|VETH_STATE_SHUTDOWN))
676 cnx->cap_event.base_event.xRc = rc;
677 HvCallEvent_ackLpEvent((struct HvLpEvent *)&cnx->cap_event);
678 if (rc == HvLpEvent_Rc_Good)
679 cnx->state |= VETH_STATE_SENTCAPACK;
684 if ((cnx->state & VETH_STATE_GOTCAPACK)
685 && (cnx->state & VETH_STATE_GOTCAPS)
686 && !(cnx->state & VETH_STATE_READY)) {
687 if (cnx->cap_ack_event.base_event.xRc == HvLpEvent_Rc_Good) {
688 /* Start the ACK timer */
689 cnx->ack_timer.expires = jiffies + cnx->ack_timeout;
690 add_timer(&cnx->ack_timer);
691 cnx->state |= VETH_STATE_READY;
693 veth_error("Caps rejected by LPAR %d, rc = %d\n",
694 rlp, cnx->cap_ack_event.base_event.xRc);
700 spin_unlock_irq(&cnx->lock);
704 /* FIXME: we get here if something happens we really can't
705 * cope with. The link will never work once we get here, and
706 * all we can do is not lock the rest of the system up */
707 veth_error("Unrecoverable error on connection to LPAR %d, shutting down"
708 " (state = 0x%04lx)\n", rlp, cnx->state);
709 cnx->state |= VETH_STATE_SHUTDOWN;
710 spin_unlock_irq(&cnx->lock);
713 static int veth_init_connection(u8 rlp)
715 struct veth_lpar_connection *cnx;
716 struct veth_msg *msgs;
719 if ( (rlp == this_lp)
720 || ! HvLpConfig_doLpsCommunicateOnVirtualLan(this_lp, rlp) )
723 cnx = kmalloc(sizeof(*cnx), GFP_KERNEL);
726 memset(cnx, 0, sizeof(*cnx));
728 cnx->remote_lp = rlp;
729 spin_lock_init(&cnx->lock);
730 INIT_WORK(&cnx->statemachine_wq, veth_statemachine, cnx);
732 init_timer(&cnx->ack_timer);
733 cnx->ack_timer.function = veth_timed_ack;
734 cnx->ack_timer.data = (unsigned long) cnx;
736 init_timer(&cnx->reset_timer);
737 cnx->reset_timer.function = veth_timed_reset;
738 cnx->reset_timer.data = (unsigned long) cnx;
739 cnx->reset_timeout = 5 * HZ * (VETH_ACKTIMEOUT / 1000000);
741 memset(&cnx->pending_acks, 0xff, sizeof (cnx->pending_acks));
745 /* This gets us 1 reference, which is held on behalf of the driver
746 * infrastructure. It's released at module unload. */
747 kobject_init(&cnx->kobject);
748 cnx->kobject.ktype = &veth_lpar_connection_ktype;
749 rc = kobject_set_name(&cnx->kobject, "cnx%.2d", rlp);
753 msgs = kmalloc(VETH_NUMBUFFERS * sizeof(struct veth_msg), GFP_KERNEL);
755 veth_error("Can't allocate buffers for LPAR %d.\n", rlp);
760 memset(msgs, 0, VETH_NUMBUFFERS * sizeof(struct veth_msg));
762 for (i = 0; i < VETH_NUMBUFFERS; i++) {
764 veth_stack_push(cnx, msgs + i);
767 cnx->num_events = veth_allocate_events(rlp, 2 + VETH_NUMBUFFERS);
769 if (cnx->num_events < (2 + VETH_NUMBUFFERS)) {
770 veth_error("Can't allocate enough events for LPAR %d.\n", rlp);
774 cnx->local_caps.num_buffers = VETH_NUMBUFFERS;
775 cnx->local_caps.ack_threshold = ACK_THRESHOLD;
776 cnx->local_caps.ack_timeout = VETH_ACKTIMEOUT;
781 static void veth_stop_connection(struct veth_lpar_connection *cnx)
786 spin_lock_irq(&cnx->lock);
787 cnx->state |= VETH_STATE_RESET | VETH_STATE_SHUTDOWN;
788 veth_kick_statemachine(cnx);
