38e3ae9155b7080fc6d0baf5c7b84eadb9f4723d
[linux-2.6.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
19 #include <asm/vio.h>
20 #include <asm/ldc.h>
21
22 #include "sunvnet.h"
23
24 #define DRV_MODULE_NAME         "sunvnet"
25 #define DRV_MODULE_VERSION      "1.0"
26 #define DRV_MODULE_RELDATE      "June 25, 2007"
27
28 static char version[] __devinitdata =
29         DRV_MODULE_NAME ".c:v" DRV_MODULE_VERSION " (" DRV_MODULE_RELDATE ")\n";
30 MODULE_AUTHOR("David S. Miller (davem@davemloft.net)");
31 MODULE_DESCRIPTION("Sun LDOM virtual network driver");
32 MODULE_LICENSE("GPL");
33 MODULE_VERSION(DRV_MODULE_VERSION);
34
35 /* Ordered from largest major to lowest */
36 static struct vio_version vnet_versions[] = {
37         { .major = 1, .minor = 0 },
38 };
39
40 static inline u32 vnet_tx_dring_avail(struct vio_dring_state *dr)
41 {
42         return vio_dring_avail(dr, VNET_TX_RING_SIZE);
43 }
44
45 static int vnet_handle_unknown(struct vnet_port *port, void *arg)
46 {
47         struct vio_msg_tag *pkt = arg;
48
49         pr_err("Received unknown msg [%02x:%02x:%04x:%08x]\n",
50                pkt->type, pkt->stype, pkt->stype_env, pkt->sid);
51         pr_err("Resetting connection\n");
52
53         ldc_disconnect(port->vio.lp);
54
55         return -ECONNRESET;
56 }
57
58 static int vnet_send_attr(struct vio_driver_state *vio)
59 {
60         struct vnet_port *port = to_vnet_port(vio);
61         struct net_device *dev = port->vp->dev;
62         struct vio_net_attr_info pkt;
63         int i;
64
65         memset(&pkt, 0, sizeof(pkt));
66         pkt.tag.type = VIO_TYPE_CTRL;
67         pkt.tag.stype = VIO_SUBTYPE_INFO;
68         pkt.tag.stype_env = VIO_ATTR_INFO;
69         pkt.tag.sid = vio_send_sid(vio);
70         pkt.xfer_mode = VIO_DRING_MODE;
71         pkt.addr_type = VNET_ADDR_ETHERMAC;
72         pkt.ack_freq = 0;
73         for (i = 0; i < 6; i++)
74                 pkt.addr |= (u64)dev->dev_addr[i] << ((5 - i) * 8);
75         pkt.mtu = ETH_FRAME_LEN;
76
77         viodbg(HS, "SEND NET ATTR xmode[0x%x] atype[0x%x] addr[%llx] "
78                "ackfreq[%u] mtu[%llu]\n",
79                pkt.xfer_mode, pkt.addr_type,
80                (unsigned long long) pkt.addr,
81                pkt.ack_freq,
82                (unsigned long long) pkt.mtu);
83
84         return vio_ldc_send(vio, &pkt, sizeof(pkt));
85 }
86
87 static int handle_attr_info(struct vio_driver_state *vio,
88                             struct vio_net_attr_info *pkt)
89 {
90         viodbg(HS, "GOT NET ATTR INFO xmode[0x%x] atype[0x%x] addr[%llx] "
91                "ackfreq[%u] mtu[%llu]\n",
92                pkt->xfer_mode, pkt->addr_type,
93                (unsigned long long) pkt->addr,
94                pkt->ack_freq,
95                (unsigned long long) pkt->mtu);
96
97         pkt->tag.sid = vio_send_sid(vio);
98
99         if (pkt->xfer_mode != VIO_DRING_MODE ||
100             pkt->addr_type != VNET_ADDR_ETHERMAC ||
101             pkt->mtu != ETH_FRAME_LEN) {
102                 viodbg(HS, "SEND NET ATTR NACK\n");
103
104                 pkt->tag.stype = VIO_SUBTYPE_NACK;
105
106                 (void) vio_ldc_send(vio, pkt, sizeof(*pkt));
107
108                 return -ECONNRESET;
109         } else {
110                 viodbg(HS, "SEND NET ATTR ACK\n");
111
112                 pkt->tag.stype = VIO_SUBTYPE_ACK;
113
114                 return vio_ldc_send(vio, pkt, sizeof(*pkt));
115         }
116
117 }
118
119 static int handle_attr_ack(struct vio_driver_state *vio,
120                            struct vio_net_attr_info *pkt)
121 {
122         viodbg(HS, "GOT NET ATTR ACK\n");
123
124         return 0;
125 }
126
127 static int handle_attr_nack(struct vio_driver_state *vio,
128                             struct vio_net_attr_info *pkt)
129 {
130         viodbg(HS, "GOT NET ATTR NACK\n");
131
132         return -ECONNRESET;
133 }
134
135 static int vnet_handle_attr(struct vio_driver_state *vio, void *arg)
136 {
137         struct vio_net_attr_info *pkt = arg;
138
139         switch (pkt->tag.stype) {
140         case VIO_SUBTYPE_INFO:
141                 return handle_attr_info(vio, pkt);
142
143         case VIO_SUBTYPE_ACK:
144                 return handle_attr_ack(vio, pkt);
145
146         case VIO_SUBTYPE_NACK:
147                 return handle_attr_nack(vio, pkt);
