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 del_timer(&cnx
->ack_timer
);
455 if (cnx
->state
& VETH_STATE_OPEN
)
456 HvCallEvent_closeLpEventPath(cnx
->remote_lp
,
457 HvLpEvent_Type_VirtualLan
);
460 memset(&cnx
->pending_acks
, 0xff, sizeof (cnx
->pending_acks
));
461 cnx
->num_pending_acks
= 0;
463 cnx
->state
&= ~(VETH_STATE_RESET
| VETH_STATE_SENTMON
464 | VETH_STATE_OPEN
| VETH_STATE_SENTCAPS
465 | VETH_STATE_GOTCAPACK
| VETH_STATE_GOTCAPS
466 | VETH_STATE_SENTCAPACK
| VETH_STATE_READY
);
468 /* Clean up any leftover messages */
470 for (i
= 0; i
< VETH_NUMBUFFERS
; ++i
)
471 veth_recycle_msg(cnx
, cnx
->msgs
+ i
);
472 spin_unlock_irq(&cnx
->lock
);
473 veth_flush_pending(cnx
);
474 spin_lock_irq(&cnx
->lock
);
475 if (cnx
->state
& VETH_STATE_RESET
)
479 if (cnx
->state
& VETH_STATE_SHUTDOWN
)
480 /* It's all over, do nothing */
483 if ( !(cnx
->state
& VETH_STATE_OPEN
) ) {
484 if (! cnx
->msgs
|| (cnx
->num_events
< (2 + VETH_NUMBUFFERS
)) )
487 HvCallEvent_openLpEventPath(rlp
, HvLpEvent_Type_VirtualLan
);
489 HvCallEvent_getSourceLpInstanceId(rlp
,
490 HvLpEvent_Type_VirtualLan
);
492 HvCallEvent_getTargetLpInstanceId(rlp
,
493 HvLpEvent_Type_VirtualLan
);
494 cnx
->state
|= VETH_STATE_OPEN
;
497 if ( (cnx
->state
& VETH_STATE_OPEN
)
498 && !(cnx
->state
& VETH_STATE_SENTMON
) ) {
499 rc
= veth_signalevent(cnx
, VethEventTypeMonitor
,
500 HvLpEvent_AckInd_DoAck
,
501 HvLpEvent_AckType_DeferredAck
,
504 if (rc
== HvLpEvent_Rc_Good
) {
505 cnx
->state
|= VETH_STATE_SENTMON
;
507 if ( (rc
!= HvLpEvent_Rc_PartitionDead
)
508 && (rc
!= HvLpEvent_Rc_PathClosed
) )
509 veth_error("Error sending monitor to LPAR %d, "
510 "rc = %d\n", rlp
, rc
);
512 /* Oh well, hope we get a cap from the other
513 * end and do better when that kicks us */
518 if ( (cnx
->state
& VETH_STATE_OPEN
)
519 && !(cnx
->state
& VETH_STATE_SENTCAPS
)) {
520 u64
*rawcap
= (u64
*)&cnx
->local_caps
;
522 rc
= veth_signalevent(cnx
, VethEventTypeCap
,
523 HvLpEvent_AckInd_DoAck
,
524 HvLpEvent_AckType_ImmediateAck
,
525 0, rawcap
[0], rawcap
[1], rawcap
[2],
526 rawcap
[3], rawcap
[4]);
528 if (rc
== HvLpEvent_Rc_Good
) {
529 cnx
->state
|= VETH_STATE_SENTCAPS
;
531 if ( (rc
!= HvLpEvent_Rc_PartitionDead
)
532 && (rc
!= HvLpEvent_Rc_PathClosed
) )
533 veth_error("Error sending caps to LPAR %d, "
534 "rc = %d\n", rlp
, rc
);
536 /* Oh well, hope we get a cap from the other
537 * end and do better when that kicks us */
542 if ((cnx
->state
& VETH_STATE_GOTCAPS
)
543 && !(cnx
->state
& VETH_STATE_SENTCAPACK
)) {
544 struct VethCapData
*remote_caps
= &cnx
->remote_caps
;
546 memcpy(remote_caps
, &cnx
->cap_event
.u
.caps_data
,
547 sizeof(*remote_caps
));
549 spin_unlock_irq(&cnx
->lock
);
550 rc
= veth_process_caps(cnx