789 spin_unlock_irq(&cnx->lock);
791 /* There's a slim chance the reset code has just queued the
792 * statemachine to run in five seconds. If so we need to cancel
793 * that and requeue the work to run now. */
794 if (cancel_delayed_work(&cnx->statemachine_wq)) {
795 spin_lock_irq(&cnx->lock);
796 veth_kick_statemachine(cnx);
797 spin_unlock_irq(&cnx->lock);
800 /* Wait for the state machine to run. */
801 flush_scheduled_work();
804 static void veth_destroy_connection(struct veth_lpar_connection *cnx)
809 if (cnx->num_events > 0)
810 mf_deallocate_lp_events(cnx->remote_lp,
811 HvLpEvent_Type_VirtualLan,
814 if (cnx->num_ack_events > 0)
815 mf_deallocate_lp_events(cnx->remote_lp,
816 HvLpEvent_Type_VirtualLan,
821 veth_cnx[cnx->remote_lp] = NULL;
825 static void veth_release_connection(struct kobject *kobj)
827 struct veth_lpar_connection *cnx;
828 cnx = container_of(kobj, struct veth_lpar_connection, kobject);
829 veth_stop_connection(cnx);
830 veth_destroy_connection(cnx);
837 static int veth_open(struct net_device *dev)
839 struct veth_port *port = (struct veth_port *) dev->priv;
841 memset(&port->stats, 0, sizeof (port->stats));
842 netif_start_queue(dev);
846 static int veth_close(struct net_device *dev)
848 netif_stop_queue(dev);
852 static struct net_device_stats *veth_get_stats(struct net_device *dev)
854 struct veth_port *port = (struct veth_port *) dev->priv;
859 static int veth_change_mtu(struct net_device *dev, int new_mtu)
861 if ((new_mtu < 68) || (new_mtu > VETH_MAX_MTU))
867 static void veth_set_multicast_list(struct net_device *dev)
869 struct veth_port *port = (struct veth_port *) dev->priv;
872 write_lock_irqsave(&port->mcast_gate, flags);
874 if ((dev->flags & IFF_PROMISC) || (dev->flags & IFF_ALLMULTI) ||
875 (dev->mc_count > VETH_MAX_MCAST)) {
876 port->promiscuous = 1;
878 struct dev_mc_list *dmi = dev->mc_list;
881 port->promiscuous = 0;
886 for (i = 0; i < dev->mc_count; i++) {
887 u8 *addr = dmi->dmi_addr;
890 if (addr[0] & 0x01) {/* multicast address? */
891 memcpy(&xaddr, addr, ETH_ALEN);
892 port->mcast_addr[port->num_mcast] = xaddr;
899 write_unlock_irqrestore(&port->mcast_gate, flags);
902 static void veth_get_drvinfo(struct net_device *dev, struct ethtool_drvinfo *info)
904 strncpy(info->driver, "veth", sizeof(info->driver) - 1);
905 info->driver[sizeof(info->driver) - 1] = '\0';
906 strncpy(info->version, "1.0", sizeof(info->version) - 1);
909 static int veth_get_settings(struct net_device *dev, struct ethtool_cmd *ecmd)
911 ecmd->supported = (SUPPORTED_1000baseT_Full
912 | SUPPORTED_Autoneg | SUPPORTED_FIBRE);
913 ecmd->advertising = (SUPPORTED_1000baseT_Full
914 | SUPPORTED_Autoneg | SUPPORTED_FIBRE);
915 ecmd->port = PORT_FIBRE;
916 ecmd->transceiver = XCVR_INTERNAL;
917 ecmd->phy_address = 0;
918 ecmd->speed = SPEED_1000;
919 ecmd->duplex = DUPLEX_FULL;
920 ecmd->autoneg = AUTONEG_ENABLE;
921 ecmd->maxtxpkt = 120;
922 ecmd->maxrxpkt = 120;
