blob: 8df645e78f2e9618bd22b83225efbfde4f43008a [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>
Tejun Heo5a0e3ad2010-03-24 17:04:11 +090072#include <linux/slab.h>
Stephen Rothwell426c1a12005-10-14 14:51:42 +100073
74#include <asm/abs_addr.h>
Kelly Dalybbc8b622005-11-02 15:10:38 +110075#include <asm/iseries/mf.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070076#include <asm/uaccess.h>
Stephen Rothwell687d18a2006-12-15 15:42:50 +110077#include <asm/firmware.h>
Kelly Daly15b17182005-11-02 11:55:28 +110078#include <asm/iseries/hv_lp_config.h>
Kelly Daly1ec65d72005-11-02 13:46:07 +110079#include <asm/iseries/hv_types.h>
Kelly Dalye45423e2005-11-02 12:08:31 +110080#include <asm/iseries/hv_lp_event.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070081#include <asm/iommu.h>
82#include <asm/vio.h>
83
Michael Ellerman61a3c692005-09-01 11:28:57 +100084#undef DEBUG
85
Linus Torvalds1da177e2005-04-16 15:20:36 -070086MODULE_AUTHOR("Kyle Lucke <klucke@us.ibm.com>");
87MODULE_DESCRIPTION("iSeries Virtual ethernet driver");
88MODULE_LICENSE("GPL");
89
Michael Ellerman59f17ae2005-09-01 11:29:27 +100090#define VETH_EVENT_CAP (0)
91#define VETH_EVENT_FRAMES (1)
92#define VETH_EVENT_MONITOR (2)
93#define VETH_EVENT_FRAMES_ACK (3)
Michael Ellerman642d1a42005-09-01 11:29:25 +100094
95#define VETH_MAX_ACKS_PER_MSG (20)
96#define VETH_MAX_FRAMES_PER_MSG (6)
97
Michael Ellerman59f17ae2005-09-01 11:29:27 +100098struct veth_frames_data {
Michael Ellerman642d1a42005-09-01 11:29:25 +100099 u32 addr[VETH_MAX_FRAMES_PER_MSG];
100 u16 len[VETH_MAX_FRAMES_PER_MSG];
101 u32 eofmask;
102};
103#define VETH_EOF_SHIFT (32-VETH_MAX_FRAMES_PER_MSG)
104
Michael Ellerman59f17ae2005-09-01 11:29:27 +1000105struct veth_frames_ack_data {
Michael Ellerman642d1a42005-09-01 11:29:25 +1000106 u16 token[VETH_MAX_ACKS_PER_MSG];
107};
108
Michael Ellerman59f17ae2005-09-01 11:29:27 +1000109struct veth_cap_data {
Michael Ellerman642d1a42005-09-01 11:29:25 +1000110 u8 caps_version;
111 u8 rsvd1;
112 u16 num_buffers;
113 u16 ack_threshold;
114 u16 rsvd2;
115 u32 ack_timeout;
116 u32 rsvd3;
117 u64 rsvd4[3];
118};
119
Michael Ellerman59f17ae2005-09-01 11:29:27 +1000120struct veth_lpevent {
Michael Ellerman642d1a42005-09-01 11:29:25 +1000121 struct HvLpEvent base_event;
122 union {
Michael Ellerman59f17ae2005-09-01 11:29:27 +1000123 struct veth_cap_data caps_data;
124 struct veth_frames_data frames_data;
125 struct veth_frames_ack_data frames_ack_data;
Michael Ellerman642d1a42005-09-01 11:29:25 +1000126 } u;
127
128};
129
Michael Ellermanee05f032005-09-01 11:29:29 +1000130#define DRV_NAME "iseries_veth"
131#define DRV_VERSION "2.0"
132
Linus Torvalds1da177e2005-04-16 15:20:36 -0700133#define VETH_NUMBUFFERS (120)
134#define VETH_ACKTIMEOUT (1000000) /* microseconds */
135#define VETH_MAX_MCAST (12)
136
137#define VETH_MAX_MTU (9000)
138
139#if VETH_NUMBUFFERS < 10
140#define ACK_THRESHOLD (1)
141#elif VETH_NUMBUFFERS < 20
142#define ACK_THRESHOLD (4)
143#elif VETH_NUMBUFFERS < 40
144#define ACK_THRESHOLD (10)
145#else
146#define ACK_THRESHOLD (20)
147#endif
148
149#define VETH_STATE_SHUTDOWN (0x0001)
150#define VETH_STATE_OPEN (0x0002)
151#define VETH_STATE_RESET (0x0004)
152#define VETH_STATE_SENTMON (0x0008)
153#define VETH_STATE_SENTCAPS (0x0010)
154#define VETH_STATE_GOTCAPACK (0x0020)
155#define VETH_STATE_GOTCAPS (0x0040)
156#define VETH_STATE_SENTCAPACK (0x0080)
157#define VETH_STATE_READY (0x0100)
158
159struct veth_msg {
160 struct veth_msg *next;
Michael Ellerman59f17ae2005-09-01 11:29:27 +1000161 struct veth_frames_data data;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700162 int token;
Michael Ellermanb08bd5c2005-09-01 11:29:06 +1000163 int in_use;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700164 struct sk_buff *skb;
165 struct device *dev;
166};
167
168struct veth_lpar_connection {
169 HvLpIndex remote_lp;
David Howellsc4028952006-11-22 14:57:56 +0000170 struct delayed_work statemachine_wq;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700171 struct veth_msg *msgs;
172 int num_events;
Michael Ellerman59f17ae2005-09-01 11:29:27 +1000173 struct veth_cap_data local_caps;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700174
Michael Ellermanf0c129c2005-09-01 11:29:09 +1000175 struct kobject kobject;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700176 struct timer_list ack_timer;
177
Michael Ellerman24562ff2005-09-01 11:29:17 +1000178 struct timer_list reset_timer;
179 unsigned int reset_timeout;
180 unsigned long last_contact;
181 int outstanding_tx;
182
Linus Torvalds1da177e2005-04-16 15:20:36 -0700183 spinlock_t lock;
184 unsigned long state;
185 HvLpInstanceId src_inst;
186 HvLpInstanceId dst_inst;
Michael Ellerman59f17ae2005-09-01 11:29:27 +1000187 struct veth_lpevent cap_event, cap_ack_event;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700188 u16 pending_acks[VETH_MAX_ACKS_PER_MSG];
189 u32 num_pending_acks;
190
191 int num_ack_events;
Michael Ellerman59f17ae2005-09-01 11:29:27 +1000192 struct veth_cap_data remote_caps;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700193 u32 ack_timeout;
194
Linus Torvalds1da177e2005-04-16 15:20:36 -0700195 struct veth_msg *msg_stack_head;
196};
197
198struct veth_port {
199 struct device *dev;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700200 u64 mac_addr;
201 HvLpIndexMap lpar_map;
202
Michael Ellermane0808492005-09-01 11:29:18 +1000203 /* queue_lock protects the stopped_map and dev's queue. */
204 spinlock_t queue_lock;
205 HvLpIndexMap stopped_map;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700206
Michael Ellermane0808492005-09-01 11:29:18 +1000207 /* mcast_gate protects promiscuous, num_mcast & mcast_addr. */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700208 rwlock_t mcast_gate;
209 int promiscuous;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700210 int num_mcast;
211 u64 mcast_addr[VETH_MAX_MCAST];
Michael Ellerman07a5c172005-09-01 11:29:21 +1000212
213 struct kobject kobject;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700214};
215
216static HvLpIndex this_lp;
217static struct veth_lpar_connection *veth_cnx[HVMAXARCHITECTEDLPS]; /* = 0 */
218static struct net_device *veth_dev[HVMAXARCHITECTEDVIRTUALLANS]; /* = 0 */
219
220static int veth_start_xmit(struct sk_buff *skb, struct net_device *dev);
221static void veth_recycle_msg(struct veth_lpar_connection *, struct veth_msg *);
Michael Ellermane0808492005-09-01 11:29:18 +1000222static void veth_wake_queues(struct veth_lpar_connection *cnx);
223static void veth_stop_queues(struct veth_lpar_connection *cnx);
Michael Ellerman59f17ae2005-09-01 11:29:27 +1000224static void veth_receive(struct veth_lpar_connection *, struct veth_lpevent *);
Michael Ellermanf0c129c2005-09-01 11:29:09 +1000225static void veth_release_connection(struct kobject *kobject);
Michael Ellerman24562ff2005-09-01 11:29:17 +1000226static void veth_timed_ack(unsigned long ptr);
227static void veth_timed_reset(unsigned long ptr);
Michael Ellermanf0c129c2005-09-01 11:29:09 +1000228
Linus Torvalds1da177e2005-04-16 15:20:36 -0700229/*
230 * Utility functions
231 */
232
Michael Ellerman61a3c692005-09-01 11:28:57 +1000233#define veth_info(fmt, args...) \
Michael Ellermanee05f032005-09-01 11:29:29 +1000234 printk(KERN_INFO DRV_NAME ": " fmt, ## args)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700235
236#define veth_error(fmt, args...) \
Michael Ellermanee05f032005-09-01 11:29:29 +1000237 printk(KERN_ERR DRV_NAME ": Error: " fmt, ## args)
Michael Ellerman61a3c692005-09-01 11:28:57 +1000238
239#ifdef DEBUG
240#define veth_debug(fmt, args...) \
Michael Ellermanee05f032005-09-01 11:29:29 +1000241 printk(KERN_DEBUG DRV_NAME ": " fmt, ## args)
Michael Ellerman61a3c692005-09-01 11:28:57 +1000242#else
243#define veth_debug(fmt, args...) do {} while (0)
244#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -0700245
Michael Ellermand7893dd2005-09-01 11:29:05 +1000246/* You must hold the connection's lock when you call this function. */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700247static inline void veth_stack_push(struct veth_lpar_connection *cnx,
248 struct veth_msg *msg)
249{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700250 msg->next = cnx->msg_stack_head;
251 cnx->msg_stack_head = msg;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700252}
253
Michael Ellermand7893dd2005-09-01 11:29:05 +1000254/* You must hold the connection's lock when you call this function. */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700255static inline struct veth_msg *veth_stack_pop(struct veth_lpar_connection *cnx)
256{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700257 struct veth_msg *msg;
258
Linus Torvalds1da177e2005-04-16 15:20:36 -0700259 msg = cnx->msg_stack_head;
260 if (msg)
261 cnx->msg_stack_head = cnx->msg_stack_head->next;
Michael Ellermand7893dd2005-09-01 11:29:05 +1000262
Linus Torvalds1da177e2005-04-16 15:20:36 -0700263 return msg;
264}
265
Michael Ellermane0808492005-09-01 11:29:18 +1000266/* You must hold the connection's lock when you call this function. */
