blob: c46864d626b28e44b7e25669ea319d1f74a2dd53 [file] [log] [blame]
Linus Torvalds1da177e2005-04-16 15:20:36 -07001/* File veth.c created by Kyle A. Lucke on Mon Aug 7 2000. */
2/*
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.
Michael Ellerman269cd382005-09-06 10:39:38 +10007 * Copyright (C) 2004-2005 Michael Ellerman, IBM Corporation.
Linus Torvalds1da177e2005-04-16 15:20:36 -07008 *
9 * This program is free software; you can redistribute it and/or
10 * modify it under the terms of the GNU General Public License as
11 * published by the Free Software Foundation; either version 2 of the
12 * License, or (at your option) any later version.
13 *
14 * This program is distributed in the hope that it will be useful, but
15 * WITHOUT ANY WARRANTY; without even the implied warranty of
16 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
17 * General Public License for more details.
18 *
19 * You should have received a copy of the GNU General Public License
20 * along with this program; if not, write to the Free Software
21 * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307
22 * USA
23 *
24 *
25 * This module implements the virtual ethernet device for iSeries LPAR
26 * Linux. It uses hypervisor message passing to implement an
27 * ethernet-like network device communicating between partitions on
28 * the iSeries.
29 *
30 * The iSeries LPAR hypervisor currently allows for up to 16 different
31 * virtual ethernets. These are all dynamically configurable on
32 * OS/400 partitions, but dynamic configuration is not supported under
33 * Linux yet. An ethXX network device will be created for each
34 * virtual ethernet this partition is connected to.
35 *
36 * - This driver is responsible for routing packets to and from other
37 * partitions. The MAC addresses used by the virtual ethernets
38 * contains meaning and must not be modified.
39 *
40 * - Having 2 virtual ethernets to the same remote partition DOES NOT
41 * double the available bandwidth. The 2 devices will share the
42 * available hypervisor bandwidth.
43 *
44 * - If you send a packet to your own mac address, it will just be
45 * dropped, you won't get it on the receive side.
46 *
47 * - Multicast is implemented by sending the frame frame to every
48 * other partition. It is the responsibility of the receiving
49 * partition to filter the addresses desired.
50 *
51 * Tunable parameters:
52 *
53 * VETH_NUMBUFFERS: This compile time option defaults to 120. It
54 * controls how much memory Linux will allocate per remote partition
55 * it is communicating with. It can be thought of as the maximum
56 * number of packets outstanding to a remote partition at a time.
57 */
58
Linus Torvalds1da177e2005-04-16 15:20:36 -070059#include <linux/module.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070060#include <linux/types.h>
61#include <linux/errno.h>
62#include <linux/ioport.h>
63#include <linux/kernel.h>
64#include <linux/netdevice.h>
65#include <linux/etherdevice.h>
66#include <linux/skbuff.h>
67#include <linux/init.h>
68#include <linux/delay.h>
69#include <linux/mm.h>
70#include <linux/ethtool.h>
Stephen Rothwell07fb3f42006-05-15 13:39:47 +100071#include <linux/if_ether.h>
Stephen Rothwell426c1a12005-10-14 14:51:42 +100072
73#include <asm/abs_addr.h>
Kelly Dalybbc8b622005-11-02 15:10:38 +110074#include <asm/iseries/mf.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070075#include <asm/uaccess.h>
Stephen Rothwell687d18a2006-12-15 15:42:50 +110076#include <asm/firmware.h>
Kelly Daly15b17182005-11-02 11:55:28 +110077#include <asm/iseries/hv_lp_config.h>
Kelly Daly1ec65d72005-11-02 13:46:07 +110078#include <asm/iseries/hv_types.h>
Kelly Dalye45423e2005-11-02 12:08:31 +110079#include <asm/iseries/hv_lp_event.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070080#include <asm/iommu.h>
81#include <asm/vio.h>
82
Michael Ellerman61a3c692005-09-01 11:28:57 +100083#undef DEBUG
84
Linus Torvalds1da177e2005-04-16 15:20:36 -070085MODULE_AUTHOR("Kyle Lucke <klucke@us.ibm.com>");
86MODULE_DESCRIPTION("iSeries Virtual ethernet driver");
87MODULE_LICENSE("GPL");
88
Michael Ellerman59f17ae2005-09-01 11:29:27 +100089#define VETH_EVENT_CAP (0)
90#define VETH_EVENT_FRAMES (1)
91#define VETH_EVENT_MONITOR (2)
92#define VETH_EVENT_FRAMES_ACK (3)
Michael Ellerman642d1a42005-09-01 11:29:25 +100093
94#define VETH_MAX_ACKS_PER_MSG (20)
95#define VETH_MAX_FRAMES_PER_MSG (6)
96
Michael Ellerman59f17ae2005-09-01 11:29:27 +100097struct veth_frames_data {
Michael Ellerman642d1a42005-09-01 11:29:25 +100098 u32 addr[VETH_MAX_FRAMES_PER_MSG];
99 u16 len[VETH_MAX_FRAMES_PER_MSG];
100 u32 eofmask;
101};
102#define VETH_EOF_SHIFT (32-VETH_MAX_FRAMES_PER_MSG)
103
Michael Ellerman59f17ae2005-09-01 11:29:27 +1000104struct veth_frames_ack_data {
Michael Ellerman642d1a42005-09-01 11:29:25 +1000105 u16 token[VETH_MAX_ACKS_PER_MSG];
106};
107
Michael Ellerman59f17ae2005-09-01 11:29:27 +1000108struct veth_cap_data {
Michael Ellerman642d1a42005-09-01 11:29:25 +1000109 u8 caps_version;
110 u8 rsvd1;
111 u16 num_buffers;
112 u16 ack_threshold;
113 u16 rsvd2;
114 u32 ack_timeout;
115 u32 rsvd3;
116 u64 rsvd4[3];
117};
118
Michael Ellerman59f17ae2005-09-01 11:29:27 +1000119struct veth_lpevent {
Michael Ellerman642d1a42005-09-01 11:29:25 +1000120 struct HvLpEvent base_event;
121 union {
Michael Ellerman59f17ae2005-09-01 11:29:27 +1000122 struct veth_cap_data caps_data;
123 struct veth_frames_data frames_data;
124 struct veth_frames_ack_data frames_ack_data;
Michael Ellerman642d1a42005-09-01 11:29:25 +1000125 } u;
126
127};
128
Michael Ellermanee05f032005-09-01 11:29:29 +1000129#define DRV_NAME "iseries_veth"
130#define DRV_VERSION "2.0"
131
Linus Torvalds1da177e2005-04-16 15:20:36 -0700132#define VETH_NUMBUFFERS (120)
133#define VETH_ACKTIMEOUT (1000000) /* microseconds */
134#define VETH_MAX_MCAST (12)
135
136#define VETH_MAX_MTU (9000)
137
138#if VETH_NUMBUFFERS < 10
139#define ACK_THRESHOLD (1)
140#elif VETH_NUMBUFFERS < 20
141#define ACK_THRESHOLD (4)
142#elif VETH_NUMBUFFERS < 40
143#define ACK_THRESHOLD (10)
144#else
145#define ACK_THRESHOLD (20)
146#endif
147
148#define VETH_STATE_SHUTDOWN (0x0001)
149#define VETH_STATE_OPEN (0x0002)
150#define VETH_STATE_RESET (0x0004)
151#define VETH_STATE_SENTMON (0x0008)
152#define VETH_STATE_SENTCAPS (0x0010)
153#define VETH_STATE_GOTCAPACK (0x0020)
154#define VETH_STATE_GOTCAPS (0x0040)
155#define VETH_STATE_SENTCAPACK (0x0080)
156#define VETH_STATE_READY (0x0100)
157
158struct veth_msg {
159 struct veth_msg *next;
Michael Ellerman59f17ae2005-09-01 11:29:27 +1000160 struct veth_frames_data data;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700161 int token;
Michael Ellermanb08bd5c2005-09-01 11:29:06 +1000162 int in_use;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700163 struct sk_buff *skb;
164 struct device *dev;
165};
166
167struct veth_lpar_connection {
168 HvLpIndex remote_lp;
David Howellsc4028952006-11-22 14:57:56 +0000169 struct delayed_work statemachine_wq;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700170 struct veth_msg *msgs;
171 int num_events;
Michael Ellerman59f17ae2005-09-01 11:29:27 +1000172 struct veth_cap_data local_caps;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700173
Michael Ellermanf0c129c2005-09-01 11:29:09 +1000174 struct kobject kobject;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700175 struct timer_list ack_timer;
176
Michael Ellerman24562ff2005-09-01 11:29:17 +1000177 struct timer_list reset_timer;
178 unsigned int reset_timeout;
179 unsigned long last_contact;
180 int outstanding_tx;
181
Linus Torvalds1da177e2005-04-16 15:20:36 -0700182 spinlock_t lock;
183 unsigned long state;
184 HvLpInstanceId src_inst;
