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 MODULE_AUTHOR("Kyle Lucke <klucke@us.ibm.com>");
85 MODULE_DESCRIPTION("iSeries Virtual ethernet driver");
86 MODULE_LICENSE("GPL");
88 #define VETH_EVENT_CAP (0)
89 #define VETH_EVENT_FRAMES (1)
90 #define VETH_EVENT_MONITOR (2)
91 #define VETH_EVENT_FRAMES_ACK (3)
93 #define VETH_MAX_ACKS_PER_MSG (20)
94 #define VETH_MAX_FRAMES_PER_MSG (6)
96 struct veth_frames_data
{
97 u32 addr
[VETH_MAX_FRAMES_PER_MSG
];
98 u16 len
[VETH_MAX_FRAMES_PER_MSG
];
101 #define VETH_EOF_SHIFT (32-VETH_MAX_FRAMES_PER_MSG)
103 struct veth_frames_ack_data
{
104 u16 token
[VETH_MAX_ACKS_PER_MSG
];
107 struct veth_cap_data
{
118 struct veth_lpevent
{
119 struct HvLpEvent base_event
;
121 struct veth_cap_data caps_data
;
122 struct veth_frames_data frames_data
;
123 struct veth_frames_ack_data frames_ack_data
;
128 #define DRV_NAME "iseries_veth"
129 #define DRV_VERSION "2.0"
131 #define VETH_NUMBUFFERS (120)
132 #define VETH_ACKTIMEOUT (1000000) /* microseconds */
133 #define VETH_MAX_MCAST (12)
135 #define VETH_MAX_MTU (9000)
137 #if VETH_NUMBUFFERS < 10
138 #define ACK_THRESHOLD (1)
139 #elif VETH_NUMBUFFERS < 20
140 #define ACK_THRESHOLD (4)
141 #elif VETH_NUMBUFFERS < 40
142 #define ACK_THRESHOLD (10)
144 #define ACK_THRESHOLD (20)
147 #define VETH_STATE_SHUTDOWN (0x0001)
148 #define VETH_STATE_OPEN (0x0002)
149 #define VETH_STATE_RESET (0x0004)
150 #define VETH_STATE_SENTMON (0x0008)
151 #define VETH_STATE_SENTCAPS (0x0010)
152 #define VETH_STATE_GOTCAPACK (0x0020)
153 #define VETH_STATE_GOTCAPS (0x0040)
154 #define VETH_STATE_SENTCAPACK (0x0080)
155 #define VETH_STATE_READY (0x0100)
158 struct veth_msg
*next
;
159 struct veth_frames_data data
;
166 struct veth_lpar_connection
{
168 struct work_struct statemachine_wq
;
169 struct veth_msg
*msgs
;
171 struct veth_cap_data local_caps
;
173 struct kobject kobject
;
174 struct timer_list ack_timer
;
176 struct timer_list reset_timer
;
177 unsigned int reset_timeout
;
178 unsigned long last_contact
;
183 HvLpInstanceId src_inst
;
184 HvLpInstanceId dst_inst
;
185 struct veth_lpevent cap_event
, cap_ack_event
;
186 u16 pending_acks
[VETH_MAX_ACKS_PER_MSG
];
187 u32 num_pending_acks
;
190 struct veth_cap_data remote_caps
;
193 struct veth_msg
*msg_stack_head
;
198 struct net_device_stats stats
;
200 HvLpIndexMap lpar_map
;
202 /* queue_lock protects the stopped_map and dev's queue. */
203 spinlock_t queue_lock
;
204 HvLpIndexMap stopped_map
;
206 /* mcast_gate protects promiscuous, num_mcast & mcast_addr. */
210 u64 mcast_addr
[VETH_MAX_MCAST
];
212 struct kobject kobject
;
215 static HvLpIndex this_lp
;
216 static struct veth_lpar_connection
*veth_cnx
[HVMAXARCHITECTEDLPS
]; /* = 0 */
217 static struct net_device
*veth_dev
[HVMAXARCHITECTEDVIRTUALLANS
]; /* = 0 */
219 static int veth_start_xmit(struct sk_buff
*skb
, struct net_device
*dev
);
220 static void veth_recycle_msg(struct veth_lpar_connection
*, struct veth_msg
*);
221 static void veth_wake_queues(struct veth_lpar_connection
*cnx
);
222 static void veth_stop_queues(struct veth_lpar_connection
*cnx
);
223 static void veth_receive(struct veth_lpar_connection
*, struct veth_lpevent
*);
224 static void veth_release_connection(struct kobject
*kobject
);
225 static void veth_timed_ack(unsigned long ptr
);
226 static void veth_timed_reset(unsigned long ptr
);
232 #define veth_info(fmt, args...) \
233 printk(KERN_INFO DRV_NAME ": " fmt, ## args)
235 #define veth_error(fmt, args...) \
236 printk(KERN_ERR DRV_NAME ": Error: " fmt, ## args)
239 #define veth_debug(fmt, args...) \
240 printk(KERN_DEBUG DRV_NAME ": " fmt, ## args)
242 #define veth_debug(fmt, args...) do {} while (0)
245 /* You must hold the connection's lock when you call this function. */
246 static inline void veth_stack_push(struct veth_lpar_connection
*cnx
,
247 struct veth_msg
*msg
)
249 msg
->next
= cnx
->msg_stack_head
;
250 cnx
->msg_stack_head
= msg
;
253 /* You must hold the connection's lock when you call this function. */
254 static inline struct veth_msg
*veth_stack_pop(struct veth_lpar_connection
*cnx
)
256 struct veth_msg
*msg
;
258 msg
= cnx
->msg_stack_head
;
260 cnx
->msg_stack_head
= cnx
->msg_stack_head
->next
;
265 /* You must hold the connection's lock when you call this function. */
266 static inline int veth_stack_is_empty(struct veth_lpar_connection
*cnx
)
268 return cnx
->msg_stack_head
== NULL
;
271 static inline HvLpEvent_Rc
272 veth_signalevent(struct veth_lpar_connection
*cnx
, u16 subtype
,
273 HvLpEvent_AckInd ackind
, HvLpEvent_AckType acktype
,
275 u64 data1
, u64 data2
, u64 data3
, u64 data4
, u64 data5
)
277 return HvCallEvent_signalLpEventFast(cnx
->remote_lp
,
278 HvLpEvent_Type_VirtualLan
,
279 subtype
, ackind
, acktype
,
282 token
, data1
, data2
, data3
,
286 static inline HvLpEvent_Rc
veth_signaldata(struct veth_lpar_connection
*cnx
,
287 u16 subtype
, u64 token
, void *data
)
289 u64
*p
= (u64
*) data
;
291 return veth_signalevent(cnx
, subtype
, HvLpEvent_AckInd_NoAck
,
292 HvLpEvent_AckType_ImmediateAck
,
293 token
, p
[0], p
[1], p
[2], p
[3], p
[4]);
296 struct veth_allocation
{
301 static void veth_complete_allocation(void *parm
, int number
)
303 struct veth_allocation
*vc
= (struct veth_allocation
*)parm
;
309 static int veth_allocate_events(HvLpIndex rlp
, int number
)
311 struct veth_allocation vc
= { COMPLETION_INITIALIZER(vc
.c
), 0 };
313 mf_allocate_lp_events(rlp
, HvLpEvent_Type_VirtualLan
,
314 sizeof(struct veth_lpevent
), number
,
315 &veth_complete_allocation
, &vc
);
316 wait_for_completion(&vc
.c
);
