projects / deliverable / linux.git / blob
? search:


966c252c3ca9b4b1e7019063ef68c62e5a12b6f5
[deliverable/linux.git] / drivers / net / ethernet / sun / sunvnet.c
1 /* sunvnet.c: Sun LDOM Virtual Network Driver.
2 *
3 * Copyright (C) 2007, 2008 David S. Miller <davem@davemloft.net>
4 */
5
6 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
7
8 #include <linux/module.h>
9 #include <linux/kernel.h>
10 #include <linux/types.h>
11 #include <linux/slab.h>
12 #include <linux/delay.h>
13 #include <linux/init.h>
14 #include <linux/netdevice.h>
15 #include <linux/ethtool.h>
16 #include <linux/etherdevice.h>
17 #include <linux/mutex.h>
18 #include <linux/if_vlan.h>
19
20 #if IS_ENABLED(CONFIG_IPV6)
21 #include <linux/icmpv6.h>
22 #endif
23
24 #include <net/icmp.h>
25 #include <net/route.h>
26
27 #include <asm/vio.h>
28 #include <asm/ldc.h>
29
30 #include "sunvnet.h"
31
32 #define DRV_MODULE_NAME "sunvnet"
33 #define DRV_MODULE_VERSION "1.0"
34 #define DRV_MODULE_RELDATE "June 25, 2007"
35
36 static char version[] =
37 DRV_MODULE_NAME ".c:v" DRV_MODULE_VERSION " (" DRV_MODULE_RELDATE ")\n";
38 MODULE_AUTHOR("David S. Miller (davem@davemloft.net)");
39 MODULE_DESCRIPTION("Sun LDOM virtual network driver");
40 MODULE_LICENSE("GPL");
41 MODULE_VERSION(DRV_MODULE_VERSION);
42
43 /* Heuristic for the number of times to exponentially backoff and
44 * retry sending an LDC trigger when EAGAIN is encountered
45 */
46 #define VNET_MAX_RETRIES 10
47
48 static int __vnet_tx_trigger(struct vnet_port *port, u32 start);
49
50 /* Ordered from largest major to lowest */
51 static struct vio_version vnet_versions[] = {
52 { .major = 1, .minor = 6 },
53 { .major = 1, .minor = 0 },
54 };
55
56 static inline u32 vnet_tx_dring_avail(struct vio_dring_state *dr)
57 {
58 return vio_dring_avail(dr, VNET_TX_RING_SIZE);
59 }
60
61 static int vnet_handle_unknown(struct vnet_port *port, void *arg)
62 {
63 struct vio_msg_tag *pkt = arg;
64
65 pr_err("Received unknown msg [%02x:%02x:%04x:%08x]\n",
66 pkt->type, pkt->stype, pkt->stype_env, pkt->sid);
67 pr_err("Resetting connection\n");
68
69 ldc_disconnect(port->vio.lp);
70
71 return -ECONNRESET;
72 }
73
74 static int vnet_send_attr(struct vio_driver_state *vio)
75 {
76 struct vnet_port *port = to_vnet_port(vio);
77 struct net_device *dev = port->vp->dev;
78 struct vio_net_attr_info pkt;
79 int framelen = ETH_FRAME_LEN;
80 int i;
81
82 memset(&pkt, 0, sizeof(pkt));
83 pkt.tag.type = VIO_TYPE_CTRL;
84 pkt.tag.stype = VIO_SUBTYPE_INFO;
85 pkt.tag.stype_env = VIO_ATTR_INFO;
86 pkt.tag.sid = vio_send_sid(vio);
87 if (vio_version_before(vio, 1, 2))
88 pkt.xfer_mode = VIO_DRING_MODE;
89 else
90 pkt.xfer_mode = VIO_NEW_DRING_MODE;
91 pkt.addr_type = VNET_ADDR_ETHERMAC;
92 pkt.ack_freq = 0;
93 for (i = 0; i < 6; i++)
94 pkt.addr |= (u64)dev->dev_addr[i] << ((5 - i) * 8);
95 if (vio_version_after(vio, 1, 3)) {
96 if (port->rmtu) {
97 port->rmtu = min(VNET_MAXPACKET, port->rmtu);
98 pkt.mtu = port->rmtu;
99 } else {
100 port->rmtu = VNET_MAXPACKET;
101 pkt.mtu = port->rmtu;
102 }
103 if (vio_version_after_eq(vio, 1, 6))
104 pkt.options = VIO_TX_DRING;
105 } else if (vio_version_before(vio, 1, 3)) {
106 pkt.mtu = framelen;
107 } else { /* v1.3 */
108 pkt.mtu = framelen + VLAN_HLEN;
109 }
110
111 pkt.plnk_updt = PHYSLINK_UPDATE_NONE;
112 pkt.cflags = 0;
113
114 viodbg(HS, "SEND NET ATTR xmode[0x%x] atype[0x%x] addr[%llx] "
115 "ackfreq[%u] plnk_updt[0x%02x] opts[0x%02x] mtu[%llu] "
116 "cflags[0x%04x] lso_max[%u]\n",
117 pkt.xfer_mode, pkt.addr_type,
118 (unsigned long long)pkt.addr,
119 pkt.ack_freq, pkt.plnk_updt, pkt.options,
120 (unsigned long long)pkt.mtu, pkt.cflags, pkt.ipv4_lso_maxlen);
121
122
123 return vio_ldc_send(vio, &pkt, sizeof(pkt));
124 }
125
126 static int handle_attr_info(struct vio_driver_state *vio,
127 struct vio_net_attr_info *pkt)
128 {
129 struct vnet_port *port = to_vnet_port(vio);
130 u64 localmtu;
131 u8 xfer_mode;
132
133 viodbg(HS, "GOT NET ATTR xmode[0x%x] atype[0x%x] addr[%llx] "
134 "ackfreq[%u] plnk_updt[0x%02x] opts[0x%02x] mtu[%llu] "
135 " (rmtu[%llu]) cflags[0x%04x] lso_max[%u]\n",
136 pkt->xfer_mode, pkt->addr_type,
137 (unsigned long long)pkt->addr,
138 pkt->ack_freq, pkt->plnk_updt, pkt->options,
139 (unsigned long long)pkt->mtu, port->rmtu, pkt->cflags,
140 pkt->ipv4_lso_maxlen);
141
142 pkt->tag.sid = vio_send_sid(vio);
143
144 xfer_mode = pkt->xfer_mode;
145 /* for version < 1.2, VIO_DRING_MODE = 0x3 and no bitmask */
146 if (vio_version_before(vio, 1, 2) && xfer_mode == VIO_DRING_MODE)
147 xfer_mode = VIO_NEW_DRING_MODE;
148
149 /* MTU negotiation:
150 * < v1.3 - ETH_FRAME_LEN exactly
151 * > v1.3 - MIN(pkt.mtu, VNET_MAXPACKET, port->rmtu) and change
152 * pkt->mtu for ACK
153 * = v1.3 - ETH_FRAME_LEN + VLAN_HLEN exactly
154 */
155 if (vio_version_before(vio, 1, 3)) {
156 localmtu = ETH_FRAME_LEN;
157 } else if (vio_version_after(vio, 1, 3)) {
158 localmtu = port->rmtu ? port->rmtu : VNET_MAXPACKET;
159 localmtu = min(pkt->mtu, localmtu);
160 pkt->mtu = localmtu;
161 } else { /* v1.3 */
162 localmtu = ETH_FRAME_LEN + VLAN_HLEN;
163 }
164 port->rmtu = localmtu;
165
166 /* for version >= 1.6, ACK packet mode we support */
167 if (vio_version_after_eq(vio, 1, 6)) {
168 pkt->xfer_mode = VIO_NEW_DRING_MODE;