148
149         default:
150                 return -ECONNRESET;
151         }
152 }
153
154 static void vnet_handshake_complete(struct vio_driver_state *vio)
155 {
156         struct vio_dring_state *dr;
157
158         dr = &vio->drings[VIO_DRIVER_RX_RING];
159         dr->snd_nxt = dr->rcv_nxt = 1;
160
161         dr = &vio->drings[VIO_DRIVER_TX_RING];
162         dr->snd_nxt = dr->rcv_nxt = 1;
163 }
164
165 /* The hypervisor interface that implements copying to/from imported
166  * memory from another domain requires that copies are done to 8-byte
167  * aligned buffers, and that the lengths of such copies are also 8-byte
168  * multiples.
169  *
170  * So we align skb->data to an 8-byte multiple and pad-out the data
171  * area so we can round the copy length up to the next multiple of
172  * 8 for the copy.
173  *
174  * The transmitter puts the actual start of the packet 6 bytes into
175  * the buffer it sends over, so that the IP headers after the ethernet
176  * header are aligned properly.  These 6 bytes are not in the descriptor
177  * length, they are simply implied.  This offset is represented using
178  * the VNET_PACKET_SKIP macro.
179  */
180 static struct sk_buff *alloc_and_align_skb(struct net_device *dev,
181                                            unsigned int len)
182 {
183         struct sk_buff *skb = netdev_alloc_skb(dev, len+VNET_PACKET_SKIP+8+8);
184         unsigned long addr, off;
185
186         if (unlikely(!skb))
187                 return NULL;
188
189         addr = (unsigned long) skb->data;
190         off = ((addr + 7UL) & ~7UL) - addr;
191         if (off)
192                 skb_reserve(skb, off);
193
194         return skb;
195 }
196
197 static int vnet_rx_one(struct vnet_port *port, unsigned int len,
198                        struct ldc_trans_cookie *cookies, int ncookies)
199 {
200         struct net_device *dev = port->vp->dev;
201         unsigned int copy_len;
202         struct sk_buff *skb;
203         int err;
204
205         err = -EMSGSIZE;
206         if (unlikely(len < ETH_ZLEN || len > ETH_FRAME_LEN)) {
207                 dev->stats.rx_length_errors++;
208                 goto out_dropped;
209         }
210
211         skb = alloc_and_align_skb(dev, len);
212         err = -ENOMEM;
213         if (unlikely(!skb)) {
214                 dev->stats.rx_missed_errors++;
215                 goto out_dropped;
216         }
217
218         copy_len = (len + VNET_PACKET_SKIP + 7U) & ~7U;
219         skb_put(skb, copy_len);
220         err = ldc_copy(port->vio.lp, LDC_COPY_IN,
221                        skb->data, copy_len, 0,
222                        cookies, ncookies);
223         if (unlikely(err < 0)) {
224                 dev->stats.rx_frame_errors++;
225                 goto out_free_skb;
226         }
227
228         skb_pull(skb, VNET_PACKET_SKIP);
229         skb_trim(skb, len);
230         skb->protocol = eth_type_trans(skb, dev);
231
232         dev->stats.rx_packets++;
233         dev->stats.rx_bytes += len;
234
235         netif_rx(skb);
236
237         return 0;
238
239 out_free_skb:
240         kfree_skb(skb);
241
242 out_dropped:
243         dev->stats.rx_dropped++;
244         return err;
245 }
246
247 static int vnet_send_ack(struct vnet_port *port, struct vio_dring_state *dr,
248                          u32 start, u32 end, u8 vio_dring_state)
249 {
250         struct vio_dring_data hdr = {
251                 .tag = {
252                         .type           = VIO_TYPE_DATA,
253                         .stype          = VIO_SUBTYPE_ACK,
254                         .stype_env      = VIO_DRING_DATA,
255                         .sid            = vio_send_sid(&port->vio),
256                 },
257                 .dring_ident            = dr->ident,
258                 .start_idx              = start,
259                 .end_idx                = end,
260                 .state                  = vio_dring_state,
261         };
262         int err, delay;
263
264         hdr.seq = dr->snd_nxt;
265         delay = 1;
266         do {
267                 err = vio_ldc_send(&port->vio, &hdr, sizeof(hdr));
268                 if (err > 0) {
269                         dr->snd_nxt++;
270                         break;
271                 }
272                 udelay(delay);
273                 if ((delay <<= 1) > 128)