325 struct veth_cnx_attribute
{
326 struct attribute attr
;
327 ssize_t (*show
)(struct veth_lpar_connection
*, char *buf
);
328 ssize_t (*store
)(struct veth_lpar_connection
*, const char *buf
);
331 static ssize_t
veth_cnx_attribute_show(struct kobject
*kobj
,
332 struct attribute
*attr
, char *buf
)
334 struct veth_cnx_attribute
*cnx_attr
;
335 struct veth_lpar_connection
*cnx
;
337 cnx_attr
= container_of(attr
, struct veth_cnx_attribute
, attr
);
338 cnx
= container_of(kobj
, struct veth_lpar_connection
, kobject
);
343 return cnx_attr
->show(cnx
, buf
);
346 #define CUSTOM_CNX_ATTR(_name, _format, _expression) \
347 static ssize_t _name##_show(struct veth_lpar_connection *cnx, char *buf)\
349 return sprintf(buf, _format, _expression); \
351 struct veth_cnx_attribute veth_cnx_attr_##_name = __ATTR_RO(_name)
353 #define SIMPLE_CNX_ATTR(_name) \
354 CUSTOM_CNX_ATTR(_name, "%lu\n", (unsigned long)cnx->_name)
356 SIMPLE_CNX_ATTR(outstanding_tx
);
357 SIMPLE_CNX_ATTR(remote_lp
);
358 SIMPLE_CNX_ATTR(num_events
);
359 SIMPLE_CNX_ATTR(src_inst
);
360 SIMPLE_CNX_ATTR(dst_inst
);
361 SIMPLE_CNX_ATTR(num_pending_acks
);
362 SIMPLE_CNX_ATTR(num_ack_events
);
363 CUSTOM_CNX_ATTR(ack_timeout
, "%d\n", jiffies_to_msecs(cnx
->ack_timeout
));
364 CUSTOM_CNX_ATTR(reset_timeout
, "%d\n", jiffies_to_msecs(cnx
->reset_timeout
));
365 CUSTOM_CNX_ATTR(state
, "0x%.4lX\n", cnx
->state
);
366 CUSTOM_CNX_ATTR(last_contact
, "%d\n", cnx
->last_contact
?
367 jiffies_to_msecs(jiffies
- cnx
->last_contact
) : 0);
369 #define GET_CNX_ATTR(_name) (&veth_cnx_attr_##_name.attr)
371 static struct attribute
*veth_cnx_default_attrs
[] = {
372 GET_CNX_ATTR(outstanding_tx
),
373 GET_CNX_ATTR(remote_lp
),
374 GET_CNX_ATTR(num_events
),
375 GET_CNX_ATTR(reset_timeout
),
376 GET_CNX_ATTR(last_contact
),
378 GET_CNX_ATTR(src_inst
),
379 GET_CNX_ATTR(dst_inst
),
380 GET_CNX_ATTR(num_pending_acks
),
381 GET_CNX_ATTR(num_ack_events
),
382 GET_CNX_ATTR(ack_timeout
),
386 static struct sysfs_ops veth_cnx_sysfs_ops
= {
387 .show
= veth_cnx_attribute_show
390 static struct kobj_type veth_lpar_connection_ktype
= {
391 .release
= veth_release_connection
,
392 .sysfs_ops
= &veth_cnx_sysfs_ops
,
393 .default_attrs
= veth_cnx_default_attrs
396 struct veth_port_attribute
{
397 struct attribute attr
;
398 ssize_t (*show
)(struct veth_port
*, char *buf
);
399 ssize_t (*store
)(struct veth_port
*, const char *buf
);
402 static ssize_t
veth_port_attribute_show(struct kobject
*kobj
,
403 struct attribute
*attr
, char *buf
)
405 struct veth_port_attribute
*port_attr
;
406 struct veth_port
*port
;
408 port_attr
= container_of(attr
, struct veth_port_attribute
, attr
);
409 port
= container_of(kobj
, struct veth_port
, kobject
);
411 if (!port_attr
->show
)
414 return port_attr
->show(port
, buf
);
417 #define CUSTOM_PORT_ATTR(_name, _format, _expression) \
418 static ssize_t _name##_show(struct veth_port *port, char *buf) \
420 return sprintf(buf, _format, _expression); \
422 struct veth_port_attribute veth_port_attr_##_name = __ATTR_RO(_name)
424 #define SIMPLE_PORT_ATTR(_name) \
425 CUSTOM_PORT_ATTR(_name, "%lu\n", (unsigned long)port->_name)
427 SIMPLE_PORT_ATTR(promiscuous
);
428 SIMPLE_PORT_ATTR(num_mcast
);
429 CUSTOM_PORT_ATTR(lpar_map
, "0x%X\n", port
->lpar_map
);
430 CUSTOM_PORT_ATTR(stopped_map
, "0x%X\n", port
->stopped_map
);
431 CUSTOM_PORT_ATTR(mac_addr
, "0x%lX\n", port
->mac_addr
);
433 #define GET_PORT_ATTR(_name) (&veth_port_attr_##_name.attr)
434 static struct attribute
*veth_port_default_attrs
[] = {
435 GET_PORT_ATTR(mac_addr
),
436 GET_PORT_ATTR(lpar_map
),
437 GET_PORT_ATTR(stopped_map
),
438 GET_PORT_ATTR(promiscuous
),
439 GET_PORT_ATTR(num_mcast
),
443 static struct sysfs_ops veth_port_sysfs_ops
= {
444 .show
= veth_port_attribute_show
447 static struct kobj_type veth_port_ktype
= {
448 .sysfs_ops
= &veth_port_sysfs_ops
,
449 .default_attrs
= veth_port_default_attrs
453 * LPAR connection code
456 static inline void veth_kick_statemachine(struct veth_lpar_connection
*cnx
)
458 schedule_work(&cnx
->statemachine_wq
);
461 static void veth_take_cap(struct veth_lpar_connection
*cnx
,
462 struct veth_lpevent
*event
)
466 spin_lock_irqsave(&cnx
->lock
, flags
);
467 /* Receiving caps may mean the other end has just come up, so
468 * we need to reload the instance ID of the far end */
470 HvCallEvent_getTargetLpInstanceId(cnx
->remote_lp
,
471 HvLpEvent_Type_VirtualLan
);
473 if (cnx
->state
& VETH_STATE_GOTCAPS
) {
474 veth_error("Received a second capabilities from LPAR %d.\n",
476 event
->base_event
.xRc
= HvLpEvent_Rc_BufferNotAvailable
;
477 HvCallEvent_ackLpEvent((struct HvLpEvent
*) event
);
479 memcpy(&cnx
->cap_event
, event
, sizeof(cnx
->cap_event
));
480 cnx
->state
|= VETH_STATE_GOTCAPS
;
481 veth_kick_statemachine(cnx
);
483 spin_unlock_irqrestore(&cnx
->lock
, flags
);
486 static void veth_take_cap_ack(struct veth_lpar_connection
*cnx
,
487 struct veth_lpevent
*event
)
491 spin_lock_irqsave(&cnx
->lock
, flags
);
492 if (cnx
->state
& VETH_STATE_GOTCAPACK
) {
493 veth_error("Received a second capabilities ack from LPAR %d.\n",
496 memcpy(&cnx
->cap_ack_event
, event
,
497 sizeof(&cnx
->cap_ack_event
));
498 cnx
->state
|= VETH_STATE_GOTCAPACK
;
499 veth_kick_statemachine(cnx
);
501 spin_unlock_irqrestore(&cnx
->lock
, flags
);
504 static void veth_take_monitor_ack(struct veth_lpar_connection
*cnx
,
505 struct veth_lpevent
*event
)
509 spin_lock_irqsave(&cnx
->lock
, flags
);
510 veth_debug("cnx %d: lost connection.\n", cnx
->remote_lp
);
512 /* Avoid kicking the statemachine once we're shutdown.