169 pkt->options = VIO_TX_DRING;
170 }
171
172 if (!(xfer_mode | VIO_NEW_DRING_MODE) ||
173 pkt->addr_type != VNET_ADDR_ETHERMAC ||
174 pkt->mtu != localmtu) {
175 viodbg(HS, "SEND NET ATTR NACK\n");
176
177 pkt->tag.stype = VIO_SUBTYPE_NACK;
178
179 (void) vio_ldc_send(vio, pkt, sizeof(*pkt));
180
181 return -ECONNRESET;
182 } else {
183 viodbg(HS, "SEND NET ATTR ACK xmode[0x%x] atype[0x%x] "
184 "addr[%llx] ackfreq[%u] plnk_updt[0x%02x] opts[0x%02x] "
185 "mtu[%llu] (rmtu[%llu]) cflags[0x%04x] lso_max[%u]\n",
186 pkt->xfer_mode, pkt->addr_type,
187 (unsigned long long)pkt->addr,
188 pkt->ack_freq, pkt->plnk_updt, pkt->options,
189 (unsigned long long)pkt->mtu, port->rmtu, pkt->cflags,
190 pkt->ipv4_lso_maxlen);
191
192 pkt->tag.stype = VIO_SUBTYPE_ACK;
193
194 return vio_ldc_send(vio, pkt, sizeof(*pkt));
195 }
196
197 }
198
199 static int handle_attr_ack(struct vio_driver_state *vio,
200 struct vio_net_attr_info *pkt)
201 {
202 viodbg(HS, "GOT NET ATTR ACK\n");
203
204 return 0;
205 }
206
207 static int handle_attr_nack(struct vio_driver_state *vio,
208 struct vio_net_attr_info *pkt)
209 {
210 viodbg(HS, "GOT NET ATTR NACK\n");
211
212 return -ECONNRESET;
213 }
214
215 static int vnet_handle_attr(struct vio_driver_state *vio, void *arg)
216 {
217 struct vio_net_attr_info *pkt = arg;
218
219 switch (pkt->tag.stype) {
220 case VIO_SUBTYPE_INFO:
221 return handle_attr_info(vio, pkt);
222
223 case VIO_SUBTYPE_ACK:
224 return handle_attr_ack(vio, pkt);
225
226 case VIO_SUBTYPE_NACK:
227 return handle_attr_nack(vio, pkt);
228
229 default:
230 return -ECONNRESET;
231 }
232 }
233
234 static void vnet_handshake_complete(struct vio_driver_state *vio)
235 {
236 struct vio_dring_state *dr;
237
238 dr = &vio->drings[VIO_DRIVER_RX_RING];
239 dr->snd_nxt = dr->rcv_nxt = 1;
240
241 dr = &vio->drings[VIO_DRIVER_TX_RING];
242 dr->snd_nxt = dr->rcv_nxt = 1;
243 }
244
245 /* The hypervisor interface that implements copying to/from imported
246 * memory from another domain requires that copies are done to 8-byte
247 * aligned buffers, and that the lengths of such copies are also 8-byte
248 * multiples.
249 *
250 * So we align skb->data to an 8-byte multiple and pad-out the data
251 * area so we can round the copy length up to the next multiple of
252 * 8 for the copy.
253 *
254 * The transmitter puts the actual start of the packet 6 bytes into
255 * the buffer it sends over, so that the IP headers after the ethernet
256 * header are aligned properly. These 6 bytes are not in the descriptor
257 * length, they are simply implied. This offset is represented using
258 * the VNET_PACKET_SKIP macro.
259 */
260 static struct sk_buff *alloc_and_align_skb(struct net_device *dev,
261 unsigned int len)
262 {
263 struct sk_buff *skb = netdev_alloc_skb(dev, len+VNET_PACKET_SKIP+8+8);
264 unsigned long addr, off;
265
266 if (unlikely(!skb))
267 return NULL;
268
269 addr = (unsigned long) skb->data;
270 off = ((addr + 7UL) & ~7UL) - addr;
271 if (off)
272 skb_reserve(skb, off);
273
274 return skb;
275 }
276
277 static int vnet_rx_one(struct vnet_port *port, unsigned int len,
278 struct ldc_trans_cookie *cookies, int ncookies)
279 {
280 struct net_device *dev = port->vp->dev;
281 unsigned int copy_len;
282 struct sk_buff *skb;
283 int err;
284
285 err = -EMSGSIZE;
286 if (unlikely(len < ETH_ZLEN || len > port->rmtu)) {
287 dev->stats.rx_length_errors++;
288 goto out_dropped;
289 }
290
291 skb = alloc_and_align_skb(dev, len);
292 err = -ENOMEM;
293 if (unlikely(!skb)) {
294 dev->stats.rx_missed_errors++;
295 goto out_dropped;
296 }
297
298 copy_len = (len + VNET_PACKET_SKIP + 7U) & ~7U;
299 skb_put(skb, copy_len);
300 err = ldc_copy(port->vio.lp, LDC_COPY_IN,
301 skb->data, copy_len, 0,
302 cookies, ncookies);
303 if (unlikely(err < 0)) {
304 dev->stats.rx_frame_errors++;
305 goto out_free_skb;
306 }
307
308 skb_pull(skb, VNET_PACKET_SKIP);
309 skb_trim(skb, len);
310 skb->protocol = eth_type_trans(skb, dev);
311
312 dev->stats.rx_packets++;
313 dev->stats.rx_bytes += len;
314 napi_gro_receive(&port->napi, skb);
315 return 0;
316
317 out_free_skb:
318 kfree_skb(skb);
319
320 out_dropped:
321 dev->stats.rx_dropped++;
322 return err;
323 }
324
325 static int vnet_send_ack(struct vnet_port *port, struct vio_dring_state *dr,
326 u32 start, u32 end, u8 vio_dring_state)
327 {
328 struct vio_dring_data hdr = {
329 .tag = {
330 .type = VIO_TYPE_DATA,
331 .stype = VIO_SUBTYPE_ACK,
332 .stype_env = VIO_DRING_DATA,
333 .sid = vio_send_sid(&port->vio),
334 },
335 .dring_ident = dr->ident,
336 .start_idx = start,
337 .end_idx = end,
338 .state = vio_dring_state,
339 };
340 int err, delay;
341 int retries = 0;
342
343 hdr.seq = dr->snd_nxt;
344 delay = 1;
345 do {
346 err = vio_ldc_send(&port->vio, &hdr, sizeof(hdr));
347 if (err > 0) {
348 dr->snd_nxt++;
349 break;
350 }
351 udelay(delay);
352 if ((delay <<= 1) > 128)
353 delay = 128;
354 if (retries++ > VNET_MAX_RETRIES) {
355 pr_info("ECONNRESET %x:%x:%x:%x:%x:%x\n",
356 port->raddr[0], port->raddr[1],
357 port->raddr[2], port->raddr[3],
358 port->raddr[4], port->raddr[5]);
359 break;
360 }
361 } while (err == -EAGAIN);
362
363 if (err <= 0 && vio_dring_state == VIO_DRING_STOPPED) {
364 port->stop_rx_idx = end;
365 port->stop_rx = true;
366 } else {
367 port->stop_rx_idx = 0;
