Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1 | /* 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. |
| 7 | * |
| 8 | * This program is free software; you can redistribute it and/or |
| 9 | * modify it under the terms of the GNU General Public License as |
| 10 | * published by the Free Software Foundation; either version 2 of the |
| 11 | * License, or (at your option) any later version. |
| 12 | * |
| 13 | * This program is distributed in the hope that it will be useful, but |
| 14 | * WITHOUT ANY WARRANTY; without even the implied warranty of |
| 15 | * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU |
| 16 | * General Public License for more details. |
| 17 | * |
| 18 | * You should have received a copy of the GNU General Public License |
| 19 | * along with this program; if not, write to the Free Software |
| 20 | * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 |
| 21 | * USA |
| 22 | * |
| 23 | * |
| 24 | * This module implements the virtual ethernet device for iSeries LPAR |
| 25 | * Linux. It uses hypervisor message passing to implement an |
| 26 | * ethernet-like network device communicating between partitions on |
| 27 | * the iSeries. |
| 28 | * |
| 29 | * The iSeries LPAR hypervisor currently allows for up to 16 different |
| 30 | * virtual ethernets. These are all dynamically configurable on |
| 31 | * OS/400 partitions, but dynamic configuration is not supported under |
| 32 | * Linux yet. An ethXX network device will be created for each |
| 33 | * virtual ethernet this partition is connected to. |
| 34 | * |
| 35 | * - This driver is responsible for routing packets to and from other |
| 36 | * partitions. The MAC addresses used by the virtual ethernets |
| 37 | * contains meaning and must not be modified. |
| 38 | * |
| 39 | * - Having 2 virtual ethernets to the same remote partition DOES NOT |
| 40 | * double the available bandwidth. The 2 devices will share the |
| 41 | * available hypervisor bandwidth. |
| 42 | * |
| 43 | * - If you send a packet to your own mac address, it will just be |
| 44 | * dropped, you won't get it on the receive side. |
| 45 | * |
| 46 | * - Multicast is implemented by sending the frame frame to every |
| 47 | * other partition. It is the responsibility of the receiving |
| 48 | * partition to filter the addresses desired. |
| 49 | * |
| 50 | * Tunable parameters: |
| 51 | * |
| 52 | * VETH_NUMBUFFERS: This compile time option defaults to 120. It |
| 53 | * controls how much memory Linux will allocate per remote partition |
| 54 | * it is communicating with. It can be thought of as the maximum |
| 55 | * number of packets outstanding to a remote partition at a time. |
| 56 | */ |
| 57 | |
| 58 | #include <linux/config.h> |
| 59 | #include <linux/module.h> |
| 60 | #include <linux/version.h> |
| 61 | #include <linux/types.h> |
| 62 | #include <linux/errno.h> |
| 63 | #include <linux/ioport.h> |
| 64 | #include <linux/kernel.h> |
| 65 | #include <linux/netdevice.h> |
| 66 | #include <linux/etherdevice.h> |
| 67 | #include <linux/skbuff.h> |
| 68 | #include <linux/init.h> |
| 69 | #include <linux/delay.h> |
| 70 | #include <linux/mm.h> |
| 71 | #include <linux/ethtool.h> |
| 72 | #include <asm/iSeries/mf.h> |
| 73 | #include <asm/iSeries/iSeries_pci.h> |
| 74 | #include <asm/uaccess.h> |
| 75 | |
| 76 | #include <asm/iSeries/HvLpConfig.h> |
| 77 | #include <asm/iSeries/HvTypes.h> |
| 78 | #include <asm/iSeries/HvLpEvent.h> |
| 79 | #include <asm/iommu.h> |
| 80 | #include <asm/vio.h> |
| 81 | |
| 82 | #include "iseries_veth.h" |
| 83 | |
| 84 | MODULE_AUTHOR("Kyle Lucke <klucke@us.ibm.com>"); |
| 85 | MODULE_DESCRIPTION("iSeries Virtual ethernet driver"); |
| 86 | MODULE_LICENSE("GPL"); |
| 87 | |
| 88 | #define VETH_NUMBUFFERS (120) |
| 89 | #define VETH_ACKTIMEOUT (1000000) /* microseconds */ |
| 90 | #define VETH_MAX_MCAST (12) |
| 91 | |
| 92 | #define VETH_MAX_MTU (9000) |
| 93 | |
| 94 | #if VETH_NUMBUFFERS < 10 |
| 95 | #define ACK_THRESHOLD (1) |
| 96 | #elif VETH_NUMBUFFERS < 20 |
| 97 | #define ACK_THRESHOLD (4) |
| 98 | #elif VETH_NUMBUFFERS < 40 |
| 99 | #define ACK_THRESHOLD (10) |
| 100 | #else |
| 101 | #define ACK_THRESHOLD (20) |
| 102 | #endif |
| 103 | |
| 104 | #define VETH_STATE_SHUTDOWN (0x0001) |
| 105 | #define VETH_STATE_OPEN (0x0002) |
| 106 | #define VETH_STATE_RESET (0x0004) |
| 107 | #define VETH_STATE_SENTMON (0x0008) |
| 108 | #define VETH_STATE_SENTCAPS (0x0010) |
| 109 | #define VETH_STATE_GOTCAPACK (0x0020) |
| 110 | #define VETH_STATE_GOTCAPS (0x0040) |
| 111 | #define VETH_STATE_SENTCAPACK (0x0080) |
| 112 | #define VETH_STATE_READY (0x0100) |
| 113 | |
| 114 | struct veth_msg { |
| 115 | struct veth_msg *next; |
| 116 | struct VethFramesData data; |
| 117 | int token; |
| 118 | unsigned long in_use; |
| 119 | struct sk_buff *skb; |
| 120 | struct device *dev; |
| 121 | }; |
| 122 | |
| 123 | struct veth_lpar_connection { |
| 124 | HvLpIndex remote_lp; |
| 125 | struct work_struct statemachine_wq; |
| 126 | struct veth_msg *msgs; |
| 127 | int num_events; |
| 128 | struct VethCapData local_caps; |
| 129 | |
| 130 | struct timer_list ack_timer; |
| 131 | |
| 132 | spinlock_t lock; |
| 133 | unsigned long state; |
| 134 | HvLpInstanceId src_inst; |
| 135 | HvLpInstanceId dst_inst; |
| 136 | struct VethLpEvent cap_event, cap_ack_event; |
| 137 | u16 pending_acks[VETH_MAX_ACKS_PER_MSG]; |
| 138 | u32 num_pending_acks; |
| 139 | |
| 140 | int num_ack_events; |
| 141 | struct VethCapData remote_caps; |
| 142 | u32 ack_timeout; |
| 143 | |
| 144 | spinlock_t msg_stack_lock; |
| 145 | struct veth_msg *msg_stack_head; |
| 146 | }; |
| 147 | |
| 148 | struct veth_port { |
| 149 | struct device *dev; |
| 150 | struct net_device_stats stats; |
| 151 | u64 mac_addr; |
| 152 | HvLpIndexMap lpar_map; |
| 153 | |
| 154 | spinlock_t pending_gate; |
| 155 | struct sk_buff *pending_skb; |
| 156 | HvLpIndexMap pending_lpmask; |
| 157 | |
| 158 | rwlock_t mcast_gate; |
| 159 | int promiscuous; |
| 160 | int all_mcast; |
| 161 | int num_mcast; |
| 162 | u64 mcast_addr[VETH_MAX_MCAST]; |
| 163 | }; |
| 164 | |
| 165 | static HvLpIndex this_lp; |
| 166 | static struct veth_lpar_connection *veth_cnx[HVMAXARCHITECTEDLPS]; /* = 0 */ |