185 HvLpInstanceId dst_inst;
Michael Ellerman59f17ae2005-09-01 11:29:27 +1000186 struct veth_lpevent cap_event, cap_ack_event;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700187 u16 pending_acks[VETH_MAX_ACKS_PER_MSG];
188 u32 num_pending_acks;
189
190 int num_ack_events;
Michael Ellerman59f17ae2005-09-01 11:29:27 +1000191 struct veth_cap_data remote_caps;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700192 u32 ack_timeout;
193
Linus Torvalds1da177e2005-04-16 15:20:36 -0700194 struct veth_msg *msg_stack_head;
195};
196
197struct veth_port {
198 struct device *dev;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700199 u64 mac_addr;
200 HvLpIndexMap lpar_map;
201
Michael Ellermane0808492005-09-01 11:29:18 +1000202 /* queue_lock protects the stopped_map and dev's queue. */
203 spinlock_t queue_lock;
204 HvLpIndexMap stopped_map;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700205
Michael Ellermane0808492005-09-01 11:29:18 +1000206 /* mcast_gate protects promiscuous, num_mcast & mcast_addr. */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700207 rwlock_t mcast_gate;
208 int promiscuous;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700209 int num_mcast;
210 u64 mcast_addr[VETH_MAX_MCAST];
Michael Ellerman07a5c172005-09-01 11:29:21 +1000211
212 struct kobject kobject;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700213};
214
215static HvLpIndex this_lp;
216static struct veth_lpar_connection *veth_cnx[HVMAXARCHITECTEDLPS]; /* = 0 */
217static struct net_device *veth_dev[HVMAXARCHITECTEDVIRTUALLANS]; /* = 0 */
218
219static int veth_start_xmit(struct sk_buff *skb, struct net_device *dev);
220static void veth_recycle_msg(struct veth_lpar_connection *, struct veth_msg *);
Michael Ellermane0808492005-09-01 11:29:18 +1000221static void veth_wake_queues(struct veth_lpar_connection *cnx);
222static void veth_stop_queues(struct veth_lpar_connection *cnx);
Michael Ellerman59f17ae2005-09-01 11:29:27 +1000223static void veth_receive(struct veth_lpar_connection *, struct veth_lpevent *);
Michael Ellermanf0c129c2005-09-01 11:29:09 +1000224static void veth_release_connection(struct kobject *kobject);
Michael Ellerman24562ff2005-09-01 11:29:17 +1000225static void veth_timed_ack(unsigned long ptr);
226static void veth_timed_reset(unsigned long ptr);
Michael Ellermanf0c129c2005-09-01 11:29:09 +1000227
Linus Torvalds1da177e2005-04-16 15:20:36 -0700228/*
229 * Utility functions
230 */
231
Michael Ellerman61a3c692005-09-01 11:28:57 +1000232#define veth_info(fmt, args...) \
Michael Ellermanee05f032005-09-01 11:29:29 +1000233 printk(KERN_INFO DRV_NAME ": " fmt, ## args)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700234
235#define veth_error(fmt, args...) \
Michael Ellermanee05f032005-09-01 11:29:29 +1000236 printk(KERN_ERR DRV_NAME ": Error: " fmt, ## args)
Michael Ellerman61a3c692005-09-01 11:28:57 +1000237
238#ifdef DEBUG
239#define veth_debug(fmt, args...) \
Michael Ellermanee05f032005-09-01 11:29:29 +1000240 printk(KERN_DEBUG DRV_NAME ": " fmt, ## args)
Michael Ellerman61a3c692005-09-01 11:28:57 +1000241#else
242#define veth_debug(fmt, args...) do {} while (0)
243#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -0700244
Michael Ellermand7893dd2005-09-01 11:29:05 +1000245/* You must hold the connection's lock when you call this function. */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700246static inline void veth_stack_push(struct veth_lpar_connection *cnx,
247 struct veth_msg *msg)
248{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700249 msg->next = cnx->msg_stack_head;
250 cnx->msg_stack_head = msg;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700251}
252
Michael Ellermand7893dd2005-09-01 11:29:05 +1000253/* You must hold the connection's lock when you call this function. */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700254static inline struct veth_msg *veth_stack_pop(struct veth_lpar_connection *cnx)
255{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700256 struct veth_msg *msg;
257
Linus Torvalds1da177e2005-04-16 15:20:36 -0700258 msg = cnx->msg_stack_head;
259 if (msg)
260 cnx->msg_stack_head = cnx->msg_stack_head->next;
Michael Ellermand7893dd2005-09-01 11:29:05 +1000261
Linus Torvalds1da177e2005-04-16 15:20:36 -0700262 return msg;
263}
264
Michael Ellermane0808492005-09-01 11:29:18 +1000265/* You must hold the connection's lock when you call this function. */
266static inline int veth_stack_is_empty(struct veth_lpar_connection *cnx)
267{
268 return cnx->msg_stack_head == NULL;
269}
270
Linus Torvalds1da177e2005-04-16 15:20:36 -0700271static inline HvLpEvent_Rc
272veth_signalevent(struct veth_lpar_connection *cnx, u16 subtype,
273 HvLpEvent_AckInd ackind, HvLpEvent_AckType acktype,
274 u64 token,
275 u64 data1, u64 data2, u64 data3, u64 data4, u64 data5)
276{
277 return HvCallEvent_signalLpEventFast(cnx->remote_lp,
278 HvLpEvent_Type_VirtualLan,
279 subtype, ackind, acktype,
280 cnx->src_inst,
281 cnx->dst_inst,
282 token, data1, data2, data3,
283 data4, data5);
284}
285
286static inline HvLpEvent_Rc veth_signaldata(struct veth_lpar_connection *cnx,
287 u16 subtype, u64 token, void *data)
288{
289 u64 *p = (u64 *) data;
290
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]);
294}
295
296struct veth_allocation {
297 struct completion c;
298 int num;
299};
300
301static void veth_complete_allocation(void *parm, int number)
302{
303 struct veth_allocation *vc = (struct veth_allocation *)parm;
304
305 vc->num = number;
306 complete(&vc->c);
307}
308
309static int veth_allocate_events(HvLpIndex rlp, int number)
310{
Benjamin Herrenschmidta2879fe2008-04-09 17:21:34 +1000311 struct veth_allocation vc =
312 { COMPLETION_INITIALIZER_ONSTACK(vc.c), 0 };
Linus Torvalds1da177e2005-04-16 15:20:36 -0700313
314 mf_allocate_lp_events(rlp, HvLpEvent_Type_VirtualLan,
Michael Ellerman59f17ae2005-09-01 11:29:27 +1000315 sizeof(struct veth_lpevent), number,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700316 &veth_complete_allocation, &vc);
317 wait_for_completion(&vc.c);
318
319 return vc.num;
320}
321
322/*
Michael Ellerman76812d8122005-09-01 11:29:20 +1000323 * sysfs support
324 */
325
326struct veth_cnx_attribute {
327 struct attribute attr;
328 ssize_t (*show)(struct veth_lpar_connection *, char *buf);
329 ssize_t (*store)(struct veth_lpar_connection *, const char *buf);
330};
331
332static ssize_t veth_cnx_attribute_show(struct kobject *kobj,
333 struct attribute *attr, char *buf)
334{
335 struct veth_cnx_attribute *cnx_attr;
336 struct veth_lpar_connection *cnx;
337
338 cnx_attr = container_of(attr, struct veth_cnx_attribute, attr);
339 cnx = container_of(kobj, struct veth_lpar_connection, kobject);
340
341 if (!cnx_attr->show)
342 return -EIO;
343
344 return cnx_attr->show(cnx, buf);
345}
346
347#define CUSTOM_CNX_ATTR(_name, _format, _expression) \
348static ssize_t _name##_show(struct veth_lpar_connection *cnx, char *buf)\
349{ \
350 return sprintf(buf, _format, _expression); \
351} \
352struct veth_cnx_attribute veth_cnx_attr_##_name = __ATTR_RO(_name)
353
354#define SIMPLE_CNX_ATTR(_name) \
355 CUSTOM_CNX_ATTR(_name, "%lu\n", (unsigned long)cnx->_name)
356
357SIMPLE_CNX_ATTR(outstanding_tx);
358SIMPLE_CNX_ATTR(remote_lp);
359SIMPLE_CNX_ATTR(num_events);
360SIMPLE_CNX_ATTR(src_inst);
361SIMPLE_CNX_ATTR(dst_inst);
362SIMPLE_CNX_ATTR(num_pending_acks);
363SIMPLE_CNX_ATTR(num_ack_events);
364CUSTOM_CNX_ATTR(ack_timeout, "%d\n", jiffies_to_msecs(cnx->ack_timeout));
365CUSTOM_CNX_ATTR(reset_timeout, "%d\n", jiffies_to_msecs(cnx->reset_timeout));
366CUSTOM_CNX_ATTR(state, "0x%.4lX\n", cnx->state);
367CUSTOM_CNX_ATTR(last_contact, "%d\n", cnx->last_contact ?