513 * It's unnecessary and it could break veth_stop_connection(). */
515 if (! (cnx
->state
& VETH_STATE_SHUTDOWN
)) {
516 cnx
->state
|= VETH_STATE_RESET
;
517 veth_kick_statemachine(cnx
);
519 spin_unlock_irqrestore(&cnx
->lock
, flags
);
522 static void veth_handle_ack(struct veth_lpevent
*event
)
524 HvLpIndex rlp
= event
->base_event
.xTargetLp
;
525 struct veth_lpar_connection
*cnx
= veth_cnx
[rlp
];
529 switch (event
->base_event
.xSubtype
) {
531 veth_take_cap_ack(cnx
, event
);
533 case VETH_EVENT_MONITOR
:
534 veth_take_monitor_ack(cnx
, event
);
537 veth_error("Unknown ack type %d from LPAR %d.\n",
538 event
->base_event
.xSubtype
, rlp
);
542 static void veth_handle_int(struct veth_lpevent
*event
)
544 HvLpIndex rlp
= event
->base_event
.xSourceLp
;
545 struct veth_lpar_connection
*cnx
= veth_cnx
[rlp
];
551 switch (event
->base_event
.xSubtype
) {
553 veth_take_cap(cnx
, event
);
555 case VETH_EVENT_MONITOR
:
556 /* do nothing... this'll hang out here til we're dead,
557 * and the hypervisor will return it for us. */
559 case VETH_EVENT_FRAMES_ACK
:
560 spin_lock_irqsave(&cnx
->lock
, flags
);
562 for (i
= 0; i
< VETH_MAX_ACKS_PER_MSG
; ++i
) {
563 u16 msgnum
= event
->u
.frames_ack_data
.token
[i
];
565 if (msgnum
< VETH_NUMBUFFERS
) {
566 veth_recycle_msg(cnx
, cnx
->msgs
+ msgnum
);
567 cnx
->outstanding_tx
--;
573 cnx
->last_contact
= jiffies
;
574 veth_wake_queues(cnx
);
577 spin_unlock_irqrestore(&cnx
->lock
, flags
);
579 case VETH_EVENT_FRAMES
:
580 veth_receive(cnx
, event
);
583 veth_error("Unknown interrupt type %d from LPAR %d.\n",
584 event
->base_event
.xSubtype
, rlp
);
588 static void veth_handle_event(struct HvLpEvent
*event
, struct pt_regs
*regs
)
590 struct veth_lpevent
*veth_event
= (struct veth_lpevent
*)event
;
592 if (event
->xFlags
.xFunction
== HvLpEvent_Function_Ack
)
593 veth_handle_ack(veth_event
);
594 else if (event
->xFlags
.xFunction
== HvLpEvent_Function_Int
)
595 veth_handle_int(veth_event
);
598 static int veth_process_caps(struct veth_lpar_connection
*cnx
)
600 struct veth_cap_data
*remote_caps
= &cnx
->remote_caps
;
603 /* Convert timer to jiffies */
604 cnx
->ack_timeout
= remote_caps
->ack_timeout
* HZ
/ 1000000;
606 if ( (remote_caps
->num_buffers
== 0)
607 || (remote_caps
->ack_threshold
> VETH_MAX_ACKS_PER_MSG
)
608 || (remote_caps
->ack_threshold
== 0)
609 || (cnx
->ack_timeout
== 0) ) {
610 veth_error("Received incompatible capabilities from LPAR %d.\n",
612 return HvLpEvent_Rc_InvalidSubtypeData
;
615 num_acks_needed
= (remote_caps
->num_buffers
616 / remote_caps
->ack_threshold
) + 1;
618 /* FIXME: locking on num_ack_events? */
619 if (cnx
->num_ack_events
< num_acks_needed
) {
622 num
= veth_allocate_events(cnx
->remote_lp
,
623 num_acks_needed
-cnx
->num_ack_events
);
625 cnx
->num_ack_events
+= num
;
627 if (cnx
->num_ack_events
< num_acks_needed
) {
628 veth_error("Couldn't allocate enough ack events "
629 "for LPAR %d.\n", cnx
->remote_lp
);
631 return HvLpEvent_Rc_BufferNotAvailable
;
636 return HvLpEvent_Rc_Good
;
639 /* FIXME: The gotos here are a bit dubious */
640 static void veth_statemachine(void *p
)
642 struct veth_lpar_connection
*cnx
= (struct veth_lpar_connection
*)p
;
643 int rlp
= cnx
->remote_lp
;
646 spin_lock_irq(&cnx
->lock
);
649 if (cnx
->state
& VETH_STATE_RESET
) {
650 if (cnx
->state
& VETH_STATE_OPEN
)
651 HvCallEvent_closeLpEventPath(cnx
->remote_lp
,
652 HvLpEvent_Type_VirtualLan
);
655 * Reset ack data. This prevents the ack_timer actually
656 * doing anything, even if it runs one more time when
657 * we drop the lock below.