267static inline int veth_stack_is_empty(struct veth_lpar_connection *cnx)
268{
269 return cnx->msg_stack_head == NULL;
270}
271
Linus Torvalds1da177e2005-04-16 15:20:36 -0700272static inline HvLpEvent_Rc
273veth_signalevent(struct veth_lpar_connection *cnx, u16 subtype,
274 HvLpEvent_AckInd ackind, HvLpEvent_AckType acktype,
275 u64 token,
276 u64 data1, u64 data2, u64 data3, u64 data4, u64 data5)
277{
278 return HvCallEvent_signalLpEventFast(cnx->remote_lp,
279 HvLpEvent_Type_VirtualLan,
280 subtype, ackind, acktype,
281 cnx->src_inst,
282 cnx->dst_inst,
283 token, data1, data2, data3,
284 data4, data5);
285}
286
287static inline HvLpEvent_Rc veth_signaldata(struct veth_lpar_connection *cnx,
288 u16 subtype, u64 token, void *data)
289{
290 u64 *p = (u64 *) data;
291
292 return veth_signalevent(cnx, subtype, HvLpEvent_AckInd_NoAck,
293 HvLpEvent_AckType_ImmediateAck,
294 token, p[0], p[1], p[2], p[3], p[4]);
295}
296
297struct veth_allocation {
298 struct completion c;
299 int num;
300};
301
302static void veth_complete_allocation(void *parm, int number)
303{
304 struct veth_allocation *vc = (struct veth_allocation *)parm;
305
306 vc->num = number;
307 complete(&vc->c);
308}
309
310static int veth_allocate_events(HvLpIndex rlp, int number)
311{
Benjamin Herrenschmidta2879fe2008-04-09 17:21:34 +1000312 struct veth_allocation vc =
313 { COMPLETION_INITIALIZER_ONSTACK(vc.c), 0 };
Linus Torvalds1da177e2005-04-16 15:20:36 -0700314
315 mf_allocate_lp_events(rlp, HvLpEvent_Type_VirtualLan,
Michael Ellerman59f17ae2005-09-01 11:29:27 +1000316 sizeof(struct veth_lpevent), number,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700317 &veth_complete_allocation, &vc);
318 wait_for_completion(&vc.c);
319
320 return vc.num;
321}
322
323/*
Michael Ellerman76812d8122005-09-01 11:29:20 +1000324 * sysfs support
325 */
326
327struct veth_cnx_attribute {
328 struct attribute attr;
329 ssize_t (*show)(struct veth_lpar_connection *, char *buf);
330 ssize_t (*store)(struct veth_lpar_connection *, const char *buf);
331};
332
333static ssize_t veth_cnx_attribute_show(struct kobject *kobj,
334 struct attribute *attr, char *buf)
335{
336 struct veth_cnx_attribute *cnx_attr;
337 struct veth_lpar_connection *cnx;
338
339 cnx_attr = container_of(attr, struct veth_cnx_attribute, attr);
340 cnx = container_of(kobj, struct veth_lpar_connection, kobject);
341
342 if (!cnx_attr->show)
343 return -EIO;
344
345 return cnx_attr->show(cnx, buf);
346}
347
348#define CUSTOM_CNX_ATTR(_name, _format, _expression) \
349static ssize_t _name##_show(struct veth_lpar_connection *cnx, char *buf)\
350{ \
351 return sprintf(buf, _format, _expression); \
352} \
353struct veth_cnx_attribute veth_cnx_attr_##_name = __ATTR_RO(_name)
354
355#define SIMPLE_CNX_ATTR(_name) \
356 CUSTOM_CNX_ATTR(_name, "%lu\n", (unsigned long)cnx->_name)
357
358SIMPLE_CNX_ATTR(outstanding_tx);
359SIMPLE_CNX_ATTR(remote_lp);
360SIMPLE_CNX_ATTR(num_events);
361SIMPLE_CNX_ATTR(src_inst);
362SIMPLE_CNX_ATTR(dst_inst);
363SIMPLE_CNX_ATTR(num_pending_acks);
364SIMPLE_CNX_ATTR(num_ack_events);
365CUSTOM_CNX_ATTR(ack_timeout, "%d\n", jiffies_to_msecs(cnx->ack_timeout));
366CUSTOM_CNX_ATTR(reset_timeout, "%d\n", jiffies_to_msecs(cnx->reset_timeout));
367CUSTOM_CNX_ATTR(state, "0x%.4lX\n", cnx->state);
368CUSTOM_CNX_ATTR(last_contact, "%d\n", cnx->last_contact ?
369 jiffies_to_msecs(jiffies - cnx->last_contact) : 0);
370
371#define GET_CNX_ATTR(_name) (&veth_cnx_attr_##_name.attr)
372
373static struct attribute *veth_cnx_default_attrs[] = {
374 GET_CNX_ATTR(outstanding_tx),
375 GET_CNX_ATTR(remote_lp),
376 GET_CNX_ATTR(num_events),
377 GET_CNX_ATTR(reset_timeout),
378 GET_CNX_ATTR(last_contact),
379 GET_CNX_ATTR(state),
380 GET_CNX_ATTR(src_inst),
381 GET_CNX_ATTR(dst_inst),
382 GET_CNX_ATTR(num_pending_acks),
383 GET_CNX_ATTR(num_ack_events),
384 GET_CNX_ATTR(ack_timeout),
385 NULL
386};
387
Emese Revfy52cf25d2010-01-19 02:58:23 +0100388static const struct sysfs_ops veth_cnx_sysfs_ops = {
Michael Ellerman76812d8122005-09-01 11:29:20 +1000389 .show = veth_cnx_attribute_show
390};
391
392static struct kobj_type veth_lpar_connection_ktype = {
393 .release = veth_release_connection,
394 .sysfs_ops = &veth_cnx_sysfs_ops,
395 .default_attrs = veth_cnx_default_attrs
396};
397
Michael Ellerman07a5c172005-09-01 11:29:21 +1000398struct veth_port_attribute {
399 struct attribute attr;
400 ssize_t (*show)(struct veth_port *, char *buf);
401 ssize_t (*store)(struct veth_port *, const char *buf);
402};
403
404static ssize_t veth_port_attribute_show(struct kobject *kobj,
405 struct attribute *attr, char *buf)
406{
407 struct veth_port_attribute *port_attr;
408 struct veth_port *port;
409
410 port_attr = container_of(attr, struct veth_port_attribute, attr);
411 port = container_of(kobj, struct veth_port, kobject);
412
413 if (!port_attr->show)
414 return -EIO;
415
416 return port_attr->show(port, buf);
417}
418
419#define CUSTOM_PORT_ATTR(_name, _format, _expression) \
420static ssize_t _name##_show(struct veth_port *port, char *buf) \
421{ \
422 return sprintf(buf, _format, _expression); \
423} \
424struct veth_port_attribute veth_port_attr_##_name = __ATTR_RO(_name)
425
426#define SIMPLE_PORT_ATTR(_name) \
427 CUSTOM_PORT_ATTR(_name, "%lu\n", (unsigned long)port->_name)
428
429SIMPLE_PORT_ATTR(promiscuous);
430SIMPLE_PORT_ATTR(num_mcast);
431CUSTOM_PORT_ATTR(lpar_map, "0x%X\n", port->lpar_map);
432CUSTOM_PORT_ATTR(stopped_map, "0x%X\n", port->stopped_map);
Stephen Rothwella1c5a892009-01-06 14:40:06 +0000433CUSTOM_PORT_ATTR(mac_addr, "0x%llX\n", port->mac_addr);
Michael Ellerman07a5c172005-09-01 11:29:21 +1000434
435#define GET_PORT_ATTR(_name) (&veth_port_attr_##_name.attr)
436static struct attribute *veth_port_default_attrs[] = {
437 GET_PORT_ATTR(mac_addr),
438 GET_PORT_ATTR(lpar_map),
439 GET_PORT_ATTR(stopped_map),
440 GET_PORT_ATTR(promiscuous),
441 GET_PORT_ATTR(num_mcast),
442 NULL
443};
444
Emese Revfy52cf25d2010-01-19 02:58:23 +0100445static const struct sysfs_ops veth_port_sysfs_ops = {
Michael Ellerman07a5c172005-09-01 11:29:21 +1000446 .show = veth_port_attribute_show
447};
448
449static struct kobj_type veth_port_ktype = {
450 .sysfs_ops = &veth_port_sysfs_ops,
451 .default_attrs = veth_port_default_attrs
452};
453
Michael Ellerman76812d8122005-09-01 11:29:20 +1000454/*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700455 * LPAR connection code
456 */
457
458static inline void veth_kick_statemachine(struct veth_lpar_connection *cnx)
459{
David Howellsc4028952006-11-22 14:57:56 +0000460 schedule_delayed_work(&cnx->statemachine_wq, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700461}
462
463static void veth_take_cap(struct veth_lpar_connection *cnx,
Michael Ellerman59f17ae2005-09-01 11:29:27 +1000464 struct veth_lpevent *event)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700465{
466 unsigned long flags;
467
468 spin_lock_irqsave(&cnx->lock, flags);
469 /* Receiving caps may mean the other end has just come up, so
470 * we need to reload the instance ID of the far end */
471 cnx->dst_inst =
472 HvCallEvent_getTargetLpInstanceId(cnx->remote_lp,
473 HvLpEvent_Type_VirtualLan);
474
475 if (cnx->state & VETH_STATE_GOTCAPS) {
Michael Ellerman61a3c692005-09-01 11:28:57 +1000476 veth_error("Received a second capabilities from LPAR %d.\n",
Linus Torvalds1da177e2005-04-16 15:20:36 -0700477 cnx->remote_lp);
478 event->base_event.xRc = HvLpEvent_Rc_BufferNotAvailable;
479 HvCallEvent_ackLpEvent((struct HvLpEvent *) event);
480 } else {
481 memcpy(&cnx->cap_event, event, sizeof(cnx->cap_event));
482 cnx->state |= VETH_STATE_GOTCAPS;
483 veth_kick_statemachine(cnx);
484 }
485 spin_unlock_irqrestore(&cnx->lock, flags);
486}
487
488static void veth_take_cap_ack(struct veth_lpar_connection *cnx,
Michael Ellerman59f17ae2005-09-01 11:29:27 +1000489 struct veth_lpevent *event)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700490{
491 unsigned long flags;
492
493 spin_lock_irqsave(&cnx->lock, flags);
494 if (cnx->state & VETH_STATE_GOTCAPACK) {
Michael Ellerman61a3c692005-09-01 11:28:57 +1000495 veth_error("Received a second capabilities ack from LPAR %d.\n",
Linus Torvalds1da177e2005-04-16 15:20:36 -0700496 cnx->remote_lp);
497 } else {
498 memcpy(&cnx->cap_ack_event, event,
Jean Delvareb607bd92009-10-02 09:55:19 -0700499 sizeof(cnx->cap_ack_event));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700500 cnx->state |= VETH_STATE_GOTCAPACK;
501 veth_kick_statemachine(cnx);
502 }
503 spin_unlock_irqrestore(&cnx->lock, flags);
504}
505
506static void veth_take_monitor_ack(struct veth_lpar_connection *cnx,
Michael Ellerman59f17ae2005-09-01 11:29:27 +1000507 struct veth_lpevent *event)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700508{
509 unsigned long flags;
510
511 spin_lock_irqsave(&cnx->lock, flags);
Michael Ellerman61a3c692005-09-01 11:28:57 +1000512 veth_debug("cnx %d: lost connection.\n", cnx->remote_lp);
Michael Ellerman58c59002005-09-01 11:29:00 +1000513
514 /* Avoid kicking the statemachine once we're shutdown.