926 static u32 veth_get_link(struct net_device *dev)
931 static struct ethtool_ops ops = {
932 .get_drvinfo = veth_get_drvinfo,
933 .get_settings = veth_get_settings,
934 .get_link = veth_get_link,
937 static struct net_device * __init veth_probe_one(int vlan, struct device *vdev)
939 struct net_device *dev;
940 struct veth_port *port;
943 dev = alloc_etherdev(sizeof (struct veth_port));
945 veth_error("Unable to allocate net_device structure!\n");
949 port = (struct veth_port *) dev->priv;
951 spin_lock_init(&port->queue_lock);
952 rwlock_init(&port->mcast_gate);
953 port->stopped_map = 0;
955 for (i = 0; i < HVMAXARCHITECTEDLPS; i++) {
956 HvLpVirtualLanIndexMap map;
960 map = HvLpConfig_getVirtualLanIndexMapForLp(i);
961 if (map & (0x8000 >> vlan))
962 port->lpar_map |= (1 << i);
966 dev->dev_addr[0] = 0x02;
967 dev->dev_addr[1] = 0x01;
968 dev->dev_addr[2] = 0xff;
969 dev->dev_addr[3] = vlan;
970 dev->dev_addr[4] = 0xff;
971 dev->dev_addr[5] = this_lp;
973 dev->mtu = VETH_MAX_MTU;
975 memcpy(&port->mac_addr, dev->dev_addr, 6);
977 dev->open = veth_open;
978 dev->hard_start_xmit = veth_start_xmit;
979 dev->stop = veth_close;
980 dev->get_stats = veth_get_stats;
981 dev->change_mtu = veth_change_mtu;
982 dev->set_mac_address = NULL;
983 dev->set_multicast_list = veth_set_multicast_list;
984 SET_ETHTOOL_OPS(dev, &ops);
986 SET_NETDEV_DEV(dev, vdev);
988 rc = register_netdev(dev);
990 veth_error("Failed registering net device for vlan%d.\n", vlan);
995 veth_info("%s attached to iSeries vlan %d (LPAR map = 0x%.4X)\n",
996 dev->name, vlan, port->lpar_map);
1005 static int veth_transmit_to_one(struct sk_buff *skb, HvLpIndex rlp,
1006 struct net_device *dev)
1008 struct veth_lpar_connection *cnx = veth_cnx[rlp];
1009 struct veth_port *port = (struct veth_port *) dev->priv;
1011 struct veth_msg *msg = NULL;
1012 unsigned long flags;
1017 spin_lock_irqsave(&cnx->lock, flags);
1019 if (! (cnx->state & VETH_STATE_READY))
1022 if ((skb->len - ETH_HLEN) > VETH_MAX_MTU)
1025 msg = veth_stack_pop(cnx);
1030 msg->skb = skb_get(skb);
1032 msg->data.addr[0] = dma_map_single(port->dev, skb->data,
1033 skb->len, DMA_TO_DEVICE);
1035 if (dma_mapping_error(msg->data.addr[0]))
1036 goto recycle_and_drop;
1038 msg->dev = port->dev;
1039 msg->data.len[0] = skb->len;
1040 msg->data.eofmask = 1 << VETH_EOF_SHIFT;
1042 rc = veth_signaldata(cnx, VethEventTypeFrames, msg->token, &msg->data);
1044 if (rc != HvLpEvent_Rc_Good)
1045 goto recycle_and_drop;
1047 /* If the timer's not already running, start it now. */
1048 if (0 == cnx->outstanding_tx)
1049 mod_timer(&cnx->reset_timer, jiffies + cnx->reset_timeout);
1051 cnx->last_contact = jiffies;
1052 cnx->outstanding_tx++;
1054 if (veth_stack_is_empty(cnx))
1055 veth_stop_queues(cnx);
1058 spin_unlock_irqrestore(&cnx->lock, flags);
1062 veth_recycle_msg(cnx, msg);
1064 spin_unlock_irqrestore(&cnx->lock, flags);
1068 static void veth_transmit_to_many(struct sk_buff *skb,
1069 HvLpIndexMap lpmask,