368 port->stop_rx = false;
369 }
370
371 return err;
372 }
373
374 static u32 next_idx(u32 idx, struct vio_dring_state *dr)
375 {
376 if (++idx == dr->num_entries)
377 idx = 0;
378 return idx;
379 }
380
381 static u32 prev_idx(u32 idx, struct vio_dring_state *dr)
382 {
383 if (idx == 0)
384 idx = dr->num_entries - 1;
385 else
386 idx--;
387
388 return idx;
389 }
390
391 static struct vio_net_desc *get_rx_desc(struct vnet_port *port,
392 struct vio_dring_state *dr,
393 u32 index)
394 {
395 struct vio_net_desc *desc = port->vio.desc_buf;
396 int err;
397
398 err = ldc_get_dring_entry(port->vio.lp, desc, dr->entry_size,
399 (index * dr->entry_size),
400 dr->cookies, dr->ncookies);
401 if (err < 0)
402 return ERR_PTR(err);
403
404 return desc;
405 }
406
407 static int put_rx_desc(struct vnet_port *port,
408 struct vio_dring_state *dr,
409 struct vio_net_desc *desc,
410 u32 index)
411 {
412 int err;
413
414 err = ldc_put_dring_entry(port->vio.lp, desc, dr->entry_size,
415 (index * dr->entry_size),
416 dr->cookies, dr->ncookies);
417 if (err < 0)
418 return err;
419
420 return 0;
421 }
422
423 static int vnet_walk_rx_one(struct vnet_port *port,
424 struct vio_dring_state *dr,
425 u32 index, int *needs_ack)
426 {
427 struct vio_net_desc *desc = get_rx_desc(port, dr, index);
428 struct vio_driver_state *vio = &port->vio;
429 int err;
430
431 BUG_ON(desc == NULL);
432 if (IS_ERR(desc))
433 return PTR_ERR(desc);
434
435 if (desc->hdr.state != VIO_DESC_READY)
436 return 1;
437
438 rmb();
439
440 viodbg(DATA, "vio_walk_rx_one desc[%02x:%02x:%08x:%08x:%llx:%llx]\n",
441 desc->hdr.state, desc->hdr.ack,
442 desc->size, desc->ncookies,
443 desc->cookies[0].cookie_addr,
444 desc->cookies[0].cookie_size);
445
446 err = vnet_rx_one(port, desc->size, desc->cookies, desc->ncookies);
447 if (err == -ECONNRESET)
448 return err;
449 desc->hdr.state = VIO_DESC_DONE;
450 err = put_rx_desc(port, dr, desc, index);
451 if (err < 0)
452 return err;
453 *needs_ack = desc->hdr.ack;
454 return 0;
455 }
456
457 static int vnet_walk_rx(struct vnet_port *port, struct vio_dring_state *dr,
458 u32 start, u32 end, int *npkts, int budget)
459 {
460 struct vio_driver_state *vio = &port->vio;
461 int ack_start = -1, ack_end = -1;
462 bool send_ack = true;
463
464 end = (end == (u32) -1) ? prev_idx(start, dr) : next_idx(end, dr);
465
466 viodbg(DATA, "vnet_walk_rx start[%08x] end[%08x]\n", start, end);
467
468 while (start != end) {
469 int ack = 0, err = vnet_walk_rx_one(port, dr, start, &ack);
470 if (err == -ECONNRESET)
471 return err;
472 if (err != 0)
473 break;
474 (*npkts)++;
475 if (ack_start == -1)
476 ack_start = start;
477 ack_end = start;
478 start = next_idx(start, dr);
479 if (ack && start != end) {
480 err = vnet_send_ack(port, dr, ack_start, ack_end,
481 VIO_DRING_ACTIVE);
482 if (err == -ECONNRESET)
483 return err;
484 ack_start = -1;
485 }
486 if ((*npkts) >= budget) {
487 send_ack = false;
488 break;
489 }
490 }
491 if (unlikely(ack_start == -1))
492 ack_start = ack_end = prev_idx(start, dr);
493 if (send_ack) {
494 port->napi_resume = false;
495 return vnet_send_ack(port, dr, ack_start, ack_end,
496 VIO_DRING_STOPPED);
497 } else {
498 port->napi_resume = true;
499 port->napi_stop_idx = ack_end;
500 return 1;
501 }
502 }
503
504 static int vnet_rx(struct vnet_port *port, void *msgbuf, int *npkts,
505 int budget)
506 {
507 struct vio_dring_data *pkt = msgbuf;
508 struct vio_dring_state *dr = &port->vio.drings[VIO_DRIVER_RX_RING];
509 struct vio_driver_state *vio = &port->vio;
510
511 viodbg(DATA, "vnet_rx stype_env[%04x] seq[%016llx] rcv_nxt[%016llx]\n",
512 pkt->tag.stype_env, pkt->seq, dr->rcv_nxt);
513
514 if (unlikely(pkt->tag.stype_env != VIO_DRING_DATA))
515 return 0;
516 if (unlikely(pkt->seq != dr->rcv_nxt)) {
517 pr_err("RX out of sequence seq[0x%llx] rcv_nxt[0x%llx]\n",
518 pkt->seq, dr->rcv_nxt);
519 return 0;
520 }
521
522 if (!port->napi_resume)
523 dr->rcv_nxt++;
524
525 /* XXX Validate pkt->start_idx and pkt->end_idx XXX */
526
527 return vnet_walk_rx(port, dr, pkt->start_idx, pkt->end_idx,
528 npkts, budget);
529 }
530
531 static int idx_is_pending(struct vio_dring_state *dr, u32 end)
532 {
533 u32 idx = dr->cons;
534 int found = 0;
535
536 while (idx != dr->prod) {
537 if (idx == end) {
538 found = 1;
539 break;
540 }
541 idx = next_idx(idx, dr);
542 }
543 return found;
544 }
545
546 static int vnet_ack(struct vnet_port *port, void *msgbuf)
547 {
548 struct vio_dring_state *dr = &port->vio.drings[VIO_DRIVER_TX_RING];
549 struct vio_dring_data *pkt = msgbuf;
550 struct net_device *dev;
551 struct vnet *vp;
552 u32 end;
553 struct vio_net_desc *desc;
554 if (unlikely(pkt->tag.stype_env != VIO_DRING_DATA))
555 return 0;
556
557 end = pkt->end_idx;
558 if (unlikely(!idx_is_pending(dr, end)))
559 return 0;
560
561 vp = port->vp;
562 dev = vp->dev;
563 /* sync for race conditions with vnet_start_xmit() and tell xmit it
564 * is time to send a trigger.
565 */
566 netif_tx_lock(dev);
567 dr->cons = next_idx(end, dr);
568 desc = vio_dring_entry(dr, dr->cons);
569 if (desc->hdr.state == VIO_DESC_READY && port->start_cons) {
570 /* vnet_start_xmit() just populated this dring but missed
571 * sending the "start" LDC message to the consumer.
572 * Send a "start" trigger on its behalf.