515 * It's unnecessary and it could break veth_stop_connection(). */
516
517 if (! (cnx->state & VETH_STATE_SHUTDOWN)) {
518 cnx->state |= VETH_STATE_RESET;
519 veth_kick_statemachine(cnx);
520 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700521 spin_unlock_irqrestore(&cnx->lock, flags);
522}
523
Michael Ellerman59f17ae2005-09-01 11:29:27 +1000524static void veth_handle_ack(struct veth_lpevent *event)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700525{
526 HvLpIndex rlp = event->base_event.xTargetLp;
527 struct veth_lpar_connection *cnx = veth_cnx[rlp];
528
529 BUG_ON(! cnx);
530
531 switch (event->base_event.xSubtype) {
Michael Ellerman59f17ae2005-09-01 11:29:27 +1000532 case VETH_EVENT_CAP:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700533 veth_take_cap_ack(cnx, event);
534 break;
Michael Ellerman59f17ae2005-09-01 11:29:27 +1000535 case VETH_EVENT_MONITOR:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700536 veth_take_monitor_ack(cnx, event);
537 break;
538 default:
Michael Ellerman61a3c692005-09-01 11:28:57 +1000539 veth_error("Unknown ack type %d from LPAR %d.\n",
540 event->base_event.xSubtype, rlp);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700541 };
542}
543
Michael Ellerman59f17ae2005-09-01 11:29:27 +1000544static void veth_handle_int(struct veth_lpevent *event)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700545{
546 HvLpIndex rlp = event->base_event.xSourceLp;
547 struct veth_lpar_connection *cnx = veth_cnx[rlp];
548 unsigned long flags;
Michael Ellerman24562ff2005-09-01 11:29:17 +1000549 int i, acked = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700550
551 BUG_ON(! cnx);
552
553 switch (event->base_event.xSubtype) {
Michael Ellerman59f17ae2005-09-01 11:29:27 +1000554 case VETH_EVENT_CAP:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700555 veth_take_cap(cnx, event);
556 break;
Michael Ellerman59f17ae2005-09-01 11:29:27 +1000557 case VETH_EVENT_MONITOR:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700558 /* do nothing... this'll hang out here til we're dead,
559 * and the hypervisor will return it for us. */
560 break;
Michael Ellerman59f17ae2005-09-01 11:29:27 +1000561 case VETH_EVENT_FRAMES_ACK:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700562 spin_lock_irqsave(&cnx->lock, flags);
Michael Ellerman24562ff2005-09-01 11:29:17 +1000563
Linus Torvalds1da177e2005-04-16 15:20:36 -0700564 for (i = 0; i < VETH_MAX_ACKS_PER_MSG; ++i) {
565 u16 msgnum = event->u.frames_ack_data.token[i];
566
Michael Ellerman24562ff2005-09-01 11:29:17 +1000567 if (msgnum < VETH_NUMBUFFERS) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700568 veth_recycle_msg(cnx, cnx->msgs + msgnum);
Michael Ellerman24562ff2005-09-01 11:29:17 +1000569 cnx->outstanding_tx--;
570 acked++;
571 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700572 }
Michael Ellerman24562ff2005-09-01 11:29:17 +1000573
Michael Ellermane0808492005-09-01 11:29:18 +1000574 if (acked > 0) {
Michael Ellerman24562ff2005-09-01 11:29:17 +1000575 cnx->last_contact = jiffies;
Michael Ellermane0808492005-09-01 11:29:18 +1000576 veth_wake_queues(cnx);
577 }
Michael Ellerman24562ff2005-09-01 11:29:17 +1000578
Linus Torvalds1da177e2005-04-16 15:20:36 -0700579 spin_unlock_irqrestore(&cnx->lock, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700580 break;
Michael Ellerman59f17ae2005-09-01 11:29:27 +1000581 case VETH_EVENT_FRAMES:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700582 veth_receive(cnx, event);
583 break;
584 default:
Michael Ellerman61a3c692005-09-01 11:28:57 +1000585 veth_error("Unknown interrupt type %d from LPAR %d.\n",
586 event->base_event.xSubtype, rlp);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700587 };
588}
589
David Howells7d12e782006-10-05 14:55:46 +0100590static void veth_handle_event(struct HvLpEvent *event)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700591{
Michael Ellerman59f17ae2005-09-01 11:29:27 +1000592 struct veth_lpevent *veth_event = (struct veth_lpevent *)event;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700593
Stephen Rothwell677f8c02006-01-12 13:47:43 +1100594 if (hvlpevent_is_ack(event))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700595 veth_handle_ack(veth_event);
Stephen Rothwell677f8c02006-01-12 13:47:43 +1100596 else
Linus Torvalds1da177e2005-04-16 15:20:36 -0700597 veth_handle_int(veth_event);
598}
599
600static int veth_process_caps(struct veth_lpar_connection *cnx)
601{
Michael Ellerman59f17ae2005-09-01 11:29:27 +1000602 struct veth_cap_data *remote_caps = &cnx->remote_caps;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700603 int num_acks_needed;
604
605 /* Convert timer to jiffies */
606 cnx->ack_timeout = remote_caps->ack_timeout * HZ / 1000000;
607
Joe Perches8e95a202009-12-03 07:58:21 +0000608 if ( (remote_caps->num_buffers == 0) ||
609 (remote_caps->ack_threshold > VETH_MAX_ACKS_PER_MSG) ||
610 (remote_caps->ack_threshold == 0) ||
611 (cnx->ack_timeout == 0) ) {
Michael Ellerman61a3c692005-09-01 11:28:57 +1000612 veth_error("Received incompatible capabilities from LPAR %d.\n",
613 cnx->remote_lp);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700614 return HvLpEvent_Rc_InvalidSubtypeData;
615 }
616
617 num_acks_needed = (remote_caps->num_buffers
618 / remote_caps->ack_threshold) + 1;
619
620 /* FIXME: locking on num_ack_events? */
621 if (cnx->num_ack_events < num_acks_needed) {
622 int num;
623
624 num = veth_allocate_events(cnx->remote_lp,
625 num_acks_needed-cnx->num_ack_events);
626 if (num > 0)
627 cnx->num_ack_events += num;
628
629 if (cnx->num_ack_events < num_acks_needed) {
Michael Ellerman61a3c692005-09-01 11:28:57 +1000630 veth_error("Couldn't allocate enough ack events "
631 "for LPAR %d.\n", cnx->remote_lp);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700632
633 return HvLpEvent_Rc_BufferNotAvailable;
634 }
635 }
636
637
638 return HvLpEvent_Rc_Good;
639}
640
641/* FIXME: The gotos here are a bit dubious */
David Howellsc4028952006-11-22 14:57:56 +0000642static void veth_statemachine(struct work_struct *work)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700643{
David Howellsc4028952006-11-22 14:57:56 +0000644 struct veth_lpar_connection *cnx =
645 container_of(work, struct veth_lpar_connection,
646 statemachine_wq.work);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700647 int rlp = cnx->remote_lp;
648 int rc;
649
650 spin_lock_irq(&cnx->lock);
651
652 restart:
653 if (cnx->state & VETH_STATE_RESET) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700654 if (cnx->state & VETH_STATE_OPEN)
655 HvCallEvent_closeLpEventPath(cnx->remote_lp,
656 HvLpEvent_Type_VirtualLan);
657
Michael Ellermanabfda472005-09-01 11:28:59 +1000658 /*
659 * Reset ack data. This prevents the ack_timer actually
660 * doing anything, even if it runs one more time when
661 * we drop the lock below.