);
551 spin_lock_irq(&cnx
->lock
);
553 /* We dropped the lock, so recheck for anything which
554 * might mess us up */
555 if (cnx
->state
& (VETH_STATE_RESET
|VETH_STATE_SHUTDOWN
))
558 cnx
->cap_event
.base_event
.xRc
= rc
;
559 HvCallEvent_ackLpEvent((struct HvLpEvent
*)&cnx
->cap_event
);
560 if (rc
== HvLpEvent_Rc_Good
)
561 cnx
->state
|= VETH_STATE_SENTCAPACK
;
566 if ((cnx
->state
& VETH_STATE_GOTCAPACK
)
567 && (cnx
->state
& VETH_STATE_GOTCAPS
)
568 && !(cnx
->state
& VETH_STATE_READY
)) {
569 if (cnx
->cap_ack_event
.base_event
.xRc
== HvLpEvent_Rc_Good
) {
570 /* Start the ACK timer */
571 cnx
->ack_timer
.expires
= jiffies
+ cnx
->ack_timeout
;
572 add_timer(&cnx
->ack_timer
);
573 cnx
->state
|= VETH_STATE_READY
;
575 veth_error("Caps rejected by LPAR %d, rc = %d\n",
576 rlp
, cnx
->cap_ack_event
.base_event
.xRc
);
582 spin_unlock_irq(&cnx
->lock
);
586 /* FIXME: we get here if something happens we really can't
587 * cope with. The link will never work once we get here, and
588 * all we can do is not lock the rest of the system up */
589 veth_error("Unrecoverable error on connection to LPAR %d, shutting down"
590 " (state = 0x%04lx)\n", rlp
, cnx
->state
);
591 cnx
->state
|= VETH_STATE_SHUTDOWN
;
592 spin_unlock_irq(&cnx
->lock
);
595 static int veth_init_connection(u8 rlp
)
597 struct veth_lpar_connection
*cnx
;
598 struct veth_msg
*msgs
;
601 if ( (rlp
== this_lp
)
602 || ! HvLpConfig_doLpsCommunicateOnVirtualLan(this_lp
, rlp
) )
605 cnx
= kmalloc(sizeof(*cnx
), GFP_KERNEL
);
608 memset(cnx
, 0, sizeof(*cnx
));
610 cnx
->remote_lp
= rlp
;
611 spin_lock_init(&cnx
->lock
);
612 INIT_WORK(&cnx
->statemachine_wq
, veth_statemachine
, cnx
);
613 init_timer(&cnx
->ack_timer
);
614 cnx
->ack_timer
.function
= veth_timed_ack
;
615 cnx
->ack_timer
.data
= (unsigned long) cnx
;
616 memset(&cnx
->pending_acks
, 0xff, sizeof (cnx
->pending_acks
));
620 msgs
= kmalloc(VETH_NUMBUFFERS
* sizeof(struct veth_msg
), GFP_KERNEL
);
622 veth_error("Can't allocate buffers for LPAR %d.\n", rlp
);
627 memset(msgs
, 0, VETH_NUMBUFFERS
* sizeof(struct veth_msg
));
628 spin_lock_init(&cnx
->msg_stack_lock
);
630 for (i
= 0; i
< VETH_NUMBUFFERS
; i
++) {
632 veth_stack_push(cnx
, msgs
+ i
);
635 cnx
->num_events
= veth_allocate_events(rlp
, 2 + VETH_NUMBUFFERS
);
637 if (cnx
->num_events
< (2 + VETH_NUMBUFFERS
)) {
638 veth_error("Can't allocate enough events for LPAR %d.\n", rlp
);
642 cnx
->local_caps
.num_buffers
= VETH_NUMBUFFERS
;
643 cnx
->local_caps
.ack_threshold
= ACK_THRESHOLD
;
644 cnx
->local_caps
.ack_timeout
= VETH_ACKTIMEOUT
;
649 static void veth_stop_connection(u8 rlp
)
651 struct veth_lpar_connection
*cnx
= veth_cnx
[rlp
];
656 spin_lock_irq(&cnx
->lock
);
657 cnx
->state