573 */
574 if (__vnet_tx_trigger(port, dr->cons) > 0)
575 port->start_cons = false;
576 else
577 port->start_cons = true;
578 } else {
579 port->start_cons = true;
580 }
581 netif_tx_unlock(dev);
582
583 if (unlikely(netif_queue_stopped(dev) &&
584 vnet_tx_dring_avail(dr) >= VNET_TX_WAKEUP_THRESH(dr)))
585 return 1;
586
587 return 0;
588 }
589
590 static int vnet_nack(struct vnet_port *port, void *msgbuf)
591 {
592 /* XXX just reset or similar XXX */
593 return 0;
594 }
595
596 static int handle_mcast(struct vnet_port *port, void *msgbuf)
597 {
598 struct vio_net_mcast_info *pkt = msgbuf;
599
600 if (pkt->tag.stype != VIO_SUBTYPE_ACK)
601 pr_err("%s: Got unexpected MCAST reply [%02x:%02x:%04x:%08x]\n",
602 port->vp->dev->name,
603 pkt->tag.type,
604 pkt->tag.stype,
605 pkt->tag.stype_env,
606 pkt->tag.sid);
607
608 return 0;
609 }
610
611 static void maybe_tx_wakeup(struct vnet *vp)
612 {
613 struct net_device *dev = vp->dev;
614
615 netif_tx_lock(dev);
616 if (likely(netif_queue_stopped(dev))) {
617 struct vnet_port *port;
618 int wake = 1;
619
620 rcu_read_lock();
621 list_for_each_entry_rcu(port, &vp->port_list, list) {
622 struct vio_dring_state *dr;
623
624 dr = &port->vio.drings[VIO_DRIVER_TX_RING];
625 if (vnet_tx_dring_avail(dr) <
626 VNET_TX_WAKEUP_THRESH(dr)) {
627 wake = 0;
628 break;
629 }
630 }
631 rcu_read_unlock();
632 if (wake)
633 netif_wake_queue(dev);
634 }
635 netif_tx_unlock(dev);
636 }
637
638 static inline bool port_is_up(struct vnet_port *vnet)
639 {
640 struct vio_driver_state *vio = &vnet->vio;
641
642 return !!(vio->hs_state & VIO_HS_COMPLETE);
643 }
644
645 static int vnet_event_napi(struct vnet_port *port, int budget)
646 {
647 struct vio_driver_state *vio = &port->vio;
648 int tx_wakeup, err;
649 int npkts = 0;
650 int event = (port->rx_event & LDC_EVENT_RESET);
651
652 ldc_ctrl:
653 if (unlikely(event == LDC_EVENT_RESET ||
654 event == LDC_EVENT_UP)) {
655 vio_link_state_change(vio, event);
656
657 if (event == LDC_EVENT_RESET) {
658 port->rmtu = 0;
659 vio_port_up(vio);
660 }
661 port->rx_event = 0;
662 return 0;
663 }
664 /* We may have multiple LDC events in rx_event. Unroll send_events() */
665 event = (port->rx_event & LDC_EVENT_UP);
666 port->rx_event &= ~(LDC_EVENT_RESET|LDC_EVENT_UP);
667 if (event == LDC_EVENT_UP)
668 goto ldc_ctrl;
669 event = port->rx_event;
670 if (!(event & LDC_EVENT_DATA_READY))
671 return 0;
672
673 /* we dont expect any other bits than RESET, UP, DATA_READY */
674 BUG_ON(event != LDC_EVENT_DATA_READY);
675
676 tx_wakeup = err = 0;
677 while (1) {
678 union {
679 struct vio_msg_tag tag;
680 u64 raw[8];
681 } msgbuf;
682
683 if (port->napi_resume) {
684 struct vio_dring_data *pkt =
685 (struct vio_dring_data *)&msgbuf;
686 struct vio_dring_state *dr =
687 &port->vio.drings[VIO_DRIVER_RX_RING];
688
689 pkt->tag.type = VIO_TYPE_DATA;
690 pkt->tag.stype = VIO_SUBTYPE_INFO;
691 pkt->tag.stype_env = VIO_DRING_DATA;
692 pkt->seq = dr->rcv_nxt;
693 pkt->start_idx = next_idx(port->napi_stop_idx, dr);
694 pkt->end_idx = -1;
695 goto napi_resume;
696 }
697 ldc_read:
698 err = ldc_read(vio->lp, &msgbuf, sizeof(msgbuf));
699 if (unlikely(err < 0)) {
700 if (err == -ECONNRESET)
701 vio_conn_reset(vio);
702 break;
703 }
704 if (err == 0)
705 break;
706 viodbg(DATA, "TAG [%02x:%02x:%04x:%08x]\n",
707 msgbuf.tag.type,
708 msgbuf.tag.stype,
709 msgbuf.tag.stype_env,
710 msgbuf.tag.sid);
711 err = vio_validate_sid(vio, &msgbuf.tag);
712 if (err < 0)
713 break;
714 napi_resume:
715 if (likely(msgbuf.tag.type == VIO_TYPE_DATA)) {
716 if (msgbuf.tag.stype == VIO_SUBTYPE_INFO) {
717 if (!port_is_up(port)) {
718 /* failures like handshake_failure()
719 * may have cleaned up dring, but
720 * NAPI polling may bring us here.
721 */
722 err = -ECONNRESET;
723 break;
724 }
725 err = vnet_rx(port, &msgbuf, &npkts, budget);
726 if (npkts >= budget)
727 break;
728 if (npkts == 0 && err != -ECONNRESET)
729 goto ldc_read;
730 } else if (msgbuf.tag.stype == VIO_SUBTYPE_ACK) {
731 err = vnet_ack(port, &msgbuf);
732 if (err > 0)
733 tx_wakeup |= err;
734 } else if (msgbuf.tag.stype == VIO_SUBTYPE_NACK) {
735 err = vnet_nack(port, &msgbuf);
736 }
737 } else if (msgbuf.tag.type == VIO_TYPE_CTRL) {
738 if (msgbuf.tag.stype_env == VNET_MCAST_INFO)
739 err = handle_mcast(port, &msgbuf);
740 else
741 err = vio_control_pkt_engine(vio, &msgbuf);
742 if (err)
743 break;
744 } else {
745 err = vnet_handle_unknown(port, &msgbuf);
746 }
747 if (err == -ECONNRESET)
748 break;
749 }
750 if (unlikely(tx_wakeup && err != -ECONNRESET))
751 maybe_tx_wakeup(port->vp);
752 return npkts;
753 }
754
755 static int vnet_poll(struct napi_struct *napi, int budget)
756 {
757 struct vnet_port *port = container_of(napi, struct vnet_port, napi);
758 struct vio_driver_state *vio = &port->vio;
759 int processed = vnet_event_napi(port, budget);
760
761 if (processed < budget) {
762 napi_complete(napi);
763 vio_set_intr(vio->vdev->rx_ino, HV_INTR_ENABLED);
764 }
765 return processed;
766 }
767
768 static void vnet_event(void *arg, int event)
769 {
770 struct vnet_port *port = arg;
771 struct vio_driver_state *vio = &port->vio;
772
773 port->rx_event |= event;
774 vio_set_intr(vio->vdev->rx_ino, HV_INTR_DISABLED);
775 napi_schedule(&port->napi);
776
777 }
778
779 static int __vnet_tx_trigger(struct vnet_port *port, u32 start)
780 {
781 struct vio_dring_state *dr = &port->vio.drings[VIO_DRIVER_TX_RING];
782 struct vio_dring_data hdr = {
783 .tag = {
784 .type = VIO_TYPE_DATA,
785 .stype = VIO_SUBTYPE_INFO,
786 .stype_env = VIO_DRING_DATA,
787 .sid = vio_send_sid(&port->vio),
788 },
789 .dring_ident = dr->ident,
790 .start_idx = start,
791 .end_idx = (u32) -1,
792 };
793 int err, delay;
794 int retries = 0;
795
796 if (port->stop_rx) {
797 err = vnet_send_ack(port,
798 &port->vio.drings[VIO_DRIVER_RX_RING],
799 port->stop_rx_idx, -1,
800 VIO_DRING_STOPPED);
801 if (err <= 0)
802 return err;
803 }
804
805 hdr.seq = dr->snd_nxt;
806 delay = 1;
807 do {
808 err = vio_ldc_send(&port->vio, &hdr, sizeof(hdr));
809 if (err > 0) {
810 dr->snd_nxt++;
811 break;
812 }
813 udelay(delay);
814 if ((delay <<= 1) > 128)
815 delay = 128;
816 if (retries++ > VNET_MAX_RETRIES)
817 break;
818 } while (err == -EAGAIN);
819
820 return err;
821 }
822
823 struct vnet_port *__tx_port_find(struct vnet *vp, struct sk_buff *skb)
824 {