368 jiffies_to_msecs(jiffies - cnx->last_contact) : 0);
369
370#define GET_CNX_ATTR(_name) (&veth_cnx_attr_##_name.attr)
371
372static struct attribute *veth_cnx_default_attrs[] = {
373 GET_CNX_ATTR(outstanding_tx),
374 GET_CNX_ATTR(remote_lp),
375 GET_CNX_ATTR(num_events),
376 GET_CNX_ATTR(reset_timeout),
377 GET_CNX_ATTR(last_contact),
378 GET_CNX_ATTR(state),
379 GET_CNX_ATTR(src_inst),
380 GET_CNX_ATTR(dst_inst),
381 GET_CNX_ATTR(num_pending_acks),
382 GET_CNX_ATTR(num_ack_events),
383 GET_CNX_ATTR(ack_timeout),
384 NULL
385};
386
387static struct sysfs_ops veth_cnx_sysfs_ops = {
388 .show = veth_cnx_attribute_show
389};
390
391static struct kobj_type veth_lpar_connection_ktype = {
392 .release = veth_release_connection,
393 .sysfs_ops = &veth_cnx_sysfs_ops,
394 .default_attrs = veth_cnx_default_attrs
395};
396
Michael Ellerman07a5c172005-09-01 11:29:21 +1000397struct veth_port_attribute {
398 struct attribute attr;
399 ssize_t (*show)(struct veth_port *, char *buf);
400 ssize_t (*store)(struct veth_port *, const char *buf);
401};
402
403static ssize_t veth_port_attribute_show(struct kobject *kobj,
404 struct attribute *attr, char *buf)
405{
406 struct veth_port_attribute *port_attr;
407 struct veth_port *port;
408
409 port_attr = container_of(attr, struct veth_port_attribute, attr);
410 port = container_of(kobj, struct veth_port, kobject);
411
412 if (!port_attr->show)
413 return -EIO;
414
415 return port_attr->show(port, buf);
416}
417
418#define CUSTOM_PORT_ATTR(_name, _format, _expression) \
419static ssize_t _name##_show(struct veth_port *port, char *buf) \
420{ \
421 return sprintf(buf, _format, _expression); \
422} \
423struct veth_port_attribute veth_port_attr_##_name = __ATTR_RO(_name)
424
425#define SIMPLE_PORT_ATTR(_name) \
426 CUSTOM_PORT_ATTR(_name, "%lu\n", (unsigned long)port->_name)
427
428SIMPLE_PORT_ATTR(promiscuous);
429SIMPLE_PORT_ATTR(num_mcast);
430CUSTOM_PORT_ATTR(lpar_map, "0x%X\n", port->lpar_map);
431CUSTOM_PORT_ATTR(stopped_map, "0x%X\n", port->stopped_map);
432CUSTOM_PORT_ATTR(mac_addr, "0x%lX\n", port->mac_addr);
433
434#define GET_PORT_ATTR(_name) (&veth_port_attr_##_name.attr)
435static struct attribute *veth_port_default_attrs[] = {
436 GET_PORT_ATTR(mac_addr),
437 GET_PORT_ATTR(lpar_map),
438 GET_PORT_ATTR(stopped_map),
439 GET_PORT_ATTR(promiscuous),
440 GET_PORT_ATTR(num_mcast),
441 NULL
442};
443
444static struct sysfs_ops veth_port_sysfs_ops = {
445 .show = veth_port_attribute_show
446};
447
448static struct kobj_type veth_port_ktype = {
449 .sysfs_ops = &veth_port_sysfs_ops,
450 .default_attrs = veth_port_default_attrs
451};
452
Michael Ellerman76812d8122005-09-01 11:29:20 +1000453/*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700454 * LPAR connection code
455 */
456
457static inline void veth_kick_statemachine(struct veth_lpar_connection *cnx)
458{
David Howellsc4028952006-11-22 14:57:56 +0000459 schedule_delayed_work(&cnx->statemachine_wq, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700460}
461
462static void veth_take_cap(struct veth_lpar_connection *cnx,
Michael Ellerman59f17ae2005-09-01 11:29:27 +1000463 struct veth_lpevent *event)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700464{
465 unsigned long flags;
466
467 spin_lock_irqsave(&cnx->lock, flags);
468 /* Receiving caps may mean the other end has just come up, so
469 * we need to reload the instance ID of the far end */
470 cnx->dst_inst =
471 HvCallEvent_getTargetLpInstanceId(cnx->remote_lp,
472 HvLpEvent_Type_VirtualLan);
473
474 if (cnx->state & VETH_STATE_GOTCAPS) {
Michael Ellerman61a3c692005-09-01 11:28:57 +1000475 veth_error("Received a second capabilities from LPAR %d.\n",
Linus Torvalds1da177e2005-04-16 15:20:36 -0700476 cnx->remote_lp);
477 event->base_event.xRc = HvLpEvent_Rc_BufferNotAvailable;
478 HvCallEvent_ackLpEvent((struct HvLpEvent *) event);
479 } else {
480 memcpy(&cnx->cap_event, event, sizeof(cnx->cap_event));
481 cnx->state |= VETH_STATE_GOTCAPS;
482 veth_kick_statemachine(cnx);
483 }
484 spin_unlock_irqrestore(&cnx->lock, flags);
485}
486
487static void veth_take_cap_ack(struct veth_lpar_connection *cnx,
Michael Ellerman59f17ae2005-09-01 11:29:27 +1000488 struct veth_lpevent *event)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700489{
490 unsigned long flags;
491
492 spin_lock_irqsave(&cnx->lock, flags);
493 if (cnx->state & VETH_STATE_GOTCAPACK) {
Michael Ellerman61a3c692005-09-01 11:28:57 +1000494 veth_error("Received a second capabilities ack from LPAR %d.\n",
Linus Torvalds1da177e2005-04-16 15:20:36 -0700495 cnx->remote_lp);
496 } else {
497 memcpy(&cnx->cap_ack_event, event,
498 sizeof(&cnx->cap_ack_event));
499 cnx->state |= VETH_STATE_GOTCAPACK;
500 veth_kick_statemachine(cnx);
501 }
502 spin_unlock_irqrestore(&cnx->lock, flags);
503}
504
505static void veth_take_monitor_ack(struct veth_lpar_connection *cnx,
Michael Ellerman59f17ae2005-09-01 11:29:27 +1000506 struct veth_lpevent *event)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700507{
508 unsigned long flags;
509
510 spin_lock_irqsave(&cnx->lock, flags);
Michael Ellerman61a3c692005-09-01 11:28:57 +1000511 veth_debug("cnx %d: lost connection.\n", cnx->remote_lp);
Michael Ellerman58c59002005-09-01 11:29:00 +1000512
513 /* Avoid kicking the statemachine once we're shutdown.
514 * It's unnecessary and it could break veth_stop_connection(). */
515
516 if (! (cnx->state & VETH_STATE_SHUTDOWN)) {
517 cnx->state |= VETH_STATE_RESET;
518 veth_kick_statemachine(cnx);
519 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700520 spin_unlock_irqrestore(&cnx->lock, flags);
521}
522
Michael Ellerman59f17ae2005-09-01 11:29:27 +1000523static void veth_handle_ack(struct veth_lpevent *event)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700524{
525 HvLpIndex rlp = event->base_event.xTargetLp;
526 struct veth_lpar_connection *cnx = veth_cnx[rlp];
527
528 BUG_ON(! cnx);
529
530 switch (event->base_event.xSubtype) {
Michael Ellerman59f17ae2005-09-01 11:29:27 +1000531 case VETH_EVENT_CAP:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700532 veth_take_cap_ack(cnx, event);
533 break;
Michael Ellerman59f17ae2005-09-01 11:29:27 +1000534 case VETH_EVENT_MONITOR:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700535 veth_take_monitor_ack(cnx, event);
536 break;
537 default:
Michael Ellerman61a3c692005-09-01 11:28:57 +1000538 veth_error("Unknown ack type %d from LPAR %d.\n",
539 event->base_event.xSubtype, rlp);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700540 };
541}
542
Michael Ellerman59f17ae2005-09-01 11:29:27 +1000543static void veth_handle_int(struct veth_lpevent *event)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700544{
545 HvLpIndex rlp = event->base_event.xSourceLp;
546 struct veth_lpar_connection *cnx = veth_cnx[rlp];
547 unsigned long flags;
Michael Ellerman24562ff2005-09-01 11:29:17 +1000548 int i, acked = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700549
550 BUG_ON(! cnx);
551
552 switch (event->base_event.xSubtype) {
Michael Ellerman59f17ae2005-09-01 11:29:27 +1000553 case VETH_EVENT_CAP:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700554 veth_take_cap(cnx, event);
555 break;
Michael Ellerman59f17ae2005-09-01 11:29:27 +1000556 case VETH_EVENT_MONITOR:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700557 /* do nothing... this'll hang out here til we're dead,
558 * and the hypervisor will return it for us. */
559 break;
Michael Ellerman59f17ae2005-09-01 11:29:27 +1000560 case VETH_EVENT_FRAMES_ACK:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700561 spin_lock_irqsave(&cnx->lock, flags);
Michael Ellerman24562ff2005-09-01 11:29:17 +1000562
Linus Torvalds1da177e2005-04-16 15:20:36 -0700563 for (i = 0; i < VETH_MAX_ACKS_PER_MSG; ++i) {
564 u16 msgnum = event->u.frames_ack_data.token[i];
565
Michael Ellerman24562ff2005-09-01 11:29:17 +1000566 if (msgnum < VETH_NUMBUFFERS) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700567 veth_recycle_msg(cnx, cnx->msgs + msgnum);
Michael Ellerman24562ff2005-09-01 11:29:17 +1000568 cnx->outstanding_tx--;
569 acked++;
570 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700571 }
Michael Ellerman24562ff2005-09-01 11:29:17 +1000572
Michael Ellermane0808492005-09-01 11:29:18 +1000573 if (acked > 0) {
Michael Ellerman24562ff2005-09-01 11:29:17 +1000574 cnx->last_contact = jiffies;
Michael Ellermane0808492005-09-01 11:29:18 +1000575 veth_wake_queues(cnx);
576 }
Michael Ellerman24562ff2005-09-01 11:29:17 +1000577
Linus Torvalds1da177e2005-04-16 15:20:36 -0700578 spin_unlock_irqrestore(&cnx->lock, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700579 break;
Michael Ellerman59f17ae2005-09-01 11:29:27 +1000580 case VETH_EVENT_FRAMES:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700581 veth_receive(cnx, event);
582 break;
583 default:
Michael Ellerman61a3c692005-09-01 11:28:57 +1000584 veth_error("Unknown interrupt type %d from LPAR %d.\n",
585 event->base_event.xSubtype, rlp);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700586 };
587}
588
David Howells7d12e782006-10-05 14:55:46 +0100589static void veth_handle_event(struct HvLpEvent *event)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700590{
Michael Ellerman59f17ae2005-09-01 11:29:27 +1000591 struct veth_lpevent *veth_event = (struct veth_lpevent *)event;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700592
Stephen Rothwell677f8c02006-01-12 13:47:43 +1100593 if (hvlpevent_is_ack(event))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700594 veth_handle_ack(veth_event);
Stephen Rothwell677f8c02006-01-12 13:47:43 +1100595 else
Linus Torvalds1da177e2005-04-16 15:20:36 -0700596 veth_handle_int(veth_event);
597}
598
599static int veth_process_caps(struct veth_lpar_connection *cnx)
600{
Michael Ellerman59f17ae2005-09-01 11:29:27 +1000601 struct veth_cap_data *remote_caps = &cnx->remote_caps;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700602 int num_acks_needed;
603
604 /* Convert timer to jiffies */
605 cnx->ack_timeout = remote_caps->ack_timeout * HZ / 1000000;
606
607 if ( (remote_caps->num_buffers == 0)
608 || (remote_caps->ack_threshold > VETH_MAX_ACKS_PER_MSG)
609 || (remote_caps->ack_threshold == 0)
610 || (cnx->ack_timeout == 0) ) {
Michael Ellerman61a3c692005-09-01 11:28:57 +1000611 veth_error("Received incompatible capabilities from LPAR %d.\n",
612 cnx->remote_lp);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700613 return HvLpEvent_Rc_InvalidSubtypeData;
614 }
615
616 num_acks_needed = (remote_caps->num_buffers
617 / remote_caps->ack_threshold) + 1;
618
619 /* FIXME: locking on num_ack_events? */
620 if (cnx->num_ack_events < num_acks_needed) {
621 int num;
622
623 num = veth_allocate_events(cnx->remote_lp,
624 num_acks_needed-cnx->num_ack_events);
625 if (num > 0)
626 cnx->num_ack_events += num;
627
628 if (cnx->num_ack_events < num_acks_needed) {
Michael Ellerman61a3c692005-09-01 11:28:57 +1000629 veth_error("Couldn't allocate enough ack events "
630 "for LPAR %d.\n", cnx->remote_lp);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700631
632 return HvLpEvent_Rc_BufferNotAvailable;
633 }
634 }
635
636
637 return HvLpEvent_Rc_Good;
638}
639
640/* FIXME: The gotos here are a bit dubious */
David Howellsc4028952006-11-22 14:57:56 +0000641static void veth_statemachine(struct work_struct *work)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700642{
David Howellsc4028952006-11-22 14:57:56 +0000643 struct veth_lpar_connection *cnx =
644 container_of(work, struct veth_lpar_connection,
645 statemachine_wq.work);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700646 int rlp = cnx->remote_lp;
647 int rc;
648
649 spin_lock_irq(&cnx->lock);
650
651 restart:
652 if (cnx->state & VETH_STATE_RESET) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700653 if (cnx->state & VETH_STATE_OPEN)
654 HvCallEvent_closeLpEventPath(cnx->remote_lp,
655 HvLpEvent_Type_VirtualLan);
656
Michael Ellermanabfda472005-09-01 11:28:59 +1000657 /*
658 * Reset ack data. This prevents the ack_timer actually
659 * doing anything, even if it runs one more time when
660 * we drop the lock below.