|= VETH_STATE_RESET
| VETH_STATE_SHUTDOWN
;
658 veth_kick_statemachine(cnx
);
659 spin_unlock_irq(&cnx
->lock
);
661 flush_scheduled_work();
663 /* FIXME: not sure if this is necessary - will already have
664 * been deleted by the state machine, just want to make sure
665 * its not running any more */
666 del_timer_sync(&cnx
->ack_timer
);
668 if (cnx
->num_events
> 0)
669 mf_deallocate_lp_events(cnx
->remote_lp
,
670 HvLpEvent_Type_VirtualLan
,
673 if (cnx
->num_ack_events
> 0)
674 mf_deallocate_lp_events(cnx
->remote_lp
,
675 HvLpEvent_Type_VirtualLan
,
680 static void veth_destroy_connection(u8 rlp
)
682 struct veth_lpar_connection
*cnx
= veth_cnx
[rlp
];
689 veth_cnx
[rlp
] = NULL
;
696 static int veth_open(struct net_device
*dev
)
698 struct veth_port
*port
= (struct veth_port
*) dev
->priv
;
700 memset(&port
->stats
, 0, sizeof (port
->stats
));
701 netif_start_queue(dev
);
705 static int veth_close(struct net_device
*dev
)
707 netif_stop_queue(dev
);
711 static struct net_device_stats
*veth_get_stats(struct net_device
*dev
)
713 struct veth_port
*port
= (struct veth_port
*) dev
->priv
;
718 static int veth_change_mtu(struct net_device
*dev
, int new_mtu
)
720 if ((new_mtu
< 68) || (new_mtu
> VETH_MAX_MTU
))
726 static void veth_set_multicast_list(struct net_device
*dev
)
728 struct veth_port
*port
= (struct veth_port
*) dev
->priv
;
731 write_lock_irqsave(&port
->mcast_gate
, flags
);
733 if (dev
->flags
& IFF_PROMISC
) { /* set promiscuous mode */
734 printk(KERN_INFO
"%s: Promiscuous mode enabled.\n",
736 port
->promiscuous
= 1;
737 } else if ( (dev
->flags
& IFF_ALLMULTI
)
738 || (dev
->mc_count
> VETH_MAX_MCAST
) ) {
741 struct dev_mc_list
*dmi
= dev
->mc_list
;
747 for (i
= 0; i
< dev
->mc_count
; i
++) {
748 u8
*addr
= dmi
->dmi_addr
;
751 if (addr
[0] & 0x01) {/* multicast address? */
752 memcpy(&xaddr
, addr
, ETH_ALEN
);
753 port
->mcast_addr
[port
->num_mcast
] = xaddr
;
760 write_unlock_irqrestore(&port
->mcast_gate
, flags
);
763 static void veth_get_drvinfo(struct net_device
*dev
, struct ethtool_drvinfo
*info
)
765 strncpy(info
->driver
, "veth", sizeof(info
->driver
) - 1);
766 info
->driver
[sizeof(info
->driver
) - 1] = '\0';
767 strncpy(info
->version
, "1.0", sizeof(info
->version
) - 1);
770 static int veth_get_settings(struct net_device
*dev
, struct ethtool_cmd
*ecmd
)
772 ecmd
->supported
= (SUPPORTED_1000baseT_Full
773 | SUPPORTED_Autoneg
| SUPPORTED_FIBRE
);
774 ecmd
->advertising
= (SUPPORTED_1000baseT_Full
775 | SUPPORTED_Autoneg
| SUPPORTED_FIBRE
);
776 ecmd
->port
= PORT_FIBRE
;
777 ecmd
->transceiver
= XCVR_INTERNAL
;
778 ecmd
->phy_address
= 0;
779 ecmd
->speed
= SPEED_1000
;
780 ecmd
->duplex
= DUPLEX_FULL
;
781 ecmd
->autoneg
= AUTONEG_ENABLE
;
782 ecmd