825 unsigned int hash = vnet_hashfn(skb->data);
826 struct hlist_head *hp = &vp->port_hash[hash];
827 struct vnet_port *port;
828
829 hlist_for_each_entry_rcu(port, hp, hash) {
830 if (!port_is_up(port))
831 continue;
832 if (ether_addr_equal(port->raddr, skb->data))
833 return port;
834 }
835 list_for_each_entry_rcu(port, &vp->port_list, list) {
836 if (!port->switch_port)
837 continue;
838 if (!port_is_up(port))
839 continue;
840 return port;
841 }
842 return NULL;
843 }
844
845 struct vnet_port *tx_port_find(struct vnet *vp, struct sk_buff *skb)
846 {
847 struct vnet_port *ret;
848 unsigned long flags;
849
850 spin_lock_irqsave(&vp->lock, flags);
851 ret = __tx_port_find(vp, skb);
852 spin_unlock_irqrestore(&vp->lock, flags);
853
854 return ret;
855 }
856
857 static struct sk_buff *vnet_clean_tx_ring(struct vnet_port *port,
858 unsigned *pending)
859 {
860 struct vio_dring_state *dr = &port->vio.drings[VIO_DRIVER_TX_RING];
861 struct sk_buff *skb = NULL;
862 int i, txi;
863
864 *pending = 0;
865
866 txi = dr->prod-1;
867 if (txi < 0)
868 txi = VNET_TX_RING_SIZE-1;
869
870 for (i = 0; i < VNET_TX_RING_SIZE; ++i) {
871 struct vio_net_desc *d;
872
873 d = vio_dring_entry(dr, txi);
874
875 if (d->hdr.state == VIO_DESC_DONE) {
876 if (port->tx_bufs[txi].skb) {
877 BUG_ON(port->tx_bufs[txi].skb->next);
878
879 port->tx_bufs[txi].skb->next = skb;
880 skb = port->tx_bufs[txi].skb;
881 port->tx_bufs[txi].skb = NULL;
882
883 ldc_unmap(port->vio.lp,
884 port->tx_bufs[txi].cookies,
885 port->tx_bufs[txi].ncookies);
886 }
887 d->hdr.state = VIO_DESC_FREE;
888 } else if (d->hdr.state == VIO_DESC_READY) {
889 (*pending)++;
890 } else if (d->hdr.state == VIO_DESC_FREE) {
891 break;
892 }
893 --txi;
894 if (txi < 0)
895 txi = VNET_TX_RING_SIZE-1;
896 }
897 return skb;
898 }
899
900 static inline void vnet_free_skbs(struct sk_buff *skb)
901 {
902 struct sk_buff *next;
903
904 while (skb) {
905 next = skb->next;
906 skb->next = NULL;
907 dev_kfree_skb(skb);
908 skb = next;
909 }
910 }
911
912 static void vnet_clean_timer_expire(unsigned long port0)
913 {
914 struct vnet_port *port = (struct vnet_port *)port0;
915 struct sk_buff *freeskbs;
916 unsigned pending;
917 unsigned long flags;
918
919 spin_lock_irqsave(&port->vio.lock, flags);
920 freeskbs = vnet_clean_tx_ring(port, &pending);
921 spin_unlock_irqrestore(&port->vio.lock, flags);
922
923 vnet_free_skbs(freeskbs);
924
925 if (pending)
926 (void)mod_timer(&port->clean_timer,
927 jiffies + VNET_CLEAN_TIMEOUT);
928 else
929 del_timer(&port->clean_timer);
930 }
931
932 static inline struct sk_buff *vnet_skb_shape(struct sk_buff *skb, void **pstart,
933 int *plen)
934 {
935 struct sk_buff *nskb;
936 int len, pad;
937
938 len = skb->len;
939 pad = 0;
940 if (len < ETH_ZLEN) {
941 pad += ETH_ZLEN - skb->len;
942 len += pad;
943 }
944 len += VNET_PACKET_SKIP;
945 pad += 8 - (len & 7);
946 len += 8 - (len & 7);
947
948 if (((unsigned long)skb->data & 7) != VNET_PACKET_SKIP ||
949 skb_tailroom(skb) < pad ||
950 skb_headroom(skb) < VNET_PACKET_SKIP) {
951 nskb = alloc_and_align_skb(skb->dev, skb->len);
952 skb_reserve(nskb, VNET_PACKET_SKIP);
953 if (skb_copy_bits(skb, 0, nskb->data, skb->len)) {
954 dev_kfree_skb(nskb);
955 dev_kfree_skb(skb);
956 return NULL;
957 }
958 (void)skb_put(nskb, skb->len);
959 dev_kfree_skb(skb);
960 skb = nskb;
961 }
962
963 *pstart = skb->data - VNET_PACKET_SKIP;
964 *plen = len;
965 return skb;
966 }
967
968 static int vnet_start_xmit(struct sk_buff *skb, struct net_device *dev)
969 {
970 struct vnet *vp = netdev_priv(dev);
971 struct vnet_port *port = NULL;
972 struct vio_dring_state *dr;
973 struct vio_net_desc *d;
974 unsigned long flags;
975 unsigned int len;
976 struct sk_buff *freeskbs = NULL;
977 int i, err, txi;
978 void *start = NULL;
979 int nlen = 0;
980 unsigned pending = 0;
981
982 skb = vnet_skb_shape(skb, &start, &nlen);
983 if (unlikely(!skb))
984 goto out_dropped;
985
986 rcu_read_lock();
987 port = tx_port_find(vp, skb);
988 if (unlikely(!port))
989 goto out_dropped;
990
991 if (skb->len > port->rmtu) {
992 unsigned long localmtu = port->rmtu - ETH_HLEN;
993
994 if (vio_version_after_eq(&port->vio, 1, 3))
995 localmtu -= VLAN_HLEN;
996
997 if (skb->protocol == htons(ETH_P_IP)) {
998 struct flowi4 fl4;
999 struct rtable *rt = NULL;
1000
1001 memset(&fl4, 0, sizeof(fl4));
1002 fl4.flowi4_oif = dev->ifindex;
1003 fl4.flowi4_tos = RT_TOS(ip_hdr(skb)->tos);
1004 fl4.daddr = ip_hdr(skb)->daddr;
1005 fl4.saddr = ip_hdr(skb)->saddr;
1006
1007 rt = ip_route_output_key(dev_net(dev), &fl4);
1008 rcu_read_unlock();
1009 if (!IS_ERR(rt)) {
1010 skb_dst_set(skb, &rt->dst);
1011 icmp_send(skb, ICMP_DEST_UNREACH,
1012 ICMP_FRAG_NEEDED,
1013 htonl(localmtu));
1014 }
1015 }
1016 #if IS_ENABLED(CONFIG_IPV6)
1017 else if (skb->protocol == htons(ETH_P_IPV6))
1018 icmpv6_send(skb, ICMPV6_PKT_TOOBIG, 0, localmtu);
1019 #endif
1020 goto out_dropped;
1021 }
1022
1023 spin_lock_irqsave(&port->vio.lock, flags);
1024
1025 dr = &port->vio.drings[VIO_DRIVER_TX_RING];
1026 if (unlikely(vnet_tx_dring_avail(dr) < 1)) {
1027 if (!netif_queue_stopped(dev)) {
1028 netif_stop_queue(dev);
1029
1030 /* This is a hard error, log it. */
1031 netdev_err(dev, "BUG! Tx Ring full when queue awake!\n");
1032 dev->stats.tx_errors++;
1033 }
1034 rcu_read_unlock();
1035 return NETDEV_TX_BUSY;
1036 }
1037
1038 d = vio_dring_cur(dr);
1039
1040 txi = dr->prod;
1041
1042 freeskbs = vnet_clean_tx_ring(port, &pending);
1043
1044 BUG_ON(port->tx_bufs[txi].skb);
1045
1046 len = skb->len;
1047 if (len < ETH_ZLEN)
1048 len = ETH_ZLEN;
1049
1050 port->tx_bufs[txi].skb = skb;
1051 skb = NULL;
1052
1053 err = ldc_map_single(port->vio.lp, start, nlen,
1054 port->tx_bufs[txi].cookies, VNET_MAXCOOKIES,
1055 (LDC_MAP_SHADOW | LDC_MAP_DIRECT | LDC_MAP_RW));
1056 if (err < 0) {
1057 netdev_info(dev, "tx buffer map error %d\n", err);
1058 goto out_dropped_unlock;
1059 }
1060 port->tx_bufs[txi].ncookies = err;
1061
1062 /* We don't rely on the ACKs to free the skb in vnet_start_xmit(),
1063 * thus it is safe to not set VIO_ACK_ENABLE for each transmission:
1064 * the protocol itself does not require it as long as the peer
1065 * sends a VIO_SUBTYPE_ACK for VIO_DRING_STOPPED.
1066 *
1067 * An ACK for every packet in the ring is expensive as the
1068 * sending of LDC messages is slow and affects performance.