274                         delay = 128;
275         } while (err == -EAGAIN);
276
277         return err;
278 }
279
280 static u32 next_idx(u32 idx, struct vio_dring_state *dr)
281 {
282         if (++idx == dr->num_entries)
283                 idx = 0;
284         return idx;
285 }
286
287 static u32 prev_idx(u32 idx, struct vio_dring_state *dr)
288 {
289         if (idx == 0)
290                 idx = dr->num_entries - 1;
291         else
292                 idx--;
293
294         return idx;
295 }
296
297 static struct vio_net_desc *get_rx_desc(struct vnet_port *port,
298                                         struct vio_dring_state *dr,
299                                         u32 index)
300 {
301         struct vio_net_desc *desc = port->vio.desc_buf;
302         int err;
303
304         err = ldc_get_dring_entry(port->vio.lp, desc, dr->entry_size,
305                                   (index * dr->entry_size),
306                                   dr->cookies, dr->ncookies);
307         if (err < 0)
308                 return ERR_PTR(err);
309
310         return desc;
311 }
312
313 static int put_rx_desc(struct vnet_port *port,
314                        struct vio_dring_state *dr,
315                        struct vio_net_desc *desc,
316                        u32 index)
317 {
318         int err;
319
320         err = ldc_put_dring_entry(port->vio.lp, desc, dr->entry_size,
321                                   (index * dr->entry_size),
322                                   dr->cookies, dr->ncookies);
323         if (err < 0)
324                 return err;
325
326         return 0;
327 }
328
329 static int vnet_walk_rx_one(struct vnet_port *port,
330                             struct vio_dring_state *dr,
331                             u32 index, int *needs_ack)
332 {
333         struct vio_net_desc *desc = get_rx_desc(port, dr, index);
334         struct vio_driver_state *vio = &port->vio;
335         int err;
336
337         if (IS_ERR(desc))
338                 return PTR_ERR(desc);
339
340         viodbg(DATA, "vio_walk_rx_one desc[%02x:%02x:%08x:%08x:%llx:%llx]\n",
341                desc->hdr.state, desc->hdr.ack,
342                desc->size, desc->ncookies,
343                desc->cookies[0].cookie_addr,
344                desc->cookies[0].cookie_size);
345
346         if (desc->hdr.state != VIO_DESC_READY)
347                 return 1;
348         err = vnet_rx_one(port, desc->size, desc->cookies, desc->ncookies);
349         if (err == -ECONNRESET)
350                 return err;
351         desc->hdr.state = VIO_DESC_DONE;
352         err = put_rx_desc(port, dr, desc, index);
353         if (err < 0)
354                 return err;
355         *needs_ack = desc->hdr.ack;
356         return 0;
357 }
358
359 static int vnet_walk_rx(struct vnet_port *port, struct vio_dring_state *dr,
360                         u32 start, u32 end)
361 {
362         struct vio_driver_state *vio = &port->vio;
363         int ack_start = -1, ack_end = -1;
364
365         end = (end == (u32) -1) ? prev_idx(start, dr) : next_idx(end, dr);
366
367         viodbg(DATA, "vnet_walk_rx start[%08x] end[%08x]\n", start, end);
368
369         while (start != end) {
370                 int ack = 0, err = vnet_walk_rx_one(port, dr, start, &ack);
371                 if (err == -ECONNRESET)
372                         return err;
373                 if (err != 0)
374                         break;
375                 if (ack_start == -1)
376                         ack_start = start;
377                 ack_end = start;
378                 start = next_idx(start, dr);
379                 if (ack && start != end) {
380                         err = vnet_send_ack(port, dr, ack_start, ack_end,
381                                             VIO_DRING_ACTIVE);
382                         if (err == -ECONNRESET)
383                                 return err;
384                         ack_start = -1;
385                 }
386         }
387         if (unlikely(ack_start == -1))
388                 ack_start = ack_end = prev_idx(start, dr);
389         return vnet_send_ack(port, dr, ack_start, ack_end, VIO_DRING_STOPPED);
390 }
391
392 static int vnet_rx(struct vnet_port *port, void *msgbuf)
393 {
394         struct vio_dring_data *pkt = msgbuf;