662 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700663 memset(&cnx->pending_acks, 0xff, sizeof (cnx->pending_acks));
664 cnx->num_pending_acks = 0;
665
666 cnx->state &= ~(VETH_STATE_RESET | VETH_STATE_SENTMON
667 | VETH_STATE_OPEN | VETH_STATE_SENTCAPS
668 | VETH_STATE_GOTCAPACK | VETH_STATE_GOTCAPS
669 | VETH_STATE_SENTCAPACK | VETH_STATE_READY);
670
671 /* Clean up any leftover messages */
Michael Ellerman24562ff2005-09-01 11:29:17 +1000672 if (cnx->msgs) {
673 int i;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700674 for (i = 0; i < VETH_NUMBUFFERS; ++i)
675 veth_recycle_msg(cnx, cnx->msgs + i);
Michael Ellerman24562ff2005-09-01 11:29:17 +1000676 }
Michael Ellermane0808492005-09-01 11:29:18 +1000677
Michael Ellerman24562ff2005-09-01 11:29:17 +1000678 cnx->outstanding_tx = 0;
Michael Ellermane0808492005-09-01 11:29:18 +1000679 veth_wake_queues(cnx);
Michael Ellermanabfda472005-09-01 11:28:59 +1000680
681 /* Drop the lock so we can do stuff that might sleep or
682 * take other locks. */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700683 spin_unlock_irq(&cnx->lock);
Michael Ellermanabfda472005-09-01 11:28:59 +1000684
685 del_timer_sync(&cnx->ack_timer);
Michael Ellerman24562ff2005-09-01 11:29:17 +1000686 del_timer_sync(&cnx->reset_timer);
687
Linus Torvalds1da177e2005-04-16 15:20:36 -0700688 spin_lock_irq(&cnx->lock);
Michael Ellermanabfda472005-09-01 11:28:59 +1000689
Linus Torvalds1da177e2005-04-16 15:20:36 -0700690 if (cnx->state & VETH_STATE_RESET)
691 goto restart;
Michael Ellerman58c59002005-09-01 11:29:00 +1000692
693 /* Hack, wait for the other end to reset itself. */
694 if (! (cnx->state & VETH_STATE_SHUTDOWN)) {
695 schedule_delayed_work(&cnx->statemachine_wq, 5 * HZ);
696 goto out;
697 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700698 }
699
700 if (cnx->state & VETH_STATE_SHUTDOWN)
701 /* It's all over, do nothing */
702 goto out;
703
704 if ( !(cnx->state & VETH_STATE_OPEN) ) {
705 if (! cnx->msgs || (cnx->num_events < (2 + VETH_NUMBUFFERS)) )
706 goto cant_cope;
707
708 HvCallEvent_openLpEventPath(rlp, HvLpEvent_Type_VirtualLan);
709 cnx->src_inst =
710 HvCallEvent_getSourceLpInstanceId(rlp,
711 HvLpEvent_Type_VirtualLan);
712 cnx->dst_inst =
713 HvCallEvent_getTargetLpInstanceId(rlp,
714 HvLpEvent_Type_VirtualLan);
715 cnx->state |= VETH_STATE_OPEN;
716 }
717
Joe Perches8e95a202009-12-03 07:58:21 +0000718 if ( (cnx->state & VETH_STATE_OPEN) &&
719 !(cnx->state & VETH_STATE_SENTMON) ) {
Michael Ellerman59f17ae2005-09-01 11:29:27 +1000720 rc = veth_signalevent(cnx, VETH_EVENT_MONITOR,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700721 HvLpEvent_AckInd_DoAck,
722 HvLpEvent_AckType_DeferredAck,
723 0, 0, 0, 0, 0, 0);
724
725 if (rc == HvLpEvent_Rc_Good) {
726 cnx->state |= VETH_STATE_SENTMON;
727 } else {
Joe Perches8e95a202009-12-03 07:58:21 +0000728 if ( (rc != HvLpEvent_Rc_PartitionDead) &&
729 (rc != HvLpEvent_Rc_PathClosed) )
Michael Ellerman61a3c692005-09-01 11:28:57 +1000730 veth_error("Error sending monitor to LPAR %d, "
731 "rc = %d\n", rlp, rc);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700732
733 /* Oh well, hope we get a cap from the other
734 * end and do better when that kicks us */
735 goto out;
736 }
737 }
738
Joe Perches8e95a202009-12-03 07:58:21 +0000739 if ( (cnx->state & VETH_STATE_OPEN) &&
740 !(cnx->state & VETH_STATE_SENTCAPS)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700741 u64 *rawcap = (u64 *)&cnx->local_caps;
742
Michael Ellerman59f17ae2005-09-01 11:29:27 +1000743 rc = veth_signalevent(cnx, VETH_EVENT_CAP,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700744 HvLpEvent_AckInd_DoAck,
745 HvLpEvent_AckType_ImmediateAck,
746 0, rawcap[0], rawcap[1], rawcap[2],
747 rawcap[3], rawcap[4]);
748
749 if (rc == HvLpEvent_Rc_Good) {
750 cnx->state |= VETH_STATE_SENTCAPS;
751 } else {
Joe Perches8e95a202009-12-03 07:58:21 +0000752 if ( (rc != HvLpEvent_Rc_PartitionDead) &&
753 (rc != HvLpEvent_Rc_PathClosed) )
Michael Ellerman61a3c692005-09-01 11:28:57 +1000754 veth_error("Error sending caps to LPAR %d, "
755 "rc = %d\n", rlp, rc);
756
Linus Torvalds1da177e2005-04-16 15:20:36 -0700757 /* Oh well, hope we get a cap from the other
758 * end and do better when that kicks us */
759 goto out;
760 }
761 }
762
Joe Perches8e95a202009-12-03 07:58:21 +0000763 if ((cnx->state & VETH_STATE_GOTCAPS) &&
764 !(cnx->state & VETH_STATE_SENTCAPACK)) {
Michael Ellerman59f17ae2005-09-01 11:29:27 +1000765 struct veth_cap_data *remote_caps = &cnx->remote_caps;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700766
767 memcpy(remote_caps, &cnx->cap_event.u.caps_data,
768 sizeof(*remote_caps));
769
770 spin_unlock_irq(&cnx->lock);
771 rc = veth_process_caps(cnx);
772 spin_lock_irq(&cnx->lock);
773
774 /* We dropped the lock, so recheck for anything which
775 * might mess us up */
776 if (cnx->state & (VETH_STATE_RESET|VETH_STATE_SHUTDOWN))
777 goto restart;
778
779 cnx->cap_event.base_event.xRc = rc;
780 HvCallEvent_ackLpEvent((struct HvLpEvent *)&cnx->cap_event);
781 if (rc == HvLpEvent_Rc_Good)
782 cnx->state |= VETH_STATE_SENTCAPACK;
783 else
784 goto cant_cope;
785 }
786
Joe Perches8e95a202009-12-03 07:58:21 +0000787 if ((cnx->state & VETH_STATE_GOTCAPACK) &&
788 (cnx->state & VETH_STATE_GOTCAPS) &&
789 !(cnx->state & VETH_STATE_READY)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700790 if (cnx->cap_ack_event.base_event.xRc == HvLpEvent_Rc_Good) {
791 /* Start the ACK timer */
792 cnx->ack_timer.expires = jiffies + cnx->ack_timeout;
793 add_timer(&cnx->ack_timer);
794 cnx->state |= VETH_STATE_READY;
795 } else {
Michael Ellerman61a3c692005-09-01 11:28:57 +1000796 veth_error("Caps rejected by LPAR %d, rc = %d\n",
797 rlp, cnx->cap_ack_event.base_event.xRc);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700798 goto cant_cope;
799 }
800 }
801
802 out:
803 spin_unlock_irq(&cnx->lock);
804 return;
805
806 cant_cope:
807 /* FIXME: we get here if something happens we really can't
808 * cope with. The link will never work once we get here, and
809 * all we can do is not lock the rest of the system up */
Michael Ellerman61a3c692005-09-01 11:28:57 +1000810 veth_error("Unrecoverable error on connection to LPAR %d, shutting down"
811 " (state = 0x%04lx)\n", rlp, cnx->state);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700812 cnx->state |= VETH_STATE_SHUTDOWN;
813 spin_unlock_irq(&cnx->lock);
814}
815
816static int veth_init_connection(u8 rlp)
817{
818 struct veth_lpar_connection *cnx;
819 struct veth_msg *msgs;
Greg Kroah-Hartman78be3d22007-12-17 23:05:35 -0700820 int i;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700821
Joe Perches8e95a202009-12-03 07:58:21 +0000822 if ( (rlp == this_lp) ||
823 ! HvLpConfig_doLpsCommunicateOnVirtualLan(this_lp, rlp) )
Linus Torvalds1da177e2005-04-16 15:20:36 -0700824 return 0;
825
Yoann Padioleaudd00cc42007-07-19 01:49:03 -0700826 cnx = kzalloc(sizeof(*cnx), GFP_KERNEL);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700827 if (! cnx)
828 return -ENOMEM;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700829
830 cnx->remote_lp = rlp;
831 spin_lock_init(&cnx->lock);
David Howellsc4028952006-11-22 14:57:56 +0000832 INIT_DELAYED_WORK(&cnx->statemachine_wq, veth_statemachine);
Michael Ellerman24562ff2005-09-01 11:29:17 +1000833
Linus Torvalds1da177e2005-04-16 15:20:36 -0700834 init_timer(&cnx->ack_timer);
835 cnx->ack_timer.function = veth_timed_ack;
836 cnx->ack_timer.data = (unsigned long) cnx;
Michael Ellerman24562ff2005-09-01 11:29:17 +1000837
838 init_timer(&cnx->reset_timer);
839 cnx->reset_timer.function = veth_timed_reset;
840 cnx->reset_timer.data = (unsigned long) cnx;
841 cnx->reset_timeout = 5 * HZ * (VETH_ACKTIMEOUT / 1000000);
842
Linus Torvalds1da177e2005-04-16 15:20:36 -0700843 memset(&cnx->pending_acks, 0xff, sizeof (cnx->pending_acks));
844
845 veth_cnx[rlp] = cnx;
846
Michael Ellermanf0c129c2005-09-01 11:29:09 +1000847 /* This gets us 1 reference, which is held on behalf of the driver
848 * infrastructure. It's released at module unload. */
Greg Kroah-Hartmanf9cb0742007-12-17 23:05:35 -0700849 kobject_init(&cnx->kobject, &veth_lpar_connection_ktype);
Michael Ellermanf0c129c2005-09-01 11:29:09 +1000850
Yoann Padioleaudd00cc42007-07-19 01:49:03 -0700851 msgs = kcalloc(VETH_NUMBUFFERS, sizeof(struct veth_msg), GFP_KERNEL);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700852 if (! msgs) {
Michael Ellerman61a3c692005-09-01 11:28:57 +1000853 veth_error("Can't allocate buffers for LPAR %d.\n", rlp);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700854 return -ENOMEM;
855 }
856
857 cnx->msgs = msgs;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700858
859 for (i = 0; i < VETH_NUMBUFFERS; i++) {
860 msgs[i].token = i;
861 veth_stack_push(cnx, msgs + i);
862 }
863
864 cnx->num_events = veth_allocate_events(rlp, 2 + VETH_NUMBUFFERS);
865
866 if (cnx->num_events < (2 + VETH_NUMBUFFERS)) {
Michael Ellerman61a3c692005-09-01 11:28:57 +1000867 veth_error("Can't allocate enough events for LPAR %d.\n", rlp);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700868 return -ENOMEM;
869 }
870
871 cnx->local_caps.num_buffers = VETH_NUMBUFFERS;
872 cnx->local_caps.ack_threshold = ACK_THRESHOLD;
873 cnx->local_caps.ack_timeout = VETH_ACKTIMEOUT;