659 memset(&cnx
->pending_acks
, 0xff, sizeof (cnx
->pending_acks
));
660 cnx
->num_pending_acks
= 0;
662 cnx
->state
&= ~(VETH_STATE_RESET
| VETH_STATE_SENTMON
663 | VETH_STATE_OPEN
| VETH_STATE_SENTCAPS
664 | VETH_STATE_GOTCAPACK
| VETH_STATE_GOTCAPS
665 | VETH_STATE_SENTCAPACK
| VETH_STATE_READY
);
667 /* Clean up any leftover messages */
670 for (i
= 0; i
< VETH_NUMBUFFERS
; ++i
)
671 veth_recycle_msg(cnx
, cnx
->msgs
+ i
);
674 cnx
->outstanding_tx
= 0;
675 veth_wake_queues(cnx
);
677 /* Drop the lock so we can do stuff that might sleep or
678 * take other locks. */
679 spin_unlock_irq(&cnx
->lock
);
681 del_timer_sync(&cnx
->ack_timer
);
682 del_timer_sync(&cnx
->reset_timer
);
684 spin_lock_irq(&cnx
->lock
);
686 if (cnx
->state
& VETH_STATE_RESET
)
689 /* Hack, wait for the other end to reset itself. */
690 if (! (cnx
->state
& VETH_STATE_SHUTDOWN
)) {
691 schedule_delayed_work(&cnx
->statemachine_wq
, 5 * HZ
);
696 if (cnx
->state
& VETH_STATE_SHUTDOWN
)
697 /* It's all over, do nothing */
700 if ( !(cnx
->state
& VETH_STATE_OPEN
) ) {
701 if (! cnx
->msgs
|| (cnx
->num_events
< (2 + VETH_NUMBUFFERS
)) )
704 HvCallEvent_openLpEventPath(rlp
, HvLpEvent_Type_VirtualLan
);
706 HvCallEvent_getSourceLpInstanceId(rlp
,
707 HvLpEvent_Type_VirtualLan
);
709 HvCallEvent_getTargetLpInstanceId(rlp
,
710 HvLpEvent_Type_VirtualLan
);
711 cnx
->state
|= VETH_STATE_OPEN
;
714 if ( (cnx
->state
& VETH_STATE_OPEN
)
715 && !(cnx
->state
& VETH_STATE_SENTMON
) ) {
716 rc
= veth_signalevent(cnx
, VETH_EVENT_MONITOR
,
717 HvLpEvent_AckInd_DoAck
,
718 HvLpEvent_AckType_DeferredAck
,
721 if (rc
== HvLpEvent_Rc_Good
) {
722 cnx
->state
|= VETH_STATE_SENTMON
;
724 if ( (rc
!= HvLpEvent_Rc_PartitionDead
)
725 && (rc
!= HvLpEvent_Rc_PathClosed
) )
726 veth_error("Error sending monitor to LPAR %d, "
727 "rc = %d\n", rlp
, rc
);
729 /* Oh well, hope we get a cap from the other
730 * end and do better when that kicks us */
735 if ( (cnx
->state
& VETH_STATE_OPEN
)
736 && !(cnx
->state
& VETH_STATE_SENTCAPS
)) {
737 u64
*rawcap
= (u64
*)&cnx
->local_caps
;
739 rc
= veth_signalevent(cnx
, VETH_EVENT_CAP
,
740 HvLpEvent_AckInd_DoAck
,
741 HvLpEvent_AckType_ImmediateAck
,
742 0, rawcap
[0], rawcap
[1], rawcap
[2],
743 rawcap
[3], rawcap
[4]);
745 if (rc
== HvLpEvent_Rc_Good
) {
746 cnx
->state
|= VETH_STATE_SENTCAPS
;
748 if ( (rc
!= HvLpEvent_Rc_PartitionDead
)
749 && (rc
!= HvLpEvent_Rc_PathClosed
) )
750 veth_error("Error sending caps to LPAR %d, "
751 "rc = %d\n", rlp
, rc
);
753 /* Oh well, hope we get a cap from the other
754 * end and do better when that kicks us */
759 if ((cnx
->state
& VETH_STATE_GOTCAPS
)
760 && !(cnx
->state
& VETH_STATE_SENTCAPACK
)) {
761 struct veth_cap_data
*remote_caps
= &cnx
->remote_caps
;
763 memcpy(remote_caps
, &cnx
->cap_event
.u
.caps_data
,
764 sizeof(*remote_caps
));
766 spin_unlock_irq(&cnx
->lock
);
767 rc
= veth_process_caps(cnx
);
768 spin_lock_irq(&cnx
->lock
);
770 /* We dropped the lock, so recheck for anything which
771 * might mess us up */
772 if (cnx
->state
& (VETH_STATE_RESET
|VETH_STATE_SHUTDOWN
))
775 cnx
->cap_event
.base_event
.xRc
= rc
;
776 HvCallEvent_ackLpEvent((struct HvLpEvent
*)&cnx
->cap_event
);
777 if (rc
== HvLpEvent_Rc_Good
)
778 cnx
->state
|= VETH_STATE_SENTCAPACK
;
783 if ((cnx
->state
& VETH_STATE_GOTCAPACK
)
784 && (cnx
->state
& VETH_STATE_GOTCAPS
)
785 && !(cnx
->state
& VETH_STATE_READY
)) {
786 if (cnx
->cap_ack_event
.base_event
.xRc
== HvLpEvent_Rc_Good
) {
787 /* Start the ACK timer */
788 cnx
->ack_timer
.expires
= jiffies
+ cnx
->ack_timeout
;
789 add_timer(&cnx
->ack_timer
);
790 cnx
->state
|= VETH_STATE_READY
;
792 veth_error("Caps rejected by LPAR %d, rc = %d\n",
793 rlp
, cnx
->cap_ack_event
.base_event
.xRc
);
799 spin_unlock_irq(&cnx
->lock
);
803 /* FIXME: we get here if something happens we really can't
804 * cope with. The link will never work once we get here, and
805 * all we can do is not lock the rest of the system up */
806 veth_error("Unrecoverable error on connection to LPAR %d, shutting down"
807 " (state = 0x%04lx)\n", rlp
, cnx
->state
);
808 cnx
->state
|= VETH_STATE_SHUTDOWN
;
809 spin_unlock_irq(&cnx
->lock
);
812 static int veth_init_connection(u8 rlp
)
814 struct veth_lpar_connection
*cnx
;
815 struct veth_msg
*msgs
;
818 if ( (rlp
== this_lp
)
819 || ! HvLpConfig_doLpsCommunicateOnVirtualLan(this_lp
, rlp
) )
822 cnx
= kmalloc(sizeof(*cnx
), GFP_KERNEL
);
825 memset(cnx
, 0, sizeof(*cnx
));
827 cnx
->remote_lp
= rlp
;
828 spin_lock_init(&cnx
->lock
);
829 INIT_WORK(&cnx
->statemachine_wq
, veth_statemachine
, cnx
);
831 init_timer(&cnx
->ack_timer
);
832 cnx
->ack_timer
.function
= veth_timed_ack
;
833 cnx
->ack_timer
.data
= (unsigned long) cnx
;
835 init_timer(&cnx
->reset_timer
);
836 cnx
->reset_timer
.function
= veth_timed_reset
;
837 cnx
->reset_timer
.data
= (unsigned long) cnx
;
838 cnx
->reset_timeout
= 5 * HZ
* (VETH_ACKTIMEOUT
/ 1000000);
840 memset(&cnx
->pending_acks
, 0xff, sizeof (cnx
->pending_acks
));
844 /* This gets us 1 reference, which is held on behalf of the driver
845 * infrastructure. It's released at module unload. */
846 kobject_init(&cnx
->kobject
);
847 cnx
->kobject
.ktype
= &veth_lpar_connection_ktype
;
848 rc
= kobject_set_name(&cnx
->kobject