395         struct vio_dring_state *dr = &port->vio.drings[VIO_DRIVER_RX_RING];
396         struct vio_driver_state *vio = &port->vio;
397
398         viodbg(DATA, "vnet_rx stype_env[%04x] seq[%016llx] rcv_nxt[%016llx]\n",
399                pkt->tag.stype_env, pkt->seq, dr->rcv_nxt);
400
401         if (unlikely(pkt->tag.stype_env != VIO_DRING_DATA))
402                 return 0;
403         if (unlikely(pkt->seq != dr->rcv_nxt)) {
404                 pr_err("RX out of sequence seq[0x%llx] rcv_nxt[0x%llx]\n",
405                        pkt->seq, dr->rcv_nxt);
406                 return 0;
407         }
408
409         dr->rcv_nxt++;
410
411         /* XXX Validate pkt->start_idx and pkt->end_idx XXX */
412
413         return vnet_walk_rx(port, dr, pkt->start_idx, pkt->end_idx);
414 }
415
416 static int idx_is_pending(struct vio_dring_state *dr, u32 end)
417 {
418         u32 idx = dr->cons;
419         int found = 0;
420
421         while (idx != dr->prod) {
422                 if (idx == end) {
423                         found = 1;
424                         break;
425                 }
426                 idx = next_idx(idx, dr);
427         }
428         return found;
429 }
430
431 static int vnet_ack(struct vnet_port *port, void *msgbuf)
432 {
433         struct vio_dring_state *dr = &port->vio.drings[VIO_DRIVER_TX_RING];
434         struct vio_dring_data *pkt = msgbuf;
435         struct net_device *dev;
436         struct vnet *vp;
437         u32 end;
438
439         if (unlikely(pkt->tag.stype_env != VIO_DRING_DATA))
440                 return 0;
441
442         end = pkt->end_idx;
443         if (unlikely(!idx_is_pending(dr, end)))
444                 return 0;
445
446         dr->cons = next_idx(end, dr);
447
448         vp = port->vp;
449         dev = vp->dev;
450         if (unlikely(netif_queue_stopped(dev) &&
451                      vnet_tx_dring_avail(dr) >= VNET_TX_WAKEUP_THRESH(dr)))
452                 return 1;
453
454         return 0;
455 }
456
457 static int vnet_nack(struct vnet_port *port, void *msgbuf)
458 {
459         /* XXX just reset or similar XXX */
460         return 0;
461 }
462
463 static int handle_mcast(struct vnet_port *port, void *msgbuf)
464 {
465         struct vio_net_mcast_info *pkt = msgbuf;
466
467         if (pkt->tag.stype != VIO_SUBTYPE_ACK)
468                 pr_err("%s: Got unexpected MCAST reply [%02x:%02x:%04x:%08x]\n",
469                        port->vp->dev->name,
470                        pkt->tag.type,
471                        pkt->tag.stype,
472                        pkt->tag.stype_env,
473                        pkt->tag.sid);
474
475         return 0;
476 }
477
478 static void maybe_tx_wakeup(struct vnet *vp)
479 {
480         struct net_device *dev = vp->dev;
481
482         netif_tx_lock(dev);
483         if (likely(netif_queue_stopped(dev))) {
484                 struct vnet_port *port;
485                 int wake = 1;
486
487                 list_for_each_entry(port, &vp->port_list, list) {
488                         struct vio_dring_state *dr;
489
490                         dr = &port->vio.drings[VIO_DRIVER_TX_RING];
491                         if (vnet_tx_dring_avail(dr) <
492                             VNET_TX_WAKEUP_THRESH(dr)) {
493                                 wake = 0;
494                                 break;
495                         }
496                 }
497                 if (wake)
498                         netif_wake_queue(dev);
499         }
500         netif_tx_unlock(dev);
501 }
502
503 static void vnet_event(void *arg, int event)
504 {
505         struct vnet_port *port = arg;
506         struct vio_driver_state *vio = &port->vio;
507         unsigned long flags;
508         int tx_wakeup, err;
509
510         spin_lock_irqsave(&vio->lock, flags);
511
512         if (unlikely(event == LDC_EVENT_RESET ||
513                      event == LDC_EVENT_UP)) {
514                 vio_link_state_change(vio, event);
515                 spin_unlock_irqrestore(&vio->lock, flags);
516
517                 if (event == LDC_EVENT_RESET)
518                         vio_port_up(vio);
519                 return;
520         }
521
522         if (unlikely(event != LDC_EVENT_DATA_READY)) {
523                 pr_warning("Unexpected LDC event %d\n", event);
524                 spin_unlock_irqrestore(&vio->lock, flags);