->maxtxpkt
= 120;
783 ecmd
->maxrxpkt
= 120;
787 static u32
veth_get_link(struct net_device
*dev
)
792 static struct ethtool_ops ops
= {
793 .get_drvinfo
= veth_get_drvinfo
,
794 .get_settings
= veth_get_settings
,
795 .get_link
= veth_get_link
,
798 static void veth_tx_timeout(struct net_device
*dev
)
800 struct veth_port
*port
= (struct veth_port
*)dev
->priv
;
801 struct net_device_stats
*stats
= &port
->stats
;
807 spin_lock_irqsave(&port
->pending_gate
, flags
);
809 if (!port
->pending_lpmask
) {
810 spin_unlock_irqrestore(&port
->pending_gate
, flags
);
814 printk(KERN_WARNING
"%s: Tx timeout! Resetting lp connections: %08x\n",
815 dev
->name
, port
->pending_lpmask
);
817 for (i
= 0; i
< HVMAXARCHITECTEDLPS
; i
++) {
818 struct veth_lpar_connection
*cnx
= veth_cnx
[i
];
820 if (! (port
->pending_lpmask
& (1<<i
)))
823 /* If we're pending on it, we must be connected to it,
824 * so we should certainly have a structure for it. */
827 /* Theoretically we could be kicking a connection
828 * which doesn't deserve it, but in practice if we've
829 * had a Tx timeout, the pending_lpmask will have
830 * exactly one bit set - the connection causing the
832 spin_lock(&cnx
->lock
);
833 cnx
->state
|= VETH_STATE_RESET
;
834 veth_kick_statemachine(cnx
);
835 spin_unlock(&cnx
->lock
);
838 spin_unlock_irqrestore(&port
->pending_gate
, flags
);
841 static struct net_device
* __init
veth_probe_one(int vlan
, struct device
*vdev
)
843 struct net_device
*dev
;
844 struct veth_port
*port
;
847 dev
= alloc_etherdev(sizeof (struct veth_port
));
849 veth_error("Unable to allocate net_device structure!\n");
853 port
= (struct veth_port
*) dev
->priv
;
855 spin_lock_init(&port
->pending_gate
);
856 rwlock_init(&port
->mcast_gate
);
858 for (i
= 0; i
< HVMAXARCHITECTEDLPS
; i
++) {
859 HvLpVirtualLanIndexMap map
;
863 map
= HvLpConfig_getVirtualLanIndexMapForLp(i
);
864 if (map
& (0x8000 >> vlan
))
865 port
->lpar_map
|= (1 << i
);
869 dev
->dev_addr
[0] = 0x02;
870 dev
->dev_addr
[1] = 0x01;
871 dev
->dev_addr
[2] = 0xff;
872 dev
->dev_addr
[3] = vlan
;
873 dev
->dev_addr
[4] = 0xff;
874 dev
->dev_addr
[5] = this_lp
;
876 dev
->mtu
= VETH_MAX_MTU
;
878 memcpy(&port
->mac_addr
, dev
->dev_addr
, 6);
880 dev
->open
= veth_open
;
881 dev
->hard_start_xmit
= veth_start_xmit
;
882 dev
->stop
= veth_close
;
883 dev
->get_stats
= veth_get_stats
;
884 dev
->change_mtu
= veth_change_mtu
;
885 dev
->set_mac_address
= NULL
;
886 dev
->set_multicast_list
= veth_set_multicast_list
;
887 SET_ETHTOOL_OPS(dev
, &ops
);
889 dev
->watchdog_timeo
= 2 * (VETH_ACKTIMEOUT
* HZ
/ 1000000);
890 dev
->tx_timeout
= veth_tx_timeout
;
892 SET_NETDEV_DEV(dev
, vdev
);
894 rc
= register_netdev(dev
);