, "cnx%.2d", rlp
);
852 msgs
= kmalloc(VETH_NUMBUFFERS
* sizeof(struct veth_msg
), GFP_KERNEL
);
854 veth_error("Can't allocate buffers for LPAR %d.\n", rlp
);
859 memset(msgs
, 0, VETH_NUMBUFFERS
* sizeof(struct veth_msg
));
861 for (i
= 0; i
< VETH_NUMBUFFERS
; i
++) {
863 veth_stack_push(cnx
, msgs
+ i
);
866 cnx
->num_events
= veth_allocate_events(rlp
, 2 + VETH_NUMBUFFERS
);
868 if (cnx
->num_events
< (2 + VETH_NUMBUFFERS
)) {
869 veth_error("Can't allocate enough events for LPAR %d.\n", rlp
);
873 cnx
->local_caps
.num_buffers
= VETH_NUMBUFFERS
;
874 cnx
->local_caps
.ack_threshold
= ACK_THRESHOLD
;
875 cnx
->local_caps
.ack_timeout
= VETH_ACKTIMEOUT
;
880 static void veth_stop_connection(struct veth_lpar_connection
*cnx
)
885 spin_lock_irq(&cnx
->lock
);
886 cnx
->state
|= VETH_STATE_RESET
| VETH_STATE_SHUTDOWN
;
887 veth_kick_statemachine(cnx
);
888 spin_unlock_irq(&cnx
->lock
);
890 /* There's a slim chance the reset code has just queued the
891 * statemachine to run in five seconds. If so we need to cancel
892 * that and requeue the work to run now. */
893 if (cancel_delayed_work(&cnx
->statemachine_wq
)) {
894 spin_lock_irq(&cnx
->lock
);
895 veth_kick_statemachine(cnx
);
896 spin_unlock_irq(&cnx
->lock
);
899 /* Wait for the state machine to run. */
900 flush_scheduled_work();
903 static void veth_destroy_connection(struct veth_lpar_connection
*cnx
)
908 if (cnx
->num_events
> 0)
909 mf_deallocate_lp_events(cnx
->remote_lp
,
910 HvLpEvent_Type_VirtualLan
,
913 if (cnx
->num_ack_events
> 0)
914 mf_deallocate_lp_events(cnx
->remote_lp
,
915 HvLpEvent_Type_VirtualLan
,
920 veth_cnx
[cnx
->remote_lp
] = NULL
;
924 static void veth_release_connection(struct kobject
*kobj
)
926 struct veth_lpar_connection
*cnx
;
927 cnx
= container_of(kobj
, struct veth_lpar_connection
, kobject
);
928 veth_stop_connection(cnx
);
929 veth_destroy_connection(cnx
);
936 static int veth_open(struct net_device
*dev
)
938 struct veth_port
*port
= (struct veth_port
*) dev
->priv
;
940 memset(&port
->stats
, 0, sizeof (port
->stats
));
941 netif_start_queue(dev
);
945 static int veth_close(struct net_device
*dev
)
947 netif_stop_queue(dev
);
951 static struct net_device_stats
*veth_get_stats(struct net_device
*dev
)
953 struct veth_port
*port
= (struct veth_port
*) dev
->priv
;
958 static int veth_change_mtu(struct net_device
*dev
, int new_mtu
)
960 if ((new_mtu
< 68) || (new_mtu
> VETH_MAX_MTU
))
966 static void veth_set_multicast_list(struct net_device
*dev
)
968 struct veth_port
*port
= (struct veth_port
*) dev
->priv
;
971 write_lock_irqsave(&port
->mcast_gate
, flags
);
973 if ((dev
->flags
& IFF_PROMISC
) || (dev
->flags
& IFF_ALLMULTI
) ||
974 (dev
->mc_count
> VETH_MAX_MCAST
)) {
975 port
->promiscuous
= 1;
977 struct dev_mc_list
*dmi
= dev
->mc_list
;
980 port
->promiscuous
= 0;
985 for (i
= 0; i
< dev
->mc_count
; i
++) {
986 u8
*addr
= dmi
->dmi_addr
;
989 if (addr
[0] & 0x01) {/* multicast address? */
990 memcpy(&xaddr
, addr
, ETH_ALEN
);
991 port
->mcast_addr
[port
->num_mcast
] = xaddr
;
998 write_unlock_irqrestore(&port
->mcast_gate
, flags
);
1001 static void veth_get_drvinfo(struct net_device
*dev
, struct ethtool_drvinfo
*info
)
1003 strncpy(info
->driver
, DRV_NAME
, sizeof(info
->driver
) - 1);
1004 info
->driver
[sizeof(info
->driver
) - 1] = '\0';
1005 strncpy(info
->version
, DRV_VERSION
, sizeof(info
->version
) - 1);
1006 info
->version
[sizeof(info
->version
) - 1] = '\0';
1009 static int veth_get_settings(struct net_device
*dev
, struct ethtool_cmd
*ecmd
)
1011 ecmd
->supported
= (SUPPORTED_1000baseT_Full
1012 | SUPPORTED_Autoneg
| SUPPORTED_FIBRE
);
1013 ecmd
->advertising
= (SUPPORTED_1000baseT_Full
1014 | SUPPORTED_Autoneg
| SUPPORTED_FIBRE
);
1015 ecmd
->port
= PORT_FIBRE
;
1016 ecmd
->transceiver
= XCVR_INTERNAL
;
1017 ecmd
->phy_address
= 0;
1018 ecmd
->speed
= SPEED_1000
;
1019 ecmd
->duplex
= DUPLEX_FULL
;
1020 ecmd
->autoneg
= AUTONEG_ENABLE
;
1021 ecmd
->maxtxpkt
= 120;
1022 ecmd
->maxrxpkt
= 120;
1026 static u32
veth_get_link(struct net_device
*dev
)
1031 static struct ethtool_ops ops
= {
1032 .get_drvinfo
= veth_get_drvinfo
,
1033 .get_settings
= veth_get_settings
,
1034 .get_link
= veth_get_link
,
1037 static struct net_device
* __init
veth_probe_one(int vlan
, struct device
*vdev
)
1039 struct net_device
*dev
;
1040 struct veth_port
*port
;
1043 dev
= alloc_etherdev(sizeof (struct veth_port
));
1045 veth_error("Unable to allocate net_device structure!\n");
1049 port
= (struct veth_port
*) dev
->priv
;
1051 spin_lock_init(&port
->queue_lock
);
1052 rwlock_init(&port
->mcast_gate
);
1053 port
->stopped_map
= 0;
1055 for (i
= 0; i
< HVMAXARCHITECTEDLPS
; i
++) {
1056 HvLpVirtualLanIndexMap map
;
1060 map
= HvLpConfig_getVirtualLanIndexMapForLp(i
);
1061 if (map
& (0x8000 >> vlan
))
1062 port
->lpar_map
|= (1 << i
);
1066 dev
->dev_addr
[0] = 0x02;
1067 dev
->dev_addr
[1] = 0x01;
1068 dev
->dev_addr
[2] = 0xff;
1069 dev
->dev_addr
[3] = vlan
;
1070 dev
->dev_addr
[4] = 0xff;
1071 dev
->dev_addr
[5] = this_lp
;
1073 dev
->mtu
= VETH_MAX_MTU
;
1075 memcpy(&port
->mac_addr
, dev
->dev_addr
, 6);
1077 dev
->open
= veth_open
;
1078 dev
->hard_start_xmit
= veth_start_xmit
;
1079 dev
->stop
= veth_close
;
1080 dev
->get_stats
= veth_get_stats