525                 return;
526         }
527
528         tx_wakeup = err = 0;
529         while (1) {
530                 union {
531                         struct vio_msg_tag tag;
532                         u64 raw[8];
533                 } msgbuf;
534
535                 err = ldc_read(vio->lp, &msgbuf, sizeof(msgbuf));
536                 if (unlikely(err < 0)) {
537                         if (err == -ECONNRESET)
538                                 vio_conn_reset(vio);
539                         break;
540                 }
541                 if (err == 0)
542                         break;
543                 viodbg(DATA, "TAG [%02x:%02x:%04x:%08x]\n",
544                        msgbuf.tag.type,
545                        msgbuf.tag.stype,
546                        msgbuf.tag.stype_env,
547                        msgbuf.tag.sid);
548                 err = vio_validate_sid(vio, &msgbuf.tag);
549                 if (err < 0)
550                         break;
551
552                 if (likely(msgbuf.tag.type == VIO_TYPE_DATA)) {
553                         if (msgbuf.tag.stype == VIO_SUBTYPE_INFO) {
554                                 err = vnet_rx(port, &msgbuf);
555                         } else if (msgbuf.tag.stype == VIO_SUBTYPE_ACK) {
556                                 err = vnet_ack(port, &msgbuf);
557                                 if (err > 0)
558                                         tx_wakeup |= err;
559                         } else if (msgbuf.tag.stype == VIO_SUBTYPE_NACK) {
560                                 err = vnet_nack(port, &msgbuf);
561                         }
562                 } else if (msgbuf.tag.type == VIO_TYPE_CTRL) {
563                         if (msgbuf.tag.stype_env == VNET_MCAST_INFO)
564                                 err = handle_mcast(port, &msgbuf);
565                         else
566                                 err = vio_control_pkt_engine(vio, &msgbuf);
567                         if (err)
568                                 break;
569                 } else {
570                         err = vnet_handle_unknown(port, &msgbuf);
571                 }
572                 if (err == -ECONNRESET)
573                         break;
574         }
575         spin_unlock(&vio->lock);
576         if (unlikely(tx_wakeup && err != -ECONNRESET))
577                 maybe_tx_wakeup(port->vp);
578         local_irq_restore(flags);
579 }
580
581 static int __vnet_tx_trigger(struct vnet_port *port)
582 {
583         struct vio_dring_state *dr = &port->vio.drings[VIO_DRIVER_TX_RING];
584         struct vio_dring_data hdr = {
585                 .tag = {
586                         .type           = VIO_TYPE_DATA,
587                         .stype          = VIO_SUBTYPE_INFO,
588                         .stype_env      = VIO_DRING_DATA,
589                         .sid            = vio_send_sid(&port->vio),
590                 },
591                 .dring_ident            = dr->ident,
592                 .start_idx              = dr->prod,
593                 .end_idx                = (u32) -1,
594         };
595         int err, delay;
596
597         hdr.seq = dr->snd_nxt;
598         delay = 1;
599         do {
600                 err = vio_ldc_send(&port->vio, &hdr, sizeof(hdr));
601                 if (err > 0) {
602                         dr->snd_nxt++;
603                         break;
604                 }
605                 udelay(delay);
606                 if ((delay <<= 1) > 128)
607                         delay = 128;
608         } while (err == -EAGAIN);
609
610         return err;
611 }
612
613 struct vnet_port *__tx_port_find(struct vnet *vp, struct sk_buff *skb)
614 {
615         unsigned int hash = vnet_hashfn(skb->data);
616         struct hlist_head *hp = &vp->port_hash[hash];
617         struct hlist_node *n;
618         struct vnet_port *port;
619
620         hlist_for_each_entry(port, n, hp, hash) {
621                 if (!compare_ether_addr(port->raddr, skb->data))
622                         return port;
623         }
624         port = NULL;
625         if (!list_empty(&vp->port_list))
626                 port = list_entry(vp->port_list.next, struct vnet_port, list);
627
628         return port;
629 }
630
631 struct vnet_port *tx_port_find(struct vnet *vp, struct sk_buff *skb)