1070 struct net_device *dev)
1072 struct veth_port *port = (struct veth_port *) dev->priv;
1073 int i, success, error;
1075 success = error = 0;
1077 for (i = 0; i < HVMAXARCHITECTEDLPS; i++) {
1078 if ((lpmask & (1 << i)) == 0)
1081 if (veth_transmit_to_one(skb, i, dev))
1088 port->stats.tx_errors++;
1091 port->stats.tx_packets++;
1092 port->stats.tx_bytes += skb->len;
1096 static int veth_start_xmit(struct sk_buff *skb, struct net_device *dev)
1098 unsigned char *frame = skb->data;
1099 struct veth_port *port = (struct veth_port *) dev->priv;
1100 HvLpIndexMap lpmask;
1102 if (! (frame[0] & 0x01)) {
1103 /* unicast packet */
1104 HvLpIndex rlp = frame[5];
1106 if ( ! ((1 << rlp) & port->lpar_map) ) {
1113 lpmask = port->lpar_map;
1116 veth_transmit_to_many(skb, lpmask, dev);
1123 /* You must hold the connection's lock when you call this function. */
1124 static void veth_recycle_msg(struct veth_lpar_connection *cnx,
1125 struct veth_msg *msg)
1127 u32 dma_address, dma_length;
1131 dma_address = msg->data.addr[0];
1132 dma_length = msg->data.len[0];
1134 if (!dma_mapping_error(dma_address))
1135 dma_unmap_single(msg->dev, dma_address, dma_length,
1139 dev_kfree_skb_any(msg->skb);
1143 memset(&msg->data, 0, sizeof(msg->data));
1144 veth_stack_push(cnx, msg);
1145 } else if (cnx->state & VETH_STATE_OPEN) {
1146 veth_error("Non-pending frame (# %d) acked by LPAR %d.\n",
1147 cnx->remote_lp, msg->token);
1151 static void veth_wake_queues(struct veth_lpar_connection *cnx)
1155 for (i = 0; i < HVMAXARCHITECTEDVIRTUALLANS; i++) {
1156 struct net_device *dev = veth_dev[i];
1157 struct veth_port *port;
1158 unsigned long flags;
1163 port = (struct veth_port *)dev->priv;
1165 if (! (port->lpar_map & (1<<cnx->remote_lp)))
1168 spin_lock_irqsave(&port->queue_lock, flags);
1170 port->stopped_map &= ~(1 << cnx->remote_lp);
1172 if (0 == port->stopped_map && netif_queue_stopped(dev)) {
1173 veth_debug("cnx %d: woke queue for %s.\n",
1174 cnx->remote_lp, dev->name);
1175 netif_wake_queue(dev);
1177 spin_unlock_irqrestore(&port->queue_lock, flags);
1181 static void veth_stop_queues(struct veth_lpar_connection *cnx)
1185 for (i = 0; i < HVMAXARCHITECTEDVIRTUALLANS; i++) {
1186 struct net_device *dev = veth_dev[i];
1187 struct veth_port *port;
1192 port = (struct veth_port *)dev->priv;
1194 /* If this cnx is not on the vlan for this port, continue */
1195 if (! (port->lpar_map & (1 << cnx->remote_lp)))
1198 spin_lock(&port->queue_lock);
1200 netif_stop_queue(dev);
1201 port->stopped_map |= (1 << cnx->remote_lp);
1203 veth_debug("cnx %d: stopped queue for %s, map = 0x%x.\n",
1204 cnx->remote_lp, dev->name, port->stopped_map);
1206 spin_unlock(&port->queue_lock);
1210 static void veth_timed_reset(unsigned long ptr)
1212 struct veth_lpar_connection *cnx = (struct veth_lpar_connection *)ptr;
1213 unsigned long trigger_time, flags;
1215 /* FIXME is it possible this fires after veth_stop_connection()?