896 veth_error("Failed registering net device for vlan%d.\n", vlan
);
901 veth_info("%s attached to iSeries vlan %d (LPAR map = 0x%.4X)\n",
902 dev
->name
, vlan
, port
->lpar_map
);
911 static int veth_transmit_to_one(struct sk_buff
*skb
, HvLpIndex rlp
,
912 struct net_device
*dev
)
914 struct veth_lpar_connection
*cnx
= veth_cnx
[rlp
];
915 struct veth_port
*port
= (struct veth_port
*) dev
->priv
;
917 u32 dma_address
, dma_length
;
918 struct veth_msg
*msg
= NULL
;
923 port
->stats
.tx_errors
++;
928 spin_lock_irqsave(&cnx
->lock
, flags
);
930 if (! (cnx
->state
& VETH_STATE_READY
))
933 if ((skb
->len
- 14) > VETH_MAX_MTU
)
936 msg
= veth_stack_pop(cnx
);
943 dma_length
= skb
->len
;
944 dma_address
= dma_map_single(port
->dev
, skb
->data
,
945 dma_length
, DMA_TO_DEVICE
);
947 if (dma_mapping_error(dma_address
))
948 goto recycle_and_drop
;
950 /* Is it really necessary to check the length and address
951 * fields of the first entry here? */
953 msg
->dev
= port
->dev
;
954 msg
->data
.addr
[0] = dma_address
;
955 msg
->data
.len
[0] = dma_length
;
956 msg
->data
.eofmask
= 1 << VETH_EOF_SHIFT
;
957 set_bit(0, &(msg
->in_use
));
958 rc
= veth_signaldata(cnx
, VethEventTypeFrames
, msg
->token
, &msg
->data
);
960 if (rc
!= HvLpEvent_Rc_Good
)
961 goto recycle_and_drop
;
963 spin_unlock_irqrestore(&cnx
->lock
, flags
);
968 /* need to set in use to make veth_recycle_msg in case this
969 * was a mapping failure */
970 set_bit(0, &msg
->in_use
);
971 veth_recycle_msg(cnx
, msg
);
973 port
->stats
.tx_errors
++;
975 spin_unlock_irqrestore(&cnx
->lock
, flags
);
979 static HvLpIndexMap
veth_transmit_to_many(struct sk_buff
*skb
,
981 struct net_device
*dev
)
983 struct veth_port
*port
= (struct veth_port
*) dev
->priv
;
987 for (i
= 0; i
< HVMAXARCHITECTEDLPS
; i
++) {
988 if ((lpmask
& (1 << i
)) == 0)
991 rc
= veth_transmit_to_one(skb_get(skb
), i
, dev
);
997 port
->stats
.tx_packets
++;
998 port
->stats
.tx_bytes
+= skb
->len
;
1004 static int veth_start_xmit(struct sk_buff
*skb
, struct net_device
*dev
)
1006 unsigned char *frame
= skb
->data
;
1007 struct veth_port
*port
= (struct veth_port
*) dev
->priv
;
1008 unsigned long flags
;
1009 HvLpIndexMap lpmask
;
1011 if (! (frame
[0] & 0x01)) {
1012 /* unicast packet */
1013 HvLpIndex rlp
= frame
[5];
1015 if ( ! ((1 << rlp
) & port
->lpar_map
) ) {
1022 lpmask
= port
->lpar_map
;
1025 spin_lock_irqsave(&port
->pending_gate
, flags
);
1027 lpmask
= veth_transmit_to_many(skb
, lpmask
, dev
);
1029 dev
->trans_start
= jiffies
;
1034 if (port
->pending_skb
) {
1035 veth_error("%s: TX while skb was pending!\n",
1038 spin_unlock_irqrestore(&port
->pending_gate
, flags
);
1042 port
->pending_skb
= skb
;
1043 port
->pending_lpmask
= lpmask
;
1044 netif_stop_queue(dev
);