632 {
633         struct vnet_port *ret;
634         unsigned long flags;
635
636         spin_lock_irqsave(&vp->lock, flags);
637         ret = __tx_port_find(vp, skb);
638         spin_unlock_irqrestore(&vp->lock, flags);
639
640         return ret;
641 }
642
643 static int vnet_start_xmit(struct sk_buff *skb, struct net_device *dev)
644 {
645         struct vnet *vp = netdev_priv(dev);
646         struct vnet_port *port = tx_port_find(vp, skb);
647         struct vio_dring_state *dr;
648         struct vio_net_desc *d;
649         unsigned long flags;
650         unsigned int len;
651         void *tx_buf;
652         int i, err;
653
654         if (unlikely(!port))
655                 goto out_dropped;
656
657         spin_lock_irqsave(&port->vio.lock, flags);
658
659         dr = &port->vio.drings[VIO_DRIVER_TX_RING];
660         if (unlikely(vnet_tx_dring_avail(dr) < 2)) {
661                 if (!netif_queue_stopped(dev)) {
662                         netif_stop_queue(dev);
663
664                         /* This is a hard error, log it. */
665                         netdev_err(dev, "BUG! Tx Ring full when queue awake!\n");
666                         dev->stats.tx_errors++;
667                 }
668                 spin_unlock_irqrestore(&port->vio.lock, flags);
669                 return NETDEV_TX_BUSY;
670         }
671
672         d = vio_dring_cur(dr);
673
674         tx_buf = port->tx_bufs[dr->prod].buf;
675         skb_copy_from_linear_data(skb, tx_buf + VNET_PACKET_SKIP, skb->len);
676
677         len = skb->len;
678         if (len < ETH_ZLEN) {
679                 len = ETH_ZLEN;
680                 memset(tx_buf+VNET_PACKET_SKIP+skb->len, 0, len - skb->len);
681         }
682
683         d->hdr.ack = VIO_ACK_ENABLE;
684         d->size = len;
685         d->ncookies = port->tx_bufs[dr->prod].ncookies;
686         for (i = 0; i < d->ncookies; i++)
687                 d->cookies[i] = port->tx_bufs[dr->prod].cookies[i];
688
689         /* This has to be a non-SMP write barrier because we are writing
690          * to memory which is shared with the peer LDOM.
691          */
692         wmb();
693
694         d->hdr.state = VIO_DESC_READY;
695
696         err = __vnet_tx_trigger(port);
697         if (unlikely(err < 0)) {
698                 netdev_info(dev, "TX trigger error %d\n", err);
699                 d->hdr.state = VIO_DESC_FREE;
700                 dev->stats.tx_carrier_errors++;
701                 goto out_dropped_unlock;
702         }
703
704         dev->stats.tx_packets++;
705         dev->stats.tx_bytes += skb->len;
706
707         dr->prod = (dr->prod + 1) & (VNET_TX_RING_SIZE - 1);
708         if (unlikely(vnet_tx_dring_avail(dr) < 2)) {
709                 netif_stop_queue(dev);
710                 if (vnet_tx_dring_avail(dr) > VNET_TX_WAKEUP_THRESH(dr))
711                         netif_wake_queue(dev);
712         }
713
714         spin_unlock_irqrestore(&port->vio.lock, flags);
715
716         dev_kfree_skb(skb);
717
718         return NETDEV_TX_OK;
719
720 out_dropped_unlock:
721         spin_unlock_irqrestore(&port->vio.lock, flags);
722
723 out_dropped:
724         dev_kfree_skb(skb);
725         dev->stats.tx_dropped++;
726         return NETDEV_TX_OK;
727 }
728
729 static void vnet_tx_timeout(struct net_device *dev)
730 {
731         /* XXX Implement me XXX */
732 }
733
734 static int vnet_open(struct net_device *dev)
735 {
736         netif_carrier_on(dev);
737         netif_start_queue(dev);
738
739         return 0;
740 }
741
742 static int vnet_close(struct net_device *dev)
743 {
744         netif_stop_queue(dev);
745         netif_carrier_off(dev);
746
747         return 0;
748 }
749
750 static struct vnet_mcast_entry *__vnet_mc_find(struct vnet *vp, u8 *addr)
751 {
752         struct vnet_mcast_entry *m;
753
754         for (m = vp->mcast_list; m; m = m->next) {
755                 if (!memcmp(m->addr, addr, ETH_ALEN))
756                         return m;
757         }
758         return NULL;
759 }
760
761 static void __update_mc_list(struct vnet *vp, struct net_device *dev)
762 {
763         struct netdev_hw_addr *ha;
764
765         netdev_for_each_mc_addr(ha, dev) {
766                 struct vnet_mcast_entry *m;
767
768                 m = __vnet_mc_find(vp, ha->addr);
769                 if (m) {