1216 * That would reschedule the statemachine for 5 seconds and probably
1217 * execute it after the module's been unloaded. Hmm. */
1219 spin_lock_irqsave(&cnx->lock, flags);
1221 if (cnx->outstanding_tx > 0) {
1222 trigger_time = cnx->last_contact + cnx->reset_timeout;
1224 if (trigger_time < jiffies) {
1225 cnx->state |= VETH_STATE_RESET;
1226 veth_kick_statemachine(cnx);
1227 veth_error("%d packets not acked by LPAR %d within %d "
1228 "seconds, resetting.\n",
1229 cnx->outstanding_tx, cnx->remote_lp,
1230 cnx->reset_timeout / HZ);
1232 /* Reschedule the timer */
1233 trigger_time = jiffies + cnx->reset_timeout;
1234 mod_timer(&cnx->reset_timer, trigger_time);
1238 spin_unlock_irqrestore(&cnx->lock, flags);
1245 static inline int veth_frame_wanted(struct veth_port *port, u64 mac_addr)
1249 unsigned long flags;
1251 if ( (mac_addr == port->mac_addr) || (mac_addr == 0xffffffffffff0000) )
1254 read_lock_irqsave(&port->mcast_gate, flags);
1256 if (port->promiscuous) {
1261 for (i = 0; i < port->num_mcast; ++i) {
1262 if (port->mcast_addr[i] == mac_addr) {
1269 read_unlock_irqrestore(&port->mcast_gate, flags);
1279 #define VETH_MAX_PAGES_PER_FRAME ( (VETH_MAX_MTU+PAGE_SIZE-2)/PAGE_SIZE + 1 )
1281 static inline void veth_build_dma_list(struct dma_chunk *list,
1282 unsigned char *p, unsigned long length)
1287 /* FIXME: skbs are continguous in real addresses. Do we
1288 * really need to break it into PAGE_SIZE chunks, or can we do
1289 * it just at the granularity of iSeries real->absolute
1290 * mapping? Indeed, given the way the allocator works, can we
1291 * count on them being absolutely contiguous? */
1292 list[0].addr = ISERIES_HV_ADDR(p);
1293 list[0].size = min(length,
1294 PAGE_SIZE - ((unsigned long)p & ~PAGE_MASK));
1296 done = list[0].size;
1297 while (done < length) {
1298 list[i].addr = ISERIES_HV_ADDR(p + done);
1299 list[i].size = min(length-done, PAGE_SIZE);
1300 done += list[i].size;
1305 static void veth_flush_acks(struct veth_lpar_connection *cnx)
1309 rc = veth_signaldata(cnx, VethEventTypeFramesAck,
1310 0, &cnx->pending_acks);
1312 if (rc != HvLpEvent_Rc_Good)
1313 veth_error("Failed acking frames from LPAR %d, rc = %d\n",
1314 cnx->remote_lp, (int)rc);
1316 cnx->num_pending_acks = 0;
1317 memset(&cnx->pending_acks, 0xff, sizeof(cnx->pending_acks));
1320 static void veth_receive(struct veth_lpar_connection *cnx,
1321 struct VethLpEvent *event)
1323 struct VethFramesData *senddata = &event->u.frames_data;
1326 unsigned long flags;
1331 struct sk_buff *skb;
1332 struct dma_chunk local_list[VETH_MAX_PAGES_PER_FRAME];
1333 struct dma_chunk remote_list[VETH_MAX_FRAMES_PER_MSG];
1335 HvLpVirtualLanIndex vlan;
1336 struct net_device *dev;
1337 struct veth_port *port;
1339 /* FIXME: do we need this? */
1340 memset(local_list, 0, sizeof(local_list));
1341 memset(remote_list, 0, sizeof(VETH_MAX_FRAMES_PER_MSG));
1343 /* a 0 address marks the end of the valid entries */
1344 if (senddata->addr[startchunk] == 0)
1347 /* make sure that we have at least 1 EOF entry in the