1047 spin_unlock_irqrestore(&port
->pending_gate
, flags
);
1052 static void veth_recycle_msg(struct veth_lpar_connection
*cnx
,
1053 struct veth_msg
*msg
)
1055 u32 dma_address
, dma_length
;
1057 if (test_and_clear_bit(0, &msg
->in_use
)) {
1058 dma_address
= msg
->data
.addr
[0];
1059 dma_length
= msg
->data
.len
[0];
1061 dma_unmap_single(msg
->dev
, dma_address
, dma_length
,
1065 dev_kfree_skb_any(msg
->skb
);
1069 memset(&msg
->data
, 0, sizeof(msg
->data
));
1070 veth_stack_push(cnx
, msg
);
1071 } else if (cnx
->state
& VETH_STATE_OPEN
) {
1072 veth_error("Non-pending frame (# %d) acked by LPAR %d.\n",
1073 cnx
->remote_lp
, msg
->token
);
1077 static void veth_flush_pending(struct veth_lpar_connection
*cnx
)
1080 for (i
= 0; i
< HVMAXARCHITECTEDVIRTUALLANS
; i
++) {
1081 struct net_device
*dev
= veth_dev
[i
];
1082 struct veth_port
*port
;
1083 unsigned long flags
;
1088 port
= (struct veth_port
*)dev
->priv
;
1090 if (! (port
->lpar_map
& (1<<cnx
->remote_lp
)))
1093 spin_lock_irqsave(&port
->pending_gate
, flags
);
1094 if (port
->pending_skb
) {
1095 port
->pending_lpmask
=
1096 veth_transmit_to_many(port
->pending_skb
,
1097 port
->pending_lpmask
,
1099 if (! port
->pending_lpmask
) {
1100 dev_kfree_skb_any(port
->pending_skb
);
1101 port
->pending_skb
= NULL
;
1102 netif_wake_queue(dev
);
1105 spin_unlock_irqrestore(&port
->pending_gate
, flags
);
1113 static inline int veth_frame_wanted(struct veth_port
*port
, u64 mac_addr
)
1117 unsigned long flags
;
1119 if ( (mac_addr
== port
->mac_addr
) || (mac_addr
== 0xffffffffffff0000) )
1122 if (! (((char *) &mac_addr
)[0] & 0x01))
1125 read_lock_irqsave(&port
->mcast_gate
, flags
);
1127 if (port
->promiscuous
|| port
->all_mcast
) {
1132 for (i
= 0; i
< port
->num_mcast
; ++i
) {
1133 if (port
->mcast_addr
[i
] == mac_addr
) {
1140 read_unlock_irqrestore(&port
->mcast_gate
, flags
);
1150 #define VETH_MAX_PAGES_PER_FRAME ( (VETH_MAX_MTU+PAGE_SIZE-2)/PAGE_SIZE + 1 )
1152 static inline void veth_build_dma_list(struct dma_chunk
*list
,
1153 unsigned char *p
, unsigned long length
)
1158 /* FIXME: skbs are continguous in real addresses. Do we
1159 * really need to break it into PAGE_SIZE chunks, or can we do
1160 * it just at the granularity of iSeries real->absolute
1161 * mapping? Indeed, given the way the allocator works, can we
1162 * count on them being absolutely contiguous? */
1163 list
[0].addr
= ISERIES_HV_ADDR(p
);
1164 list
[0].size
= min(length
,
1165 PAGE_SIZE
- ((unsigned long)p
& ~PAGE_MASK
));
1167 done
= list
[0].size
;
1168 while (done
< length
) {
1169 list
[i
].addr
= ISERIES_HV_ADDR(p
+ done
);
1170 list
[i
].size
= min(length
-done
, PAGE_SIZE
);
1171 done
+= list
[i
].size
;
1176 static void veth_flush_acks(struct veth_lpar_connection
