1 /* File veth.c created by Kyle A. Lucke on Mon Aug 7 2000. */
3 * IBM eServer iSeries Virtual Ethernet Device Driver
4 * Copyright (C) 2001 Kyle A. Lucke (klucke@us.ibm.com), IBM Corp.
5 * Substantially cleaned up by:
6 * Copyright (C) 2003 David Gibson <dwg@au1.ibm.com>, IBM Corporation.
8 * This program is free software; you can redistribute it and/or
9 * modify it under the terms of the GNU General Public License as
10 * published by the Free Software Foundation; either version 2 of the
11 * License, or (at your option) any later version.
13 * This program is distributed in the hope that it will be useful, but
14 * WITHOUT ANY WARRANTY; without even the implied warranty of
15 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
16 * General Public License for more details.
18 * You should have received a copy of the GNU General Public License
19 * along with this program; if not, write to the Free Software
20 * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307
24 * This module implements the virtual ethernet device for iSeries LPAR
25 * Linux. It uses hypervisor message passing to implement an
26 * ethernet-like network device communicating between partitions on
29 * The iSeries LPAR hypervisor currently allows for up to 16 different
30 * virtual ethernets. These are all dynamically configurable on
31 * OS/400 partitions, but dynamic configuration is not supported under
32 * Linux yet. An ethXX network device will be created for each
33 * virtual ethernet this partition is connected to.
35 * - This driver is responsible for routing packets to and from other
36 * partitions. The MAC addresses used by the virtual ethernets
37 * contains meaning and must not be modified.
39 * - Having 2 virtual ethernets to the same remote partition DOES NOT
40 * double the available bandwidth. The 2 devices will share the
41 * available hypervisor bandwidth.
43 * - If you send a packet to your own mac address, it will just be
44 * dropped, you won't get it on the receive side.
46 * - Multicast is implemented by sending the frame frame to every
47 * other partition. It is the responsibility of the receiving
48 * partition to filter the addresses desired.
52 * VETH_NUMBUFFERS: This compile time option defaults to 120. It
53 * controls how much memory Linux will allocate per remote partition
54 * it is communicating with. It can be thought of as the maximum
55 * number of packets outstanding to a remote partition at a time.
58 #include <linux/config.h>
59 #include <linux/module.h>
60 #include <linux/version.h>
61 #include <linux/types.h>
62 #include <linux/errno.h>
63 #include <linux/ioport.h>
64 #include <linux/kernel.h>
65 #include <linux/netdevice.h>
66 #include <linux/etherdevice.h>
67 #include <linux/skbuff.h>
68 #include <linux/init.h>
69 #include <linux/delay.h>
71 #include <linux/ethtool.h>
72 #include <asm/iSeries/mf.h>
73 #include <asm/iSeries/iSeries_pci.h>
74 #include <asm/uaccess.h>
76 #include <asm/iSeries/HvLpConfig.h>
77 #include <asm/iSeries/HvTypes.h>
78 #include <asm/iSeries/HvLpEvent.h>
79 #include <asm/iommu.h>
82 #include "iseries_veth.h"
84 MODULE_AUTHOR("Kyle Lucke <klucke@us.ibm.com>");
85 MODULE_DESCRIPTION("iSeries Virtual ethernet driver");
86 MODULE_LICENSE("GPL");
88 #define VETH_NUMBUFFERS (120)
89 #define VETH_ACKTIMEOUT (1000000) /* microseconds */
90 #define VETH_MAX_MCAST (12)
92 #define VETH_MAX_MTU (9000)
94 #if VETH_NUMBUFFERS < 10
95 #define ACK_THRESHOLD (1)
96 #elif VETH_NUMBUFFERS < 20
97 #define ACK_THRESHOLD (4)
98 #elif VETH_NUMBUFFERS < 40
99 #define ACK_THRESHOLD (10)
101 #define ACK_THRESHOLD (20)
104 #define VETH_STATE_SHUTDOWN (0x0001)
105 #define VETH_STATE_OPEN (0x0002)
106 #define VETH_STATE_RESET (0x0004)
107 #define VETH_STATE_SENTMON (0x0008)
108 #define VETH_STATE_SENTCAPS (0x0010)
109 #define VETH_STATE_GOTCAPACK (0x0020)
110 #define VETH_STATE_GOTCAPS (0x0040)
111 #define VETH_STATE_SENTCAPACK (0x0080)
112 #define VETH_STATE_READY (0x0100)
115 struct veth_msg
*next
;
116 struct VethFramesData data
;
118 unsigned long in_use
;
123 struct veth_lpar_connection
{
125 struct work_struct statemachine_wq
;
126 struct veth_msg
*msgs
;
128 struct VethCapData local_caps
;
130 struct timer_list ack_timer
;
134 HvLpInstanceId src_inst
;
135 HvLpInstanceId dst_inst
;
136 struct VethLpEvent cap_event
, cap_ack_event
;
137 u16 pending_acks
[VETH_MAX_ACKS_PER_MSG
];
138 u32 num_pending_acks
;
141 struct VethCapData remote_caps