874
875 return 0;
876}
877
Michael Ellermanf0c129c2005-09-01 11:29:09 +1000878static void veth_stop_connection(struct veth_lpar_connection *cnx)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700879{
Michael Ellermanf0c129c2005-09-01 11:29:09 +1000880 if (!cnx)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700881 return;
882
883 spin_lock_irq(&cnx->lock);
884 cnx->state |= VETH_STATE_RESET | VETH_STATE_SHUTDOWN;
885 veth_kick_statemachine(cnx);
886 spin_unlock_irq(&cnx->lock);
887
Michael Ellerman58c59002005-09-01 11:29:00 +1000888 /* There's a slim chance the reset code has just queued the
889 * statemachine to run in five seconds. If so we need to cancel
890 * that and requeue the work to run now. */
891 if (cancel_delayed_work(&cnx->statemachine_wq)) {
892 spin_lock_irq(&cnx->lock);
893 veth_kick_statemachine(cnx);
894 spin_unlock_irq(&cnx->lock);
895 }
896
Michael Ellermanabfda472005-09-01 11:28:59 +1000897 /* Wait for the state machine to run. */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700898 flush_scheduled_work();
Michael Ellermanec60bee2005-09-01 11:29:08 +1000899}
900
Michael Ellermanf0c129c2005-09-01 11:29:09 +1000901static void veth_destroy_connection(struct veth_lpar_connection *cnx)
Michael Ellermanec60bee2005-09-01 11:29:08 +1000902{
Michael Ellermanf0c129c2005-09-01 11:29:09 +1000903 if (!cnx)
Michael Ellermanec60bee2005-09-01 11:29:08 +1000904 return;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700905
Linus Torvalds1da177e2005-04-16 15:20:36 -0700906 if (cnx->num_events > 0)
907 mf_deallocate_lp_events(cnx->remote_lp,
908 HvLpEvent_Type_VirtualLan,
909 cnx->num_events,
910 NULL, NULL);
911 if (cnx->num_ack_events > 0)
912 mf_deallocate_lp_events(cnx->remote_lp,
913 HvLpEvent_Type_VirtualLan,
914 cnx->num_ack_events,
915 NULL, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700916
917 kfree(cnx->msgs);
Michael Ellermanf0c129c2005-09-01 11:29:09 +1000918 veth_cnx[cnx->remote_lp] = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700919 kfree(cnx);
Michael Ellermanf0c129c2005-09-01 11:29:09 +1000920}
921
922static void veth_release_connection(struct kobject *kobj)
923{
924 struct veth_lpar_connection *cnx;
925 cnx = container_of(kobj, struct veth_lpar_connection, kobject);
926 veth_stop_connection(cnx);
927 veth_destroy_connection(cnx);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700928}
929
930/*
931 * net_device code
932 */
933
934static int veth_open(struct net_device *dev)
935{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700936 netif_start_queue(dev);
937 return 0;
938}
939
940static int veth_close(struct net_device *dev)
941{
942 netif_stop_queue(dev);
943 return 0;
944}
945
Linus Torvalds1da177e2005-04-16 15:20:36 -0700946static int veth_change_mtu(struct net_device *dev, int new_mtu)
947{
948 if ((new_mtu < 68) || (new_mtu > VETH_MAX_MTU))
949 return -EINVAL;
950 dev->mtu = new_mtu;
951 return 0;
952}
953
954static void veth_set_multicast_list(struct net_device *dev)
955{
Wang Chen454d7c92008-11-12 23:37:49 -0800956 struct veth_port *port = netdev_priv(dev);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700957 unsigned long flags;
958
959 write_lock_irqsave(&port->mcast_gate, flags);
960
Michael Ellerman2a5391a2005-09-01 11:29:02 +1000961 if ((dev->flags & IFF_PROMISC) || (dev->flags & IFF_ALLMULTI) ||
Jiri Pirko4cd24ea2010-02-08 04:30:35 +0000962 (netdev_mc_count(dev) > VETH_MAX_MCAST)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700963 port->promiscuous = 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700964 } else {
Jiri Pirko22bedad32010-04-01 21:22:57 +0000965 struct netdev_hw_addr *ha;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700966
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
Jiri Pirko22bedad32010-04-01 21:22:57 +0000972 netdev_for_each_mc_addr(ha, dev) {
973 u8 *addr = ha->addr;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700974 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 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700981 }
982 }
983
984 write_unlock_irqrestore(&port->mcast_gate, flags);
985}
986
987static void veth_get_drvinfo(struct net_device *dev, struct ethtool_drvinfo *info)
988{
Michael Ellermanee05f032005-09-01 11:29:29 +1000989 strncpy(info->driver, DRV_NAME, sizeof(info->driver) - 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700990 info->driver[sizeof(info->driver) - 1] = '\0';
Michael Ellermanee05f032005-09-01 11:29:29 +1000991 strncpy(info->version, DRV_VERSION, sizeof(info->version) - 1);
992 info->version[sizeof(info->version) - 1] = '\0';
Linus Torvalds1da177e2005-04-16 15:20:36 -0700993}
994
995static int veth_get_settings(struct net_device *dev, struct ethtool_cmd *ecmd)
996{
997 ecmd->supported = (SUPPORTED_1000baseT_Full
998 | SUPPORTED_Autoneg | SUPPORTED_FIBRE);
999 ecmd->advertising = (SUPPORTED_1000baseT_Full
1000 | SUPPORTED_Autoneg | SUPPORTED_FIBRE);
1001 ecmd->port = PORT_FIBRE;
1002 ecmd->transceiver = XCVR_INTERNAL;
1003 ecmd->phy_address = 0;
1004 ecmd->speed = SPEED_1000;
1005 ecmd->duplex = DUPLEX_FULL;
1006 ecmd->autoneg = AUTONEG_ENABLE;
1007 ecmd->maxtxpkt = 120;
1008 ecmd->maxrxpkt = 120;
1009 return 0;
1010}
1011
1012static u32 veth_get_link(struct net_device *dev)
1013{
1014 return 1;
1015}
1016
Jeff Garzik7282d492006-09-13 14:30:00 -04001017static const struct ethtool_ops ops = {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001018 .get_drvinfo = veth_get_drvinfo,
1019 .get_settings = veth_get_settings,
1020 .get_link = veth_get_link,
1021};
1022
Alexander Beregalovc2775362009-04-15 12:52:43 +00001023static const struct net_device_ops veth_netdev_ops = {
1024 .ndo_open = veth_open,
1025 .ndo_stop = veth_close,
1026 .ndo_start_xmit = veth_start_xmit,
1027 .ndo_change_mtu = veth_change_mtu,
1028 .ndo_set_multicast_list = veth_set_multicast_list,
1029 .ndo_set_mac_address = NULL,
1030 .ndo_validate_addr = eth_validate_addr,
1031};
1032
Stephen Rothwell57f78ab32008-02-05 14:19:40 +11001033static struct net_device *veth_probe_one(int vlan,
Stephen Rothwell07fb3f42006-05-15 13:39:47 +10001034 struct vio_dev *vio_dev)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001035{
1036 struct net_device *dev;
1037 struct veth_port *port;
Stephen Rothwell07fb3f42006-05-15 13:39:47 +10001038 struct device *vdev = &vio_dev->dev;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001039 int i, rc;
Stephen Rothwell07fb3f42006-05-15 13:39:47 +10001040 const unsigned char *mac_addr;
1041
1042 mac_addr = vio_get_attribute(vio_dev, "local-mac-address", NULL);
1043 if (mac_addr == NULL)
1044 mac_addr = vio_get_attribute(vio_dev, "mac-address", NULL);
1045 if (mac_addr == NULL) {
1046 veth_error("Unable to fetch MAC address from device tree.\n");
1047 return NULL;
1048 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001049
1050 dev = alloc_etherdev(sizeof (struct veth_port));
1051 if (! dev) {
1052 veth_error("Unable to allocate net_device structure!\n");
1053 return NULL;
1054 }
1055
Wang Chen454d7c92008-11-12 23:37:49 -08001056 port = netdev_priv(dev);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001057
Michael Ellermane0808492005-09-01 11:29:18 +10001058 spin_lock_init(&port->queue_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001059 rwlock_init(&port->mcast_gate);
Michael Ellermane0808492005-09-01 11:29:18 +10001060 port->stopped_map = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001061
1062 for (i = 0; i < HVMAXARCHITECTEDLPS; i++) {
1063 HvLpVirtualLanIndexMap map;
1064
1065 if (i == this_lp)
1066 continue;
1067 map = HvLpConfig_getVirtualLanIndexMapForLp(i);
1068 if (map & (0x8000 >> vlan))
1069 port->lpar_map |= (1 << i);
1070 }
1071 port->dev = vdev;
1072
Stephen Rothwell07fb3f42006-05-15 13:39:47 +10001073 memcpy(dev->dev_addr, mac_addr, ETH_ALEN);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001074
1075 dev->mtu = VETH_MAX_MTU;
1076
Stephen Rothwell07fb3f42006-05-15 13:39:47 +10001077 memcpy(&port->mac_addr, mac_addr, ETH_ALEN);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001078
Alexander Beregalov3ef8e4e2009-04-19 18:42:58 -07001079 dev->netdev_ops = &veth_netdev_ops;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001080 SET_ETHTOOL_OPS(dev, &ops);
1081
Linus Torvalds1da177e2005-04-16 15:20:36 -07001082 SET_NETDEV_DEV(dev, vdev);
1083
1084 rc = register_netdev(dev);
1085 if (rc != 0) {
Michael Ellerman61a3c692005-09-01 11:28:57 +10001086 veth_error("Failed registering net device for vlan%d.\n", vlan);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001087 free_netdev(dev);
1088 return NULL;
1089 }
1090
Greg Kroah-Hartmanf9cb0742007-12-17 23:05:35 -07001091 kobject_init(&port->kobject, &veth_port_ktype);
Greg Kroah-Hartmanb2d6db52007-12-17 23:05:35 -07001092 if (0 != kobject_add(&port->kobject, &dev->dev.kobj, "veth_port"))
Michael Ellerman07a5c172005-09-01 11:29:21 +10001093 veth_error("Failed adding port for %s to sysfs.\n", dev->name);
1094
Michael Ellerman61a3c692005-09-01 11:28:57 +10001095 veth_info("%s attached to iSeries vlan %d (LPAR map = 0x%.4X)\n",
1096 dev->name, vlan, port->lpar_map);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001097
1098 return dev;
1099}
1100
1101/*
1102 * Tx path
1103 */
1104