661 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700662 memset(&cnx->pending_acks, 0xff, sizeof (cnx->pending_acks));
663 cnx->num_pending_acks = 0;
664
665 cnx->state &= ~(VETH_STATE_RESET | VETH_STATE_SENTMON
666 | VETH_STATE_OPEN | VETH_STATE_SENTCAPS
667 | VETH_STATE_GOTCAPACK | VETH_STATE_GOTCAPS
668 | VETH_STATE_SENTCAPACK | VETH_STATE_READY);
669
670 /* Clean up any leftover messages */
Michael Ellerman24562ff2005-09-01 11:29:17 +1000671 if (cnx->msgs) {
672 int i;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700673 for (i = 0; i < VETH_NUMBUFFERS; ++i)
674 veth_recycle_msg(cnx, cnx->msgs + i);
Michael Ellerman24562ff2005-09-01 11:29:17 +1000675 }
Michael Ellermane0808492005-09-01 11:29:18 +1000676
Michael Ellerman24562ff2005-09-01 11:29:17 +1000677 cnx->outstanding_tx = 0;
Michael Ellermane0808492005-09-01 11:29:18 +1000678 veth_wake_queues(cnx);
Michael Ellermanabfda472005-09-01 11:28:59 +1000679
680 /* Drop the lock so we can do stuff that might sleep or
681 * take other locks. */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700682 spin_unlock_irq(&cnx->lock);
Michael Ellermanabfda472005-09-01 11:28:59 +1000683
684 del_timer_sync(&cnx->ack_timer);
Michael Ellerman24562ff2005-09-01 11:29:17 +1000685 del_timer_sync(&cnx->reset_timer);
686
Linus Torvalds1da177e2005-04-16 15:20:36 -0700687 spin_lock_irq(&cnx->lock);
Michael Ellermanabfda472005-09-01 11:28:59 +1000688
Linus Torvalds1da177e2005-04-16 15:20:36 -0700689 if (cnx->state & VETH_STATE_RESET)
690 goto restart;
Michael Ellerman58c59002005-09-01 11:29:00 +1000691
692 /* Hack, wait for the other end to reset itself. */
693 if (! (cnx->state & VETH_STATE_SHUTDOWN)) {
694 schedule_delayed_work(&cnx->statemachine_wq, 5 * HZ);
695 goto out;
696 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700697 }
698
699 if (cnx->state & VETH_STATE_SHUTDOWN)
700 /* It's all over, do nothing */
701 goto out;
702
703 if ( !(cnx->state & VETH_STATE_OPEN) ) {
704 if (! cnx->msgs || (cnx->num_events < (2 + VETH_NUMBUFFERS)) )
705 goto cant_cope;
706
707 HvCallEvent_openLpEventPath(rlp, HvLpEvent_Type_VirtualLan);
708 cnx->src_inst =
709 HvCallEvent_getSourceLpInstanceId(rlp,
710 HvLpEvent_Type_VirtualLan);
711 cnx->dst_inst =
712 HvCallEvent_getTargetLpInstanceId(rlp,
713 HvLpEvent_Type_VirtualLan);
714 cnx->state |= VETH_STATE_OPEN;
715 }
716
717 if ( (cnx->state & VETH_STATE_OPEN)
718 && !(cnx->state & VETH_STATE_SENTMON) ) {
Michael Ellerman59f17ae2005-09-01 11:29:27 +1000719 rc = veth_signalevent(cnx, VETH_EVENT_MONITOR,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700720 HvLpEvent_AckInd_DoAck,
721 HvLpEvent_AckType_DeferredAck,
722 0, 0, 0, 0, 0, 0);
723
724 if (rc == HvLpEvent_Rc_Good) {
725 cnx->state |= VETH_STATE_SENTMON;
726 } else {
727 if ( (rc != HvLpEvent_Rc_PartitionDead)
728 && (rc != HvLpEvent_Rc_PathClosed) )
Michael Ellerman61a3c692005-09-01 11:28:57 +1000729 veth_error("Error sending monitor to LPAR %d, "
730 "rc = %d\n", rlp, rc);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700731
732 /* Oh well, hope we get a cap from the other
733 * end and do better when that kicks us */
734 goto out;
735 }
736 }
737
738 if ( (cnx->state & VETH_STATE_OPEN)
739 && !(cnx->state & VETH_STATE_SENTCAPS)) {
740 u64 *rawcap = (u64 *)&cnx->local_caps;
741
Michael Ellerman59f17ae2005-09-01 11:29:27 +1000742 rc = veth_signalevent(cnx, VETH_EVENT_CAP,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700743 HvLpEvent_AckInd_DoAck,
744 HvLpEvent_AckType_ImmediateAck,
745 0, rawcap[0], rawcap[1], rawcap[2],
746 rawcap[3], rawcap[4]);
747
748 if (rc == HvLpEvent_Rc_Good) {
749 cnx->state |= VETH_STATE_SENTCAPS;
750 } else {
751 if ( (rc != HvLpEvent_Rc_PartitionDead)
752 && (rc != HvLpEvent_Rc_PathClosed) )
Michael Ellerman61a3c692005-09-01 11:28:57 +1000753 veth_error("Error sending caps to LPAR %d, "
754 "rc = %d\n", rlp, rc);
755
Linus Torvalds1da177e2005-04-16 15:20:36 -0700756 /* Oh well, hope we get a cap from the other
757 * end and do better when that kicks us */
758 goto out;
759 }
760 }
761
762 if ((cnx->state & VETH_STATE_GOTCAPS)
763 && !(cnx->state & VETH_STATE_SENTCAPACK)) {
Michael Ellerman59f17ae2005-09-01 11:29:27 +1000764 struct veth_cap_data *remote_caps = &cnx->remote_caps;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700765
766 memcpy(remote_caps, &cnx->cap_event.u.caps_data,
767 sizeof(*remote_caps));
768
769 spin_unlock_irq(&cnx->lock);
770 rc = veth_process_caps(cnx);
771 spin_lock_irq(&cnx->lock);
772
773 /* We dropped the lock, so recheck for anything which
774 * might mess us up */
775 if (cnx->state & (VETH_STATE_RESET|VETH_STATE_SHUTDOWN))
776 goto restart;
777
778 cnx->cap_event.base_event.xRc = rc;
779 HvCallEvent_ackLpEvent((struct HvLpEvent *)&cnx->cap_event);
780 if (rc == HvLpEvent_Rc_Good)
781 cnx->state |= VETH_STATE_SENTCAPACK;
782 else
783 goto cant_cope;
784 }
785
786 if ((cnx->state & VETH_STATE_GOTCAPACK)
787 && (cnx->state & VETH_STATE_GOTCAPS)
788 && !(cnx->state & VETH_STATE_READY)) {
789 if (cnx->cap_ack_event.base_event.xRc == HvLpEvent_Rc_Good) {
790 /* Start the ACK timer */
791 cnx->ack_timer.expires = jiffies + cnx->ack_timeout;
792 add_timer(&cnx->ack_timer);
793 cnx->state |= VETH_STATE_READY;
794 } else {
Michael Ellerman61a3c692005-09-01 11:28:57 +1000795 veth_error("Caps rejected by LPAR %d, rc = %d\n",
796 rlp, cnx->cap_ack_event.base_event.xRc);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700797 goto cant_cope;
798 }
799 }
800
801 out:
802 spin_unlock_irq(&cnx->lock);
803 return;
804
805 cant_cope:
806 /* FIXME: we get here if something happens we really can't
807 * cope with. The link will never work once we get here, and
808 * all we can do is not lock the rest of the system up */
Michael Ellerman61a3c692005-09-01 11:28:57 +1000809 veth_error("Unrecoverable error on connection to LPAR %d, shutting down"
810 " (state = 0x%04lx)\n", rlp, cnx->state);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700811 cnx->state |= VETH_STATE_SHUTDOWN;
812 spin_unlock_irq(&cnx->lock);
813}
814
815static int veth_init_connection(u8 rlp)
816{
817 struct veth_lpar_connection *cnx;
818 struct veth_msg *msgs;
Greg Kroah-Hartman78be3d22007-12-17 23:05:35 -0700819 int i;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700820
821 if ( (rlp == this_lp)
822 || ! HvLpConfig_doLpsCommunicateOnVirtualLan(this_lp, rlp) )
823 return 0;
824
Yoann Padioleaudd00cc42007-07-19 01:49:03 -0700825 cnx = kzalloc(sizeof(*cnx), GFP_KERNEL);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700826 if (! cnx)
827 return -ENOMEM;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700828
829 cnx->remote_lp = rlp;
830 spin_lock_init(&cnx->lock);
David Howellsc4028952006-11-22 14:57:56 +0000831 INIT_DELAYED_WORK(&cnx->statemachine_wq, veth_statemachine);
Michael Ellerman24562ff2005-09-01 11:29:17 +1000832
Linus Torvalds1da177e2005-04-16 15:20:36 -0700833 init_timer(&cnx->ack_timer);
834 cnx->ack_timer.function = veth_timed_ack;
835 cnx->ack_timer.data = (unsigned long) cnx;
Michael Ellerman24562ff2005-09-01 11:29:17 +1000836
837 init_timer(&cnx->reset_timer);
838 cnx->reset_timer.function = veth_timed_reset;
839 cnx->reset_timer.data = (unsigned long) cnx;
840 cnx->reset_timeout = 5 * HZ * (VETH_ACKTIMEOUT / 1000000);
841
Linus Torvalds1da177e2005-04-16 15:20:36 -0700842 memset(&cnx->pending_acks, 0xff, sizeof (cnx->pending_acks));
843
844 veth_cnx[rlp] = cnx;
845
Michael Ellermanf0c129c2005-09-01 11:29:09 +1000846 /* This gets us 1 reference, which is held on behalf of the driver
847 * infrastructure. It's released at module unload. */
Greg Kroah-Hartmanf9cb0742007-12-17 23:05:35 -0700848 kobject_init(&cnx->kobject, &veth_lpar_connection_ktype);
Michael Ellermanf0c129c2005-09-01 11:29:09 +1000849
Yoann Padioleaudd00cc42007-07-19 01:49:03 -0700850 msgs = kcalloc(VETH_NUMBUFFERS, sizeof(struct veth_msg), GFP_KERNEL);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700851 if (! msgs) {
Michael Ellerman61a3c692005-09-01 11:28:57 +1000852 veth_error("Can't allocate buffers for LPAR %d.\n", rlp);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700853 return -ENOMEM;
854 }
855
856 cnx->msgs = msgs;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700857
858 for (i = 0; i < VETH_NUMBUFFERS; i++) {
859 msgs[i].token = i;
860 veth_stack_push(cnx, msgs + i);
861 }
862
863 cnx->num_events = veth_allocate_events(rlp, 2 + VETH_NUMBUFFERS);
864
865 if (cnx->num_events < (2 + VETH_NUMBUFFERS)) {
Michael Ellerman61a3c692005-09-01 11:28:57 +1000866 veth_error("Can't allocate enough events for LPAR %d.\n", rlp);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700867 return -ENOMEM;
868 }
869
870 cnx->local_caps.num_buffers = VETH_NUMBUFFERS;
871 cnx->local_caps.ack_threshold = ACK_THRESHOLD;
872 cnx->local_caps.ack_timeout = VETH_ACKTIMEOUT;