;
1081 dev
->change_mtu
= veth_change_mtu
;
1082 dev
->set_mac_address
= NULL
;
1083 dev
->set_multicast_list
= veth_set_multicast_list
;
1084 SET_ETHTOOL_OPS(dev
, &ops
);
1086 SET_NETDEV_DEV(dev
, vdev
);
1088 rc
= register_netdev(dev
);
1090 veth_error("Failed registering net device for vlan%d.\n", vlan
);
1095 kobject_init(&port
->kobject
);
1096 port
->kobject
.parent
= &dev
->class_dev
.kobj
;
1097 port
->kobject
.ktype
= &veth_port_ktype
;
1098 kobject_set_name(&port
->kobject
, "veth_port");
1099 if (0 != kobject_add(&port
->kobject
))
1100 veth_error("Failed adding port for %s to sysfs.\n", dev
->name
);
1102 veth_info("%s attached to iSeries vlan %d (LPAR map = 0x%.4X)\n",
1103 dev
->name
, vlan
, port
->lpar_map
);
1112 static int veth_transmit_to_one(struct sk_buff
*skb
, HvLpIndex rlp
,
1113 struct net_device
*dev
)
1115 struct veth_lpar_connection
*cnx
= veth_cnx
[rlp
];
1116 struct veth_port
*port
= (struct veth_port
*) dev
->priv
;
1118 struct veth_msg
*msg
= NULL
;
1119 unsigned long flags
;
1124 spin_lock_irqsave(&cnx
->lock
, flags
);
1126 if (! (cnx
->state
& VETH_STATE_READY
))
1129 if ((skb
->len
- ETH_HLEN
) > VETH_MAX_MTU
)
1132 msg
= veth_stack_pop(cnx
);
1137 msg
->skb
= skb_get(skb
);
1139 msg
->data
.addr
[0] = dma_map_single(port
->dev
, skb
->data
,
1140 skb
->len
, DMA_TO_DEVICE
);
1142 if (dma_mapping_error(msg
->data
.addr
[0]))
1143 goto recycle_and_drop
;
1145 msg
->dev
= port
->dev
;
1146 msg
->data
.len
[0] = skb
->len
;
1147 msg
->data
.eofmask
= 1 << VETH_EOF_SHIFT
;
1149 rc
= veth_signaldata(cnx
, VETH_EVENT_FRAMES
, msg
->token
, &msg
->data
);
1151 if (rc
!= HvLpEvent_Rc_Good
)
1152 goto recycle_and_drop
;
1154 /* If the timer's not already running, start it now. */
1155 if (0 == cnx
->outstanding_tx
)
1156 mod_timer(&cnx
->reset_timer
, jiffies
+ cnx
->reset_timeout
);
1158 cnx
->last_contact
= jiffies
;
1159 cnx
->outstanding_tx
++;
1161 if (veth_stack_is_empty(cnx
))
1162 veth_stop_queues(cnx
);
1165 spin_unlock_irqrestore(&cnx
->lock
, flags
);
1169 veth_recycle_msg(cnx
, msg
);
1171 spin_unlock_irqrestore(&cnx
->lock
, flags
);
1175 static void veth_transmit_to_many(struct sk_buff
*skb
,
1176 HvLpIndexMap lpmask
,
1177 struct net_device
*dev
)
1179 struct veth_port
*port
= (struct veth_port
*) dev
->priv
;
1180 int i
, success
, error
;
1182 success
= error
= 0;
1184 for (i
= 0; i
< HVMAXARCHITECTEDLPS
; i
++) {
1185 if ((lpmask
& (1 << i
)) == 0)
1188 if (veth_transmit_to_one(skb
, i
, dev
))
1195 port
->stats
.tx_errors
++;
1198 port
->stats
.tx_packets
++;
1199 port
->stats
.tx_bytes
+= skb
->len
;
1203 static int veth_start_xmit(struct sk_buff
*skb
, struct net_device
*dev
)
1205 unsigned char *frame
= skb
->data
;
1206 struct veth_port
*port
= (struct veth_port
*) dev
->priv
;
1207 HvLpIndexMap lpmask
;
1209 if (! (frame
[0] & 0x01)) {
1210 /* unicast packet */
1211 HvLpIndex rlp
= frame
[5];
1213 if ( ! ((1 << rlp
) & port
->lpar_map
) ) {
1220 lpmask
= port
->lpar_map
;
1223 veth_transmit_to_many(skb
, lpmask
, dev
);
1230 /* You must hold the connection's lock when you call this function. */
1231 static void veth_recycle_msg(struct veth_lpar_connection
*cnx
,
1232 struct veth_msg
*msg
)
1234 u32 dma_address
, dma_length
;
1238 dma_address
= msg
->data
.addr
[0];
1239 dma_length
= msg
->data
.len
[0];
1241 if (!dma_mapping_error(dma_address
))
1242 dma_unmap_single(msg
->dev
, dma_address
, dma_length
,
1246 dev_kfree_skb_any(msg
->skb
);
1250 memset(&msg
->data
, 0, sizeof(msg
->data
));
1251 veth_stack_push(cnx
, msg
);
1252 } else if (cnx
->state
& VETH_STATE_OPEN
) {
1253 veth_error("Non-pending frame (# %d) acked by LPAR %d.\n",
1254 cnx
->remote_lp
, msg
->token
);
1258 static void veth_wake_queues(struct veth_lpar_connection
*cnx
)
1262 for (i
= 0; i
< HVMAXARCHITECTEDVIRTUALLANS
; i
++) {
1263 struct net_device
*dev
= veth_dev
[i
];
1264 struct veth_port
*port
;
1265 unsigned long flags
;
1270 port
= (struct veth_port
*)dev
->priv
;
1272 if (! (port
->lpar_map
& (1<<cnx
->remote_lp
)))
1275 spin_lock_irqsave(&port
->queue_lock
, flags
);
1277 port
->stopped_map
&= ~(1 << cnx
->remote_lp
);
1279 if (0 == port
->stopped_map
&& netif_queue_stopped(dev
)) {
1280 veth_debug("cnx %d: woke queue for %s.\n",
1281 cnx
->remote_lp
, dev
->name
);
1282 netif_wake_queue(dev
);
1284 spin_unlock_irqrestore(&port
->queue_lock
, flags
);
1288 static void veth_stop_queues(struct veth_lpar_connection
*cnx
)
1292 for (i
= 0; i
< HVMAXARCHITECTEDVIRTUALLANS
; i
++) {
1293 struct net_device
*dev
= veth_dev
[i
];
1294 struct veth_port
*port
;
1299 port
= (struct veth_port
*)dev
->priv
;
1301 /* If this cnx is not on the vlan for this port, continue */
1302 if (! (port
->lpar_map
& (1 << cnx
->remote_lp
)))
1305 spin_lock(&port
->queue_lock
);
1307 netif_stop_queue(dev
);
1308 port
->stopped_map
|= (1 << cnx
->remote_lp
);
1310 veth_debug("cnx %d: stopped queue for %s, map = 0x%x.\n",
1311 cnx
->remote_lp
, dev
->name
, port
->stopped_map
);
1313 spin_unlock(&port
->queue_lock
);
1317 static void veth_timed_reset(unsigned long ptr
)
1319 struct veth_lpar_connection
*cnx
= (struct veth_lpar_connection
*)ptr
;
1320 unsigned long trigger_time
, flags
;
1322 /* FIXME is it possible this fires after veth_stop_connection()?