| 167 | static struct net_device *veth_dev[HVMAXARCHITECTEDVIRTUALLANS]; /* = 0 */ |
| 168 | |
| 169 | static int veth_start_xmit(struct sk_buff *skb, struct net_device *dev); |
| 170 | static void veth_recycle_msg(struct veth_lpar_connection *, struct veth_msg *); |
| 171 | static void veth_flush_pending(struct veth_lpar_connection *cnx); |
| 172 | static void veth_receive(struct veth_lpar_connection *, struct VethLpEvent *); |
| 173 | static void veth_timed_ack(unsigned long connectionPtr); |
| 174 | |
| 175 | /* |
| 176 | * Utility functions |
| 177 | */ |
| 178 | |
| 179 | #define veth_printk(prio, fmt, args...) \ |
| 180 | printk(prio "%s: " fmt, __FILE__, ## args) |
| 181 | |
| 182 | #define veth_error(fmt, args...) \ |
| 183 | printk(KERN_ERR "(%s:%3.3d) ERROR: " fmt, __FILE__, __LINE__ , ## args) |
| 184 | |
| 185 | static inline void veth_stack_push(struct veth_lpar_connection *cnx, |
| 186 | struct veth_msg *msg) |
| 187 | { |
| 188 | unsigned long flags; |
| 189 | |
| 190 | spin_lock_irqsave(&cnx->msg_stack_lock, flags); |
| 191 | msg->next = cnx->msg_stack_head; |
| 192 | cnx->msg_stack_head = msg; |
| 193 | spin_unlock_irqrestore(&cnx->msg_stack_lock, flags); |
| 194 | } |
| 195 | |
| 196 | static inline struct veth_msg *veth_stack_pop(struct veth_lpar_connection *cnx) |
| 197 | { |
| 198 | unsigned long flags; |
| 199 | struct veth_msg *msg; |
| 200 | |
| 201 | spin_lock_irqsave(&cnx->msg_stack_lock, flags); |
| 202 | msg = cnx->msg_stack_head; |
| 203 | if (msg) |
| 204 | cnx->msg_stack_head = cnx->msg_stack_head->next; |
| 205 | spin_unlock_irqrestore(&cnx->msg_stack_lock, flags); |
| 206 | return msg; |
| 207 | } |
| 208 | |
| 209 | static inline HvLpEvent_Rc |
| 210 | veth_signalevent(struct veth_lpar_connection *cnx, u16 subtype, |
| 211 | HvLpEvent_AckInd ackind, HvLpEvent_AckType acktype, |
| 212 | u64 token, |
| 213 | u64 data1, u64 data2, u64 data3, u64 data4, u64 data5) |
| 214 | { |
| 215 | return HvCallEvent_signalLpEventFast(cnx->remote_lp, |
| 216 | HvLpEvent_Type_VirtualLan, |
| 217 | subtype, ackind, acktype, |
| 218 | cnx->src_inst, |
| 219 | cnx->dst_inst, |
| 220 | token, data1, data2, data3, |
| 221 | data4, data5); |
| 222 | } |
| 223 | |
| 224 | static inline HvLpEvent_Rc veth_signaldata(struct veth_lpar_connection *cnx, |
| 225 | u16 subtype, u64 token, void *data) |
| 226 | { |
| 227 | u64 *p = (u64 *) data; |
| 228 | |
| 229 | return veth_signalevent(cnx, subtype, HvLpEvent_AckInd_NoAck, |
| 230 | HvLpEvent_AckType_ImmediateAck, |
| 231 | token, p[0], p[1], p[2], p[3], p[4]); |
| 232 | } |
| 233 | |
| 234 | struct veth_allocation { |
| 235 | struct completion c; |
| 236 | int num; |
| 237 | }; |
| 238 | |
| 239 | static void veth_complete_allocation(void *parm, int number) |
| 240 | { |
| 241 | struct veth_allocation *vc = (struct veth_allocation *)parm; |
| 242 | |
| 243 | vc->num = number; |
| 244 | complete(&vc->c); |
| 245 | } |
| 246 | |
| 247 | static int veth_allocate_events(HvLpIndex rlp, int number) |
| 248 | { |
| 249 | struct veth_allocation vc = { COMPLETION_INITIALIZER(vc.c), 0 }; |
| 250 | |
| 251 | mf_allocate_lp_events(rlp, HvLpEvent_Type_VirtualLan, |
| 252 | sizeof(struct VethLpEvent), number, |
| 253 | &veth_complete_allocation, &vc); |
| 254 | wait_for_completion(&vc.c); |
| 255 | |
| 256 | return vc.num; |
| 257 | } |
| 258 | |
| 259 | /* |
| 260 | * LPAR connection code |
| 261 | */ |
| 262 | |
| 263 | static inline void veth_kick_statemachine(struct veth_lpar_connection *cnx) |
| 264 | { |
| 265 | schedule_work(&cnx->statemachine_wq); |
| 266 | } |
| 267 | |
| 268 | static void veth_take_cap(struct veth_lpar_connection *cnx, |
| 269 | struct VethLpEvent *event) |
| 270 | { |
| 271 | unsigned long flags; |
| 272 | |
| 273 | spin_lock_irqsave(&cnx->lock, flags); |
| 274 | /* Receiving caps may mean the other end has just come up, so |
| 275 | * we need to reload the instance ID of the far end */ |
| 276 | cnx->dst_inst = |
| 277 | HvCallEvent_getTargetLpInstanceId(cnx->remote_lp, |
| 278 | HvLpEvent_Type_VirtualLan); |
| 279 | |
| 280 | if (cnx->state & VETH_STATE_GOTCAPS) { |
| 281 | veth_error("Received a second capabilities from lpar %d\n", |
| 282 | cnx->remote_lp); |
| 283 | event->base_event.xRc = HvLpEvent_Rc_BufferNotAvailable; |
| 284 | HvCallEvent_ackLpEvent((struct HvLpEvent *) event); |
| 285 | } else { |
| 286 | memcpy(&cnx->cap_event, event, sizeof(cnx->cap_event)); |
| 287 | cnx->state |= VETH_STATE_GOTCAPS; |
| 288 | veth_kick_statemachine(cnx); |
| 289 | } |
| 290 | spin_unlock_irqrestore(&cnx->lock, flags); |
| 291 | } |
| 292 | |
| 293 | static void veth_take_cap_ack(struct veth_lpar_connection *cnx, |
| 294 | struct VethLpEvent *event) |
| 295 | { |
| 296 | unsigned long flags; |
| 297 | |
| 298 | spin_lock_irqsave(&cnx->lock, flags); |
| 299 | if (cnx->state & VETH_STATE_GOTCAPACK) { |
| 300 | veth_error("Received a second capabilities ack from lpar %d\n", |
| 301 | cnx->remote_lp); |
| 302 | } else { |
| 303 | memcpy(&cnx->cap_ack_event, event, |
| 304 | sizeof(&cnx->cap_ack_event)); |
| 305 | cnx->state |= VETH_STATE_GOTCAPACK; |
| 306 | veth_kick_statemachine(cnx); |
| 307 | } |
| 308 | spin_unlock_irqrestore(&cnx->lock, flags); |
| 309 | } |
| 310 | |
| 311 | static void veth_take_monitor_ack(struct veth_lpar_connection *cnx, |
| 312 | struct VethLpEvent *event) |
| 313 | { |
| 314 | unsigned long flags; |
| 315 | |
| 316 | spin_lock_irqsave(&cnx->lock, flags); |
| 317 | veth_printk(KERN_DEBUG, "Monitor ack returned for lpar %d\n", |
| 318 | cnx->remote_lp); |
| 319 | cnx->state |= VETH_STATE_RESET; |
| 320 | veth_kick_statemachine(cnx); |
| 321 | spin_unlock_irqrestore(&cnx->lock, flags); |
| 322 | } |
| 323 | |
| 324 | static void veth_handle_ack(struct VethLpEvent *event) |
| 325 | { |
| 326 | HvLpIndex rlp = event->base_event.xTargetLp; |
| 327 | struct veth_lpar_connection *cnx = veth_cnx[rlp]; |
| 328 | |
| 329 | BUG_ON(! cnx); |
| 330 | |
| 331 | switch (event->base_event.xSubtype) { |
| 332 | case VethEventTypeCap: |
| 333 | veth_take_cap_ack(cnx, event); |
| 334 | break; |
| 335 | case VethEventTypeMonitor: |
| 336 | veth_take_monitor_ack(cnx, event); |
| 337 | break; |
| 338 | default: |
| 339 | veth_error("Unknown ack type %d from lpar %d\n", |
| 340 | event->base_event.xSubtype, rlp); |