;
144 spinlock_t msg_stack_lock
;
145 struct veth_msg
*msg_stack_head
;
150 struct net_device_stats stats
;
152 HvLpIndexMap lpar_map
;
154 spinlock_t pending_gate
;
155 struct sk_buff
*pending_skb
;
156 HvLpIndexMap pending_lpmask
;
162 u64 mcast_addr
[VETH_MAX_MCAST
];
165 static HvLpIndex this_lp
;
166 static struct veth_lpar_connection
*veth_cnx
[HVMAXARCHITECTEDLPS
]; /* = 0 */
167 static struct net_device
*veth_dev
[HVMAXARCHITECTEDVIRTUALLANS
]; /* = 0 */
169 static int veth_start_xmit(struct sk_buff
*skb
, struct net_device
*dev
);
170 static void veth_recycle_msg(struct veth_lpar_connection
*, struct veth_msg
*);
171 static void veth_flush_pending(struct veth_lpar_connection
*cnx
);
172 static void veth_receive(struct veth_lpar_connection
*, struct VethLpEvent
*);
173 static void veth_timed_ack(unsigned long connectionPtr
);
179 #define veth_printk(prio, fmt, args...) \
180 printk(prio "%s: " fmt, __FILE__, ## args)
182 #define veth_error(fmt, args...) \
183 printk(KERN_ERR "(%s:%3.3d) ERROR: " fmt, __FILE__, __LINE__ , ## args)
185 static inline void veth_stack_push(struct veth_lpar_connection
*cnx
,
186 struct veth_msg
*msg
)
190 spin_lock_irqsave(&cnx
->msg_stack_lock
, flags
);
191 msg
->next
= cnx
->msg_stack_head
;
192 cnx
->msg_stack_head
= msg
;
193 spin_unlock_irqrestore(&cnx
->msg_stack_lock
, flags
);
196 static inline struct veth_msg
*veth_stack_pop(struct veth_lpar_connection
*cnx
)
199 struct veth_msg
*msg
;
201 spin_lock_irqsave(&cnx
->msg_stack_lock
, flags
);
202 msg
= cnx
->msg_stack_head
;
204 cnx
->msg_stack_head
= cnx
->msg_stack_head
->next
;
205 spin_unlock_irqrestore(&cnx
->msg_stack_lock
, flags
);
209 static inline HvLpEvent_Rc
210 veth_signalevent(struct veth_lpar_connection
*cnx
, u16 subtype
,
211 HvLpEvent_AckInd ackind
, HvLpEvent_AckType acktype
,
213 u64 data1
, u64 data2
, u64 data3
, u64 data4
, u64 data5
)
215 return HvCallEvent_signalLpEventFast(cnx
->remote_lp
,
216 HvLpEvent_Type_VirtualLan
,
217 subtype
, ackind
, acktype
,
220 token
, data1
, data2
, data3
,
224 static inline HvLpEvent_Rc
veth_signaldata(struct veth_lpar_connection
*cnx
,
225 u16 subtype
, u64 token
, void *data
)
227 u64
*p
= (u64
*) data
;
229 return veth_signalevent(cnx
, subtype
, HvLpEvent_AckInd_NoAck
,
230 HvLpEvent_AckType_ImmediateAck
,
231 token
, p
[0], p
[1], p
[2], p
[3], p
[4]);
234 struct veth_allocation
{
239 static void veth_complete_allocation(void *parm
, int number
)
241 struct veth_allocation
*vc
= (struct veth_allocation
*)parm
;
247 static int veth_allocate_events(HvLpIndex rlp
, int number
)
249 struct veth_allocation vc
= { COMPLETION_INITIALIZER(vc
.c
), 0 };
251 mf_allocate_lp_events(rlp
, HvLpEvent_Type_VirtualLan
,
252 sizeof(struct VethLpEvent
), number
,
253 &veth_complete_allocation
, &vc
);
254 wait_for_completion(&vc
.c
);
260 * LPAR connection code
263 static inline void veth_kick_statemachine(struct veth_lpar_connection
*cnx
)
265 schedule_work(&cnx
->statemachine_wq
);
268 static void veth_take_cap(struct veth_lpar_connection
*cnx
,
269 struct VethLpEvent
*event
)
273 spin_lock_irqsave(&cnx
->lock
, flags
);
274 /* Receiving caps may mean the other end has just come up, so
275 * we need to reload the instance ID of the far end */
277 HvCallEvent_getTargetLpInstanceId(cnx
->remote_lp
,
278 HvLpEvent_Type_VirtualLan
);
280 if (cnx
->state
& VETH_STATE_GOTCAPS
) {
281 veth_error("Received a second capabilities from lpar %d\n",
283 event
->base_event
.xRc
= HvLpEvent_Rc_BufferNotAvailable
;
284 HvCallEvent_ackLpEvent((struct HvLpEvent
*) event
);
286 memcpy(&cnx
->cap_event
, event
, sizeof(cnx
->cap_event
));
287 cnx
->state
|= VETH_STATE_GOTCAPS
;
288 veth_kick_statemachine(cnx
);
290 spin_unlock_irqrestore(&cnx
->lock
, flags
);
293 static void veth_take_cap_ack(struct veth_lpar_connection
*cnx
,
294 struct VethLpEvent
*event
)
298 spin_lock_irqsave(&cnx
->lock
, flags
);
299 if (cnx
->state
& VETH_STATE_GOTCAPACK
) {
300 veth_error("Received a second capabilities ack from lpar %d\n",
303 memcpy(&cnx
->cap_ack_event
, event
,
304 sizeof(&cnx
->cap_ack_event
));
305 cnx
->state
|= VETH_STATE_GOTCAPACK
;
306 veth_kick_statemachine(cnx
);
308 spin_unlock_irqrestore(&cnx
->lock
, flags
);
311 static void veth_take_monitor_ack(struct veth_lpar_connection
*cnx
,
312 struct VethLpEvent
*event
)
316 spin_lock_irqsave(&cnx
->lock
, flags
);
317 veth_printk(KERN_DEBUG
, "Monitor ack returned for lpar %d\n",