770                         m->hit = 1;
771                         continue;
772                 }
773
774                 if (!m) {
775                         m = kzalloc(sizeof(*m), GFP_ATOMIC);
776                         if (!m)
777                                 continue;
778                         memcpy(m->addr, ha->addr, ETH_ALEN);
779                         m->hit = 1;
780
781                         m->next = vp->mcast_list;
782                         vp->mcast_list = m;
783                 }
784         }
785 }
786
787 static void __send_mc_list(struct vnet *vp, struct vnet_port *port)
788 {
789         struct vio_net_mcast_info info;
790         struct vnet_mcast_entry *m, **pp;
791         int n_addrs;
792
793         memset(&info, 0, sizeof(info));
794
795         info.tag.type = VIO_TYPE_CTRL;
796         info.tag.stype = VIO_SUBTYPE_INFO;
797         info.tag.stype_env = VNET_MCAST_INFO;
798         info.tag.sid = vio_send_sid(&port->vio);
799         info.set = 1;
800
801         n_addrs = 0;
802         for (m = vp->mcast_list; m; m = m->next) {
803                 if (m->sent)
804                         continue;
805                 m->sent = 1;
806                 memcpy(&info.mcast_addr[n_addrs * ETH_ALEN],
807                        m->addr, ETH_ALEN);
808                 if (++n_addrs == VNET_NUM_MCAST) {
809                         info.count = n_addrs;
810
811                         (void) vio_ldc_send(&port->vio, &info,
812                                             sizeof(info));
813                         n_addrs = 0;
814                 }
815         }
816         if (n_addrs) {
817                 info.count = n_addrs;
818                 (void) vio_ldc_send(&port->vio, &info, sizeof(info));
819         }
820
821         info.set = 0;
822
823         n_addrs = 0;
824         pp = &vp->mcast_list;
825         while ((m = *pp) != NULL) {
826                 if (m->hit) {
827                         m->hit = 0;
828                         pp = &m->next;
829                         continue;
830                 }
831
832                 memcpy(&info.mcast_addr[n_addrs * ETH_ALEN],
833                        m->addr, ETH_ALEN);
834                 if (++n_addrs == VNET_NUM_MCAST) {
835                         info.count = n_addrs;
836                         (void) vio_ldc_send(&port->vio, &info,
837                                             sizeof(info));
838                         n_addrs = 0;
839                 }
840
841                 *pp = m->next;
842                 kfree(m);
843         }
844         if (n_addrs) {
845                 info.count = n_addrs;
846                 (void) vio_ldc_send(&port->vio, &info, sizeof(info));
847         }
848 }
849
850 static void vnet_set_rx_mode(struct net_device *dev)
851 {
852         struct vnet *vp = netdev_priv(dev);
853         struct vnet_port *port;
854         unsigned long flags;
855
856         spin_lock_irqsave(&vp->lock, flags);
857         if (!list_empty(&vp->port_list)) {
858                 port = list_entry(vp->port_list.next, struct vnet_port, list);
859
860                 if (port->switch_port) {
861                         __update_mc_list(vp, dev);
862                         __send_mc_list(vp, port);
863                 }
864         }
865         spin_unlock_irqrestore(&vp->lock, flags);
866 }
867
868 static int vnet_change_mtu(struct net_device *dev, int new_mtu)
869 {
870         if (new_mtu != ETH_DATA_LEN)
871                 return -EINVAL;
872
873         dev->mtu = new_mtu;
874         return 0;
875 }
876
877 static int vnet_set_mac_addr(struct net_device *dev, void *p)
878 {
879         return -EINVAL;
880 }
881
882 static void vnet_get_drvinfo(struct net_device *dev,
883                              struct ethtool_drvinfo *info)
884 {
885         strcpy(info->driver, DRV_MODULE_NAME);
886         strcpy(info->version, DRV_MODULE_VERSION);
887 }
888
889 static u32 vnet_get_msglevel(struct net_device *dev)
890 {
891         struct vnet *vp = netdev_priv(dev);
892         return vp->msg_enable;
893 }
894
895 static void vnet_set_msglevel(struct net_device *dev, u32 value)
896 {
897         struct vnet *vp = netdev_priv(dev);
898         vp->msg_enable = value;
899 }
900
901 static const struct ethtool_ops vnet_ethtool_ops = {
902         .get_drvinfo            = vnet_get_drvinfo,