1348 * remaining entries */
1349 if (! (senddata->eofmask >> (startchunk + VETH_EOF_SHIFT))) {
1350 veth_error("Missing EOF fragment in event "
1351 "eofmask = 0x%x startchunk = %d\n",
1352 (unsigned)senddata->eofmask,
1357 /* build list of chunks in this frame */
1360 remote_list[nchunks].addr =
1361 (u64) senddata->addr[startchunk+nchunks] << 32;
1362 remote_list[nchunks].size =
1363 senddata->len[startchunk+nchunks];
1364 length += remote_list[nchunks].size;
1365 } while (! (senddata->eofmask &
1366 (1 << (VETH_EOF_SHIFT + startchunk + nchunks++))));
1368 /* length == total length of all chunks */
1369 /* nchunks == # of chunks in this frame */
1371 if ((length - ETH_HLEN) > VETH_MAX_MTU) {
1372 veth_error("Received oversize frame from LPAR %d "
1374 cnx->remote_lp, length);
1378 skb = alloc_skb(length, GFP_ATOMIC);
1382 veth_build_dma_list(local_list, skb->data, length);
1384 rc = HvCallEvent_dmaBufList(HvLpEvent_Type_VirtualLan,
1385 event->base_event.xSourceLp,
1386 HvLpDma_Direction_RemoteToLocal,
1389 HvLpDma_AddressType_RealAddress,
1390 HvLpDma_AddressType_TceIndex,
1391 ISERIES_HV_ADDR(&local_list),
1392 ISERIES_HV_ADDR(&remote_list),
1394 if (rc != HvLpDma_Rc_Good) {
1395 dev_kfree_skb_irq(skb);
1399 vlan = skb->data[9];
1400 dev = veth_dev[vlan];
1403 * Some earlier versions of the driver sent
1404 * broadcasts down all connections, even to lpars
1405 * that weren't on the relevant vlan. So ignore
1406 * packets belonging to a vlan we're not on.
1407 * We can also be here if we receive packets while
1408 * the driver is going down, because then dev is NULL.
1410 dev_kfree_skb_irq(skb);
1414 port = (struct veth_port *)dev->priv;
1415 dest = *((u64 *) skb->data) & 0xFFFFFFFFFFFF0000;
1417 if ((vlan > HVMAXARCHITECTEDVIRTUALLANS) || !port) {
1418 dev_kfree_skb_irq(skb);
1421 if (! veth_frame_wanted(port, dest)) {
1422 dev_kfree_skb_irq(skb);
1426 skb_put(skb, length);
1428 skb->protocol = eth_type_trans(skb, dev);
1429 skb->ip_summed = CHECKSUM_NONE;
1430 netif_rx(skb); /* send it up */
1431 port->stats.rx_packets++;
1432 port->stats.rx_bytes += length;
1433 } while (startchunk += nchunks, startchunk < VETH_MAX_FRAMES_PER_MSG);
1436 spin_lock_irqsave(&cnx->lock, flags);
1437 BUG_ON(cnx->num_pending_acks > VETH_MAX_ACKS_PER_MSG);
1439 cnx->pending_acks[cnx->num_pending_acks++] =
1440 event->base_event.xCorrelationToken;
1442 if ( (cnx->num_pending_acks >= cnx->remote_caps.ack_threshold)
1443 || (cnx->num_pending_acks >= VETH_MAX_ACKS_PER_MSG) )
1444 veth_flush_acks(cnx);
1446 spin_unlock_irqrestore(&cnx->lock, flags);
1449 static void veth_timed_ack(unsigned long ptr)
1451 struct veth_lpar_connection *cnx = (struct veth_lpar_connection *) ptr;
1452 unsigned long flags;
1454 /* Ack all the events */
1455 spin_lock_irqsave(&cnx->lock, flags);
1456 if (cnx->num_pending_acks > 0)
1457 veth_flush_acks(cnx);
1459 /* Reschedule the timer */
1460 cnx->ack_timer.expires = jiffies + cnx->ack_timeout;
1461 add_timer(&cnx->ack_timer);