| 341 | }; |
| 342 | } |
| 343 | |
| 344 | static void veth_handle_int(struct VethLpEvent *event) |
| 345 | { |
| 346 | HvLpIndex rlp = event->base_event.xSourceLp; |
| 347 | struct veth_lpar_connection *cnx = veth_cnx[rlp]; |
| 348 | unsigned long flags; |
| 349 | int i; |
| 350 | |
| 351 | BUG_ON(! cnx); |
| 352 | |
| 353 | switch (event->base_event.xSubtype) { |
| 354 | case VethEventTypeCap: |
| 355 | veth_take_cap(cnx, event); |
| 356 | break; |
| 357 | case VethEventTypeMonitor: |
| 358 | /* do nothing... this'll hang out here til we're dead, |
| 359 | * and the hypervisor will return it for us. */ |
| 360 | break; |
| 361 | case VethEventTypeFramesAck: |
| 362 | spin_lock_irqsave(&cnx->lock, flags); |
| 363 | for (i = 0; i < VETH_MAX_ACKS_PER_MSG; ++i) { |
| 364 | u16 msgnum = event->u.frames_ack_data.token[i]; |
| 365 | |
| 366 | if (msgnum < VETH_NUMBUFFERS) |
| 367 | veth_recycle_msg(cnx, cnx->msgs + msgnum); |
| 368 | } |
| 369 | spin_unlock_irqrestore(&cnx->lock, flags); |
| 370 | veth_flush_pending(cnx); |
| 371 | break; |
| 372 | case VethEventTypeFrames: |
| 373 | veth_receive(cnx, event); |
| 374 | break; |
| 375 | default: |
| 376 | veth_error("Unknown interrupt type %d from lpar %d\n", |
| 377 | event->base_event.xSubtype, rlp); |
| 378 | }; |
| 379 | } |
| 380 | |
| 381 | static void veth_handle_event(struct HvLpEvent *event, struct pt_regs *regs) |
| 382 | { |
| 383 | struct VethLpEvent *veth_event = (struct VethLpEvent *)event; |
| 384 | |
| 385 | if (event->xFlags.xFunction == HvLpEvent_Function_Ack) |
| 386 | veth_handle_ack(veth_event); |
| 387 | else if (event->xFlags.xFunction == HvLpEvent_Function_Int) |
| 388 | veth_handle_int(veth_event); |
| 389 | } |
| 390 | |
| 391 | static int veth_process_caps(struct veth_lpar_connection *cnx) |
| 392 | { |
| 393 | struct VethCapData *remote_caps = &cnx->remote_caps; |
| 394 | int num_acks_needed; |
| 395 | |
| 396 | /* Convert timer to jiffies */ |
| 397 | cnx->ack_timeout = remote_caps->ack_timeout * HZ / 1000000; |
| 398 | |
| 399 | if ( (remote_caps->num_buffers == 0) |
| 400 | || (remote_caps->ack_threshold > VETH_MAX_ACKS_PER_MSG) |
| 401 | || (remote_caps->ack_threshold == 0) |
| 402 | || (cnx->ack_timeout == 0) ) { |
| 403 | veth_error("Received incompatible capabilities from lpar %d\n", |
| 404 | cnx->remote_lp); |
| 405 | return HvLpEvent_Rc_InvalidSubtypeData; |
| 406 | } |
| 407 | |
| 408 | num_acks_needed = (remote_caps->num_buffers |
| 409 | / remote_caps->ack_threshold) + 1; |
| 410 | |
| 411 | /* FIXME: locking on num_ack_events? */ |
| 412 | if (cnx->num_ack_events < num_acks_needed) { |
| 413 | int num; |
| 414 | |
| 415 | num = veth_allocate_events(cnx->remote_lp, |
| 416 | num_acks_needed-cnx->num_ack_events); |
| 417 | if (num > 0) |
| 418 | cnx->num_ack_events += num; |
| 419 | |
| 420 | if (cnx->num_ack_events < num_acks_needed) { |
| 421 | veth_error("Couldn't allocate enough ack events for lpar %d\n", |
| 422 | cnx->remote_lp); |
| 423 | |
| 424 | return HvLpEvent_Rc_BufferNotAvailable; |
| 425 | } |
| 426 | } |
| 427 | |
| 428 | |
| 429 | return HvLpEvent_Rc_Good; |
| 430 | } |
| 431 | |
| 432 | /* FIXME: The gotos here are a bit dubious */ |
| 433 | static void veth_statemachine(void *p) |
| 434 | { |
| 435 | struct veth_lpar_connection *cnx = (struct veth_lpar_connection *)p; |
| 436 | int rlp = cnx->remote_lp; |
| 437 | int rc; |
| 438 | |
| 439 | spin_lock_irq(&cnx->lock); |
| 440 | |
| 441 | restart: |
| 442 | if (cnx->state & VETH_STATE_RESET) { |
| 443 | int i; |
| 444 | |
| 445 | del_timer(&cnx->ack_timer); |
| 446 | |
| 447 | if (cnx->state & VETH_STATE_OPEN) |
| 448 | HvCallEvent_closeLpEventPath(cnx->remote_lp, |
| 449 | HvLpEvent_Type_VirtualLan); |
| 450 | |
| 451 | /* reset ack data */ |
| 452 | memset(&cnx->pending_acks, 0xff, sizeof (cnx->pending_acks)); |
| 453 | cnx->num_pending_acks = 0; |
| 454 | |
| 455 | cnx->state &= ~(VETH_STATE_RESET | VETH_STATE_SENTMON |
| 456 | | VETH_STATE_OPEN | VETH_STATE_SENTCAPS |
| 457 | | VETH_STATE_GOTCAPACK | VETH_STATE_GOTCAPS |
| 458 | | VETH_STATE_SENTCAPACK | VETH_STATE_READY); |
| 459 | |
| 460 | /* Clean up any leftover messages */ |
| 461 | if (cnx->msgs) |
| 462 | for (i = 0; i < VETH_NUMBUFFERS; ++i) |
| 463 | veth_recycle_msg(cnx, cnx->msgs + i); |
| 464 | spin_unlock_irq(&cnx->lock); |
| 465 | veth_flush_pending(cnx); |
| 466 | spin_lock_irq(&cnx->lock); |
| 467 | if (cnx->state & VETH_STATE_RESET) |
| 468 | goto restart; |
| 469 | } |
| 470 | |
| 471 | if (cnx->state & VETH_STATE_SHUTDOWN) |
| 472 | /* It's all over, do nothing */ |
| 473 | goto out; |
| 474 | |
| 475 | if ( !(cnx->state & VETH_STATE_OPEN) ) { |
| 476 | if (! cnx->msgs || (cnx->num_events < (2 + VETH_NUMBUFFERS)) ) |
| 477 | goto cant_cope; |
| 478 | |
| 479 | HvCallEvent_openLpEventPath(rlp, HvLpEvent_Type_VirtualLan); |
| 480 | cnx->src_inst = |
| 481 | HvCallEvent_getSourceLpInstanceId(rlp, |
| 482 | HvLpEvent_Type_VirtualLan); |
| 483 | cnx->dst_inst = |
| 484 | HvCallEvent_getTargetLpInstanceId(rlp, |
| 485 | HvLpEvent_Type_VirtualLan); |
| 486 | cnx->state |= VETH_STATE_OPEN; |
| 487 | } |
| 488 | |
| 489 | if ( (cnx->state & VETH_STATE_OPEN) |
| 490 | && !(cnx->state & VETH_STATE_SENTMON) ) { |
| 491 | rc = veth_signalevent(cnx, VethEventTypeMonitor, |
| 492 | HvLpEvent_AckInd_DoAck, |
| 493 | HvLpEvent_AckType_DeferredAck, |
| 494 | 0, 0, 0, 0, 0, 0); |
| 495 | |
| 496 | if (rc == HvLpEvent_Rc_Good) { |
| 497 | cnx->state |= VETH_STATE_SENTMON; |
| 498 | } else { |
| 499 | if ( (rc != HvLpEvent_Rc_PartitionDead) |
| 500 | && (rc != HvLpEvent_Rc_PathClosed) ) |
| 501 | veth_error("Error sending monitor to " |
| 502 | "lpar %d, rc=%x\n", |
| 503 | rlp, (int) rc); |
| 504 | |
| 505 | /* Oh well, hope we get a cap from the other |
| 506 | * end and do better when that kicks us */ |
| 507 | goto out; |
| 508 | } |
| 509 | } |
| 510 | |
| 511 | if ( (cnx->state & VETH_STATE_OPEN) |
| 512 | && !(cnx->state & VETH_STATE_SENTCAPS)) { |
| 513 | u64 *rawcap = (u64 *)&cnx->local_caps; |
| 514 | |
| 515 | rc = veth_signalevent(cnx, VethEventTypeCap, |
| 516 | HvLpEvent_AckInd_DoAck, |
| 517 | HvLpEvent_AckType_ImmediateAck, |
| 518 | 0, rawcap[0], rawcap[1], rawcap[2], |
| 519 | rawcap[3], rawcap[4]); |
| 520 | |