*cnx
)
1180 rc
= veth_signaldata(cnx
, VethEventTypeFramesAck
,
1181 0, &cnx
->pending_acks
);
1183 if (rc
!= HvLpEvent_Rc_Good
)
1184 veth_error("Failed acking frames from LPAR %d, rc = %d\n",
1185 cnx
->remote_lp
, (int)rc
);
1187 cnx
->num_pending_acks
= 0;
1188 memset(&cnx
->pending_acks
, 0xff, sizeof(cnx
->pending_acks
));
1191 static void veth_receive(struct veth_lpar_connection
*cnx
,
1192 struct VethLpEvent
*event
)
1194 struct VethFramesData
*senddata
= &event
->u
.frames_data
;
1197 unsigned long flags
;
1202 struct sk_buff
*skb
;
1203 struct dma_chunk local_list
[VETH_MAX_PAGES_PER_FRAME
];
1204 struct dma_chunk remote_list
[VETH_MAX_FRAMES_PER_MSG
];
1206 HvLpVirtualLanIndex vlan
;
1207 struct net_device
*dev
;
1208 struct veth_port
*port
;
1210 /* FIXME: do we need this? */
1211 memset(local_list
, 0, sizeof(local_list
));
1212 memset(remote_list
, 0, sizeof(VETH_MAX_FRAMES_PER_MSG
));
1214 /* a 0 address marks the end of the valid entries */
1215 if (senddata
->addr
[startchunk
] == 0)
1218 /* make sure that we have at least 1 EOF entry in the
1219 * remaining entries */
1220 if (! (senddata
->eofmask
>> (startchunk
+ VETH_EOF_SHIFT
))) {
1221 veth_error("Missing EOF fragment in event "
1222 "eofmask = 0x%x startchunk = %d\n",
1223 (unsigned)senddata
->eofmask
,
1228 /* build list of chunks in this frame */
1231 remote_list
[nchunks
].addr
=
1232 (u64
) senddata
->addr
[startchunk
+nchunks
] << 32;
1233 remote_list
[nchunks
].size
=
1234 senddata
->len
[startchunk
+nchunks
];
1235 length
+= remote_list
[nchunks
].size
;
1236 } while (! (senddata
->eofmask
&
1237 (1 << (VETH_EOF_SHIFT
+ startchunk
+ nchunks
++))));
1239 /* length == total length of all chunks */
1240 /* nchunks == # of chunks in this frame */
1242 if ((length
- ETH_HLEN
) > VETH_MAX_MTU
) {
1243 veth_error("Received oversize frame from LPAR %d "
1245 cnx
->remote_lp
, length
);
1249 skb
= alloc_skb(length
, GFP_ATOMIC
);
1253 veth_build_dma_list(local_list
, skb
->data
, length
);
1255 rc
= HvCallEvent_dmaBufList(HvLpEvent_Type_VirtualLan
,
1256 event
->base_event
.xSourceLp
,
1257 HvLpDma_Direction_RemoteToLocal
,
1260 HvLpDma_AddressType_RealAddress
,
1261 HvLpDma_AddressType_TceIndex
,
1262 ISERIES_HV_ADDR(&local_list
),
1263 ISERIES_HV_ADDR(&remote_list
),
1265 if (rc
!= HvLpDma_Rc_Good
) {
1266 dev_kfree_skb_irq(skb
);
1270 vlan
= skb
->data
[9];
1271 dev
= veth_dev
[vlan
];
1274 * Some earlier versions of the driver sent
1275 * broadcasts down all connections, even to lpars
1276 * that weren't on the relevant vlan. So ignore
1277 * packets belonging to a vlan we're not on.
1278 * We can also be here if we receive packets while
1279 * the driver is going down, because then dev is NULL.