319 cnx
->state
|= VETH_STATE_RESET
;
320 veth_kick_statemachine(cnx
);
321 spin_unlock_irqrestore(&cnx
->lock
, flags
);
324 static void veth_handle_ack(struct VethLpEvent
*event
)
326 HvLpIndex rlp
= event
->base_event
.xTargetLp
;
327 struct veth_lpar_connection
*cnx
= veth_cnx
[rlp
];
331 switch (event
->base_event
.xSubtype
) {
332 case VethEventTypeCap
:
333 veth_take_cap_ack(cnx
, event
);
335 case VethEventTypeMonitor
:
336 veth_take_monitor_ack(cnx
, event
);
339 veth_error("Unknown ack type %d from lpar %d\n",
340 event
->base_event
.xSubtype
, rlp
);
344 static void veth_handle_int(struct VethLpEvent
*event
)
346 HvLpIndex rlp
= event
->base_event
.xSourceLp
;
347 struct veth_lpar_connection
*cnx
= veth_cnx
[rlp
];
353 switch (event
->base_event
.xSubtype
) {
354 case VethEventTypeCap
:
355 veth_take_cap(cnx
, event
);
357 case VethEventTypeMonitor
:
358 /* do nothing... this'll hang out here til we're dead,
359 * and the hypervisor will return it for us. */
361 case VethEventTypeFramesAck
:
362 spin_lock_irqsave(&cnx
->lock
, flags
);
363 for (i
= 0; i
< VETH_MAX_ACKS_PER_MSG
; ++i
) {
364 u16 msgnum
= event
->u
.frames_ack_data
.token
[i
];
366 if (msgnum
< VETH_NUMBUFFERS
)
367 veth_recycle_msg(cnx
, cnx
->msgs
+ msgnum
);
369 spin_unlock_irqrestore(&cnx
->lock
, flags
);
370 veth_flush_pending(cnx
);
372 case VethEventTypeFrames
:
373 veth_receive(cnx
, event
);
376 veth_error("Unknown interrupt type %d from lpar %d\n",
377 event
->base_event
.xSubtype
, rlp
);
381 static void veth_handle_event(struct HvLpEvent
*event
, struct pt_regs
*regs
)
383 struct VethLpEvent
*veth_event
= (struct VethLpEvent
*)event
;
385 if (event
->xFlags
.xFunction
== HvLpEvent_Function_Ack
)
386 veth_handle_ack(veth_event
);
387 else if (event
->xFlags
.xFunction
== HvLpEvent_Function_Int
)
388 veth_handle_int(veth_event
);
391 static int veth_process_caps(struct veth_lpar_connection
*cnx
)
393 struct VethCapData
*remote_caps
= &cnx
->remote_caps
;
396 /* Convert timer to jiffies */
397 cnx
->ack_timeout
= remote_caps
->ack_timeout
* HZ
/ 1000000;
399 if ( (remote_caps
->num_buffers
== 0)
400 || (remote_caps
->ack_threshold
> VETH_MAX_ACKS_PER_MSG
)
401 || (remote_caps
->ack_threshold
== 0)
402 || (cnx
->ack_timeout
== 0) ) {
403 veth_error("Received incompatible capabilities from lpar %d\n",
405 return HvLpEvent_Rc_InvalidSubtypeData
;
408 num_acks_needed
= (remote_caps
->num_buffers
409 / remote_caps
->ack_threshold
) + 1;
411 /* FIXME: locking on num_ack_events? */
412 if (cnx
->num_ack_events
< num_acks_needed
) {
415 num
= veth_allocate_events(cnx
->remote_lp
,
416 num_acks_needed
-cnx
->num_ack_events
);
418 cnx
->num_ack_events
+= num
;
420 if (cnx
->num_ack_events
< num_acks_needed
) {
421 veth_error("Couldn't allocate enough ack events for lpar %d\n",
424 return HvLpEvent_Rc_BufferNotAvailable
;
429 return HvLpEvent_Rc_Good
;
432 /* FIXME: The gotos here are a bit dubious */
433 static void veth_statemachine(void *p
)
435 struct veth_lpar_connection
*cnx
= (struct veth_lpar_connection
*)p
;
436 int rlp
= cnx
->remote_lp
;
439 spin_lock_irq(&cnx
->lock
);
442 if (cnx
->state
& VETH_STATE_RESET
) {
445 del_timer(&cnx
->ack_timer
);
447 if (cnx
->state
& VETH_STATE_OPEN
)
448 HvCallEvent_closeLpEventPath(cnx
->remote_lp
,
449 HvLpEvent_Type_VirtualLan
);
452 memset(&cnx
->pending_acks
, 0xff, sizeof (cnx
->pending_acks
));
453 cnx
->num_pending_acks
= 0;
455 cnx
->state
&= ~(VETH_STATE_RESET
| VETH_STATE_SENTMON
456 | VETH_STATE_OPEN
| VETH_STATE_SENTCAPS
457 | VETH_STATE_GOTCAPACK
| VETH_STATE_GOTCAPS
458 | VETH_STATE_SENTCAPACK
| VETH_STATE_READY
);
460 /* Clean up any leftover messages */
462 for (i
= 0; i
< VETH_NUMBUFFERS
; ++i
)
463 veth_recycle_msg(cnx
, cnx
->msgs
+ i
);
464 spin_unlock_irq(&cnx
->lock
);
465 veth_flush_pending(cnx
);
466 spin_lock_irq(&cnx
->lock
);
467 if (cnx
->state
& VETH_STATE_RESET
)
471 if (cnx
->state
& VETH_STATE_SHUTDOWN
)
472 /* It's all over, do nothing */
475 if ( !(cnx
->state
& VETH_STATE_OPEN
) ) {
476 if (! cnx
->msgs
|| (cnx
->num_events
< (2 + VETH_NUMBUFFERS
)) )
479 HvCallEvent_openLpEventPath(rlp
, HvLpEvent_Type_VirtualLan
);
481 HvCallEvent_getSourceLpInstanceId(rlp
,
482 HvLpEvent_Type_VirtualLan
);
484 HvCallEvent_getTargetLpInstanceId(rlp
,
485 HvLpEvent_Type_VirtualLan
);
486 cnx
->state
|= VETH_STATE_OPEN
;
489 if ( (cnx
->state
& VETH_STATE_OPEN
)
490 && !(cnx
->state
& VETH_STATE_SENTMON