| 521 | if (rc == HvLpEvent_Rc_Good) { |
| 522 | cnx->state |= VETH_STATE_SENTCAPS; |
| 523 | } else { |
| 524 | if ( (rc != HvLpEvent_Rc_PartitionDead) |
| 525 | && (rc != HvLpEvent_Rc_PathClosed) ) |
| 526 | veth_error("Error sending caps to " |
| 527 | "lpar %d, rc=%x\n", |
| 528 | rlp, (int) rc); |
| 529 | /* Oh well, hope we get a cap from the other |
| 530 | * end and do better when that kicks us */ |
| 531 | goto out; |
| 532 | } |
| 533 | } |
| 534 | |
| 535 | if ((cnx->state & VETH_STATE_GOTCAPS) |
| 536 | && !(cnx->state & VETH_STATE_SENTCAPACK)) { |
| 537 | struct VethCapData *remote_caps = &cnx->remote_caps; |
| 538 | |
| 539 | memcpy(remote_caps, &cnx->cap_event.u.caps_data, |
| 540 | sizeof(*remote_caps)); |
| 541 | |
| 542 | spin_unlock_irq(&cnx->lock); |
| 543 | rc = veth_process_caps(cnx); |
| 544 | spin_lock_irq(&cnx->lock); |
| 545 | |
| 546 | /* We dropped the lock, so recheck for anything which |
| 547 | * might mess us up */ |
| 548 | if (cnx->state & (VETH_STATE_RESET|VETH_STATE_SHUTDOWN)) |
| 549 | goto restart; |
| 550 | |
| 551 | cnx->cap_event.base_event.xRc = rc; |
| 552 | HvCallEvent_ackLpEvent((struct HvLpEvent *)&cnx->cap_event); |
| 553 | if (rc == HvLpEvent_Rc_Good) |
| 554 | cnx->state |= VETH_STATE_SENTCAPACK; |
| 555 | else |
| 556 | goto cant_cope; |
| 557 | } |
| 558 | |
| 559 | if ((cnx->state & VETH_STATE_GOTCAPACK) |
| 560 | && (cnx->state & VETH_STATE_GOTCAPS) |
| 561 | && !(cnx->state & VETH_STATE_READY)) { |
| 562 | if (cnx->cap_ack_event.base_event.xRc == HvLpEvent_Rc_Good) { |
| 563 | /* Start the ACK timer */ |
| 564 | cnx->ack_timer.expires = jiffies + cnx->ack_timeout; |
| 565 | add_timer(&cnx->ack_timer); |
| 566 | cnx->state |= VETH_STATE_READY; |
| 567 | } else { |
| 568 | veth_printk(KERN_ERR, "Caps rejected (rc=%d) by " |
| 569 | "lpar %d\n", |
| 570 | cnx->cap_ack_event.base_event.xRc, |
| 571 | rlp); |
| 572 | goto cant_cope; |
| 573 | } |
| 574 | } |
| 575 | |
| 576 | out: |
| 577 | spin_unlock_irq(&cnx->lock); |
| 578 | return; |
| 579 | |
| 580 | cant_cope: |
| 581 | /* FIXME: we get here if something happens we really can't |
| 582 | * cope with. The link will never work once we get here, and |
| 583 | * all we can do is not lock the rest of the system up */ |
| 584 | veth_error("Badness on connection to lpar %d (state=%04lx) " |
| 585 | " - shutting down\n", rlp, cnx->state); |
| 586 | cnx->state |= VETH_STATE_SHUTDOWN; |
| 587 | spin_unlock_irq(&cnx->lock); |
| 588 | } |
| 589 | |
| 590 | static int veth_init_connection(u8 rlp) |
| 591 | { |
| 592 | struct veth_lpar_connection *cnx; |
| 593 | struct veth_msg *msgs; |
| 594 | int i; |
| 595 | |
| 596 | if ( (rlp == this_lp) |
| 597 | || ! HvLpConfig_doLpsCommunicateOnVirtualLan(this_lp, rlp) ) |
| 598 | return 0; |
| 599 | |
| 600 | cnx = kmalloc(sizeof(*cnx), GFP_KERNEL); |
| 601 | if (! cnx) |
| 602 | return -ENOMEM; |
| 603 | memset(cnx, 0, sizeof(*cnx)); |
| 604 | |
| 605 | cnx->remote_lp = rlp; |
| 606 | spin_lock_init(&cnx->lock); |
| 607 | INIT_WORK(&cnx->statemachine_wq, veth_statemachine, cnx); |
| 608 | init_timer(&cnx->ack_timer); |
| 609 | cnx->ack_timer.function = veth_timed_ack; |
| 610 | cnx->ack_timer.data = (unsigned long) cnx; |
| 611 | memset(&cnx->pending_acks, 0xff, sizeof (cnx->pending_acks)); |
| 612 | |
| 613 | veth_cnx[rlp] = cnx; |
| 614 | |
| 615 | msgs = kmalloc(VETH_NUMBUFFERS * sizeof(struct veth_msg), GFP_KERNEL); |
| 616 | if (! msgs) { |
| 617 | veth_error("Can't allocate buffers for lpar %d\n", rlp); |
| 618 | return -ENOMEM; |
| 619 | } |
| 620 | |
| 621 | cnx->msgs = msgs; |
| 622 | memset(msgs, 0, VETH_NUMBUFFERS * sizeof(struct veth_msg)); |
| 623 | spin_lock_init(&cnx->msg_stack_lock); |
| 624 | |
| 625 | for (i = 0; i < VETH_NUMBUFFERS; i++) { |
| 626 | msgs[i].token = i; |
| 627 | veth_stack_push(cnx, msgs + i); |
| 628 | } |
| 629 | |
| 630 | cnx->num_events = veth_allocate_events(rlp, 2 + VETH_NUMBUFFERS); |
| 631 | |
| 632 | if (cnx->num_events < (2 + VETH_NUMBUFFERS)) { |
| 633 | veth_error("Can't allocate events for lpar %d, only got %d\n", |
| 634 | rlp, cnx->num_events); |
| 635 | return -ENOMEM; |
| 636 | } |
| 637 | |
| 638 | cnx->local_caps.num_buffers = VETH_NUMBUFFERS; |
| 639 | cnx->local_caps.ack_threshold = ACK_THRESHOLD; |
| 640 | cnx->local_caps.ack_timeout = VETH_ACKTIMEOUT; |
| 641 | |
| 642 | return 0; |
| 643 | } |
| 644 | |
| 645 | static void veth_stop_connection(u8 rlp) |
| 646 | { |
| 647 | struct veth_lpar_connection *cnx = veth_cnx[rlp]; |
| 648 | |
| 649 | if (! cnx) |
| 650 | return; |
| 651 | |
| 652 | spin_lock_irq(&cnx->lock); |
| 653 | cnx->state |= VETH_STATE_RESET | VETH_STATE_SHUTDOWN; |
| 654 | veth_kick_statemachine(cnx); |
| 655 | spin_unlock_irq(&cnx->lock); |
| 656 | |
| 657 | flush_scheduled_work(); |
| 658 | |
| 659 | /* FIXME: not sure if this is necessary - will already have |
| 660 | * been deleted by the state machine, just want to make sure |
| 661 | * its not running any more */ |
| 662 | del_timer_sync(&cnx->ack_timer); |
| 663 | |
| 664 | if (cnx->num_events > 0) |
| 665 | mf_deallocate_lp_events(cnx->remote_lp, |
| 666 | HvLpEvent_Type_VirtualLan, |
| 667 | cnx->num_events, |
| 668 | NULL, NULL); |
| 669 | if (cnx->num_ack_events > 0) |
| 670 | mf_deallocate_lp_events(cnx->remote_lp, |
| 671 | HvLpEvent_Type_VirtualLan, |
| 672 | cnx->num_ack_events, |
| 673 | NULL, NULL); |
| 674 | } |
| 675 | |
| 676 | static void veth_destroy_connection(u8 rlp) |
| 677 | { |
| 678 | struct veth_lpar_connection *cnx = veth_cnx[rlp]; |
| 679 | |
| 680 | if (! cnx) |
| 681 | return; |
| 682 | |
| 683 | kfree(cnx->msgs); |
| 684 | kfree(cnx); |
| 685 | veth_cnx[rlp] = NULL; |
| 686 | } |
| 687 | |
| 688 | /* |
| 689 | * net_device code |
| 690 | */ |
| 691 | |
| 692 | static int veth_open(struct net_device *dev) |
| 693 | { |
| 694 | struct veth_port *port = (struct veth_port *) dev->priv; |
| 695 | |
| 696 | memset(&port->stats, 0, sizeof (port->stats)); |
| 697 | netif_start_queue(dev); |
| 698 | return 0; |
| 699 | } |
| 700 | |
| 701 | static int veth_close(struct net_device *dev) |
| 702 | { |
| 703 | netif_stop_queue(dev); |
| 704 | return 0; |
| 705 | } |
| 706 | |
| 707 | static struct net_device_stats *veth_get_stats(struct net_device *dev) |
| 708 | { |