1281 dev_kfree_skb_irq(skb
);
1285 port
= (struct veth_port
*)dev
->priv
;
1286 dest
= *((u64
*) skb
->data
) & 0xFFFFFFFFFFFF0000;
1288 if ((vlan
> HVMAXARCHITECTEDVIRTUALLANS
) || !port
) {
1289 dev_kfree_skb_irq(skb
);
1292 if (! veth_frame_wanted(port
, dest
)) {
1293 dev_kfree_skb_irq(skb
);
1297 skb_put(skb
, length
);
1299 skb
->protocol
= eth_type_trans(skb
, dev
);
1300 skb
->ip_summed
= CHECKSUM_NONE
;
1301 netif_rx(skb
); /* send it up */
1302 port
->stats
.rx_packets
++;
1303 port
->stats
.rx_bytes
+= length
;
1304 } while (startchunk
+= nchunks
, startchunk
< VETH_MAX_FRAMES_PER_MSG
);
1307 spin_lock_irqsave(&cnx
->lock
, flags
);
1308 BUG_ON(cnx
->num_pending_acks
> VETH_MAX_ACKS_PER_MSG
);
1310 cnx
->pending_acks
[cnx
->num_pending_acks
++] =
1311 event
->base_event
.xCorrelationToken
;
1313 if ( (cnx
->num_pending_acks
>= cnx
->remote_caps
.ack_threshold
)
1314 || (cnx
->num_pending_acks
>= VETH_MAX_ACKS_PER_MSG
) )
1315 veth_flush_acks(cnx
);
1317 spin_unlock_irqrestore(&cnx
->lock
, flags
);
1320 static void veth_timed_ack(unsigned long ptr
)
1322 struct veth_lpar_connection
*cnx
= (struct veth_lpar_connection
*) ptr
;
1323 unsigned long flags
;
1325 /* Ack all the events */
1326 spin_lock_irqsave(&cnx
->lock
, flags
);
1327 if (cnx
->num_pending_acks
> 0)
1328 veth_flush_acks(cnx
);
1330 /* Reschedule the timer */
1331 cnx
->ack_timer
.expires
= jiffies
+ cnx
->ack_timeout
;
1332 add_timer(&cnx
->ack_timer
);
1333 spin_unlock_irqrestore(&cnx
->lock
, flags
);
1336 static int veth_remove(struct vio_dev
*vdev
)
1338 int i
= vdev
->unit_address
;
1339 struct net_device
*dev
;
1344 unregister_netdev(dev
);
1350 static int veth_probe(struct vio_dev
*vdev
, const struct vio_device_id
*id
)
1352 int i
= vdev
->unit_address
;
1353 struct net_device
*dev
;
1355 dev
= veth_probe_one(i
, &vdev
->dev
);
1362 /* Start the state machine on each connection, to commence
1363 * link negotiation */
1364 for (i
= 0; i
< HVMAXARCHITECTEDLPS
; i
++)
1366 veth_kick_statemachine(veth_cnx
[i
]);
1372 * veth_device_table: Used by vio.c to match devices that we
1375 static struct vio_device_id veth_device_table
[] __devinitdata
= {
1379 MODULE_DEVICE_TABLE(vio
, veth_device_table
);
1381 static struct vio_driver veth_driver
= {
1382 .name
= "iseries_veth",
1383 .id_table
= veth_device_table
,
1384 .probe
= veth_probe
,
1385 .remove
= veth_remove
1389 * Module initialization/cleanup
1392 void __exit
veth_module_cleanup(void)
1396 /* Stop the queues first to stop any new packets being sent. */
1397 for (i
= 0; i
< HVMAXARCHITECTEDVIRTUALLANS
; i
++)
1399 netif_stop_queue(veth_dev
[i
]);
1401 /* Stop the connections before we unregister the driver. This
1402 * ensures there's no skbs lying around holding the device open. */
1403 for (i
= 0; i
< HVMAXARCHITECTEDLPS
; ++i
)
1404 veth_stop_connection(i
);
1406 HvLpEvent_unregisterHandler(HvLpEvent_Type_VirtualLan
);
1408 /* Hypervisor callbacks may have scheduled more work while we
1409 * were stoping connections. Now that we've disconnected from
1410 * the hypervisor make sure everything's finished. */
1411 flush_scheduled_work();
1413 vio_unregister_driver(&veth_driver
);
1415 for (i
= 0; i
< HVMAXARCHITECTEDLPS
; ++i
)
1416 veth_destroy_connection(i
);
1419 module_exit(veth_module_cleanup
);
1421 int __init
veth_module_init(void)
1426 this_lp
= HvLpConfig_getLpIndex_outline();
1428 for (i
= 0; i
< HVMAXARCHITECTEDLPS
; ++i
) {
1429 rc
= veth_init_connection(i
);
1431 veth_module_cleanup();
1436 HvLpEvent_registerHandler(HvLpEvent_Type_VirtualLan
,
1437 &veth_handle_event
);
1439 return vio_register_driver(&veth_driver
);
1441 module_init(veth_module_init
);