) ) {
491 rc
= veth_signalevent(cnx
, VethEventTypeMonitor
,
492 HvLpEvent_AckInd_DoAck
,
493 HvLpEvent_AckType_DeferredAck
,
496 if (rc
== HvLpEvent_Rc_Good
) {
497 cnx
->state
|= VETH_STATE_SENTMON
;
499 if ( (rc
!= HvLpEvent_Rc_PartitionDead
)
500 && (rc
!= HvLpEvent_Rc_PathClosed
) )
501 veth_error("Error sending monitor to "
505 /* Oh well, hope we get a cap from the other
506 * end and do better when that kicks us */
511 if ( (cnx
->state
& VETH_STATE_OPEN
)
512 && !(cnx
->state
& VETH_STATE_SENTCAPS
)) {
513 u64
*rawcap
= (u64
*)&cnx
->local_caps
;
515 rc
= veth_signalevent(cnx
, VethEventTypeCap
,
516 HvLpEvent_AckInd_DoAck
,
517 HvLpEvent_AckType_ImmediateAck
,
518 0, rawcap
[0], rawcap
[1], rawcap
[2],
519 rawcap
[3], rawcap
[4]);
521 if (rc
== HvLpEvent_Rc_Good
) {
522 cnx
->state
|= VETH_STATE_SENTCAPS
;
524 if ( (rc
!= HvLpEvent_Rc_PartitionDead
)
525 && (rc
!= HvLpEvent_Rc_PathClosed
) )
526 veth_error("Error sending caps to "
529 /* Oh well, hope we get a cap from the other
530 * end and do better when that kicks us */
535 if ((cnx
->state
& VETH_STATE_GOTCAPS
)
536 && !(cnx
->state
& VETH_STATE_SENTCAPACK
)) {
537 struct VethCapData
*remote_caps
= &cnx
->remote_caps
;
539 memcpy(remote_caps
, &cnx
->cap_event
.u
.caps_data
,
540 sizeof(*remote_caps
));
542 spin_unlock_irq(&cnx
->lock
);
543 rc
= veth_process_caps(cnx
);
544 spin_lock_irq(&cnx
->lock
);
546 /* We dropped the lock, so recheck for anything which
547 * might mess us up */
548 if (cnx
->state
& (VETH_STATE_RESET
|VETH_STATE_SHUTDOWN
))
551 cnx
->cap_event
.base_event
.xRc
= rc
;
552 HvCallEvent_ackLpEvent((struct HvLpEvent
*)&cnx
->cap_event
);
553 if (rc
== HvLpEvent_Rc_Good
)
554 cnx
->state
|= VETH_STATE_SENTCAPACK
;
559 if ((cnx
->state
& VETH_STATE_GOTCAPACK
)
560 && (cnx
->state
& VETH_STATE_GOTCAPS
)
561 && !(cnx
->state
& VETH_STATE_READY
)) {
562 if (cnx
->cap_ack_event
.base_event
.xRc
== HvLpEvent_Rc_Good
) {
563 /* Start the ACK timer */
564 cnx
->ack_timer
.expires
= jiffies
+ cnx
->ack_timeout
;
565 add_timer(&cnx
->ack_timer
);
566 cnx
->state
|= VETH_STATE_READY
;
568 veth_printk(KERN_ERR
, "Caps rejected (rc=%d) by "
570 cnx
->cap_ack_event
.base_event
.xRc
,
577 spin_unlock_irq(&cnx
->lock
);
581 /* FIXME: we get here if something happens we really can't
582 * cope with. The link will never work once we get here, and
583 * all we can do is not lock the rest of the system up */
584 veth_error("Badness on connection to lpar %d (state=%04lx) "
585 " - shutting down\n", rlp
, cnx
->state
);
586 cnx
->state
|= VETH_STATE_SHUTDOWN
;
587 spin_unlock_irq(&cnx
->lock
);
590 static int veth_init_connection(u8 rlp
)
592 struct veth_lpar_connection
*cnx
;
593 struct veth_msg
*msgs
;
596 if ( (rlp
== this_lp
)
597 || ! HvLpConfig_doLpsCommunicateOnVirtualLan(this_lp
, rlp
) )
600 cnx
= kmalloc(sizeof(*cnx
), GFP_KERNEL
);
603 memset(cnx
, 0, sizeof(*cnx
));
605 cnx
->remote_lp
= rlp
;
606 spin_lock_init(&cnx
->lock
);
607 INIT_WORK(&cnx
->statemachine_wq
, veth_statemachine
, cnx
);
608 init_timer(&cnx
->ack_timer
);
609 cnx
->ack_timer
.function
= veth_timed_ack
;
610 cnx
->ack_timer
.data
= (unsigned long) cnx
;
611 memset(&cnx
->pending_acks
, 0xff, sizeof (cnx
->pending_acks
));
615 msgs
= kmalloc(VETH_NUMBUFFERS
* sizeof(struct veth_msg
), GFP_KERNEL
);
617 veth_error("Can't allocate buffers for lpar %d\n", rlp
);
622 memset(msgs
, 0, VETH_NUMBUFFERS
* sizeof(struct veth_msg
));
623 spin_lock_init(&cnx
->msg_stack_lock
);
625 for (i
= 0; i
< VETH_NUMBUFFERS
; i
++) {
627 veth_stack_push(cnx
, msgs
+ i
);
630 cnx
->num_events
= veth_allocate_events(rlp
, 2 + VETH_NUMBUFFERS
);
632 if (cnx
->num_events
< (2 + VETH_NUMBUFFERS
)) {
633 veth_error("Can't allocate events for lpar %d, only got %d\n",
634 rlp
, cnx
->num_events
);
638 cnx
->local_caps
.num_buffers
= VETH_NUMBUFFERS
;
639 cnx
->local_caps
.ack_threshold
= ACK_THRESHOLD
;
640 cnx
->local_caps
.ack_timeout
= VETH_ACKTIMEOUT
;
645 static void veth_stop_connection(u8 rlp
)
647 struct veth_lpar_connection
*cnx
= veth_cnx
[rlp
];
652 spin_lock_irq(&cnx
->lock
);
653 cnx
->state
|= VETH_STATE_RESET
| VETH_STATE_SHUTDOWN
;
654 veth_kick_statemachine(cnx
);
655 spin_unlock_irq(&cnx
->lock
);
657 flush_scheduled_work();
659 /* FIXME: not sure if this is necessary - will already have
660 * been deleted by the state machine, just want to make sure
661 * its not running any more */
662 del_timer_sync(&cnx