1323 * That would reschedule the statemachine for 5 seconds and probably
1324 * execute it after the module's been unloaded. Hmm. */
1326 spin_lock_irqsave(&cnx
->lock
, flags
);
1328 if (cnx
->outstanding_tx
> 0) {
1329 trigger_time
= cnx
->last_contact
+ cnx
->reset_timeout
;
1331 if (trigger_time
< jiffies
) {
1332 cnx
->state
|= VETH_STATE_RESET
;
1333 veth_kick_statemachine(cnx
);
1334 veth_error("%d packets not acked by LPAR %d within %d "
1335 "seconds, resetting.\n",
1336 cnx
->outstanding_tx
, cnx
->remote_lp
,
1337 cnx
->reset_timeout
/ HZ
);
1339 /* Reschedule the timer */
1340 trigger_time
= jiffies
+ cnx
->reset_timeout
;
1341 mod_timer(&cnx
->reset_timer
, trigger_time
);
1345 spin_unlock_irqrestore(&cnx
->lock
, flags
);
1352 static inline int veth_frame_wanted(struct veth_port
*port
, u64 mac_addr
)
1356 unsigned long flags
;
1358 if ( (mac_addr
== port
->mac_addr
) || (mac_addr
== 0xffffffffffff0000) )
1361 read_lock_irqsave(&port
->mcast_gate
, flags
);
1363 if (port
->promiscuous
) {
1368 for (i
= 0; i
< port
->num_mcast
; ++i
) {
1369 if (port
->mcast_addr
[i
] == mac_addr
) {
1376 read_unlock_irqrestore(&port
->mcast_gate
, flags
);
1386 #define VETH_MAX_PAGES_PER_FRAME ( (VETH_MAX_MTU+PAGE_SIZE-2)/PAGE_SIZE + 1 )
1388 static inline void veth_build_dma_list(struct dma_chunk
*list
,
1389 unsigned char *p
, unsigned long length
)
1394 /* FIXME: skbs are continguous in real addresses. Do we
1395 * really need to break it into PAGE_SIZE chunks, or can we do
1396 * it just at the granularity of iSeries real->absolute
1397 * mapping? Indeed, given the way the allocator works, can we
1398 * count on them being absolutely contiguous? */
1399 list
[0].addr
= ISERIES_HV_ADDR(p
);
1400 list
[0].size
= min(length
,
1401 PAGE_SIZE
- ((unsigned long)p
& ~PAGE_MASK
));
1403 done
= list
[0].size
;
1404 while (done
< length
) {
1405 list
[i
].addr
= ISERIES_HV_ADDR(p
+ done
);
1406 list
[i
].size
= min(length
-done
, PAGE_SIZE
);
1407 done
+= list
[i
].size
;
1412 static void veth_flush_acks(struct veth_lpar_connection
*cnx
)
1416 rc
= veth_signaldata(cnx
, VETH_EVENT_FRAMES_ACK
,
1417 0, &cnx
->pending_acks
);
1419 if (rc
!= HvLpEvent_Rc_Good
)
1420 veth_error("Failed acking frames from LPAR %d, rc = %d\n",
1421 cnx
->remote_lp
, (int)rc
);
1423 cnx
->num_pending_acks
= 0;
1424 memset(&cnx
->pending_acks
, 0xff, sizeof(cnx
->pending_acks
));
1427 static void veth_receive(struct veth_lpar_connection
*cnx
,
1428 struct veth_lpevent
*event
)
1430 struct veth_frames_data
*senddata
= &event
->u
.frames_data
;
1433 unsigned long flags
;
1438 struct sk_buff
*skb
;
1439 struct dma_chunk local_list
[VETH_MAX_PAGES_PER_FRAME
];
1440 struct dma_chunk remote_list
[VETH_MAX_FRAMES_PER_MSG
];
1442 HvLpVirtualLanIndex vlan
;
1443 struct net_device
*dev
;
1444 struct veth_port
*port
;
1446 /* FIXME: do we need this? */
1447 memset(local_list
, 0, sizeof(local_list
));
1448 memset(remote_list
, 0, sizeof(VETH_MAX_FRAMES_PER_MSG
));
1450 /* a 0 address marks the end of the valid entries */
1451 if (senddata
->addr
[startchunk
] == 0)
1454 /* make sure that we have at least 1 EOF entry in the
1455 * remaining entries */
1456 if (! (senddata
->eofmask
>> (startchunk
+ VETH_EOF_SHIFT
))) {
1457 veth_error("Missing EOF fragment in event "
1458 "eofmask = 0x%x startchunk = %d\n",
1459 (unsigned)senddata
->eofmask
,
1464 /* build list of chunks in this frame */
1467 remote_list
[nchunks
].addr
=
1468 (u64
) senddata
->addr
[startchunk
+nchunks
] << 32;
1469 remote_list
[nchunks
].size
=
1470 senddata
->len
[startchunk
+nchunks
];
1471 length
+= remote_list
[nchunks
].size
;
1472 } while (! (senddata
->eofmask
&
1473 (1 << (VETH_EOF_SHIFT
+ startchunk
+ nchunks
++))));
1475 /* length == total length of all chunks */
1476 /* nchunks == # of chunks in this frame */
1478 if ((length
- ETH_HLEN
) > VETH_MAX_MTU
) {
1479 veth_error("Received oversize frame from LPAR %d "
1481 cnx
->remote_lp
, length
);
1485 skb
= alloc_skb(length
, GFP_ATOMIC
);
1489 veth_build_dma_list(local_list
, skb
->data
, length
);
1491 rc
= HvCallEvent_dmaBufList(HvLpEvent_Type_VirtualLan
,
1492 event
->base_event
.xSourceLp
,
1493 HvLpDma_Direction_RemoteToLocal
,
1496 HvLpDma_AddressType_RealAddress
,
1497 HvLpDma_AddressType_TceIndex
,
1498 ISERIES_HV_ADDR(&local_list
),
1499 ISERIES_HV_ADDR(&remote_list
),
1501 if (rc
!= HvLpDma_Rc_Good
) {
1502 dev_kfree_skb_irq(skb
);
1506 vlan
= skb
->data
[9];
1507 dev
= veth_dev
[vlan
];
1510 * Some earlier versions of the driver sent
1511 * broadcasts down all connections, even to lpars
1512 * that weren't on the relevant vlan. So ignore
1513 * packets belonging to a vlan we're not on.