1069 */
1070 d->hdr.ack = VIO_ACK_DISABLE;
1071 d->size = len;
1072 d->ncookies = port->tx_bufs[txi].ncookies;
1073 for (i = 0; i < d->ncookies; i++)
1074 d->cookies[i] = port->tx_bufs[txi].cookies[i];
1075
1076 /* This has to be a non-SMP write barrier because we are writing
1077 * to memory which is shared with the peer LDOM.
1078 */
1079 wmb();
1080
1081 d->hdr.state = VIO_DESC_READY;
1082
1083 /* Exactly one ldc "start" trigger (for dr->cons) needs to be sent
1084 * to notify the consumer that some descriptors are READY.
1085 * After that "start" trigger, no additional triggers are needed until
1086 * a DRING_STOPPED is received from the consumer. The dr->cons field
1087 * (set up by vnet_ack()) has the value of the next dring index
1088 * that has not yet been ack-ed. We send a "start" trigger here
1089 * if, and only if, start_cons is true (reset it afterward). Conversely,
1090 * vnet_ack() should check if the dring corresponding to cons
1091 * is marked READY, but start_cons was false.
1092 * If so, vnet_ack() should send out the missed "start" trigger.
1093 *
1094 * Note that the wmb() above makes sure the cookies et al. are
1095 * not globally visible before the VIO_DESC_READY, and that the
1096 * stores are ordered correctly by the compiler. The consumer will
1097 * not proceed until the VIO_DESC_READY is visible assuring that
1098 * the consumer does not observe anything related to descriptors
1099 * out of order. The HV trap from the LDC start trigger is the
1100 * producer to consumer announcement that work is available to the
1101 * consumer
1102 */
1103 if (!port->start_cons)
1104 goto ldc_start_done; /* previous trigger suffices */
1105
1106 err = __vnet_tx_trigger(port, dr->cons);
1107 if (unlikely(err < 0)) {
1108 netdev_info(dev, "TX trigger error %d\n", err);
1109 d->hdr.state = VIO_DESC_FREE;
1110 dev->stats.tx_carrier_errors++;
1111 goto out_dropped_unlock;
1112 }
1113
1114 ldc_start_done:
1115 port->start_cons = false;
1116
1117 dev->stats.tx_packets++;
1118 dev->stats.tx_bytes += port->tx_bufs[txi].skb->len;
1119
1120 dr->prod = (dr->prod + 1) & (VNET_TX_RING_SIZE - 1);
1121 if (unlikely(vnet_tx_dring_avail(dr) < 1)) {
1122 netif_stop_queue(dev);
1123 if (vnet_tx_dring_avail(dr) > VNET_TX_WAKEUP_THRESH(dr))
1124 netif_wake_queue(dev);
1125 }
1126
1127 spin_unlock_irqrestore(&port->vio.lock, flags);
1128 (void)mod_timer(&port->clean_timer, jiffies + VNET_CLEAN_TIMEOUT);
1129 rcu_read_unlock();
1130
1131 vnet_free_skbs(freeskbs);
1132
1133 return NETDEV_TX_OK;
1134
1135 out_dropped_unlock:
1136 spin_unlock_irqrestore(&port->vio.lock, flags);
1137
1138 out_dropped:
1139 if (pending)
1140 (void)mod_timer(&port->clean_timer,
1141 jiffies + VNET_CLEAN_TIMEOUT);
1142 else if (port)
1143 del_timer(&port->clean_timer);
1144 if (port)
1145 rcu_read_unlock();
1146 if (skb)
1147 dev_kfree_skb(skb);
1148 vnet_free_skbs(freeskbs);
1149 dev->stats.tx_dropped++;
1150 return NETDEV_TX_OK;
1151 }
1152
1153 static void vnet_tx_timeout(struct net_device *dev)
1154 {
1155 /* XXX Implement me XXX */
1156 }
1157
1158 static int vnet_open(struct net_device *dev)
1159 {
1160 netif_carrier_on(dev);
1161 netif_start_queue(dev);
1162
1163 return 0;
1164 }
1165
1166 static int vnet_close(struct net_device *dev)
1167 {
1168 netif_stop_queue(dev);
1169 netif_carrier_off(dev);
1170
1171 return 0;
1172 }
1173
1174 static struct vnet_mcast_entry *__vnet_mc_find(struct vnet *vp, u8 *addr)
1175 {
1176 struct vnet_mcast_entry *m;
1177
1178 for (m = vp->mcast_list; m; m = m->next) {
1179 if (ether_addr_equal(m->addr, addr))
1180 return m;
1181 }
1182 return NULL;
1183 }
1184
1185 static void __update_mc_list(struct vnet *vp, struct net_device *dev)
1186 {
1187 struct netdev_hw_addr *ha;
1188
1189 netdev_for_each_mc_addr(ha, dev) {
1190 struct vnet_mcast_entry *m;
1191
1192 m = __vnet_mc_find(vp, ha->addr);
1193 if (m) {
1194 m->hit = 1;
1195 continue;
1196 }
1197
1198 if (!m) {
1199 m = kzalloc(sizeof(*m), GFP_ATOMIC);
1200 if (!m)
1201 continue;
1202 memcpy(m->addr, ha->addr, ETH_ALEN);
1203 m->hit = 1;
1204
1205 m->next = vp->mcast_list;
1206 vp->mcast_list = m;
1207 }
1208 }
1209 }
1210
1211 static void __send_mc_list(struct vnet *vp, struct vnet_port *port)
1212 {
1213 struct vio_net_mcast_info info;
1214 struct vnet_mcast_entry *m, **pp;
1215 int n_addrs;
1216
1217 memset(&info, 0, sizeof(info));
1218
1219 info.tag.type = VIO_TYPE_CTRL;
1220 info.tag.stype = VIO_SUBTYPE_INFO;
1221 info.tag.stype_env = VNET_MCAST_INFO;
1222 info.tag.sid = vio_send_sid(&port->vio);
1223 info.set = 1;
1224
1225 n_addrs = 0;
1226 for (m = vp->mcast_list; m; m = m->next) {
1227 if (m->sent)
1228 continue;
1229 m->sent = 1;
1230 memcpy(&info.mcast_addr[n_addrs * ETH_ALEN],
1231 m->addr, ETH_ALEN);
1232 if (++n_addrs == VNET_NUM_MCAST) {
1233 info.count = n_addrs;
1234
1235 (void) vio_ldc_send(&port->vio, &info,
1236 sizeof(info));
1237 n_addrs = 0;
1238 }
1239 }
1240 if (n_addrs) {
1241 info.count = n_addrs;
1242 (void) vio_ldc_send(&port->vio, &info, sizeof(info));
1243 }
1244
1245 info.set = 0;
1246
1247 n_addrs = 0;
1248 pp = &vp->mcast_list;
1249 while ((m = *pp) != NULL) {
1250 if (m->hit) {
1251 m->hit = 0;
1252 pp = &m->next;
1253 continue;
1254 }
1255
1256 memcpy(&info.mcast_addr[n_addrs * ETH_ALEN],
1257 m->addr, ETH_ALEN);
1258 if (++n_addrs == VNET_NUM_MCAST) {
1259 info.count = n_addrs;
1260 (void) vio_ldc_send(&port->vio, &info,
1261 sizeof(info));
1262 n_addrs = 0;
1263 }
1264
1265 *pp = m->next;
1266 kfree(m);
1267 }
1268 if (n_addrs) {
1269 info.count = n_addrs;
1270 (void) vio_ldc_send(&port->vio, &info, sizeof(info));
1271 }
1272 }
1273
1274 static void vnet_set_rx_mode(struct net_device *dev)
1275 {
1276 struct vnet *vp = netdev_priv(dev);
1277 struct vnet_port *port;
1278
1279 rcu_read_lock();