->ack_timer
);
664 if (cnx
->num_events
> 0)
665 mf_deallocate_lp_events(cnx
->remote_lp
,
666 HvLpEvent_Type_VirtualLan
,
669 if (cnx
->num_ack_events
> 0)
670 mf_deallocate_lp_events(cnx
->remote_lp
,
671 HvLpEvent_Type_VirtualLan
,
676 static void veth_destroy_connection(u8 rlp
)
678 struct veth_lpar_connection
*cnx
= veth_cnx
[rlp
];
685 veth_cnx
[rlp
] = NULL
;
692 static int veth_open(struct net_device
*dev
)
694 struct veth_port
*port
= (struct veth_port
*) dev
->priv
;
696 memset(&port
->stats
, 0, sizeof (port
->stats
));
697 netif_start_queue(dev
);
701 static int veth_close(struct net_device
*dev
)
703 netif_stop_queue(dev
);
707 static struct net_device_stats
*veth_get_stats(struct net_device
*dev
)
709 struct veth_port
*port
= (struct veth_port
*) dev
->priv
;
714 static int veth_change_mtu(struct net_device
*dev
, int new_mtu
)
716 if ((new_mtu
< 68) || (new_mtu
> VETH_MAX_MTU
))
722 static void veth_set_multicast_list(struct net_device
*dev
)
724 struct veth_port
*port
= (struct veth_port
*) dev
->priv
;
727 write_lock_irqsave(&port
->mcast_gate
, flags
);
729 if (dev
->flags
& IFF_PROMISC
) { /* set promiscuous mode */
730 printk(KERN_INFO
"%s: Promiscuous mode enabled.\n",
732 port
->promiscuous
= 1;
733 } else if ( (dev
->flags
& IFF_ALLMULTI
)
734 || (dev
->mc_count
> VETH_MAX_MCAST
) ) {
737 struct dev_mc_list
*dmi
= dev
->mc_list
;
743 for (i
= 0; i
< dev
->mc_count
; i
++) {
744 u8
*addr
= dmi
->dmi_addr
;
747 if (addr
[0] & 0x01) {/* multicast address? */
748 memcpy(&xaddr
, addr
, ETH_ALEN
);
749 port
->mcast_addr
[port
->num_mcast
] = xaddr
;
756 write_unlock_irqrestore(&port
->mcast_gate
, flags
);
759 static void veth_get_drvinfo(struct net_device
*dev
, struct ethtool_drvinfo
*info
)
761 strncpy(info
->driver
, "veth", sizeof(info
->driver
) - 1);
762 info
->driver
[sizeof(info
->driver
) - 1] = '\0';
763 strncpy(info
->version
, "1.0", sizeof(info
->version
) - 1);
766 static int veth_get_settings(struct net_device
*dev
, struct ethtool_cmd
*ecmd
)
768 ecmd
->supported
= (SUPPORTED_1000baseT_Full
769 | SUPPORTED_Autoneg
| SUPPORTED_FIBRE
);
770 ecmd
->advertising
= (SUPPORTED_1000baseT_Full
771 | SUPPORTED_Autoneg
| SUPPORTED_FIBRE
);
772 ecmd
->port
= PORT_FIBRE
;
773 ecmd
->transceiver
= XCVR_INTERNAL
;
774 ecmd
->phy_address
= 0;
775 ecmd
->speed
= SPEED_1000
;
776 ecmd
->duplex
= DUPLEX_FULL
;
777 ecmd
->autoneg
= AUTONEG_ENABLE
;
778 ecmd
->maxtxpkt
= 120;
779 ecmd
->maxrxpkt
= 120;
783 static u32
veth_get_link(struct net_device
*dev
)
788 static struct ethtool_ops ops
= {
789 .get_drvinfo
= veth_get_drvinfo
,
790 .get_settings
= veth_get_settings
,
791 .get_link
= veth_get_link
,
794 static void veth_tx_timeout(struct net_device
*dev
)
796 struct veth_port
*port
= (struct veth_port
*)dev
->priv
;
797 struct net_device_stats
*stats
= &port
->stats
;
803 spin_lock_irqsave(&port
->pending_gate
, flags
);
805 if (!port
->pending_lpmask
) {
806 spin_unlock_irqrestore(&port
->pending_gate
, flags
);
810 printk(KERN_WARNING
"%s: Tx timeout! Resetting lp connections: %08x\n",
811 dev
->name
, port
->pending_lpmask
);
813 for (i
= 0; i
< HVMAXARCHITECTEDLPS
; i
++) {
814 struct veth_lpar_connection
*cnx
= veth_cnx
[i
];
816 if (! (port
->pending_lpmask
& (1<<i
)))
819 /* If we're pending on it, we must be connected to it,
820 * so we should certainly have a structure for it. */
823 /* Theoretically we could be kicking a connection
824 * which doesn't deserve it, but in practice if we've
825 * had a Tx timeout, the pending_lpmask will have
826 * exactly one bit set - the connection causing the
828 spin_lock(&cnx
->lock
);
829 cnx
->state
|= VETH_STATE_RESET
;
830 veth_kick_statemachine(cnx
);
831 spin_unlock(&cnx
->lock
);
834 spin_unlock_irqrestore(&port
->pending_gate
, flags
);
837 static struct net_device
* __init
veth_probe_one(int vlan
, struct device
*vdev
)
839 struct net_device
*dev
;
840 struct veth_port
*port
;
843 dev
= alloc_etherdev(sizeof (struct veth_port
));
845 veth_error("Unable to allocate net_device structure!\n");
849 port
= (struct veth_port
*) dev
->priv
;
851 spin_lock_init(&port
->pending_gate
);
852 rwlock_init(&port
->mcast_gate
);
854 for (i
= 0; i
< HVMAXARCHITECTEDLPS
; i
++) {
855 HvLpVirtualLanIndexMap map
;
859 map
= HvLpConfig_getVirtualLanIndexMapForLp(i
);
860 if (map
& (0x8000 >> vlan
))
861 port
->lpar_map
|= (1 << i
);
865 dev
->dev_addr
[0] = 0x02;
866 dev
->dev_addr
[1] = 0x01;
867 dev