| 709 | struct veth_port *port = (struct veth_port *) dev->priv; |
| 710 | |
| 711 | return &port->stats; |
| 712 | } |
| 713 | |
| 714 | static int veth_change_mtu(struct net_device *dev, int new_mtu) |
| 715 | { |
| 716 | if ((new_mtu < 68) || (new_mtu > VETH_MAX_MTU)) |
| 717 | return -EINVAL; |
| 718 | dev->mtu = new_mtu; |
| 719 | return 0; |
| 720 | } |
| 721 | |
| 722 | static void veth_set_multicast_list(struct net_device *dev) |
| 723 | { |
| 724 | struct veth_port *port = (struct veth_port *) dev->priv; |
| 725 | unsigned long flags; |
| 726 | |
| 727 | write_lock_irqsave(&port->mcast_gate, flags); |
| 728 | |
| 729 | if (dev->flags & IFF_PROMISC) { /* set promiscuous mode */ |
| 730 | printk(KERN_INFO "%s: Promiscuous mode enabled.\n", |
| 731 | dev->name); |
| 732 | port->promiscuous = 1; |
| 733 | } else if ( (dev->flags & IFF_ALLMULTI) |
| 734 | || (dev->mc_count > VETH_MAX_MCAST) ) { |
| 735 | port->all_mcast = 1; |
| 736 | } else { |
| 737 | struct dev_mc_list *dmi = dev->mc_list; |
| 738 | int i; |
| 739 | |
| 740 | /* Update table */ |
| 741 | port->num_mcast = 0; |
| 742 | |
| 743 | for (i = 0; i < dev->mc_count; i++) { |
| 744 | u8 *addr = dmi->dmi_addr; |
| 745 | u64 xaddr = 0; |
| 746 | |
| 747 | if (addr[0] & 0x01) {/* multicast address? */ |
| 748 | memcpy(&xaddr, addr, ETH_ALEN); |
| 749 | port->mcast_addr[port->num_mcast] = xaddr; |
| 750 | port->num_mcast++; |
| 751 | } |
| 752 | dmi = dmi->next; |
| 753 | } |
| 754 | } |
| 755 | |
| 756 | write_unlock_irqrestore(&port->mcast_gate, flags); |
| 757 | } |
| 758 | |
| 759 | static void veth_get_drvinfo(struct net_device *dev, struct ethtool_drvinfo *info) |
| 760 | { |
| 761 | strncpy(info->driver, "veth", sizeof(info->driver) - 1); |
| 762 | info->driver[sizeof(info->driver) - 1] = '\0'; |
| 763 | strncpy(info->version, "1.0", sizeof(info->version) - 1); |
| 764 | } |
| 765 | |
| 766 | static int veth_get_settings(struct net_device *dev, struct ethtool_cmd *ecmd) |
| 767 | { |
| 768 | ecmd->supported = (SUPPORTED_1000baseT_Full |
| 769 | | SUPPORTED_Autoneg | SUPPORTED_FIBRE); |
| 770 | ecmd->advertising = (SUPPORTED_1000baseT_Full |
| 771 | | SUPPORTED_Autoneg | SUPPORTED_FIBRE); |
| 772 | ecmd->port = PORT_FIBRE; |
| 773 | ecmd->transceiver = XCVR_INTERNAL; |
| 774 | ecmd->phy_address = 0; |
| 775 | ecmd->speed = SPEED_1000; |
| 776 | ecmd->duplex = DUPLEX_FULL; |
| 777 | ecmd->autoneg = AUTONEG_ENABLE; |
| 778 | ecmd->maxtxpkt = 120; |
| 779 | ecmd->maxrxpkt = 120; |
| 780 | return 0; |
| 781 | } |
| 782 | |
| 783 | static u32 veth_get_link(struct net_device *dev) |
| 784 | { |
| 785 | return 1; |
| 786 | } |
| 787 | |
| 788 | static struct ethtool_ops ops = { |
| 789 | .get_drvinfo = veth_get_drvinfo, |
| 790 | .get_settings = veth_get_settings, |
| 791 | .get_link = veth_get_link, |
| 792 | }; |
| 793 | |
| 794 | static void veth_tx_timeout(struct net_device *dev) |
| 795 | { |
| 796 | struct veth_port *port = (struct veth_port *)dev->priv; |
| 797 | struct net_device_stats *stats = &port->stats; |
| 798 | unsigned long flags; |
| 799 | int i; |
| 800 | |
| 801 | stats->tx_errors++; |
| 802 | |
| 803 | spin_lock_irqsave(&port->pending_gate, flags); |
| 804 | |
Michael Ellerman | 243cd55 | 2005-06-09 12:36:33 -0700 | [diff] [blame^] | 805 | if (!port->pending_lpmask) { |
| 806 | spin_unlock_irqrestore(&port->pending_gate, flags); |
| 807 | return; |
| 808 | } |
| 809 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 810 | printk(KERN_WARNING "%s: Tx timeout! Resetting lp connections: %08x\n", |
| 811 | dev->name, port->pending_lpmask); |
| 812 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 813 | for (i = 0; i < HVMAXARCHITECTEDLPS; i++) { |
| 814 | struct veth_lpar_connection *cnx = veth_cnx[i]; |
| 815 | |
| 816 | if (! (port->pending_lpmask & (1<<i))) |
| 817 | continue; |
| 818 | |
| 819 | /* If we're pending on it, we must be connected to it, |
| 820 | * so we should certainly have a structure for it. */ |
| 821 | BUG_ON(! cnx); |
| 822 | |
| 823 | /* Theoretically we could be kicking a connection |
| 824 | * which doesn't deserve it, but in practice if we've |
| 825 | * had a Tx timeout, the pending_lpmask will have |
| 826 | * exactly one bit set - the connection causing the |
| 827 | * problem. */ |
| 828 | spin_lock(&cnx->lock); |
| 829 | cnx->state |= VETH_STATE_RESET; |
| 830 | veth_kick_statemachine(cnx); |
| 831 | spin_unlock(&cnx->lock); |
| 832 | } |
| 833 | |
| 834 | spin_unlock_irqrestore(&port->pending_gate, flags); |
| 835 | } |
| 836 | |
| 837 | static struct net_device * __init veth_probe_one(int vlan, struct device *vdev) |
| 838 | { |
| 839 | struct net_device *dev; |
| 840 | struct veth_port *port; |
| 841 | int i, rc; |
| 842 | |
| 843 | dev = alloc_etherdev(sizeof (struct veth_port)); |
| 844 | if (! dev) { |
| 845 | veth_error("Unable to allocate net_device structure!\n"); |
| 846 | return NULL; |
| 847 | } |
| 848 | |
| 849 | port = (struct veth_port *) dev->priv; |
| 850 | |
| 851 | spin_lock_init(&port->pending_gate); |
| 852 | rwlock_init(&port->mcast_gate); |
| 853 | |
| 854 | for (i = 0; i < HVMAXARCHITECTEDLPS; i++) { |
| 855 | HvLpVirtualLanIndexMap map; |
| 856 | |
| 857 | if (i == this_lp) |
| 858 | continue; |
| 859 | map = HvLpConfig_getVirtualLanIndexMapForLp(i); |
| 860 | if (map & (0x8000 >> vlan)) |
| 861 | port->lpar_map |= (1 << i); |
| 862 | } |
| 863 | port->dev = vdev; |
| 864 | |
| 865 | dev->dev_addr[0] = 0x02; |
| 866 | dev->dev_addr[1] = 0x01; |
| 867 | dev->dev_addr[2] = 0xff; |
| 868 | dev->dev_addr[3] = vlan; |
| 869 | dev->dev_addr[4] = 0xff; |
| 870 | dev->dev_addr[5] = this_lp; |
| 871 | |
| 872 | dev->mtu = VETH_MAX_MTU; |
| 873 | |
| 874 | memcpy(&port->mac_addr, dev->dev_addr, 6); |
| 875 | |
| 876 | dev->open = veth_open; |
| 877 | dev->hard_start_xmit = veth_start_xmit; |
| 878 | dev->stop = veth_close; |
| 879 | dev->get_stats = veth_get_stats; |
| 880 | dev->change_mtu = veth_change_mtu; |
| 881 | dev->set_mac_address = NULL; |
| 882 | dev->set_multicast_list = veth_set_multicast_list; |
| 883 | SET_ETHTOOL_OPS(dev, &ops); |
| 884 | |
| 885 | dev->watchdog_timeo = 2 * (VETH_ACKTIMEOUT * HZ / 1000000); |
| 886 | dev->tx_timeout = veth_tx_timeout; |
| 887 | |
| 888 | SET_NETDEV_DEV(dev, vdev); |