1280 list_for_each_entry_rcu(port, &vp->port_list, list) {
1281
1282 if (port->switch_port) {
1283 __update_mc_list(vp, dev);
1284 __send_mc_list(vp, port);
1285 break;
1286 }
1287 }
1288 rcu_read_unlock();
1289 }
1290
1291 static int vnet_change_mtu(struct net_device *dev, int new_mtu)
1292 {
1293 if (new_mtu < 68 || new_mtu > 65535)
1294 return -EINVAL;
1295
1296 dev->mtu = new_mtu;
1297 return 0;
1298 }
1299
1300 static int vnet_set_mac_addr(struct net_device *dev, void *p)
1301 {
1302 return -EINVAL;
1303 }
1304
1305 static void vnet_get_drvinfo(struct net_device *dev,
1306 struct ethtool_drvinfo *info)
1307 {
1308 strlcpy(info->driver, DRV_MODULE_NAME, sizeof(info->driver));
1309 strlcpy(info->version, DRV_MODULE_VERSION, sizeof(info->version));
1310 }
1311
1312 static u32 vnet_get_msglevel(struct net_device *dev)
1313 {
1314 struct vnet *vp = netdev_priv(dev);
1315 return vp->msg_enable;
1316 }
1317
1318 static void vnet_set_msglevel(struct net_device *dev, u32 value)
1319 {
1320 struct vnet *vp = netdev_priv(dev);
1321 vp->msg_enable = value;
1322 }
1323
1324 static const struct ethtool_ops vnet_ethtool_ops = {
1325 .get_drvinfo = vnet_get_drvinfo,
1326 .get_msglevel = vnet_get_msglevel,
1327 .set_msglevel = vnet_set_msglevel,
1328 .get_link = ethtool_op_get_link,
1329 };
1330
1331 static void vnet_port_free_tx_bufs(struct vnet_port *port)
1332 {
1333 struct vio_dring_state *dr;
1334 int i;
1335
1336 dr = &port->vio.drings[VIO_DRIVER_TX_RING];
1337 if (dr->base) {
1338 ldc_free_exp_dring(port->vio.lp, dr->base,
1339 (dr->entry_size * dr->num_entries),
1340 dr->cookies, dr->ncookies);
1341 dr->base = NULL;
1342 dr->entry_size = 0;
1343 dr->num_entries = 0;
1344 dr->pending = 0;
1345 dr->ncookies = 0;
1346 }
1347
1348 for (i = 0; i < VNET_TX_RING_SIZE; i++) {
1349 struct vio_net_desc *d;
1350 void *skb = port->tx_bufs[i].skb;
1351
1352 if (!skb)
1353 continue;
1354
1355 d = vio_dring_entry(dr, i);
1356 if (d->hdr.state == VIO_DESC_READY)
1357 pr_warn("active transmit buffers freed\n");
1358
1359 ldc_unmap(port->vio.lp,
1360 port->tx_bufs[i].cookies,
1361 port->tx_bufs[i].ncookies);
1362 dev_kfree_skb(skb);
1363 port->tx_bufs[i].skb = NULL;
1364 d->hdr.state = VIO_DESC_FREE;
1365 }
1366 }
1367
1368 static int vnet_port_alloc_tx_bufs(struct vnet_port *port)
1369 {
1370 struct vio_dring_state *dr;
1371 unsigned long len;
1372 int i, err, ncookies;
1373 void *dring;
1374
1375 dr = &port->vio.drings[VIO_DRIVER_TX_RING];
1376
1377 len = (VNET_TX_RING_SIZE *
1378 (sizeof(struct vio_net_desc) +
1379 (sizeof(struct ldc_trans_cookie) * 2)));
1380
1381 ncookies = VIO_MAX_RING_COOKIES;
1382 dring = ldc_alloc_exp_dring(port->vio.lp, len,
1383 dr->cookies, &ncookies,
1384 (LDC_MAP_SHADOW |
1385 LDC_MAP_DIRECT |
1386 LDC_MAP_RW));
1387 if (IS_ERR(dring)) {
1388 err = PTR_ERR(dring);
1389 goto err_out;
1390 }
1391
1392 dr->base = dring;
1393 dr->entry_size = (sizeof(struct vio_net_desc) +
1394 (sizeof(struct ldc_trans_cookie) * 2));
1395 dr->num_entries = VNET_TX_RING_SIZE;
1396 dr->prod = dr->cons = 0;
1397 port->start_cons = true; /* need an initial trigger */
1398 dr->pending = VNET_TX_RING_SIZE;
1399 dr->ncookies = ncookies;
1400
1401 for (i = 0; i < VNET_TX_RING_SIZE; ++i) {
1402 struct vio_net_desc *d;
1403
1404 d = vio_dring_entry(dr, i);
1405 d->hdr.state = VIO_DESC_FREE;
1406 }
1407 return 0;
1408
1409 err_out:
1410 vnet_port_free_tx_bufs(port);
1411
1412 return err;
1413 }
1414
1415 #ifdef CONFIG_NET_POLL_CONTROLLER
1416 static void vnet_poll_controller(struct net_device *dev)
1417 {
1418 struct vnet *vp = netdev_priv(dev);
1419 struct vnet_port *port;
1420 unsigned long flags;
1421
1422 spin_lock_irqsave(&vp->lock, flags);
1423 if (!list_empty(&vp->port_list)) {
1424 port = list_entry(vp->port_list.next, struct vnet_port, list);
1425 napi_schedule(&port->napi);
1426 }
1427 spin_unlock_irqrestore(&vp->lock, flags);
1428 }
1429 #endif
1430 static LIST_HEAD(vnet_list);
1431 static DEFINE_MUTEX(vnet_list_mutex);
1432
1433 static const struct net_device_ops vnet_ops = {
1434 .ndo_open = vnet_open,
1435 .ndo_stop = vnet_close,
1436 .ndo_set_rx_mode = vnet_set_rx_mode,
1437 .ndo_set_mac_address = vnet_set_mac_addr,
1438 .ndo_validate_addr = eth_validate_addr,
1439 .ndo_tx_timeout = vnet_tx_timeout,
1440 .ndo_change_mtu = vnet_change_mtu,
1441 .ndo_start_xmit = vnet_start_xmit,
1442 #ifdef CONFIG_NET_POLL_CONTROLLER
1443 .ndo_poll_controller = vnet_poll_controller,
1444 #endif
1445 };
1446
1447 static struct vnet *vnet_new(const u64 *local_mac)
1448 {
1449 struct net_device *dev;
1450 struct vnet *vp;
1451 int err, i;
1452
1453 dev = alloc_etherdev(sizeof(*vp));
1454 if (!dev)
1455 return ERR_PTR(-ENOMEM);
1456 dev->needed_headroom = VNET_PACKET_SKIP + 8;
1457 dev->needed_tailroom = 8;
1458
1459 for (i = 0; i < ETH_ALEN; i++)
1460 dev->dev_addr[i] = (*local_mac >> (5 - i) * 8) & 0xff;
1461
1462 vp = netdev_priv(dev);
1463
1464 spin_lock_init(&vp->lock);
1465 vp->dev = dev;
1466
1467 INIT_LIST_HEAD(&vp->port_list);
1468 for (i = 0; i < VNET_PORT_HASH_SIZE; i++)
1469 INIT_HLIST_HEAD(&vp->port_hash[i]);
1470 INIT_LIST_HEAD(&vp->list);
1471 vp->local_mac = *local_mac;
1472
1473 dev->netdev_ops = &vnet_ops;
1474 dev->ethtool_ops = &vnet_ethtool_ops;
1475 dev->watchdog_timeo = VNET_TX_TIMEOUT;
1476
1477 err = register_netdev(dev);
1478 if (err) {
1479 pr_err("Cannot register net device, aborting\n");
1480 goto err_out_free_dev;
1481 }
1482
1483 netdev_info(dev, "Sun LDOM vnet %pM\n", dev->dev_addr);
1484
1485 list_add(&vp->list, &vnet_list);
1486
1487 return vp;
1488
1489 err_out_free_dev:
1490 free_netdev(dev);
1491
1492 return ERR_PTR(err);
1493 }
1494
1495 static struct vnet *vnet_find_or_create(const u64 *local_mac)
1496 {
1497 struct vnet *iter, *vp;
1498
1499 mutex_lock(&vnet_list_mutex);
1500 vp = NULL;