->dev_addr
[2] = 0xff;
868 dev
->dev_addr
[3] = vlan
;
869 dev
->dev_addr
[4] = 0xff;
870 dev
->dev_addr
[5] = this_lp
;
872 dev
->mtu
= VETH_MAX_MTU
;
874 memcpy(&port
->mac_addr
, dev
->dev_addr
, 6);
876 dev
->open
= veth_open
;
877 dev
->hard_start_xmit
= veth_start_xmit
;
878 dev
->stop
= veth_close
;
879 dev
->get_stats
= veth_get_stats
;
880 dev
->change_mtu
= veth_change_mtu
;
881 dev
->set_mac_address
= NULL
;
882 dev
->set_multicast_list
= veth_set_multicast_list
;
883 SET_ETHTOOL_OPS(dev
, &ops
);
885 dev
->watchdog_timeo
= 2 * (VETH_ACKTIMEOUT
* HZ
/ 1000000);
886 dev
->tx_timeout
= veth_tx_timeout
;
888 SET_NETDEV_DEV(dev
, vdev
);
890 rc
= register_netdev(dev
);
892 veth_printk(KERN_ERR
,
893 "Failed to register ethernet device for vlan %d\n",
899 veth_printk(KERN_DEBUG
, "%s attached to iSeries vlan %d (lpar_map=0x%04x)\n",
900 dev
->name
, vlan
, port
->lpar_map
);
909 static int veth_transmit_to_one(struct sk_buff
*skb
, HvLpIndex rlp
,
910 struct net_device
*dev
)
912 struct veth_lpar_connection
*cnx
= veth_cnx
[rlp
];
913 struct veth_port
*port
= (struct veth_port
*) dev
->priv
;
915 u32 dma_address
, dma_length
;
916 struct veth_msg
*msg
= NULL
;
921 port
->stats
.tx_errors
++;
926 spin_lock_irqsave(&cnx
->lock
, flags
);
928 if (! (cnx
->state
& VETH_STATE_READY
))
931 if ((skb
->len
- 14) > VETH_MAX_MTU
)
934 msg
= veth_stack_pop(cnx
);
941 dma_length
= skb
->len
;
942 dma_address
= dma_map_single(port
->dev
, skb
->data
,
943 dma_length
, DMA_TO_DEVICE
);
945 if (dma_mapping_error(dma_address
))
946 goto recycle_and_drop
;
948 /* Is it really necessary to check the length and address
949 * fields of the first entry here? */
951 msg
->dev
= port
->dev
;
952 msg
->data
.addr
[0] = dma_address
;
953 msg
->data
.len
[0] = dma_length
;
954 msg
->data
.eofmask
= 1 << VETH_EOF_SHIFT
;
955 set_bit(0, &(msg
->in_use
));
956 rc
= veth_signaldata(cnx
, VethEventTypeFrames
, msg
->token
, &msg
->data
);
958 if (rc
!= HvLpEvent_Rc_Good
)
959 goto recycle_and_drop
;
961 spin_unlock_irqrestore(&cnx
->lock
, flags
);
966 /* need to set in use to make veth_recycle_msg in case this
967 * was a mapping failure */
968 set_bit(0, &msg
->in_use
);
969 veth_recycle_msg(cnx
, msg
);
971 port
->stats
.tx_errors
++;
973 spin_unlock_irqrestore(&cnx
->lock
, flags
);
977 static HvLpIndexMap
veth_transmit_to_many(struct sk_buff
*skb
,
979 struct net_device
*dev
)
981 struct veth_port
*port
= (struct veth_port
*) dev
->priv
;
985 for (i
= 0; i
< HVMAXARCHITECTEDLPS
; i
++) {
986 if ((lpmask
& (1 << i
)) == 0)
989 rc
= veth_transmit_to_one(skb_get(skb
), i
, dev
);
995 port
->stats
.tx_packets
++;
996 port
->stats
.tx_bytes
+= skb
->len
;
1002 static int veth_start_xmit(struct sk_buff
*skb
, struct net_device
*dev
)
1004 unsigned char *frame
= skb
->data
;
1005 struct veth_port
*port
= (struct veth_port
*) dev
->priv
;
1006 unsigned long flags
;
1007 HvLpIndexMap lpmask
;
1009 if (! (frame
[0] & 0x01)) {
1010 /* unicast packet */
1011 HvLpIndex rlp
= frame
[5];
1013 if ( ! ((1 << rlp
) & port
->lpar_map
) ) {
1020 lpmask
= port
->lpar_map
;
1023 spin_lock_irqsave(&port
->pending_gate
, flags
);
1025 lpmask
= veth_transmit_to_many(skb
, lpmask
, dev
);
1027 dev
->trans_start
= jiffies
;
1032 if (port
->pending_skb
) {
1033 veth_error("%s: Tx while skb was pending!\n",
1036 spin_unlock_irqrestore(&port
->pending_gate
, flags
);
1040 port
->pending_skb
= skb
;
1041 port
->pending_lpmask
= lpmask
;
1042 netif_stop_queue(dev
);
1045 spin_unlock_irqrestore(&port
->pending_gate
, flags
);
1050 static void veth_recycle_msg(struct veth_lpar_connection
*cnx
,
1051 struct veth_msg
*msg
)
1053 u32 dma_address
, dma_length
;
1055 if (test_and_clear_bit(0, &msg
->in_use
)) {
1056 dma_address
= msg
->data
.addr
[0];
1057 dma_length
= msg
->data
.len
[0];
1059 dma_unmap_single(msg
->dev
, dma_address
, dma_length
,
1063 dev_kfree_skb_any(msg
->skb
);
1067 memset(&msg
->data
, 0, sizeof(msg
->data
));
1068 veth_stack_push(cnx
, msg
);
1070 if (cnx
->state
& VETH_STATE_OPEN
)
1071 veth_error("Bogus frames ack from lpar %d (#%d)\n",
1072 cnx
->remote_lp
, msg
->token
);
1075 static void veth_flush_pending(struct veth_lpar_connection
*cnx
)
1078 for (i
= 0; i
< HVMAXARCHITECTEDVIRTUALLANS
; i
++) {
1079 struct net_device
*dev
= veth_dev
[i
];
1080 struct veth_port
*port
;
1081 unsigned long flags
;
1086 port
= (struct veth_port
*)dev
->priv
;
1088 if (! (port
->lpar_map