1514 * We can also be here if we receive packets while
1515 * the driver is going down, because then dev is NULL.
1517 dev_kfree_skb_irq(skb
);
1521 port
= (struct veth_port
*)dev
->priv
;
1522 dest
= *((u64
*) skb
->data
) & 0xFFFFFFFFFFFF0000;
1524 if ((vlan
> HVMAXARCHITECTEDVIRTUALLANS
) || !port
) {
1525 dev_kfree_skb_irq(skb
);
1528 if (! veth_frame_wanted(port
, dest
)) {
1529 dev_kfree_skb_irq(skb
);
1533 skb_put(skb
, length
);
1535 skb
->protocol
= eth_type_trans(skb
, dev
);
1536 skb
->ip_summed
= CHECKSUM_NONE
;
1537 netif_rx(skb
); /* send it up */
1538 port
->stats
.rx_packets
++;
1539 port
->stats
.rx_bytes
+= length
;
1540 } while (startchunk
+= nchunks
, startchunk
< VETH_MAX_FRAMES_PER_MSG
);
1543 spin_lock_irqsave(&cnx
->lock
, flags
);
1544 BUG_ON(cnx
->num_pending_acks
> VETH_MAX_ACKS_PER_MSG
);
1546 cnx
->pending_acks
[cnx
->num_pending_acks
++] =
1547 event
->base_event
.xCorrelationToken
;
1549 if ( (cnx
->num_pending_acks
>= cnx
->remote_caps
.ack_threshold
)
1550 || (cnx
->num_pending_acks
>= VETH_MAX_ACKS_PER_MSG
) )
1551 veth_flush_acks(cnx
);
1553 spin_unlock_irqrestore(&cnx
->lock
, flags
);
1556 static void veth_timed_ack(unsigned long ptr
)
1558 struct veth_lpar_connection
*cnx
= (struct veth_lpar_connection
*) ptr
;
1559 unsigned long flags
;
1561 /* Ack all the events */
1562 spin_lock_irqsave(&cnx
->lock
, flags
);
1563 if (cnx
->num_pending_acks
> 0)
1564 veth_flush_acks(cnx
);
1566 /* Reschedule the timer */
1567 cnx
->ack_timer
.expires
= jiffies
+ cnx
->ack_timeout
;
1568 add_timer(&cnx
->ack_timer
);
1569 spin_unlock_irqrestore(&cnx
->lock
, flags
);
1572 static int veth_remove(struct vio_dev
*vdev
)
1574 struct veth_lpar_connection
*cnx
;
1575 struct net_device
*dev
;
1576 struct veth_port
*port
;
1579 dev
= veth_dev
[vdev
->unit_address
];
1584 port
= netdev_priv(dev
);
1586 for (i
= 0; i
< HVMAXARCHITECTEDLPS
; i
++) {
1589 if (cnx
&& (port
->lpar_map
& (1 << i
))) {
1590 /* Drop our reference to connections on our VLAN */
1591 kobject_put(&cnx
->kobject
);
1595 veth_dev
[vdev
->unit_address
] = NULL
;
1596 kobject_del(&port
->kobject
);
1597 kobject_put(&port
->kobject
);
1598 unregister_netdev(dev
);
1604 static int veth_probe(struct vio_dev
*vdev
, const struct vio_device_id
*id
)
1606 int i
= vdev
->unit_address
;
1607 struct net_device
*dev
;
1608 struct veth_port
*port
;
1610 dev
= veth_probe_one(i
, &vdev
->dev
);
1617 port
= (struct veth_port
*)netdev_priv(dev
);
1619 /* Start the state machine on each connection on this vlan. If we're
1620 * the first dev to do so this will commence link negotiation */
1621 for (i
= 0; i
< HVMAXARCHITECTEDLPS
; i
++) {
1622 struct veth_lpar_connection
*cnx
;
1624 if (! (port
->lpar_map
& (1 << i
)))
1631 kobject_get(&cnx
->kobject
);
1632 veth_kick_statemachine(cnx
);
1639 * veth_device_table: Used by vio.c to match devices that we
1642 static struct vio_device_id veth_device_table
[] __devinitdata
= {
1646 MODULE_DEVICE_TABLE(vio
, veth_device_table
);
1648 static struct vio_driver veth_driver
= {
1650 .id_table
= veth_device_table
,
1651 .probe
= veth_probe
,
1652 .remove
= veth_remove
1656 * Module initialization/cleanup
1659 void __exit
veth_module_cleanup(void)
1662 struct veth_lpar_connection
*cnx
;
1664 /* Disconnect our "irq" to stop events coming from the Hypervisor. */
1665 HvLpEvent_unregisterHandler(HvLpEvent_Type_VirtualLan
);
1667 /* Make sure any work queued from Hypervisor callbacks is finished. */
1668 flush_scheduled_work();
1670 for (i
= 0; i
< HVMAXARCHITECTEDLPS
; ++i
) {
1676 /* Remove the connection from sysfs */
1677 kobject_del(&cnx
->kobject
);
1678 /* Drop the driver's reference to the connection */
1679 kobject_put(&cnx
->kobject
);
1682 /* Unregister the driver, which will close all the netdevs and stop
1683 * the connections when they're no longer referenced. */
1684 vio_unregister_driver(&veth_driver
);
1686 module_exit(veth_module_cleanup
);
1688 int __init
veth_module_init(void)
1693 this_lp
= HvLpConfig_getLpIndex_outline();
1695 for (i
= 0; i
< HVMAXARCHITECTEDLPS
; ++i
) {
1696 rc
= veth_init_connection(i
);
1701 HvLpEvent_registerHandler(HvLpEvent_Type_VirtualLan
,
1702 &veth_handle_event
);
1704 rc
= vio_register_driver(&veth_driver
);
1708 for (i
= 0; i
< HVMAXARCHITECTEDLPS
; ++i
) {
1709 struct kobject
*kobj
;
1714 kobj
= &veth_cnx
[i
]->kobject
;
1715 kobj
->parent
= &veth_driver
.driver
.kobj
;
1716 /* If the add failes, complain but otherwise continue */
1717 if (0 != kobject_add(kobj
))
1718 veth_error("cnx %d: Failed adding to sysfs.\n", i
);
1724 for (i
= 0; i
< HVMAXARCHITECTEDLPS
; ++i
) {
1725 veth_destroy_connection(veth_cnx
[i
]);
1730 module_init(veth_module_init
);