1462 spin_unlock_irqrestore(&cnx->lock, flags);
1465 static int veth_remove(struct vio_dev *vdev)
1467 struct veth_lpar_connection *cnx;
1468 struct net_device *dev;
1469 struct veth_port *port;
1472 dev = veth_dev[vdev->unit_address];
1477 port = netdev_priv(dev);
1479 for (i = 0; i < HVMAXARCHITECTEDLPS; i++) {
1482 if (cnx && (port->lpar_map & (1 << i))) {
1483 /* Drop our reference to connections on our VLAN */
1484 kobject_put(&cnx->kobject);
1488 veth_dev[vdev->unit_address] = NULL;
1489 unregister_netdev(dev);
1495 static int veth_probe(struct vio_dev *vdev, const struct vio_device_id *id)
1497 int i = vdev->unit_address;
1498 struct net_device *dev;
1499 struct veth_port *port;
1501 dev = veth_probe_one(i, &vdev->dev);
1508 port = (struct veth_port*)netdev_priv(dev);
1510 /* Start the state machine on each connection on this vlan. If we're
1511 * the first dev to do so this will commence link negotiation */
1512 for (i = 0; i < HVMAXARCHITECTEDLPS; i++) {
1513 struct veth_lpar_connection *cnx;
1515 if (! (port->lpar_map & (1 << i)))
1522 kobject_get(&cnx->kobject);
1523 veth_kick_statemachine(cnx);
1530 * veth_device_table: Used by vio.c to match devices that we
1533 static struct vio_device_id veth_device_table[] __devinitdata = {
1537 MODULE_DEVICE_TABLE(vio, veth_device_table);
1539 static struct vio_driver veth_driver = {
1540 .name = "iseries_veth",
1541 .id_table = veth_device_table,
1542 .probe = veth_probe,
1543 .remove = veth_remove
1547 * Module initialization/cleanup
1550 void __exit veth_module_cleanup(void)
1553 struct veth_lpar_connection *cnx;
1555 /* Disconnect our "irq" to stop events coming from the Hypervisor. */
1556 HvLpEvent_unregisterHandler(HvLpEvent_Type_VirtualLan);
1558 /* Make sure any work queued from Hypervisor callbacks is finished. */
1559 flush_scheduled_work();
1561 for (i = 0; i < HVMAXARCHITECTEDLPS; ++i) {
1567 /* Remove the connection from sysfs */
1568 kobject_del(&cnx->kobject);
1569 /* Drop the driver's reference to the connection */
1570 kobject_put(&cnx->kobject);
1573 /* Unregister the driver, which will close all the netdevs and stop
1574 * the connections when they're no longer referenced. */
1575 vio_unregister_driver(&veth_driver);
1577 module_exit(veth_module_cleanup);
1579 int __init veth_module_init(void)
1584 this_lp = HvLpConfig_getLpIndex_outline();
1586 for (i = 0; i < HVMAXARCHITECTEDLPS; ++i) {
1587 rc = veth_init_connection(i);
1592 HvLpEvent_registerHandler(HvLpEvent_Type_VirtualLan,
1593 &veth_handle_event);
1595 rc = vio_register_driver(&veth_driver);
1599 for (i = 0; i < HVMAXARCHITECTEDLPS; ++i) {
1600 struct kobject *kobj;
1605 kobj = &veth_cnx[i]->kobject;
1606 kobj->parent = &veth_driver.driver.kobj;
1607 /* If the add failes, complain but otherwise continue */
1608 if (0 != kobject_add(kobj))
1609 veth_error("cnx %d: Failed adding to sysfs.\n", i);
1615 for (i = 0; i < HVMAXARCHITECTEDLPS; ++i) {
1616 veth_destroy_connection(veth_cnx[i]);
1621 module_init(veth_module_init);