873
874 return 0;
875}
876
Michael Ellermanf0c129c2005-09-01 11:29:09 +1000877static void veth_stop_connection(struct veth_lpar_connection *cnx)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700878{
Michael Ellermanf0c129c2005-09-01 11:29:09 +1000879 if (!cnx)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700880 return;
881
882 spin_lock_irq(&cnx->lock);
883 cnx->state |= VETH_STATE_RESET | VETH_STATE_SHUTDOWN;
884 veth_kick_statemachine(cnx);
885 spin_unlock_irq(&cnx->lock);
886
Michael Ellerman58c59002005-09-01 11:29:00 +1000887 /* There's a slim chance the reset code has just queued the
888 * statemachine to run in five seconds. If so we need to cancel
889 * that and requeue the work to run now. */
890 if (cancel_delayed_work(&cnx->statemachine_wq)) {
891 spin_lock_irq(&cnx->lock);
892 veth_kick_statemachine(cnx);
893 spin_unlock_irq(&cnx->lock);
894 }
895
Michael Ellermanabfda472005-09-01 11:28:59 +1000896 /* Wait for the state machine to run. */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700897 flush_scheduled_work();
Michael Ellermanec60bee2005-09-01 11:29:08 +1000898}
899
Michael Ellermanf0c129c2005-09-01 11:29:09 +1000900static void veth_destroy_connection(struct veth_lpar_connection *cnx)
Michael Ellermanec60bee2005-09-01 11:29:08 +1000901{
Michael Ellermanf0c129c2005-09-01 11:29:09 +1000902 if (!cnx)
Michael Ellermanec60bee2005-09-01 11:29:08 +1000903 return;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700904
Linus Torvalds1da177e2005-04-16 15:20:36 -0700905 if (cnx->num_events > 0)
906 mf_deallocate_lp_events(cnx->remote_lp,
907 HvLpEvent_Type_VirtualLan,
908 cnx->num_events,
909 NULL, NULL);
910 if (cnx->num_ack_events > 0)
911 mf_deallocate_lp_events(cnx->remote_lp,
912 HvLpEvent_Type_VirtualLan,
913 cnx->num_ack_events,
914 NULL, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700915
916 kfree(cnx->msgs);
Michael Ellermanf0c129c2005-09-01 11:29:09 +1000917 veth_cnx[cnx->remote_lp] = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700918 kfree(cnx);
Michael Ellermanf0c129c2005-09-01 11:29:09 +1000919}
920
921static void veth_release_connection(struct kobject *kobj)
922{
923 struct veth_lpar_connection *cnx;
924 cnx = container_of(kobj, struct veth_lpar_connection, kobject);
925 veth_stop_connection(cnx);
926 veth_destroy_connection(cnx);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700927}
928
929/*
930 * net_device code
931 */
932
933static int veth_open(struct net_device *dev)
934{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700935 netif_start_queue(dev);
936 return 0;
937}
938
939static int veth_close(struct net_device *dev)
940{
941 netif_stop_queue(dev);
942 return 0;
943}
944
Linus Torvalds1da177e2005-04-16 15:20:36 -0700945static int veth_change_mtu(struct net_device *dev, int new_mtu)
946{
947 if ((new_mtu < 68) || (new_mtu > VETH_MAX_MTU))
948 return -EINVAL;
949 dev->mtu = new_mtu;
950 return 0;
951}
952
953static void veth_set_multicast_list(struct net_device *dev)
954{
955 struct veth_port *port = (struct veth_port *) dev->priv;
956 unsigned long flags;
957
958 write_lock_irqsave(&port->mcast_gate, flags);
959
Michael Ellerman2a5391a2005-09-01 11:29:02 +1000960 if ((dev->flags & IFF_PROMISC) || (dev->flags & IFF_ALLMULTI) ||
961 (dev->mc_count > VETH_MAX_MCAST)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700962 port->promiscuous = 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700963 } else {
964 struct dev_mc_list *dmi = dev->mc_list;
965 int i;
966
Michael Ellerman2a5391a2005-09-01 11:29:02 +1000967 port->promiscuous = 0;
968
Linus Torvalds1da177e2005-04-16 15:20:36 -0700969 /* Update table */
970 port->num_mcast = 0;
971
972 for (i = 0; i < dev->mc_count; i++) {
973 u8 *addr = dmi->dmi_addr;
974 u64 xaddr = 0;
975
976 if (addr[0] & 0x01) {/* multicast address? */
977 memcpy(&xaddr, addr, ETH_ALEN);
978 port->mcast_addr[port->num_mcast] = xaddr;
979 port->num_mcast++;
980 }
981 dmi = dmi->next;
982 }
983 }
984
985 write_unlock_irqrestore(&port->mcast_gate, flags);
986}
987
988static void veth_get_drvinfo(struct net_device *dev, struct ethtool_drvinfo *info)
989{
Michael Ellermanee05f032005-09-01 11:29:29 +1000990 strncpy(info->driver, DRV_NAME, sizeof(info->driver) - 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700991 info->driver[sizeof(info->driver) - 1] = '\0';
Michael Ellermanee05f032005-09-01 11:29:29 +1000992 strncpy(info->version, DRV_VERSION, sizeof(info->version) - 1);
993 info->version[sizeof(info->version) - 1] = '\0';
Linus Torvalds1da177e2005-04-16 15:20:36 -0700994}
995
996static int veth_get_settings(struct net_device *dev, struct ethtool_cmd *ecmd)
997{
998 ecmd->supported = (SUPPORTED_1000baseT_Full
999 | SUPPORTED_Autoneg | SUPPORTED_FIBRE);
1000 ecmd->advertising = (SUPPORTED_1000baseT_Full
1001 | SUPPORTED_Autoneg | SUPPORTED_FIBRE);
1002 ecmd->port = PORT_FIBRE;
1003 ecmd->transceiver = XCVR_INTERNAL;
1004 ecmd->phy_address = 0;
1005 ecmd->speed = SPEED_1000;
1006 ecmd->duplex = DUPLEX_FULL;
1007 ecmd->autoneg = AUTONEG_ENABLE;
1008 ecmd->maxtxpkt = 120;
1009 ecmd->maxrxpkt = 120;
1010 return 0;
1011}
1012
1013static u32 veth_get_link(struct net_device *dev)
1014{
1015 return 1;
1016}
1017
Jeff Garzik7282d492006-09-13 14:30:00 -04001018static const struct ethtool_ops ops = {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001019 .get_drvinfo = veth_get_drvinfo,
1020 .get_settings = veth_get_settings,
1021 .get_link = veth_get_link,
1022};
1023
Stephen Rothwell57f78ab2008-02-05 14:19:40 +11001024static struct net_device *veth_probe_one(int vlan,
Stephen Rothwell07fb3f42006-05-15 13:39:47 +10001025 struct vio_dev *vio_dev)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001026{
1027 struct net_device *dev;
1028 struct veth_port *port;
Stephen Rothwell07fb3f42006-05-15 13:39:47 +10001029 struct device *vdev = &vio_dev->dev;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001030 int i, rc;
Stephen Rothwell07fb3f42006-05-15 13:39:47 +10001031 const unsigned char *mac_addr;
1032
1033 mac_addr = vio_get_attribute(vio_dev, "local-mac-address", NULL);
1034 if (mac_addr == NULL)
1035 mac_addr = vio_get_attribute(vio_dev, "mac-address", NULL);
1036 if (mac_addr == NULL) {
1037 veth_error("Unable to fetch MAC address from device tree.\n");
1038 return NULL;
1039 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001040
1041 dev = alloc_etherdev(sizeof (struct veth_port));
1042 if (! dev) {
1043 veth_error("Unable to allocate net_device structure!\n");
1044 return NULL;
1045 }
1046
1047 port = (struct veth_port *) dev->priv;
1048
Michael Ellermane0808492005-09-01 11:29:18 +10001049 spin_lock_init(&port->queue_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001050 rwlock_init(&port->mcast_gate);
Michael Ellermane0808492005-09-01 11:29:18 +10001051 port->stopped_map = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001052
1053 for (i = 0; i < HVMAXARCHITECTEDLPS; i++) {
1054 HvLpVirtualLanIndexMap map;
1055
1056 if (i == this_lp)
1057 continue;
1058 map = HvLpConfig_getVirtualLanIndexMapForLp(i);
1059 if (map & (0x8000 >> vlan))
1060 port->lpar_map |= (1 << i);
1061 }
1062 port->dev = vdev;
1063
Stephen Rothwell07fb3f42006-05-15 13:39:47 +10001064 memcpy(dev->dev_addr, mac_addr, ETH_ALEN);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001065
1066 dev->mtu = VETH_MAX_MTU;
1067
Stephen Rothwell07fb3f42006-05-15 13:39:47 +10001068 memcpy(&port->mac_addr, mac_addr, ETH_ALEN);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001069
1070 dev->open = veth_open;
1071 dev->hard_start_xmit = veth_start_xmit;
1072 dev->stop = veth_close;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001073 dev->change_mtu = veth_change_mtu;
1074 dev->set_mac_address = NULL;
1075 dev->set_multicast_list = veth_set_multicast_list;
1076 SET_ETHTOOL_OPS(dev, &ops);
1077
Linus Torvalds1da177e2005-04-16 15:20:36 -07001078 SET_NETDEV_DEV(dev, vdev);
1079
1080 rc = register_netdev(dev);
1081 if (rc != 0) {
Michael Ellerman61a3c692005-09-01 11:28:57 +10001082 veth_error("Failed registering net device for vlan%d.\n", vlan);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001083 free_netdev(dev);
1084 return NULL;
1085 }
1086
Greg Kroah-Hartmanf9cb0742007-12-17 23:05:35 -07001087 kobject_init(&port->kobject, &veth_port_ktype);
Greg Kroah-Hartmanb2d6db52007-12-17 23:05:35 -07001088 if (0 != kobject_add(&port->kobject, &dev->dev.kobj, "veth_port"))
Michael Ellerman07a5c172005-09-01 11:29:21 +10001089 veth_error("Failed adding port for %s to sysfs.\n", dev->name);
1090
Michael Ellerman61a3c692005-09-01 11:28:57 +10001091 veth_info("%s attached to iSeries vlan %d (LPAR map = 0x%.4X)\n",
1092 dev->name, vlan, port->lpar_map);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001093
1094 return dev;
1095}
1096
1097/*