& (1<<cnx
->remote_lp
)))
1091 spin_lock_irqsave(&port
->pending_gate
, flags
);
1092 if (port
->pending_skb
) {
1093 port
->pending_lpmask
=
1094 veth_transmit_to_many(port
->pending_skb
,
1095 port
->pending_lpmask
,
1097 if (! port
->pending_lpmask
) {
1098 dev_kfree_skb_any(port
->pending_skb
);
1099 port
->pending_skb
= NULL
;
1100 netif_wake_queue(dev
);
1103 spin_unlock_irqrestore(&port
->pending_gate
, flags
);
1111 static inline int veth_frame_wanted(struct veth_port
*port
, u64 mac_addr
)
1115 unsigned long flags
;
1117 if ( (mac_addr
== port
->mac_addr
) || (mac_addr
== 0xffffffffffff0000) )
1120 if (! (((char *) &mac_addr
)[0] & 0x01))
1123 read_lock_irqsave(&port
->mcast_gate
, flags
);
1125 if (port
->promiscuous
|| port
->all_mcast
) {
1130 for (i
= 0; i
< port
->num_mcast
; ++i
) {
1131 if (port
->mcast_addr
[i
] == mac_addr
) {
1138 read_unlock_irqrestore(&port
->mcast_gate
, flags
);
1148 #define VETH_MAX_PAGES_PER_FRAME ( (VETH_MAX_MTU+PAGE_SIZE-2)/PAGE_SIZE + 1 )
1150 static inline void veth_build_dma_list(struct dma_chunk
*list
,
1151 unsigned char *p
, unsigned long length
)
1156 /* FIXME: skbs are continguous in real addresses. Do we
1157 * really need to break it into PAGE_SIZE chunks, or can we do
1158 * it just at the granularity of iSeries real->absolute
1159 * mapping? Indeed, given the way the allocator works, can we
1160 * count on them being absolutely contiguous? */
1161 list
[0].addr
= ISERIES_HV_ADDR(p
);
1162 list
[0].size
= min(length
,
1163 PAGE_SIZE
- ((unsigned long)p
& ~PAGE_MASK
));
1165 done
= list
[0].size
;
1166 while (done
< length
) {
1167 list
[i
].addr
= ISERIES_HV_ADDR(p
+ done
);
1168 list
[i
].size
= min(length
-done
, PAGE_SIZE
);
1169 done
+= list
[i
].size
;
1174 static void veth_flush_acks(struct veth_lpar_connection
*cnx
)
1178 rc
= veth_signaldata(cnx
, VethEventTypeFramesAck
,
1179 0, &cnx
->pending_acks
);
1181 if (rc
!= HvLpEvent_Rc_Good
)
1182 veth_error("Error 0x%x acking frames from lpar %d!\n",
1183 (unsigned)rc
, cnx
->remote_lp
);
1185 cnx
->num_pending_acks
= 0;
1186 memset(&cnx
->pending_acks
, 0xff, sizeof(cnx
->pending_acks
));
1189 static void veth_receive(struct veth_lpar_connection
*cnx
,
1190 struct VethLpEvent
*event
)
1192 struct VethFramesData
*senddata
= &event
->u
.frames_data
;
1195 unsigned long flags
;
1200 struct sk_buff
*skb
;
1201 struct dma_chunk local_list
[VETH_MAX_PAGES_PER_FRAME
];
1202 struct dma_chunk remote_list
[VETH_MAX_FRAMES_PER_MSG
];
1204 HvLpVirtualLanIndex vlan
;
1205 struct net_device
*dev
;
1206 struct veth_port
*port
;
1208 /* FIXME: do we need this? */
1209 memset(local_list
, 0, sizeof(local_list
));
1210 memset(remote_list
, 0, sizeof(VETH_MAX_FRAMES_PER_MSG
));
1212 /* a 0 address marks the end of the valid entries */
1213 if (senddata
->addr
[startchunk
] == 0)
1216 /* make sure that we have at least 1 EOF entry in the
1217 * remaining entries */
1218 if (! (senddata
->eofmask
>> (startchunk
+ VETH_EOF_SHIFT
))) {
1219 veth_error("missing EOF frag in event "
1220 "eofmask=0x%x startchunk=%d\n",
1221 (unsigned) senddata
->eofmask
, startchunk
);
1225 /* build list of chunks in this frame */
1228 remote_list
[nchunks
].addr
=
1229 (u64
) senddata
->addr
[startchunk
+nchunks
] << 32;
1230 remote_list
[nchunks
].size
=
1231 senddata
->len
[startchunk
+nchunks
];
1232 length
+= remote_list
[nchunks
].size
;
1233 } while (! (senddata
->eofmask
&
1234 (1 << (VETH_EOF_SHIFT
+ startchunk
+ nchunks
++))));
1236 /* length == total length of all chunks */
1237 /* nchunks == # of chunks in this frame */
1239 if ((length
- ETH_HLEN
) > VETH_MAX_MTU
) {
1240 veth_error("Received oversize frame from lpar %d "
1241 "(length=%d)\n", cnx
->remote_lp
, length
);
1245 skb
= alloc_skb(length
, GFP_ATOMIC
);
1249 veth_build_dma_list(local_list
, skb
->data
, length
);
1251 rc
= HvCallEvent_dmaBufList(HvLpEvent_Type_VirtualLan
,
1252 event
->base_event
.xSourceLp
,
1253 HvLpDma_Direction_RemoteToLocal
,
1256 HvLpDma_AddressType_RealAddress
,
1257 HvLpDma_AddressType_TceIndex
,
1258 ISERIES_HV_ADDR(&local_list
),
1259 ISERIES_HV_ADDR(&remote_list
),
1261 if (rc
!= HvLpDma_Rc_Good
) {
1262 dev_kfree_skb_irq(skb
);
1266 vlan
= skb
->data
[9];
1267 dev
= veth_dev
[vlan
];
1270 * Some earlier versions of the driver sent
1271 * broadcasts down all connections, even to lpars
1272 * that weren't on the relevant vlan. So ignore
1273 * packets belonging to a vlan we're not on.