1105static int veth_transmit_to_one(struct sk_buff *skb, HvLpIndex rlp,
1106 struct net_device *dev)
1107{
1108 struct veth_lpar_connection *cnx = veth_cnx[rlp];
Wang Chen454d7c92008-11-12 23:37:49 -08001109 struct veth_port *port = netdev_priv(dev);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001110 HvLpEvent_Rc rc;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001111 struct veth_msg *msg = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001112 unsigned long flags;
1113
Michael Ellermandb5e8712005-09-01 11:29:19 +10001114 if (! cnx)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001115 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001116
1117 spin_lock_irqsave(&cnx->lock, flags);
1118
Michael Ellermanf27eff12005-05-12 17:47:27 +10001119 if (! (cnx->state & VETH_STATE_READY))
Michael Ellermandb5e8712005-09-01 11:29:19 +10001120 goto no_error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001121
Michael Ellermandb5e8712005-09-01 11:29:19 +10001122 if ((skb->len - ETH_HLEN) > VETH_MAX_MTU)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001123 goto drop;
1124
1125 msg = veth_stack_pop(cnx);
Michael Ellermandb5e8712005-09-01 11:29:19 +10001126 if (! msg)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001127 goto drop;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001128
Michael Ellermanb08bd5c2005-09-01 11:29:06 +10001129 msg->in_use = 1;
Michael Ellermandb5e8712005-09-01 11:29:19 +10001130 msg->skb = skb_get(skb);
Michael Ellermanb08bd5c2005-09-01 11:29:06 +10001131
Michael Ellermancbf90742005-09-01 11:29:07 +10001132 msg->data.addr[0] = dma_map_single(port->dev, skb->data,
1133 skb->len, DMA_TO_DEVICE);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001134
FUJITA Tomonori8d8bb392008-07-25 19:44:49 -07001135 if (dma_mapping_error(port->dev, msg->data.addr[0]))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001136 goto recycle_and_drop;
1137
Linus Torvalds1da177e2005-04-16 15:20:36 -07001138 msg->dev = port->dev;
Michael Ellermancbf90742005-09-01 11:29:07 +10001139 msg->data.len[0] = skb->len;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001140 msg->data.eofmask = 1 << VETH_EOF_SHIFT;
Michael Ellermancbf90742005-09-01 11:29:07 +10001141
Michael Ellerman59f17ae2005-09-01 11:29:27 +10001142 rc = veth_signaldata(cnx, VETH_EVENT_FRAMES, msg->token, &msg->data);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001143
1144 if (rc != HvLpEvent_Rc_Good)
1145 goto recycle_and_drop;
1146
Michael Ellerman24562ff2005-09-01 11:29:17 +10001147 /* If the timer's not already running, start it now. */
1148 if (0 == cnx->outstanding_tx)
1149 mod_timer(&cnx->reset_timer, jiffies + cnx->reset_timeout);
1150
1151 cnx->last_contact = jiffies;
1152 cnx->outstanding_tx++;
1153
Michael Ellermane0808492005-09-01 11:29:18 +10001154 if (veth_stack_is_empty(cnx))
1155 veth_stop_queues(cnx);
1156
Michael Ellermandb5e8712005-09-01 11:29:19 +10001157 no_error:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001158 spin_unlock_irqrestore(&cnx->lock, flags);
1159 return 0;
1160
1161 recycle_and_drop:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001162 veth_recycle_msg(cnx, msg);
1163 drop:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001164 spin_unlock_irqrestore(&cnx->lock, flags);
Michael Ellermandb5e8712005-09-01 11:29:19 +10001165 return 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001166}
1167
Michael Ellermandb5e8712005-09-01 11:29:19 +10001168static void veth_transmit_to_many(struct sk_buff *skb,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001169 HvLpIndexMap lpmask,
1170 struct net_device *dev)
1171{
Michael Ellermandb5e8712005-09-01 11:29:19 +10001172 int i, success, error;
1173
1174 success = error = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001175
1176 for (i = 0; i < HVMAXARCHITECTEDLPS; i++) {
1177 if ((lpmask & (1 << i)) == 0)
1178 continue;
1179
Michael Ellermandb5e8712005-09-01 11:29:19 +10001180 if (veth_transmit_to_one(skb, i, dev))
1181 error = 1;
1182 else
1183 success = 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001184 }
1185
Michael Ellermandb5e8712005-09-01 11:29:19 +10001186 if (error)
Jeff Garzik09f75cd2007-10-03 17:41:50 -07001187 dev->stats.tx_errors++;
Michael Ellermandb5e8712005-09-01 11:29:19 +10001188
1189 if (success) {
Jeff Garzik09f75cd2007-10-03 17:41:50 -07001190 dev->stats.tx_packets++;
1191 dev->stats.tx_bytes += skb->len;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001192 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001193}
1194
1195static int veth_start_xmit(struct sk_buff *skb, struct net_device *dev)
1196{
1197 unsigned char *frame = skb->data;
Wang Chen454d7c92008-11-12 23:37:49 -08001198 struct veth_port *port = netdev_priv(dev);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001199 HvLpIndexMap lpmask;
1200
1201 if (! (frame[0] & 0x01)) {
1202 /* unicast packet */
1203 HvLpIndex rlp = frame[5];
1204
1205 if ( ! ((1 << rlp) & port->lpar_map) ) {
1206 dev_kfree_skb(skb);
Patrick McHardy6ed10652009-06-23 06:03:08 +00001207 return NETDEV_TX_OK;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001208 }
1209
1210 lpmask = 1 << rlp;
1211 } else {
1212 lpmask = port->lpar_map;
1213 }
1214
Michael Ellermane0808492005-09-01 11:29:18 +10001215 veth_transmit_to_many(skb, lpmask, dev);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001216
Michael Ellermane0808492005-09-01 11:29:18 +10001217 dev_kfree_skb(skb);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001218
Patrick McHardy6ed10652009-06-23 06:03:08 +00001219 return NETDEV_TX_OK;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001220}
1221
Michael Ellermanb08bd5c2005-09-01 11:29:06 +10001222/* You must hold the connection's lock when you call this function. */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001223static void veth_recycle_msg(struct veth_lpar_connection *cnx,
1224 struct veth_msg *msg)
1225{
1226 u32 dma_address, dma_length;
1227
Michael Ellermanb08bd5c2005-09-01 11:29:06 +10001228 if (msg->in_use) {
1229 msg->in_use = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001230 dma_address = msg->data.addr[0];
1231 dma_length = msg->data.len[0];
1232
FUJITA Tomonori8d8bb392008-07-25 19:44:49 -07001233 if (!dma_mapping_error(msg->dev, dma_address))
Michael Ellermancbf90742005-09-01 11:29:07 +10001234 dma_unmap_single(msg->dev, dma_address, dma_length,
1235 DMA_TO_DEVICE);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001236
1237 if (msg->skb) {
1238 dev_kfree_skb_any(msg->skb);
1239 msg->skb = NULL;
1240 }
1241
1242 memset(&msg->data, 0, sizeof(msg->data));
1243 veth_stack_push(cnx, msg);
Michael Ellerman61a3c692005-09-01 11:28:57 +10001244 } else if (cnx->state & VETH_STATE_OPEN) {
1245 veth_error("Non-pending frame (# %d) acked by LPAR %d.\n",
1246 cnx->remote_lp, msg->token);
1247 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001248}
1249
Michael Ellermane0808492005-09-01 11:29:18 +10001250static void veth_wake_queues(struct veth_lpar_connection *cnx)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001251{
1252 int i;
Michael Ellermane0808492005-09-01 11:29:18 +10001253
Linus Torvalds1da177e2005-04-16 15:20:36 -07001254 for (i = 0; i < HVMAXARCHITECTEDVIRTUALLANS; i++) {
1255 struct net_device *dev = veth_dev[i];
1256 struct veth_port *port;
1257 unsigned long flags;
1258
1259 if (! dev)
1260 continue;
1261
Wang Chen454d7c92008-11-12 23:37:49 -08001262 port = netdev_priv(dev);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001263
1264 if (! (port->lpar_map & (1<<cnx->remote_lp)))
1265 continue;
1266
Michael Ellermane0808492005-09-01 11:29:18 +10001267 spin_lock_irqsave(&port->queue_lock, flags);
1268
1269 port->stopped_map &= ~(1 << cnx->remote_lp);
1270
1271 if (0 == port->stopped_map && netif_queue_stopped(dev)) {
1272 veth_debug("cnx %d: woke queue for %s.\n",
1273 cnx->remote_lp, dev->name);
1274 netif_wake_queue(dev);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001275 }
Michael Ellermane0808492005-09-01 11:29:18 +10001276 spin_unlock_irqrestore(&port->queue_lock, flags);
1277 }
1278}
1279
1280static void veth_stop_queues(struct veth_lpar_connection *cnx)
1281{
1282 int i;
1283
1284 for (i = 0; i < HVMAXARCHITECTEDVIRTUALLANS; i++) {
1285 struct net_device *dev = veth_dev[i];
1286 struct veth_port *port;
1287
1288 if (! dev)
1289 continue;
1290
Wang Chen454d7c92008-11-12 23:37:49 -08001291 port = netdev_priv(dev);
Michael Ellermane0808492005-09-01 11:29:18 +10001292
1293 /* If this cnx is not on the vlan for this port, continue */
1294 if (! (port->lpar_map & (1 << cnx->remote_lp)))
1295 continue;
1296
1297 spin_lock(&port->queue_lock);
1298
1299 netif_stop_queue(dev);
1300 port->stopped_map |= (1 << cnx->remote_lp);
1301
1302 veth_debug("cnx %d: stopped queue for %s, map = 0x%x.\n",
1303 cnx->remote_lp, dev->name, port->stopped_map);
1304
1305 spin_unlock(&port->queue_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001306 }
1307}
1308
Michael Ellerman24562ff2005-09-01 11:29:17 +10001309static void veth_timed_reset(unsigned long ptr)
1310{
1311 struct veth_lpar_connection *cnx = (struct veth_lpar_connection *)ptr;
1312 unsigned long trigger_time, flags;
1313
1314 /* FIXME is it possible this fires after veth_stop_connection()?