1098 * Tx path
1099 */
1100
1101static int veth_transmit_to_one(struct sk_buff *skb, HvLpIndex rlp,
1102 struct net_device *dev)
1103{
1104 struct veth_lpar_connection *cnx = veth_cnx[rlp];
1105 struct veth_port *port = (struct veth_port *) dev->priv;
1106 HvLpEvent_Rc rc;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001107 struct veth_msg *msg = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001108 unsigned long flags;
1109
Michael Ellermandb5e8712005-09-01 11:29:19 +10001110 if (! cnx)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001111 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001112
1113 spin_lock_irqsave(&cnx->lock, flags);
1114
Michael Ellermanf27eff12005-05-12 17:47:27 +10001115 if (! (cnx->state & VETH_STATE_READY))
Michael Ellermandb5e8712005-09-01 11:29:19 +10001116 goto no_error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001117
Michael Ellermandb5e8712005-09-01 11:29:19 +10001118 if ((skb->len - ETH_HLEN) > VETH_MAX_MTU)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001119 goto drop;
1120
1121 msg = veth_stack_pop(cnx);
Michael Ellermandb5e8712005-09-01 11:29:19 +10001122 if (! msg)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001123 goto drop;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001124
Michael Ellermanb08bd5c2005-09-01 11:29:06 +10001125 msg->in_use = 1;
Michael Ellermandb5e8712005-09-01 11:29:19 +10001126 msg->skb = skb_get(skb);
Michael Ellermanb08bd5c2005-09-01 11:29:06 +10001127
Michael Ellermancbf90742005-09-01 11:29:07 +10001128 msg->data.addr[0] = dma_map_single(port->dev, skb->data,
1129 skb->len, DMA_TO_DEVICE);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001130
FUJITA Tomonori8d8bb392008-07-25 19:44:49 -07001131 if (dma_mapping_error(port->dev, msg->data.addr[0]))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001132 goto recycle_and_drop;
1133
Linus Torvalds1da177e2005-04-16 15:20:36 -07001134 msg->dev = port->dev;
Michael Ellermancbf90742005-09-01 11:29:07 +10001135 msg->data.len[0] = skb->len;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001136 msg->data.eofmask = 1 << VETH_EOF_SHIFT;
Michael Ellermancbf90742005-09-01 11:29:07 +10001137
Michael Ellerman59f17ae2005-09-01 11:29:27 +10001138 rc = veth_signaldata(cnx, VETH_EVENT_FRAMES, msg->token, &msg->data);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001139
1140 if (rc != HvLpEvent_Rc_Good)
1141 goto recycle_and_drop;
1142
Michael Ellerman24562ff2005-09-01 11:29:17 +10001143 /* If the timer's not already running, start it now. */
1144 if (0 == cnx->outstanding_tx)
1145 mod_timer(&cnx->reset_timer, jiffies + cnx->reset_timeout);
1146
1147 cnx->last_contact = jiffies;
1148 cnx->outstanding_tx++;
1149
Michael Ellermane0808492005-09-01 11:29:18 +10001150 if (veth_stack_is_empty(cnx))
1151 veth_stop_queues(cnx);
1152
Michael Ellermandb5e8712005-09-01 11:29:19 +10001153 no_error:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001154 spin_unlock_irqrestore(&cnx->lock, flags);
1155 return 0;
1156
1157 recycle_and_drop:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001158 veth_recycle_msg(cnx, msg);
1159 drop:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001160 spin_unlock_irqrestore(&cnx->lock, flags);
Michael Ellermandb5e8712005-09-01 11:29:19 +10001161 return 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001162}
1163
Michael Ellermandb5e8712005-09-01 11:29:19 +10001164static void veth_transmit_to_many(struct sk_buff *skb,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001165 HvLpIndexMap lpmask,
1166 struct net_device *dev)
1167{
Michael Ellermandb5e8712005-09-01 11:29:19 +10001168 int i, success, error;
1169
1170 success = error = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001171
1172 for (i = 0; i < HVMAXARCHITECTEDLPS; i++) {
1173 if ((lpmask & (1 << i)) == 0)
1174 continue;
1175
Michael Ellermandb5e8712005-09-01 11:29:19 +10001176 if (veth_transmit_to_one(skb, i, dev))
1177 error = 1;
1178 else
1179 success = 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001180 }
1181
Michael Ellermandb5e8712005-09-01 11:29:19 +10001182 if (error)
Jeff Garzik09f75cd2007-10-03 17:41:50 -07001183 dev->stats.tx_errors++;
Michael Ellermandb5e8712005-09-01 11:29:19 +10001184
1185 if (success) {
Jeff Garzik09f75cd2007-10-03 17:41:50 -07001186 dev->stats.tx_packets++;
1187 dev->stats.tx_bytes += skb->len;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001188 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001189}
1190
1191static int veth_start_xmit(struct sk_buff *skb, struct net_device *dev)
1192{
1193 unsigned char *frame = skb->data;
1194 struct veth_port *port = (struct veth_port *) dev->priv;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001195 HvLpIndexMap lpmask;
1196
1197 if (! (frame[0] & 0x01)) {
1198 /* unicast packet */
1199 HvLpIndex rlp = frame[5];
1200
1201 if ( ! ((1 << rlp) & port->lpar_map) ) {
1202 dev_kfree_skb(skb);
1203 return 0;
1204 }
1205
1206 lpmask = 1 << rlp;
1207 } else {
1208 lpmask = port->lpar_map;
1209 }
1210
Michael Ellermane0808492005-09-01 11:29:18 +10001211 veth_transmit_to_many(skb, lpmask, dev);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001212
Michael Ellermane0808492005-09-01 11:29:18 +10001213 dev_kfree_skb(skb);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001214
1215 return 0;
1216}
1217
Michael Ellermanb08bd5c2005-09-01 11:29:06 +10001218/* You must hold the connection's lock when you call this function. */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001219static void veth_recycle_msg(struct veth_lpar_connection *cnx,
1220 struct veth_msg *msg)
1221{
1222 u32 dma_address, dma_length;
1223
Michael Ellermanb08bd5c2005-09-01 11:29:06 +10001224 if (msg->in_use) {
1225 msg->in_use = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001226 dma_address = msg->data.addr[0];
1227 dma_length = msg->data.len[0];
1228
FUJITA Tomonori8d8bb392008-07-25 19:44:49 -07001229 if (!dma_mapping_error(msg->dev, dma_address))
Michael Ellermancbf90742005-09-01 11:29:07 +10001230 dma_unmap_single(msg->dev, dma_address, dma_length,
1231 DMA_TO_DEVICE);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001232
1233 if (msg->skb) {
1234 dev_kfree_skb_any(msg->skb);
1235 msg->skb = NULL;
1236 }
1237
1238 memset(&msg->data, 0, sizeof(msg->data));
1239 veth_stack_push(cnx, msg);
Michael Ellerman61a3c692005-09-01 11:28:57 +10001240 } else if (cnx->state & VETH_STATE_OPEN) {
1241 veth_error("Non-pending frame (# %d) acked by LPAR %d.\n",
1242 cnx->remote_lp, msg->token);
1243 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001244}
1245
Michael Ellermane0808492005-09-01 11:29:18 +10001246static void veth_wake_queues(struct veth_lpar_connection *cnx)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001247{
1248 int i;
Michael Ellermane0808492005-09-01 11:29:18 +10001249
Linus Torvalds1da177e2005-04-16 15:20:36 -07001250 for (i = 0; i < HVMAXARCHITECTEDVIRTUALLANS; i++) {
1251 struct net_device *dev = veth_dev[i];
1252 struct veth_port *port;
1253 unsigned long flags;
1254
1255 if (! dev)
1256 continue;
1257
1258 port = (struct veth_port *)dev->priv;
1259
1260 if (! (port->lpar_map & (1<<cnx->remote_lp)))
1261 continue;
1262
Michael Ellermane0808492005-09-01 11:29:18 +10001263 spin_lock_irqsave(&port->queue_lock, flags);
1264
1265 port->stopped_map &= ~(1 << cnx->remote_lp);
1266
1267 if (0 == port->stopped_map && netif_queue_stopped(dev)) {
1268 veth_debug("cnx %d: woke queue for %s.\n",
1269 cnx->remote_lp, dev->name);
1270 netif_wake_queue(dev);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001271 }
Michael Ellermane0808492005-09-01 11:29:18 +10001272 spin_unlock_irqrestore(&port->queue_lock, flags);
1273 }
1274}
1275
1276static void veth_stop_queues(struct veth_lpar_connection *cnx)
1277{
1278 int i;
1279
1280 for (i = 0; i < HVMAXARCHITECTEDVIRTUALLANS; i++) {
1281 struct net_device *dev = veth_dev[i];
1282 struct veth_port *port;
1283
1284 if (! dev)
1285 continue;
1286
1287 port = (struct veth_port *)dev->priv;
1288
1289 /* If this cnx is not on the vlan for this port, continue */
1290 if (! (port->lpar_map & (1 << cnx->remote_lp)))
1291 continue;
1292
1293 spin_lock(&port->queue_lock);
1294
1295 netif_stop_queue(dev);
1296 port->stopped_map |= (1 << cnx->remote_lp);
1297
1298 veth_debug("cnx %d: stopped queue for %s, map = 0x%x.\n",
1299 cnx->remote_lp, dev->name, port->stopped_map);
1300
1301 spin_unlock(&port->queue_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001302 }
1303}
1304
Michael Ellerman24562ff2005-09-01 11:29:17 +10001305static void veth_timed_reset(unsigned long ptr)
1306{
1307 struct veth_lpar_connection *cnx = (struct veth_lpar_connection *)ptr;
1308 unsigned long trigger_time, flags;
1309
1310 /* FIXME is it possible this fires after veth_stop_connection()?