1274 * We can also be here if we receive packets while
1275 * the driver is going down, because then dev is NULL.
1277 dev_kfree_skb_irq(skb
);
1281 port
= (struct veth_port
*)dev
->priv
;
1282 dest
= *((u64
*) skb
->data
) & 0xFFFFFFFFFFFF0000;
1284 if ((vlan
> HVMAXARCHITECTEDVIRTUALLANS
) || !port
) {
1285 dev_kfree_skb_irq(skb
);
1288 if (! veth_frame_wanted(port
, dest
)) {
1289 dev_kfree_skb_irq(skb
);
1293 skb_put(skb
, length
);
1295 skb
->protocol
= eth_type_trans(skb
, dev
);
1296 skb
->ip_summed
= CHECKSUM_NONE
;
1297 netif_rx(skb
); /* send it up */
1298 port
->stats
.rx_packets
++;
1299 port
->stats
.rx_bytes
+= length
;
1300 } while (startchunk
+= nchunks
, startchunk
< VETH_MAX_FRAMES_PER_MSG
);
1303 spin_lock_irqsave(&cnx
->lock
, flags
);
1304 BUG_ON(cnx
->num_pending_acks
> VETH_MAX_ACKS_PER_MSG
);
1306 cnx
->pending_acks
[cnx
->num_pending_acks
++] =
1307 event
->base_event
.xCorrelationToken
;
1309 if ( (cnx
->num_pending_acks
>= cnx
->remote_caps
.ack_threshold
)
1310 || (cnx
->num_pending_acks
>= VETH_MAX_ACKS_PER_MSG
) )
1311 veth_flush_acks(cnx
);
1313 spin_unlock_irqrestore(&cnx
->lock
, flags
);
1316 static void veth_timed_ack(unsigned long ptr
)
1318 struct veth_lpar_connection
*cnx
= (struct veth_lpar_connection
*) ptr
;
1319 unsigned long flags
;
1321 /* Ack all the events */
1322 spin_lock_irqsave(&cnx
->lock
, flags
);
1323 if (cnx
->num_pending_acks
> 0)
1324 veth_flush_acks(cnx
);
1326 /* Reschedule the timer */
1327 cnx
->ack_timer
.expires
= jiffies
+ cnx
->ack_timeout
;
1328 add_timer(&cnx
->ack_timer
);
1329 spin_unlock_irqrestore(&cnx
->lock
, flags
);
1332 static int veth_remove(struct vio_dev
*vdev
)
1334 int i
= vdev
->unit_address
;
1335 struct net_device
*dev
;
1340 unregister_netdev(dev
);
1346 static int veth_probe(struct vio_dev
*vdev
, const struct vio_device_id
*id
)
1348 int i
= vdev
->unit_address
;
1349 struct net_device
*dev
;
1351 dev
= veth_probe_one(i
, &vdev
->dev
);
1358 /* Start the state machine on each connection, to commence
1359 * link negotiation */
1360 for (i
= 0; i
< HVMAXARCHITECTEDLPS
; i
++)
1362 veth_kick_statemachine(veth_cnx
[i
]);
1368 * veth_device_table: Used by vio.c to match devices that we
1371 static struct vio_device_id veth_device_table
[] __devinitdata
= {
1375 MODULE_DEVICE_TABLE(vio
, veth_device_table
);
1377 static struct vio_driver veth_driver
= {
1378 .name
= "iseries_veth",
1379 .id_table
= veth_device_table
,
1380 .probe
= veth_probe
,
1381 .remove
= veth_remove
1385 * Module initialization/cleanup
1388 void __exit
veth_module_cleanup(void)
1392 /* Stop the queues first to stop any new packets being sent. */
1393 for (i
= 0; i
< HVMAXARCHITECTEDVIRTUALLANS
; i
++)
1395 netif_stop_queue(veth_dev
[i
]);
1397 /* Stop the connections before we unregister the driver. This
1398 * ensures there's no skbs lying around holding the device open. */
1399 for (i
= 0; i
< HVMAXARCHITECTEDLPS
; ++i
)
1400 veth_stop_connection(i
);
1402 HvLpEvent_unregisterHandler(HvLpEvent_Type_VirtualLan
);
1404 /* Hypervisor callbacks may have scheduled more work while we
1405 * were stoping connections. Now that we've disconnected from
1406 * the hypervisor make sure everything's finished. */
1407 flush_scheduled_work();
1409 vio_unregister_driver(&veth_driver
);
1411 for (i
= 0; i
< HVMAXARCHITECTEDLPS
; ++i
)
1412 veth_destroy_connection(i
);
1415 module_exit(veth_module_cleanup
);
1417 int __init
veth_module_init(void)
1422 this_lp
= HvLpConfig_getLpIndex_outline();
1424 for (i
= 0; i
< HVMAXARCHITECTEDLPS
; ++i
) {
1425 rc
= veth_init_connection(i
);
1427 veth_module_cleanup();
1432 HvLpEvent_registerHandler(HvLpEvent_Type_VirtualLan
,
1433 &veth_handle_event
);
1435 return vio_register_driver(&veth_driver
);
1437 module_init(veth_module_init
);