1315 * That would reschedule the statemachine for 5 seconds and probably
1316 * execute it after the module's been unloaded. Hmm. */
1317
1318 spin_lock_irqsave(&cnx->lock, flags);
1319
1320 if (cnx->outstanding_tx > 0) {
1321 trigger_time = cnx->last_contact + cnx->reset_timeout;
1322
1323 if (trigger_time < jiffies) {
1324 cnx->state |= VETH_STATE_RESET;
1325 veth_kick_statemachine(cnx);
1326 veth_error("%d packets not acked by LPAR %d within %d "
1327 "seconds, resetting.\n",
1328 cnx->outstanding_tx, cnx->remote_lp,
1329 cnx->reset_timeout / HZ);
1330 } else {
1331 /* Reschedule the timer */
1332 trigger_time = jiffies + cnx->reset_timeout;
1333 mod_timer(&cnx->reset_timer, trigger_time);
1334 }
1335 }
1336
1337 spin_unlock_irqrestore(&cnx->lock, flags);
1338}
1339
Linus Torvalds1da177e2005-04-16 15:20:36 -07001340/*
1341 * Rx path
1342 */
1343
1344static inline int veth_frame_wanted(struct veth_port *port, u64 mac_addr)
1345{
1346 int wanted = 0;
1347 int i;
1348 unsigned long flags;
1349
1350 if ( (mac_addr == port->mac_addr) || (mac_addr == 0xffffffffffff0000) )
1351 return 1;
1352
Linus Torvalds1da177e2005-04-16 15:20:36 -07001353 read_lock_irqsave(&port->mcast_gate, flags);
1354
Michael Ellerman2a5391a2005-09-01 11:29:02 +10001355 if (port->promiscuous) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001356 wanted = 1;
1357 goto out;
1358 }
1359
1360 for (i = 0; i < port->num_mcast; ++i) {
1361 if (port->mcast_addr[i] == mac_addr) {
1362 wanted = 1;
1363 break;
1364 }
1365 }
1366
1367 out:
1368 read_unlock_irqrestore(&port->mcast_gate, flags);
1369
1370 return wanted;
1371}
1372
1373struct dma_chunk {
1374 u64 addr;
1375 u64 size;
1376};
1377
1378#define VETH_MAX_PAGES_PER_FRAME ( (VETH_MAX_MTU+PAGE_SIZE-2)/PAGE_SIZE + 1 )
1379
1380static inline void veth_build_dma_list(struct dma_chunk *list,
1381 unsigned char *p, unsigned long length)
1382{
1383 unsigned long done;
1384 int i = 1;
1385
André Goddard Rosaaf901ca2009-11-14 13:09:05 -02001386 /* FIXME: skbs are contiguous in real addresses. Do we
Linus Torvalds1da177e2005-04-16 15:20:36 -07001387 * really need to break it into PAGE_SIZE chunks, or can we do
1388 * it just at the granularity of iSeries real->absolute
1389 * mapping? Indeed, given the way the allocator works, can we
1390 * count on them being absolutely contiguous? */
Stephen Rothwell426c1a12005-10-14 14:51:42 +10001391 list[0].addr = iseries_hv_addr(p);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001392 list[0].size = min(length,
1393 PAGE_SIZE - ((unsigned long)p & ~PAGE_MASK));
1394
1395 done = list[0].size;
1396 while (done < length) {
Stephen Rothwell426c1a12005-10-14 14:51:42 +10001397 list[i].addr = iseries_hv_addr(p + done);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001398 list[i].size = min(length-done, PAGE_SIZE);
1399 done += list[i].size;
1400 i++;
1401 }
1402}
1403
1404static void veth_flush_acks(struct veth_lpar_connection *cnx)
1405{
1406 HvLpEvent_Rc rc;
1407
Michael Ellerman59f17ae2005-09-01 11:29:27 +10001408 rc = veth_signaldata(cnx, VETH_EVENT_FRAMES_ACK,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001409 0, &cnx->pending_acks);
1410
1411 if (rc != HvLpEvent_Rc_Good)
Michael Ellerman61a3c692005-09-01 11:28:57 +10001412 veth_error("Failed acking frames from LPAR %d, rc = %d\n",
1413 cnx->remote_lp, (int)rc);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001414
1415 cnx->num_pending_acks = 0;
1416 memset(&cnx->pending_acks, 0xff, sizeof(cnx->pending_acks));
1417}
1418
1419static void veth_receive(struct veth_lpar_connection *cnx,
Michael Ellerman59f17ae2005-09-01 11:29:27 +10001420 struct veth_lpevent *event)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001421{
Michael Ellerman59f17ae2005-09-01 11:29:27 +10001422 struct veth_frames_data *senddata = &event->u.frames_data;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001423 int startchunk = 0;
1424 int nchunks;
1425 unsigned long flags;
1426 HvLpDma_Rc rc;
1427
1428 do {
1429 u16 length = 0;
1430 struct sk_buff *skb;
1431 struct dma_chunk local_list[VETH_MAX_PAGES_PER_FRAME];
1432 struct dma_chunk remote_list[VETH_MAX_FRAMES_PER_MSG];
1433 u64 dest;
1434 HvLpVirtualLanIndex vlan;
1435 struct net_device *dev;
1436 struct veth_port *port;
1437
1438 /* FIXME: do we need this? */
1439 memset(local_list, 0, sizeof(local_list));
1440 memset(remote_list, 0, sizeof(VETH_MAX_FRAMES_PER_MSG));
1441
1442 /* a 0 address marks the end of the valid entries */
1443 if (senddata->addr[startchunk] == 0)
1444 break;
1445
1446 /* make sure that we have at least 1 EOF entry in the
1447 * remaining entries */
1448 if (! (senddata->eofmask >> (startchunk + VETH_EOF_SHIFT))) {
Michael Ellerman61a3c692005-09-01 11:28:57 +10001449 veth_error("Missing EOF fragment in event "
1450 "eofmask = 0x%x startchunk = %d\n",
1451 (unsigned)senddata->eofmask,
1452 startchunk);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001453 break;
1454 }
1455
1456 /* build list of chunks in this frame */
1457 nchunks = 0;
1458 do {
1459 remote_list[nchunks].addr =
1460 (u64) senddata->addr[startchunk+nchunks] << 32;
1461 remote_list[nchunks].size =
1462 senddata->len[startchunk+nchunks];
1463 length += remote_list[nchunks].size;
1464 } while (! (senddata->eofmask &
1465 (1 << (VETH_EOF_SHIFT + startchunk + nchunks++))));
1466
1467 /* length == total length of all chunks */
1468 /* nchunks == # of chunks in this frame */
1469
1470 if ((length - ETH_HLEN) > VETH_MAX_MTU) {
Michael Ellerman61a3c692005-09-01 11:28:57 +10001471 veth_error("Received oversize frame from LPAR %d "
1472 "(length = %d)\n",
1473 cnx->remote_lp, length);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001474 continue;
1475 }
1476
1477 skb = alloc_skb(length, GFP_ATOMIC);
1478 if (!skb)
1479 continue;
1480
1481 veth_build_dma_list(local_list, skb->data, length);
1482
1483 rc = HvCallEvent_dmaBufList(HvLpEvent_Type_VirtualLan,
1484 event->base_event.xSourceLp,
1485 HvLpDma_Direction_RemoteToLocal,
1486 cnx->src_inst,
1487 cnx->dst_inst,
1488 HvLpDma_AddressType_RealAddress,
1489 HvLpDma_AddressType_TceIndex,
Stephen Rothwell426c1a12005-10-14 14:51:42 +10001490 iseries_hv_addr(&local_list),
1491 iseries_hv_addr(&remote_list),
Linus Torvalds1da177e2005-04-16 15:20:36 -07001492 length);
1493 if (rc != HvLpDma_Rc_Good) {
1494 dev_kfree_skb_irq(skb);
1495 continue;
1496 }
1497
1498 vlan = skb->data[9];
1499 dev = veth_dev[vlan];
Michael Ellerman41664c02005-05-12 17:55:08 +10001500 if (! dev) {
1501 /*
1502 * Some earlier versions of the driver sent
1503 * broadcasts down all connections, even to lpars
1504 * that weren't on the relevant vlan. So ignore
1505 * packets belonging to a vlan we're not on.
1506 * We can also be here if we receive packets while
1507 * the driver is going down, because then dev is NULL.