1311 * That would reschedule the statemachine for 5 seconds and probably
1312 * execute it after the module's been unloaded. Hmm. */
1313
1314 spin_lock_irqsave(&cnx->lock, flags);
1315
1316 if (cnx->outstanding_tx > 0) {
1317 trigger_time = cnx->last_contact + cnx->reset_timeout;
1318
1319 if (trigger_time < jiffies) {
1320 cnx->state |= VETH_STATE_RESET;
1321 veth_kick_statemachine(cnx);
1322 veth_error("%d packets not acked by LPAR %d within %d "
1323 "seconds, resetting.\n",
1324 cnx->outstanding_tx, cnx->remote_lp,
1325 cnx->reset_timeout / HZ);
1326 } else {
1327 /* Reschedule the timer */
1328 trigger_time = jiffies + cnx->reset_timeout;
1329 mod_timer(&cnx->reset_timer, trigger_time);
1330 }
1331 }
1332
1333 spin_unlock_irqrestore(&cnx->lock, flags);
1334}
1335
Linus Torvalds1da177e2005-04-16 15:20:36 -07001336/*
1337 * Rx path
1338 */
1339
1340static inline int veth_frame_wanted(struct veth_port *port, u64 mac_addr)
1341{
1342 int wanted = 0;
1343 int i;
1344 unsigned long flags;
1345
1346 if ( (mac_addr == port->mac_addr) || (mac_addr == 0xffffffffffff0000) )
1347 return 1;
1348
Linus Torvalds1da177e2005-04-16 15:20:36 -07001349 read_lock_irqsave(&port->mcast_gate, flags);
1350
Michael Ellerman2a5391a2005-09-01 11:29:02 +10001351 if (port->promiscuous) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001352 wanted = 1;
1353 goto out;
1354 }
1355
1356 for (i = 0; i < port->num_mcast; ++i) {
1357 if (port->mcast_addr[i] == mac_addr) {
1358 wanted = 1;
1359 break;
1360 }
1361 }
1362
1363 out:
1364 read_unlock_irqrestore(&port->mcast_gate, flags);
1365
1366 return wanted;
1367}
1368
1369struct dma_chunk {
1370 u64 addr;
1371 u64 size;
1372};
1373
1374#define VETH_MAX_PAGES_PER_FRAME ( (VETH_MAX_MTU+PAGE_SIZE-2)/PAGE_SIZE + 1 )
1375
1376static inline void veth_build_dma_list(struct dma_chunk *list,
1377 unsigned char *p, unsigned long length)
1378{
1379 unsigned long done;
1380 int i = 1;
1381
1382 /* FIXME: skbs are continguous in real addresses. Do we
1383 * really need to break it into PAGE_SIZE chunks, or can we do
1384 * it just at the granularity of iSeries real->absolute
1385 * mapping? Indeed, given the way the allocator works, can we
1386 * count on them being absolutely contiguous? */
Stephen Rothwell426c1a12005-10-14 14:51:42 +10001387 list[0].addr = iseries_hv_addr(p);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001388 list[0].size = min(length,
1389 PAGE_SIZE - ((unsigned long)p & ~PAGE_MASK));
1390
1391 done = list[0].size;
1392 while (done < length) {
Stephen Rothwell426c1a12005-10-14 14:51:42 +10001393 list[i].addr = iseries_hv_addr(p + done);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001394 list[i].size = min(length-done, PAGE_SIZE);
1395 done += list[i].size;
1396 i++;
1397 }
1398}
1399
1400static void veth_flush_acks(struct veth_lpar_connection *cnx)
1401{
1402 HvLpEvent_Rc rc;
1403
Michael Ellerman59f17ae2005-09-01 11:29:27 +10001404 rc = veth_signaldata(cnx, VETH_EVENT_FRAMES_ACK,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001405 0, &cnx->pending_acks);
1406
1407 if (rc != HvLpEvent_Rc_Good)
Michael Ellerman61a3c692005-09-01 11:28:57 +10001408 veth_error("Failed acking frames from LPAR %d, rc = %d\n",
1409 cnx->remote_lp, (int)rc);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001410
1411 cnx->num_pending_acks = 0;
1412 memset(&cnx->pending_acks, 0xff, sizeof(cnx->pending_acks));
1413}
1414
1415static void veth_receive(struct veth_lpar_connection *cnx,
Michael Ellerman59f17ae2005-09-01 11:29:27 +10001416 struct veth_lpevent *event)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001417{
Michael Ellerman59f17ae2005-09-01 11:29:27 +10001418 struct veth_frames_data *senddata = &event->u.frames_data;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001419 int startchunk = 0;
1420 int nchunks;
1421 unsigned long flags;
1422 HvLpDma_Rc rc;
1423
1424 do {
1425 u16 length = 0;
1426 struct sk_buff *skb;
1427 struct dma_chunk local_list[VETH_MAX_PAGES_PER_FRAME];
1428 struct dma_chunk remote_list[VETH_MAX_FRAMES_PER_MSG];
1429 u64 dest;
1430 HvLpVirtualLanIndex vlan;
1431 struct net_device *dev;
1432 struct veth_port *port;
1433
1434 /* FIXME: do we need this? */
1435 memset(local_list, 0, sizeof(local_list));
1436 memset(remote_list, 0, sizeof(VETH_MAX_FRAMES_PER_MSG));
1437
1438 /* a 0 address marks the end of the valid entries */
1439 if (senddata->addr[startchunk] == 0)
1440 break;
1441
1442 /* make sure that we have at least 1 EOF entry in the
1443 * remaining entries */
1444 if (! (senddata->eofmask >> (startchunk + VETH_EOF_SHIFT))) {
Michael Ellerman61a3c692005-09-01 11:28:57 +10001445 veth_error("Missing EOF fragment in event "
1446 "eofmask = 0x%x startchunk = %d\n",
1447 (unsigned)senddata->eofmask,
1448 startchunk);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001449 break;
1450 }
1451
1452 /* build list of chunks in this frame */
1453 nchunks = 0;
1454 do {
1455 remote_list[nchunks].addr =
1456 (u64) senddata->addr[startchunk+nchunks] << 32;
1457 remote_list[nchunks].size =
1458 senddata->len[startchunk+nchunks];
1459 length += remote_list[nchunks].size;
1460 } while (! (senddata->eofmask &
1461 (1 << (VETH_EOF_SHIFT + startchunk + nchunks++))));
1462
1463 /* length == total length of all chunks */
1464 /* nchunks == # of chunks in this frame */
1465
1466 if ((length - ETH_HLEN) > VETH_MAX_MTU) {
Michael Ellerman61a3c692005-09-01 11:28:57 +10001467 veth_error("Received oversize frame from LPAR %d "
1468 "(length = %d)\n",
1469 cnx->remote_lp, length);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001470 continue;
1471 }
1472
1473 skb = alloc_skb(length, GFP_ATOMIC);
1474 if (!skb)
1475 continue;
1476
1477 veth_build_dma_list(local_list, skb->data, length);
1478
1479 rc = HvCallEvent_dmaBufList(HvLpEvent_Type_VirtualLan,
1480 event->base_event.xSourceLp,
1481 HvLpDma_Direction_RemoteToLocal,
1482 cnx->src_inst,
1483 cnx->dst_inst,
1484 HvLpDma_AddressType_RealAddress,
1485 HvLpDma_AddressType_TceIndex,
Stephen Rothwell426c1a12005-10-14 14:51:42 +10001486 iseries_hv_addr(&local_list),
1487 iseries_hv_addr(&remote_list),
Linus Torvalds1da177e2005-04-16 15:20:36 -07001488 length);
1489 if (rc != HvLpDma_Rc_Good) {
1490 dev_kfree_skb_irq(skb);
1491 continue;
1492 }
1493
1494 vlan = skb->data[9];
1495 dev = veth_dev[vlan];
Michael Ellerman41664c02005-05-12 17:55:08 +10001496 if (! dev) {
1497 /*
1498 * Some earlier versions of the driver sent
1499 * broadcasts down all connections, even to lpars
1500 * that weren't on the relevant vlan. So ignore
1501 * packets belonging to a vlan we're not on.
1502 * We can also be here if we receive packets while
1503 * the driver is going down, because then dev is NULL.