| 889 | |
| 890 | rc = register_netdev(dev); |
| 891 | if (rc != 0) { |
| 892 | veth_printk(KERN_ERR, |
| 893 | "Failed to register ethernet device for vlan %d\n", |
| 894 | vlan); |
| 895 | free_netdev(dev); |
| 896 | return NULL; |
| 897 | } |
| 898 | |
| 899 | veth_printk(KERN_DEBUG, "%s attached to iSeries vlan %d (lpar_map=0x%04x)\n", |
| 900 | dev->name, vlan, port->lpar_map); |
| 901 | |
| 902 | return dev; |
| 903 | } |
| 904 | |
| 905 | /* |
| 906 | * Tx path |
| 907 | */ |
| 908 | |
| 909 | static int veth_transmit_to_one(struct sk_buff *skb, HvLpIndex rlp, |
| 910 | struct net_device *dev) |
| 911 | { |
| 912 | struct veth_lpar_connection *cnx = veth_cnx[rlp]; |
| 913 | struct veth_port *port = (struct veth_port *) dev->priv; |
| 914 | HvLpEvent_Rc rc; |
| 915 | u32 dma_address, dma_length; |
| 916 | struct veth_msg *msg = NULL; |
| 917 | int err = 0; |
| 918 | unsigned long flags; |
| 919 | |
| 920 | if (! cnx) { |
| 921 | port->stats.tx_errors++; |
| 922 | dev_kfree_skb(skb); |
| 923 | return 0; |
| 924 | } |
| 925 | |
| 926 | spin_lock_irqsave(&cnx->lock, flags); |
| 927 | |
Michael Ellerman | f27eff1 | 2005-05-12 17:47:27 +1000 | [diff] [blame] | 928 | if (! (cnx->state & VETH_STATE_READY)) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 929 | goto drop; |
| 930 | |
| 931 | if ((skb->len - 14) > VETH_MAX_MTU) |
| 932 | goto drop; |
| 933 | |
| 934 | msg = veth_stack_pop(cnx); |
| 935 | |
| 936 | if (! msg) { |
| 937 | err = 1; |
| 938 | goto drop; |
| 939 | } |
| 940 | |
| 941 | dma_length = skb->len; |
| 942 | dma_address = dma_map_single(port->dev, skb->data, |
| 943 | dma_length, DMA_TO_DEVICE); |
| 944 | |
| 945 | if (dma_mapping_error(dma_address)) |
| 946 | goto recycle_and_drop; |
| 947 | |
| 948 | /* Is it really necessary to check the length and address |
| 949 | * fields of the first entry here? */ |
| 950 | msg->skb = skb; |
| 951 | msg->dev = port->dev; |
| 952 | msg->data.addr[0] = dma_address; |
| 953 | msg->data.len[0] = dma_length; |
| 954 | msg->data.eofmask = 1 << VETH_EOF_SHIFT; |
| 955 | set_bit(0, &(msg->in_use)); |
| 956 | rc = veth_signaldata(cnx, VethEventTypeFrames, msg->token, &msg->data); |
| 957 | |
| 958 | if (rc != HvLpEvent_Rc_Good) |
| 959 | goto recycle_and_drop; |
| 960 | |
| 961 | spin_unlock_irqrestore(&cnx->lock, flags); |
| 962 | return 0; |
| 963 | |
| 964 | recycle_and_drop: |
| 965 | msg->skb = NULL; |
| 966 | /* need to set in use to make veth_recycle_msg in case this |
| 967 | * was a mapping failure */ |
| 968 | set_bit(0, &msg->in_use); |
| 969 | veth_recycle_msg(cnx, msg); |
| 970 | drop: |
| 971 | port->stats.tx_errors++; |
| 972 | dev_kfree_skb(skb); |
| 973 | spin_unlock_irqrestore(&cnx->lock, flags); |
| 974 | return err; |
| 975 | } |
| 976 | |
| 977 | static HvLpIndexMap veth_transmit_to_many(struct sk_buff *skb, |
| 978 | HvLpIndexMap lpmask, |
| 979 | struct net_device *dev) |
| 980 | { |
| 981 | struct veth_port *port = (struct veth_port *) dev->priv; |
| 982 | int i; |
| 983 | int rc; |
| 984 | |
| 985 | for (i = 0; i < HVMAXARCHITECTEDLPS; i++) { |
| 986 | if ((lpmask & (1 << i)) == 0) |
| 987 | continue; |
| 988 | |
| 989 | rc = veth_transmit_to_one(skb_get(skb), i, dev); |
| 990 | if (! rc) |
| 991 | lpmask &= ~(1<<i); |
| 992 | } |
| 993 | |
| 994 | if (! lpmask) { |
| 995 | port->stats.tx_packets++; |
| 996 | port->stats.tx_bytes += skb->len; |
| 997 | } |
| 998 | |
| 999 | return lpmask; |
| 1000 | } |
| 1001 | |
| 1002 | static int veth_start_xmit(struct sk_buff *skb, struct net_device *dev) |
| 1003 | { |
| 1004 | unsigned char *frame = skb->data; |
| 1005 | struct veth_port *port = (struct veth_port *) dev->priv; |
| 1006 | unsigned long flags; |
| 1007 | HvLpIndexMap lpmask; |
| 1008 | |
| 1009 | if (! (frame[0] & 0x01)) { |
| 1010 | /* unicast packet */ |
| 1011 | HvLpIndex rlp = frame[5]; |
| 1012 | |
| 1013 | if ( ! ((1 << rlp) & port->lpar_map) ) { |
| 1014 | dev_kfree_skb(skb); |
| 1015 | return 0; |
| 1016 | } |
| 1017 | |
| 1018 | lpmask = 1 << rlp; |
| 1019 | } else { |
| 1020 | lpmask = port->lpar_map; |
| 1021 | } |
| 1022 | |
| 1023 | spin_lock_irqsave(&port->pending_gate, flags); |
| 1024 | |
| 1025 | lpmask = veth_transmit_to_many(skb, lpmask, dev); |
| 1026 | |
Michael Ellerman | eb235ae | 2005-05-12 17:53:18 +1000 | [diff] [blame] | 1027 | dev->trans_start = jiffies; |
| 1028 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1029 | if (! lpmask) { |
| 1030 | dev_kfree_skb(skb); |
| 1031 | } else { |
| 1032 | if (port->pending_skb) { |
| 1033 | veth_error("%s: Tx while skb was pending!\n", |
| 1034 | dev->name); |
| 1035 | dev_kfree_skb(skb); |
| 1036 | spin_unlock_irqrestore(&port->pending_gate, flags); |
| 1037 | return 1; |
| 1038 | } |
| 1039 | |
| 1040 | port->pending_skb = skb; |
| 1041 | port->pending_lpmask = lpmask; |
| 1042 | netif_stop_queue(dev); |
| 1043 | } |
| 1044 | |
| 1045 | spin_unlock_irqrestore(&port->pending_gate, flags); |
| 1046 | |
| 1047 | return 0; |
| 1048 | } |
| 1049 | |
| 1050 | static void veth_recycle_msg(struct veth_lpar_connection *cnx, |
| 1051 | struct veth_msg *msg) |
| 1052 | { |
| 1053 | u32 dma_address, dma_length; |
| 1054 | |
| 1055 | if (test_and_clear_bit(0, &msg->in_use)) { |
| 1056 | dma_address = msg->data.addr[0]; |
| 1057 | dma_length = msg->data.len[0]; |
| 1058 | |
| 1059 | dma_unmap_single(msg->dev, dma_address, dma_length, |
| 1060 | DMA_TO_DEVICE); |
| 1061 | |
| 1062 | if (msg->skb) { |
| 1063 | dev_kfree_skb_any(msg->skb); |
| 1064 | msg->skb = NULL; |
| 1065 | } |
| 1066 | |
| 1067 | memset(&msg->data, 0, sizeof(msg->data)); |
| 1068 | veth_stack_push(cnx, msg); |
| 1069 | } else |
| 1070 | if (cnx->state & VETH_STATE_OPEN) |
| 1071 | veth_error("Bogus frames ack from lpar %d (#%d)\n", |
| 1072 | cnx->remote_lp, msg->token); |
| 1073 | } |
| 1074 | |
| 1075 | static void veth_flush_pending(struct veth_lpar_connection *cnx) |
| 1076 | { |
| 1077 | int i; |
| 1078 | for (i = 0; i < HVMAXARCHITECTEDVIRTUALLANS; i++) { |
| 1079 | struct net_device *dev = veth_dev[i]; |
| 1080 | struct veth_port *port; |
| 1081 | unsigned long flags; |
| 1082 | |
| 1083 | if (! dev) |
| 1084 | continue; |
| 1085 | |
| 1086 | port = (struct veth_port *)dev->priv; |
| 1087 | |