1508 */
1509 dev_kfree_skb_irq(skb);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001510 continue;
Michael Ellerman41664c02005-05-12 17:55:08 +10001511 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001512
Wang Chen454d7c92008-11-12 23:37:49 -08001513 port = netdev_priv(dev);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001514 dest = *((u64 *) skb->data) & 0xFFFFFFFFFFFF0000;
1515
1516 if ((vlan > HVMAXARCHITECTEDVIRTUALLANS) || !port) {
1517 dev_kfree_skb_irq(skb);
1518 continue;
1519 }
1520 if (! veth_frame_wanted(port, dest)) {
1521 dev_kfree_skb_irq(skb);
1522 continue;
1523 }
1524
1525 skb_put(skb, length);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001526 skb->protocol = eth_type_trans(skb, dev);
Eric Dumazetbc8acf22010-09-02 13:07:41 -07001527 skb_checksum_none_assert(skb);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001528 netif_rx(skb); /* send it up */
Jeff Garzik09f75cd2007-10-03 17:41:50 -07001529 dev->stats.rx_packets++;
1530 dev->stats.rx_bytes += length;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001531 } while (startchunk += nchunks, startchunk < VETH_MAX_FRAMES_PER_MSG);
1532
1533 /* Ack it */
1534 spin_lock_irqsave(&cnx->lock, flags);
1535 BUG_ON(cnx->num_pending_acks > VETH_MAX_ACKS_PER_MSG);
1536
1537 cnx->pending_acks[cnx->num_pending_acks++] =
1538 event->base_event.xCorrelationToken;
1539
Joe Perches8e95a202009-12-03 07:58:21 +00001540 if ( (cnx->num_pending_acks >= cnx->remote_caps.ack_threshold) ||
1541 (cnx->num_pending_acks >= VETH_MAX_ACKS_PER_MSG) )
Linus Torvalds1da177e2005-04-16 15:20:36 -07001542 veth_flush_acks(cnx);
1543
1544 spin_unlock_irqrestore(&cnx->lock, flags);
1545}
1546
1547static void veth_timed_ack(unsigned long ptr)
1548{
1549 struct veth_lpar_connection *cnx = (struct veth_lpar_connection *) ptr;
1550 unsigned long flags;
1551
1552 /* Ack all the events */
1553 spin_lock_irqsave(&cnx->lock, flags);
1554 if (cnx->num_pending_acks > 0)
1555 veth_flush_acks(cnx);
1556
1557 /* Reschedule the timer */
1558 cnx->ack_timer.expires = jiffies + cnx->ack_timeout;
1559 add_timer(&cnx->ack_timer);
1560 spin_unlock_irqrestore(&cnx->lock, flags);
1561}
1562
1563static int veth_remove(struct vio_dev *vdev)
1564{
Michael Ellermanf0c129c2005-09-01 11:29:09 +10001565 struct veth_lpar_connection *cnx;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001566 struct net_device *dev;
Michael Ellermanf0c129c2005-09-01 11:29:09 +10001567 struct veth_port *port;
1568 int i;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001569
Michael Ellermanf0c129c2005-09-01 11:29:09 +10001570 dev = veth_dev[vdev->unit_address];
1571
1572 if (! dev)
1573 return 0;
1574
1575 port = netdev_priv(dev);
1576
1577 for (i = 0; i < HVMAXARCHITECTEDLPS; i++) {
1578 cnx = veth_cnx[i];
1579
1580 if (cnx && (port->lpar_map & (1 << i))) {
1581 /* Drop our reference to connections on our VLAN */
1582 kobject_put(&cnx->kobject);
1583 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001584 }
Michael Ellermanf0c129c2005-09-01 11:29:09 +10001585
1586 veth_dev[vdev->unit_address] = NULL;
Michael Ellerman07a5c172005-09-01 11:29:21 +10001587 kobject_del(&port->kobject);
1588 kobject_put(&port->kobject);
Michael Ellermanf0c129c2005-09-01 11:29:09 +10001589 unregister_netdev(dev);
1590 free_netdev(dev);
1591
Linus Torvalds1da177e2005-04-16 15:20:36 -07001592 return 0;
1593}
1594
1595static int veth_probe(struct vio_dev *vdev, const struct vio_device_id *id)
1596{
1597 int i = vdev->unit_address;
1598 struct net_device *dev;
Michael Ellermanf0c129c2005-09-01 11:29:09 +10001599 struct veth_port *port;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001600
Stephen Rothwell07fb3f42006-05-15 13:39:47 +10001601 dev = veth_probe_one(i, vdev);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001602 if (dev == NULL) {
1603 veth_remove(vdev);
1604 return 1;
1605 }
1606 veth_dev[i] = dev;
1607
Michael Ellermanf0c129c2005-09-01 11:29:09 +10001608 port = (struct veth_port*)netdev_priv(dev);
1609
1610 /* Start the state machine on each connection on this vlan. If we're
1611 * the first dev to do so this will commence link negotiation */
1612 for (i = 0; i < HVMAXARCHITECTEDLPS; i++) {
1613 struct veth_lpar_connection *cnx;
1614
1615 if (! (port->lpar_map & (1 << i)))
1616 continue;
1617
1618 cnx = veth_cnx[i];
1619 if (!cnx)
1620 continue;
1621
1622 kobject_get(&cnx->kobject);
1623 veth_kick_statemachine(cnx);
1624 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001625
1626 return 0;
1627}
1628
1629/**
1630 * veth_device_table: Used by vio.c to match devices that we
1631 * support.
1632 */
1633static struct vio_device_id veth_device_table[] __devinitdata = {
Stephen Rothwell07fb3f42006-05-15 13:39:47 +10001634 { "network", "IBM,iSeries-l-lan" },
Stephen Rothwellfb120da2005-08-17 16:42:59 +10001635 { "", "" }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001636};
1637MODULE_DEVICE_TABLE(vio, veth_device_table);
1638
1639static struct vio_driver veth_driver = {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001640 .id_table = veth_device_table,
1641 .probe = veth_probe,
Stephen Rothwell6fdf5392005-10-24 14:53:21 +10001642 .remove = veth_remove,
1643 .driver = {
1644 .name = DRV_NAME,
Stephen Rothwell915124d2005-10-24 15:12:22 +10001645 .owner = THIS_MODULE,
Stephen Rothwell6fdf5392005-10-24 14:53:21 +10001646 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001647};
1648
1649/*
1650 * Module initialization/cleanup
1651 */
1652
Stephen Rothwell687d18a2006-12-15 15:42:50 +11001653static void __exit veth_module_cleanup(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001654{
1655 int i;
Michael Ellermanf0c129c2005-09-01 11:29:09 +10001656 struct veth_lpar_connection *cnx;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001657
Michael Ellermanf0c129c2005-09-01 11:29:09 +10001658 /* Disconnect our "irq" to stop events coming from the Hypervisor. */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001659 HvLpEvent_unregisterHandler(HvLpEvent_Type_VirtualLan);
1660
Michael Ellermanf0c129c2005-09-01 11:29:09 +10001661 /* Make sure any work queued from Hypervisor callbacks is finished. */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001662 flush_scheduled_work();
1663
Michael Ellermanf0c129c2005-09-01 11:29:09 +10001664 for (i = 0; i < HVMAXARCHITECTEDLPS; ++i) {
1665 cnx = veth_cnx[i];
1666
1667 if (!cnx)
1668 continue;
1669
Michael Ellerman76812d8122005-09-01 11:29:20 +10001670 /* Remove the connection from sysfs */
1671 kobject_del(&cnx->kobject);
Michael Ellermanf0c129c2005-09-01 11:29:09 +10001672 /* Drop the driver's reference to the connection */
1673 kobject_put(&cnx->kobject);
1674 }
1675
1676 /* Unregister the driver, which will close all the netdevs and stop
1677 * the connections when they're no longer referenced. */
Michael Ellermanb2e08522005-05-12 18:09:45 +10001678 vio_unregister_driver(&veth_driver);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001679}
1680module_exit(veth_module_cleanup);
1681
Stephen Rothwell687d18a2006-12-15 15:42:50 +11001682static int __init veth_module_init(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001683{
1684 int i;
1685 int rc;
1686
Stephen Rothwell687d18a2006-12-15 15:42:50 +11001687 if (!firmware_has_feature(FW_FEATURE_ISERIES))
1688 return -ENODEV;
1689
Linus Torvalds1da177e2005-04-16 15:20:36 -07001690 this_lp = HvLpConfig_getLpIndex_outline();
1691
1692 for (i = 0; i < HVMAXARCHITECTEDLPS; ++i) {
1693 rc = veth_init_connection(i);
Michael Ellermanec60bee2005-09-01 11:29:08 +10001694 if (rc != 0)
1695 goto error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001696 }
1697
1698 HvLpEvent_registerHandler(HvLpEvent_Type_VirtualLan,
1699 &veth_handle_event);
1700
Michael Ellermanec60bee2005-09-01 11:29:08 +10001701 rc = vio_register_driver(&veth_driver);
1702 if (rc != 0)
1703 goto error;
1704
Michael Ellerman76812d8122005-09-01 11:29:20 +10001705 for (i = 0; i < HVMAXARCHITECTEDLPS; ++i) {
1706 struct kobject *kobj;
1707
1708 if (!veth_cnx[i])
1709 continue;
1710
1711 kobj = &veth_cnx[i]->kobject;
Michael Ellerman76812d8122005-09-01 11:29:20 +10001712 /* If the add failes, complain but otherwise continue */
Greg Kroah-Hartmancbe9c592007-12-19 15:54:39 -04001713 if (0 != driver_add_kobj(&veth_driver.driver, kobj,
Greg Kroah-Hartman78be3d22007-12-17 23:05:35 -07001714 "cnx%.2d", veth_cnx[i]->remote_lp))
Michael Ellerman76812d8122005-09-01 11:29:20 +10001715 veth_error("cnx %d: Failed adding to sysfs.\n", i);
1716 }
1717
Michael Ellermanec60bee2005-09-01 11:29:08 +10001718 return 0;
1719
1720error:
1721 for (i = 0; i < HVMAXARCHITECTEDLPS; ++i) {
Michael Ellermanf0c129c2005-09-01 11:29:09 +10001722 veth_destroy_connection(veth_cnx[i]);
Michael Ellermanec60bee2005-09-01 11:29:08 +10001723 }
1724
1725 return rc;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001726}
1727module_init(veth_module_init);