1504 */
1505 dev_kfree_skb_irq(skb);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001506 continue;
Michael Ellerman41664c02005-05-12 17:55:08 +10001507 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001508
1509 port = (struct veth_port *)dev->priv;
1510 dest = *((u64 *) skb->data) & 0xFFFFFFFFFFFF0000;
1511
1512 if ((vlan > HVMAXARCHITECTEDVIRTUALLANS) || !port) {
1513 dev_kfree_skb_irq(skb);
1514 continue;
1515 }
1516 if (! veth_frame_wanted(port, dest)) {
1517 dev_kfree_skb_irq(skb);
1518 continue;
1519 }
1520
1521 skb_put(skb, length);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001522 skb->protocol = eth_type_trans(skb, dev);
1523 skb->ip_summed = CHECKSUM_NONE;
1524 netif_rx(skb); /* send it up */
Jeff Garzik09f75cd2007-10-03 17:41:50 -07001525 dev->stats.rx_packets++;
1526 dev->stats.rx_bytes += length;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001527 } while (startchunk += nchunks, startchunk < VETH_MAX_FRAMES_PER_MSG);
1528
1529 /* Ack it */
1530 spin_lock_irqsave(&cnx->lock, flags);
1531 BUG_ON(cnx->num_pending_acks > VETH_MAX_ACKS_PER_MSG);
1532
1533 cnx->pending_acks[cnx->num_pending_acks++] =
1534 event->base_event.xCorrelationToken;
1535
1536 if ( (cnx->num_pending_acks >= cnx->remote_caps.ack_threshold)
1537 || (cnx->num_pending_acks >= VETH_MAX_ACKS_PER_MSG) )
1538 veth_flush_acks(cnx);
1539
1540 spin_unlock_irqrestore(&cnx->lock, flags);
1541}
1542
1543static void veth_timed_ack(unsigned long ptr)
1544{
1545 struct veth_lpar_connection *cnx = (struct veth_lpar_connection *) ptr;
1546 unsigned long flags;
1547
1548 /* Ack all the events */
1549 spin_lock_irqsave(&cnx->lock, flags);
1550 if (cnx->num_pending_acks > 0)
1551 veth_flush_acks(cnx);
1552
1553 /* Reschedule the timer */
1554 cnx->ack_timer.expires = jiffies + cnx->ack_timeout;
1555 add_timer(&cnx->ack_timer);
1556 spin_unlock_irqrestore(&cnx->lock, flags);
1557}
1558
1559static int veth_remove(struct vio_dev *vdev)
1560{
Michael Ellermanf0c129c2005-09-01 11:29:09 +10001561 struct veth_lpar_connection *cnx;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001562 struct net_device *dev;
Michael Ellermanf0c129c2005-09-01 11:29:09 +10001563 struct veth_port *port;
1564 int i;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001565
Michael Ellermanf0c129c2005-09-01 11:29:09 +10001566 dev = veth_dev[vdev->unit_address];
1567
1568 if (! dev)
1569 return 0;
1570
1571 port = netdev_priv(dev);
1572
1573 for (i = 0; i < HVMAXARCHITECTEDLPS; i++) {
1574 cnx = veth_cnx[i];
1575
1576 if (cnx && (port->lpar_map & (1 << i))) {
1577 /* Drop our reference to connections on our VLAN */
1578 kobject_put(&cnx->kobject);
1579 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001580 }
Michael Ellermanf0c129c2005-09-01 11:29:09 +10001581
1582 veth_dev[vdev->unit_address] = NULL;
Michael Ellerman07a5c172005-09-01 11:29:21 +10001583 kobject_del(&port->kobject);
1584 kobject_put(&port->kobject);
Michael Ellermanf0c129c2005-09-01 11:29:09 +10001585 unregister_netdev(dev);
1586 free_netdev(dev);
1587
Linus Torvalds1da177e2005-04-16 15:20:36 -07001588 return 0;
1589}
1590
1591static int veth_probe(struct vio_dev *vdev, const struct vio_device_id *id)
1592{
1593 int i = vdev->unit_address;
1594 struct net_device *dev;
Michael Ellermanf0c129c2005-09-01 11:29:09 +10001595 struct veth_port *port;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001596
Stephen Rothwell07fb3f42006-05-15 13:39:47 +10001597 dev = veth_probe_one(i, vdev);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001598 if (dev == NULL) {
1599 veth_remove(vdev);
1600 return 1;
1601 }
1602 veth_dev[i] = dev;
1603
Michael Ellermanf0c129c2005-09-01 11:29:09 +10001604 port = (struct veth_port*)netdev_priv(dev);
1605
1606 /* Start the state machine on each connection on this vlan. If we're
1607 * the first dev to do so this will commence link negotiation */
1608 for (i = 0; i < HVMAXARCHITECTEDLPS; i++) {
1609 struct veth_lpar_connection *cnx;
1610
1611 if (! (port->lpar_map & (1 << i)))
1612 continue;
1613
1614 cnx = veth_cnx[i];
1615 if (!cnx)
1616 continue;
1617
1618 kobject_get(&cnx->kobject);
1619 veth_kick_statemachine(cnx);
1620 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001621
1622 return 0;
1623}
1624
1625/**
1626 * veth_device_table: Used by vio.c to match devices that we
1627 * support.
1628 */
1629static struct vio_device_id veth_device_table[] __devinitdata = {
Stephen Rothwell07fb3f42006-05-15 13:39:47 +10001630 { "network", "IBM,iSeries-l-lan" },
Stephen Rothwellfb120da2005-08-17 16:42:59 +10001631 { "", "" }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001632};
1633MODULE_DEVICE_TABLE(vio, veth_device_table);
1634
1635static struct vio_driver veth_driver = {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001636 .id_table = veth_device_table,
1637 .probe = veth_probe,
Stephen Rothwell6fdf5392005-10-24 14:53:21 +10001638 .remove = veth_remove,
1639 .driver = {
1640 .name = DRV_NAME,
Stephen Rothwell915124d2005-10-24 15:12:22 +10001641 .owner = THIS_MODULE,
Stephen Rothwell6fdf5392005-10-24 14:53:21 +10001642 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001643};
1644
1645/*
1646 * Module initialization/cleanup
1647 */
1648
Stephen Rothwell687d18a2006-12-15 15:42:50 +11001649static void __exit veth_module_cleanup(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001650{
1651 int i;
Michael Ellermanf0c129c2005-09-01 11:29:09 +10001652 struct veth_lpar_connection *cnx;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001653
Michael Ellermanf0c129c2005-09-01 11:29:09 +10001654 /* Disconnect our "irq" to stop events coming from the Hypervisor. */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001655 HvLpEvent_unregisterHandler(HvLpEvent_Type_VirtualLan);
1656
Michael Ellermanf0c129c2005-09-01 11:29:09 +10001657 /* Make sure any work queued from Hypervisor callbacks is finished. */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001658 flush_scheduled_work();
1659
Michael Ellermanf0c129c2005-09-01 11:29:09 +10001660 for (i = 0; i < HVMAXARCHITECTEDLPS; ++i) {
1661 cnx = veth_cnx[i];
1662
1663 if (!cnx)
1664 continue;
1665
Michael Ellerman76812d8122005-09-01 11:29:20 +10001666 /* Remove the connection from sysfs */
1667 kobject_del(&cnx->kobject);
Michael Ellermanf0c129c2005-09-01 11:29:09 +10001668 /* Drop the driver's reference to the connection */
1669 kobject_put(&cnx->kobject);
1670 }
1671
1672 /* Unregister the driver, which will close all the netdevs and stop
1673 * the connections when they're no longer referenced. */
Michael Ellermanb2e08522005-05-12 18:09:45 +10001674 vio_unregister_driver(&veth_driver);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001675}
1676module_exit(veth_module_cleanup);
1677
Stephen Rothwell687d18a2006-12-15 15:42:50 +11001678static int __init veth_module_init(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001679{
1680 int i;
1681 int rc;
1682
Stephen Rothwell687d18a2006-12-15 15:42:50 +11001683 if (!firmware_has_feature(FW_FEATURE_ISERIES))
1684 return -ENODEV;
1685
Linus Torvalds1da177e2005-04-16 15:20:36 -07001686 this_lp = HvLpConfig_getLpIndex_outline();
1687
1688 for (i = 0; i < HVMAXARCHITECTEDLPS; ++i) {
1689 rc = veth_init_connection(i);
Michael Ellermanec60bee2005-09-01 11:29:08 +10001690 if (rc != 0)
1691 goto error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001692 }
1693
1694 HvLpEvent_registerHandler(HvLpEvent_Type_VirtualLan,
1695 &veth_handle_event);
1696
Michael Ellermanec60bee2005-09-01 11:29:08 +10001697 rc = vio_register_driver(&veth_driver);
1698 if (rc != 0)
1699 goto error;
1700
Michael Ellerman76812d8122005-09-01 11:29:20 +10001701 for (i = 0; i < HVMAXARCHITECTEDLPS; ++i) {
1702 struct kobject *kobj;
1703
1704 if (!veth_cnx[i])
1705 continue;
1706
1707 kobj = &veth_cnx[i]->kobject;
Michael Ellerman76812d8122005-09-01 11:29:20 +10001708 /* If the add failes, complain but otherwise continue */
Greg Kroah-Hartmancbe9c592007-12-19 15:54:39 -04001709 if (0 != driver_add_kobj(&veth_driver.driver, kobj,
Greg Kroah-Hartman78be3d22007-12-17 23:05:35 -07001710 "cnx%.2d", veth_cnx[i]->remote_lp))
Michael Ellerman76812d8122005-09-01 11:29:20 +10001711 veth_error("cnx %d: Failed adding to sysfs.\n", i);
1712 }
1713
Michael Ellermanec60bee2005-09-01 11:29:08 +10001714 return 0;
1715
1716error:
1717 for (i = 0; i < HVMAXARCHITECTEDLPS; ++i) {
Michael Ellermanf0c129c2005-09-01 11:29:09 +10001718 veth_destroy_connection(veth_cnx[i]);
Michael Ellermanec60bee2005-09-01 11:29:08 +10001719 }
1720
1721 return rc;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001722}
1723module_init(veth_module_init);