| 1088 | if (! (port->lpar_map & (1<<cnx->remote_lp))) |
| 1089 | continue; |
| 1090 | |
| 1091 | spin_lock_irqsave(&port->pending_gate, flags); |
| 1092 | if (port->pending_skb) { |
| 1093 | port->pending_lpmask = |
| 1094 | veth_transmit_to_many(port->pending_skb, |
| 1095 | port->pending_lpmask, |
| 1096 | dev); |
| 1097 | if (! port->pending_lpmask) { |
| 1098 | dev_kfree_skb_any(port->pending_skb); |
| 1099 | port->pending_skb = NULL; |
| 1100 | netif_wake_queue(dev); |
| 1101 | } |
| 1102 | } |
| 1103 | spin_unlock_irqrestore(&port->pending_gate, flags); |
| 1104 | } |
| 1105 | } |
| 1106 | |
| 1107 | /* |
| 1108 | * Rx path |
| 1109 | */ |
| 1110 | |
| 1111 | static inline int veth_frame_wanted(struct veth_port *port, u64 mac_addr) |
| 1112 | { |
| 1113 | int wanted = 0; |
| 1114 | int i; |
| 1115 | unsigned long flags; |
| 1116 | |
| 1117 | if ( (mac_addr == port->mac_addr) || (mac_addr == 0xffffffffffff0000) ) |
| 1118 | return 1; |
| 1119 | |
| 1120 | if (! (((char *) &mac_addr)[0] & 0x01)) |
| 1121 | return 0; |
| 1122 | |
| 1123 | read_lock_irqsave(&port->mcast_gate, flags); |
| 1124 | |
| 1125 | if (port->promiscuous || port->all_mcast) { |
| 1126 | wanted = 1; |
| 1127 | goto out; |
| 1128 | } |
| 1129 | |
| 1130 | for (i = 0; i < port->num_mcast; ++i) { |
| 1131 | if (port->mcast_addr[i] == mac_addr) { |
| 1132 | wanted = 1; |
| 1133 | break; |
| 1134 | } |
| 1135 | } |
| 1136 | |
| 1137 | out: |
| 1138 | read_unlock_irqrestore(&port->mcast_gate, flags); |
| 1139 | |
| 1140 | return wanted; |
| 1141 | } |
| 1142 | |
| 1143 | struct dma_chunk { |
| 1144 | u64 addr; |
| 1145 | u64 size; |
| 1146 | }; |
| 1147 | |
| 1148 | #define VETH_MAX_PAGES_PER_FRAME ( (VETH_MAX_MTU+PAGE_SIZE-2)/PAGE_SIZE + 1 ) |
| 1149 | |
| 1150 | static inline void veth_build_dma_list(struct dma_chunk *list, |
| 1151 | unsigned char *p, unsigned long length) |
| 1152 | { |
| 1153 | unsigned long done; |
| 1154 | int i = 1; |
| 1155 | |
| 1156 | /* FIXME: skbs are continguous in real addresses. Do we |
| 1157 | * really need to break it into PAGE_SIZE chunks, or can we do |
| 1158 | * it just at the granularity of iSeries real->absolute |
| 1159 | * mapping? Indeed, given the way the allocator works, can we |
| 1160 | * count on them being absolutely contiguous? */ |
| 1161 | list[0].addr = ISERIES_HV_ADDR(p); |
| 1162 | list[0].size = min(length, |
| 1163 | PAGE_SIZE - ((unsigned long)p & ~PAGE_MASK)); |
| 1164 | |
| 1165 | done = list[0].size; |
| 1166 | while (done < length) { |
| 1167 | list[i].addr = ISERIES_HV_ADDR(p + done); |
| 1168 | list[i].size = min(length-done, PAGE_SIZE); |
| 1169 | done += list[i].size; |
| 1170 | i++; |
| 1171 | } |
| 1172 | } |
| 1173 | |
| 1174 | static void veth_flush_acks(struct veth_lpar_connection *cnx) |
| 1175 | { |
| 1176 | HvLpEvent_Rc rc; |
| 1177 | |
| 1178 | rc = veth_signaldata(cnx, VethEventTypeFramesAck, |
| 1179 | 0, &cnx->pending_acks); |
| 1180 | |
| 1181 | if (rc != HvLpEvent_Rc_Good) |
| 1182 | veth_error("Error 0x%x acking frames from lpar %d!\n", |
| 1183 | (unsigned)rc, cnx->remote_lp); |
| 1184 | |
| 1185 | cnx->num_pending_acks = 0; |
| 1186 | memset(&cnx->pending_acks, 0xff, sizeof(cnx->pending_acks)); |
| 1187 | } |
| 1188 | |
| 1189 | static void veth_receive(struct veth_lpar_connection *cnx, |
| 1190 | struct VethLpEvent *event) |
| 1191 | { |
| 1192 | struct VethFramesData *senddata = &event->u.frames_data; |
| 1193 | int startchunk = 0; |
| 1194 | int nchunks; |
| 1195 | unsigned long flags; |
| 1196 | HvLpDma_Rc rc; |
| 1197 | |
| 1198 | do { |
| 1199 | u16 length = 0; |
| 1200 | struct sk_buff *skb; |
| 1201 | struct dma_chunk local_list[VETH_MAX_PAGES_PER_FRAME]; |
| 1202 | struct dma_chunk remote_list[VETH_MAX_FRAMES_PER_MSG]; |
| 1203 | u64 dest; |
| 1204 | HvLpVirtualLanIndex vlan; |
| 1205 | struct net_device *dev; |
| 1206 | struct veth_port *port; |
| 1207 | |
| 1208 | /* FIXME: do we need this? */ |
| 1209 | memset(local_list, 0, sizeof(local_list)); |
| 1210 | memset(remote_list, 0, sizeof(VETH_MAX_FRAMES_PER_MSG)); |
| 1211 | |
| 1212 | /* a 0 address marks the end of the valid entries */ |
| 1213 | if (senddata->addr[startchunk] == 0) |
| 1214 | break; |
| 1215 | |
| 1216 | /* make sure that we have at least 1 EOF entry in the |
| 1217 | * remaining entries */ |
| 1218 | if (! (senddata->eofmask >> (startchunk + VETH_EOF_SHIFT))) { |
| 1219 | veth_error("missing EOF frag in event " |
| 1220 | "eofmask=0x%x startchunk=%d\n", |
| 1221 | (unsigned) senddata->eofmask, startchunk); |
| 1222 | break; |
| 1223 | } |
| 1224 | |
| 1225 | /* build list of chunks in this frame */ |
| 1226 | nchunks = 0; |
| 1227 | do { |
| 1228 | remote_list[nchunks].addr = |
| 1229 | (u64) senddata->addr[startchunk+nchunks] << 32; |
| 1230 | remote_list[nchunks].size = |
| 1231 | senddata->len[startchunk+nchunks]; |
| 1232 | length += remote_list[nchunks].size; |
| 1233 | } while (! (senddata->eofmask & |
| 1234 | (1 << (VETH_EOF_SHIFT + startchunk + nchunks++)))); |
| 1235 | |
| 1236 | /* length == total length of all chunks */ |
| 1237 | /* nchunks == # of chunks in this frame */ |
| 1238 | |
| 1239 | if ((length - ETH_HLEN) > VETH_MAX_MTU) { |
| 1240 | veth_error("Received oversize frame from lpar %d " |
| 1241 | "(length=%d)\n", cnx->remote_lp, length); |
| 1242 | continue; |
| 1243 | } |
| 1244 | |
| 1245 | skb = alloc_skb(length, GFP_ATOMIC); |
| 1246 | if (!skb) |
| 1247 | continue; |
| 1248 | |
| 1249 | veth_build_dma_list(local_list, skb->data, length); |
| 1250 | |
| 1251 | rc = HvCallEvent_dmaBufList(HvLpEvent_Type_VirtualLan, |
| 1252 | event->base_event.xSourceLp, |
| 1253 | HvLpDma_Direction_RemoteToLocal, |
| 1254 | cnx->src_inst, |
| 1255 | cnx->dst_inst, |
| 1256 | HvLpDma_AddressType_RealAddress, |
| 1257 | HvLpDma_AddressType_TceIndex, |
| 1258 | ISERIES_HV_ADDR(&local_list), |
| 1259 | ISERIES_HV_ADDR(&remote_list), |
| 1260 | length); |
| 1261 | if (rc != HvLpDma_Rc_Good) { |
| 1262 | dev_kfree_skb_irq(skb); |
| 1263 | continue; |
| 1264 | } |
| 1265 | |
| 1266 | vlan = skb->data[9]; |
| 1267 | dev = veth_dev[vlan]; |
Michael Ellerman | 41664c0 | 2005-05-12 17:55:08 +1000 | [diff] [blame] | 1268 | if (! dev) { |
| 1269 | /* |
| 1270 | * Some earlier versions of the driver sent |