1501 list_for_each_entry(iter, &vnet_list, list) {
1502 if (iter->local_mac == *local_mac) {
1503 vp = iter;
1504 break;
1505 }
1506 }
1507 if (!vp)
1508 vp = vnet_new(local_mac);
1509 mutex_unlock(&vnet_list_mutex);
1510
1511 return vp;
1512 }
1513
1514 static void vnet_cleanup(void)
1515 {
1516 struct vnet *vp;
1517 struct net_device *dev;
1518
1519 mutex_lock(&vnet_list_mutex);
1520 while (!list_empty(&vnet_list)) {
1521 vp = list_first_entry(&vnet_list, struct vnet, list);
1522 list_del(&vp->list);
1523 dev = vp->dev;
1524 /* vio_unregister_driver() should have cleaned up port_list */
1525 BUG_ON(!list_empty(&vp->port_list));
1526 unregister_netdev(dev);
1527 free_netdev(dev);
1528 }
1529 mutex_unlock(&vnet_list_mutex);
1530 }
1531
1532 static const char *local_mac_prop = "local-mac-address";
1533
1534 static struct vnet *vnet_find_parent(struct mdesc_handle *hp,
1535 u64 port_node)
1536 {
1537 const u64 *local_mac = NULL;
1538 u64 a;
1539
1540 mdesc_for_each_arc(a, hp, port_node, MDESC_ARC_TYPE_BACK) {
1541 u64 target = mdesc_arc_target(hp, a);
1542 const char *name;
1543
1544 name = mdesc_get_property(hp, target, "name", NULL);
1545 if (!name || strcmp(name, "network"))
1546 continue;
1547
1548 local_mac = mdesc_get_property(hp, target,
1549 local_mac_prop, NULL);
1550 if (local_mac)
1551 break;
1552 }
1553 if (!local_mac)
1554 return ERR_PTR(-ENODEV);
1555
1556 return vnet_find_or_create(local_mac);
1557 }
1558
1559 static struct ldc_channel_config vnet_ldc_cfg = {
1560 .event = vnet_event,
1561 .mtu = 64,
1562 .mode = LDC_MODE_UNRELIABLE,
1563 };
1564
1565 static struct vio_driver_ops vnet_vio_ops = {
1566 .send_attr = vnet_send_attr,
1567 .handle_attr = vnet_handle_attr,
1568 .handshake_complete = vnet_handshake_complete,
1569 };
1570
1571 static void print_version(void)
1572 {
1573 printk_once(KERN_INFO "%s", version);
1574 }
1575
1576 const char *remote_macaddr_prop = "remote-mac-address";
1577
1578 static int vnet_port_probe(struct vio_dev *vdev, const struct vio_device_id *id)
1579 {
1580 struct mdesc_handle *hp;
1581 struct vnet_port *port;
1582 unsigned long flags;
1583 struct vnet *vp;
1584 const u64 *rmac;
1585 int len, i, err, switch_port;
1586
1587 print_version();
1588
1589 hp = mdesc_grab();
1590
1591 vp = vnet_find_parent(hp, vdev->mp);
1592 if (IS_ERR(vp)) {
1593 pr_err("Cannot find port parent vnet\n");
1594 err = PTR_ERR(vp);
1595 goto err_out_put_mdesc;
1596 }
1597
1598 rmac = mdesc_get_property(hp, vdev->mp, remote_macaddr_prop, &len);
1599 err = -ENODEV;
1600 if (!rmac) {
1601 pr_err("Port lacks %s property\n", remote_macaddr_prop);
1602 goto err_out_put_mdesc;
1603 }
1604
1605 port = kzalloc(sizeof(*port), GFP_KERNEL);
1606 err = -ENOMEM;
1607 if (!port)
1608 goto err_out_put_mdesc;
1609
1610 for (i = 0; i < ETH_ALEN; i++)
1611 port->raddr[i] = (*rmac >> (5 - i) * 8) & 0xff;
1612
1613 port->vp = vp;
1614
1615 err = vio_driver_init(&port->vio, vdev, VDEV_NETWORK,
1616 vnet_versions, ARRAY_SIZE(vnet_versions),
1617 &vnet_vio_ops, vp->dev->name);
1618 if (err)
1619 goto err_out_free_port;
1620
1621 err = vio_ldc_alloc(&port->vio, &vnet_ldc_cfg, port);
1622 if (err)
1623 goto err_out_free_port;
1624
1625 netif_napi_add(port->vp->dev, &port->napi, vnet_poll, NAPI_POLL_WEIGHT);
1626
1627 err = vnet_port_alloc_tx_bufs(port);
1628 if (err)
1629 goto err_out_free_ldc;
1630
1631 INIT_HLIST_NODE(&port->hash);
1632 INIT_LIST_HEAD(&port->list);
1633
1634 switch_port = 0;
1635 if (mdesc_get_property(hp, vdev->mp, "switch-port", NULL) != NULL)
1636 switch_port = 1;
1637 port->switch_port = switch_port;
1638
1639 spin_lock_irqsave(&vp->lock, flags);
1640 if (switch_port)
1641 list_add_rcu(&port->list, &vp->port_list);
1642 else
1643 list_add_tail_rcu(&port->list, &vp->port_list);
1644 hlist_add_head_rcu(&port->hash,
1645 &vp->port_hash[vnet_hashfn(port->raddr)]);
1646 spin_unlock_irqrestore(&vp->lock, flags);
1647
1648 dev_set_drvdata(&vdev->dev, port);
1649
1650 pr_info("%s: PORT ( remote-mac %pM%s )\n",
1651 vp->dev->name, port->raddr, switch_port ? " switch-port" : "");
1652
1653 setup_timer(&port->clean_timer, vnet_clean_timer_expire,
1654 (unsigned long)port);
1655
1656 napi_enable(&port->napi);
1657 vio_port_up(&port->vio);
1658
1659 mdesc_release(hp);
1660
1661 return 0;
1662
1663 err_out_free_ldc:
1664 netif_napi_del(&port->napi);
1665 vio_ldc_free(&port->vio);
1666
1667 err_out_free_port:
1668 kfree(port);
1669
1670 err_out_put_mdesc:
1671 mdesc_release(hp);
1672 return err;
1673 }
1674
1675 static int vnet_port_remove(struct vio_dev *vdev)
1676 {
1677 struct vnet_port *port = dev_get_drvdata(&vdev->dev);
1678
1679 if (port) {
1680
1681 del_timer_sync(&port->vio.timer);
1682
1683 napi_disable(&port->napi);
1684
1685 list_del_rcu(&port->list);
1686 hlist_del_rcu(&port->hash);
1687
1688 synchronize_rcu();
1689 del_timer_sync(&port->clean_timer);
1690 netif_napi_del(&port->napi);
1691 vnet_port_free_tx_bufs(port);
1692 vio_ldc_free(&port->vio);
1693
1694 dev_set_drvdata(&vdev->dev, NULL);
1695
1696 kfree(port);
1697
1698 }
1699 return 0;
1700 }
1701
1702 static const struct vio_device_id vnet_port_match[] = {
1703 {
1704 .type = "vnet-port",
1705 },
1706 {},
1707 };
1708 MODULE_DEVICE_TABLE(vio, vnet_port_match);
1709
1710 static struct vio_driver vnet_port_driver = {
1711 .id_table = vnet_port_match,
1712 .probe = vnet_port_probe,
1713 .remove = vnet_port_remove,
1714 .name = "vnet_port",
1715 };
1716
1717 static int __init vnet_init(void)
1718 {
1719 return vio_register_driver(&vnet_port_driver);
1720 }
1721
1722 static void __exit vnet_exit(void)
1723 {
1724 vio_unregister_driver(&vnet_port_driver);
1725 vnet_cleanup();
1726 }
1727
1728 module_init(vnet_init);
1729 module_exit(vnet_exit);
Linux kernel - Deliverables
This page took 0.063964 seconds and 4 git commands to generate.