| 1271 | * broadcasts down all connections, even to lpars |
| 1272 | * that weren't on the relevant vlan. So ignore |
| 1273 | * packets belonging to a vlan we're not on. |
| 1274 | * We can also be here if we receive packets while |
| 1275 | * the driver is going down, because then dev is NULL. |
| 1276 | */ |
| 1277 | dev_kfree_skb_irq(skb); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1278 | continue; |
Michael Ellerman | 41664c0 | 2005-05-12 17:55:08 +1000 | [diff] [blame] | 1279 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1280 | |
| 1281 | port = (struct veth_port *)dev->priv; |
| 1282 | dest = *((u64 *) skb->data) & 0xFFFFFFFFFFFF0000; |
| 1283 | |
| 1284 | if ((vlan > HVMAXARCHITECTEDVIRTUALLANS) || !port) { |
| 1285 | dev_kfree_skb_irq(skb); |
| 1286 | continue; |
| 1287 | } |
| 1288 | if (! veth_frame_wanted(port, dest)) { |
| 1289 | dev_kfree_skb_irq(skb); |
| 1290 | continue; |
| 1291 | } |
| 1292 | |
| 1293 | skb_put(skb, length); |
| 1294 | skb->dev = dev; |
| 1295 | skb->protocol = eth_type_trans(skb, dev); |
| 1296 | skb->ip_summed = CHECKSUM_NONE; |
| 1297 | netif_rx(skb); /* send it up */ |
| 1298 | port->stats.rx_packets++; |
| 1299 | port->stats.rx_bytes += length; |
| 1300 | } while (startchunk += nchunks, startchunk < VETH_MAX_FRAMES_PER_MSG); |
| 1301 | |
| 1302 | /* Ack it */ |
| 1303 | spin_lock_irqsave(&cnx->lock, flags); |
| 1304 | BUG_ON(cnx->num_pending_acks > VETH_MAX_ACKS_PER_MSG); |
| 1305 | |
| 1306 | cnx->pending_acks[cnx->num_pending_acks++] = |
| 1307 | event->base_event.xCorrelationToken; |
| 1308 | |
| 1309 | if ( (cnx->num_pending_acks >= cnx->remote_caps.ack_threshold) |
| 1310 | || (cnx->num_pending_acks >= VETH_MAX_ACKS_PER_MSG) ) |
| 1311 | veth_flush_acks(cnx); |
| 1312 | |
| 1313 | spin_unlock_irqrestore(&cnx->lock, flags); |
| 1314 | } |
| 1315 | |
| 1316 | static void veth_timed_ack(unsigned long ptr) |
| 1317 | { |
| 1318 | struct veth_lpar_connection *cnx = (struct veth_lpar_connection *) ptr; |
| 1319 | unsigned long flags; |
| 1320 | |
| 1321 | /* Ack all the events */ |
| 1322 | spin_lock_irqsave(&cnx->lock, flags); |
| 1323 | if (cnx->num_pending_acks > 0) |
| 1324 | veth_flush_acks(cnx); |
| 1325 | |
| 1326 | /* Reschedule the timer */ |
| 1327 | cnx->ack_timer.expires = jiffies + cnx->ack_timeout; |
| 1328 | add_timer(&cnx->ack_timer); |
| 1329 | spin_unlock_irqrestore(&cnx->lock, flags); |
| 1330 | } |
| 1331 | |
| 1332 | static int veth_remove(struct vio_dev *vdev) |
| 1333 | { |
| 1334 | int i = vdev->unit_address; |
| 1335 | struct net_device *dev; |
| 1336 | |
| 1337 | dev = veth_dev[i]; |
| 1338 | if (dev != NULL) { |
| 1339 | veth_dev[i] = NULL; |
| 1340 | unregister_netdev(dev); |
| 1341 | free_netdev(dev); |
| 1342 | } |
| 1343 | return 0; |
| 1344 | } |
| 1345 | |
| 1346 | static int veth_probe(struct vio_dev *vdev, const struct vio_device_id *id) |
| 1347 | { |
| 1348 | int i = vdev->unit_address; |
| 1349 | struct net_device *dev; |
| 1350 | |
| 1351 | dev = veth_probe_one(i, &vdev->dev); |
| 1352 | if (dev == NULL) { |
| 1353 | veth_remove(vdev); |
| 1354 | return 1; |
| 1355 | } |
| 1356 | veth_dev[i] = dev; |
| 1357 | |
| 1358 | /* Start the state machine on each connection, to commence |
| 1359 | * link negotiation */ |
| 1360 | for (i = 0; i < HVMAXARCHITECTEDLPS; i++) |
| 1361 | if (veth_cnx[i]) |
| 1362 | veth_kick_statemachine(veth_cnx[i]); |
| 1363 | |
| 1364 | return 0; |
| 1365 | } |
| 1366 | |
| 1367 | /** |
| 1368 | * veth_device_table: Used by vio.c to match devices that we |
| 1369 | * support. |
| 1370 | */ |
| 1371 | static struct vio_device_id veth_device_table[] __devinitdata = { |
| 1372 | { "vlan", "" }, |
| 1373 | { NULL, NULL } |
| 1374 | }; |
| 1375 | MODULE_DEVICE_TABLE(vio, veth_device_table); |
| 1376 | |
| 1377 | static struct vio_driver veth_driver = { |
| 1378 | .name = "iseries_veth", |
| 1379 | .id_table = veth_device_table, |
| 1380 | .probe = veth_probe, |
| 1381 | .remove = veth_remove |
| 1382 | }; |
| 1383 | |
| 1384 | /* |
| 1385 | * Module initialization/cleanup |
| 1386 | */ |
| 1387 | |
| 1388 | void __exit veth_module_cleanup(void) |
| 1389 | { |
| 1390 | int i; |
| 1391 | |
Michael Ellerman | b2e0852 | 2005-05-12 18:09:45 +1000 | [diff] [blame] | 1392 | /* Stop the queues first to stop any new packets being sent. */ |
| 1393 | for (i = 0; i < HVMAXARCHITECTEDVIRTUALLANS; i++) |
| 1394 | if (veth_dev[i]) |
| 1395 | netif_stop_queue(veth_dev[i]); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1396 | |
Michael Ellerman | b2e0852 | 2005-05-12 18:09:45 +1000 | [diff] [blame] | 1397 | /* Stop the connections before we unregister the driver. This |
| 1398 | * ensures there's no skbs lying around holding the device open. */ |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1399 | for (i = 0; i < HVMAXARCHITECTEDLPS; ++i) |
| 1400 | veth_stop_connection(i); |
| 1401 | |
| 1402 | HvLpEvent_unregisterHandler(HvLpEvent_Type_VirtualLan); |
| 1403 | |
| 1404 | /* Hypervisor callbacks may have scheduled more work while we |
Michael Ellerman | b2e0852 | 2005-05-12 18:09:45 +1000 | [diff] [blame] | 1405 | * were stoping connections. Now that we've disconnected from |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1406 | * the hypervisor make sure everything's finished. */ |
| 1407 | flush_scheduled_work(); |
| 1408 | |
Michael Ellerman | b2e0852 | 2005-05-12 18:09:45 +1000 | [diff] [blame] | 1409 | vio_unregister_driver(&veth_driver); |
| 1410 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1411 | for (i = 0; i < HVMAXARCHITECTEDLPS; ++i) |
| 1412 | veth_destroy_connection(i); |
| 1413 | |
| 1414 | } |
| 1415 | module_exit(veth_module_cleanup); |
| 1416 | |
| 1417 | int __init veth_module_init(void) |
| 1418 | { |
| 1419 | int i; |
| 1420 | int rc; |
| 1421 | |
| 1422 | this_lp = HvLpConfig_getLpIndex_outline(); |
| 1423 | |
| 1424 | for (i = 0; i < HVMAXARCHITECTEDLPS; ++i) { |
| 1425 | rc = veth_init_connection(i); |
| 1426 | if (rc != 0) { |
| 1427 | veth_module_cleanup(); |
| 1428 | return rc; |
| 1429 | } |
| 1430 | } |
| 1431 | |
| 1432 | HvLpEvent_registerHandler(HvLpEvent_Type_VirtualLan, |
| 1433 | &veth_handle_event); |
| 1434 | |
| 1435 | return vio_register_driver(&veth_driver); |
| 1436 | } |
| 1437 | module_init(veth_module_init); |