903         .get_msglevel           = vnet_get_msglevel,
904         .set_msglevel           = vnet_set_msglevel,
905         .get_link               = ethtool_op_get_link,
906 };
907
908 static void vnet_port_free_tx_bufs(struct vnet_port *port)
909 {
910         struct vio_dring_state *dr;
911         int i;
912
913         dr = &port->vio.drings[VIO_DRIVER_TX_RING];
914         if (dr->base) {
915                 ldc_free_exp_dring(port->vio.lp, dr->base,
916                                    (dr->entry_size * dr->num_entries),
917                                    dr->cookies, dr->ncookies);
918                 dr->base = NULL;
919                 dr->entry_size = 0;
920                 dr->num_entries = 0;
921                 dr->pending = 0;
922                 dr->ncookies = 0;
923         }
924
925         for (i = 0; i < VNET_TX_RING_SIZE; i++) {
926                 void *buf = port->tx_bufs[i].buf;
927
928                 if (!buf)
929                         continue;
930
931                 ldc_unmap(port->vio.lp,
932                           port->tx_bufs[i].cookies,
933                           port->tx_bufs[i].ncookies);
934
935                 kfree(buf);
936                 port->tx_bufs[i].buf = NULL;
937         }
938 }
939
940 static int __devinit vnet_port_alloc_tx_bufs(struct vnet_port *port)
941 {
942         struct vio_dring_state *dr;
943         unsigned long len;
944         int i, err, ncookies;
945         void *dring;
946
947         for (i = 0; i < VNET_TX_RING_SIZE; i++) {
948                 void *buf = kzalloc(ETH_FRAME_LEN + 8, GFP_KERNEL);
949                 int map_len = (ETH_FRAME_LEN + 7) & ~7;
950
951                 err = -ENOMEM;
952                 if (!buf)
953                         goto err_out;
954
955                 err = -EFAULT;
956                 if ((unsigned long)buf & (8UL - 1)) {
957                         pr_err("TX buffer misaligned\n");
958                         kfree(buf);
959                         goto err_out;
960                 }
961
962                 err = ldc_map_single(port->vio.lp, buf, map_len,
963                                      port->tx_bufs[i].cookies, 2,
964                                      (LDC_MAP_SHADOW |
965                                       LDC_MAP_DIRECT |
966                                       LDC_MAP_RW));
967                 if (err < 0) {
968                         kfree(buf);
969                         goto err_out;
970                 }
971                 port->tx_bufs[i].buf = buf;
972                 port->tx_bufs[i].ncookies = err;
973         }
974
975         dr = &port->vio.drings[VIO_DRIVER_TX_RING];
976
977         len = (VNET_TX_RING_SIZE *
978                (sizeof(struct vio_net_desc) +
979                 (sizeof(struct ldc_trans_cookie) * 2)));
980
981         ncookies = VIO_MAX_RING_COOKIES;
982         dring = ldc_alloc_exp_dring(port->vio.lp, len,
983                                     dr->cookies, &ncookies,
984                                     (LDC_MAP_SHADOW |
985                                      LDC_MAP_DIRECT |
986                                      LDC_MAP_RW));
987         if (IS_ERR(dring)) {
988                 err = PTR_ERR(dring);
989                 goto err_out;
990         }
991
992         dr->base = dring;
993         dr->entry_size = (sizeof(struct vio_net_desc) +
994                           (sizeof(struct ldc_trans_cookie) * 2));
995         dr->num_entries = VNET_TX_RING_SIZE;
996         dr->prod = dr->cons = 0;
997         dr->pending = VNET_TX_RING_SIZE;
998         dr->ncookies = ncookies;
999
1000         return 0;
1001
1002 err_out:
1003         vnet_port_free_tx_bufs(port);
1004
1005         return err;
1006 }
1007
1008 static LIST_HEAD(vnet_list);
1009 static DEFINE_MUTEX(vnet_list_mutex);
1010
1011 static const struct net_device_ops vnet_ops = {
1012         .ndo_open               = vnet_open,
1013         .ndo_stop               = vnet_close,
1014         .ndo_set_rx_mode        = vnet_set_rx_mode,
1015         .ndo_set_mac_address    = vnet_set_mac_addr,
1016         .ndo_validate_addr      = eth_validate_addr,
1017         .ndo_tx_timeout         = vnet_tx_timeout,
1018         .ndo_change_mtu         = vnet_change_mtu,
1019         .ndo_start_xmit         = vnet_start_xmit,
1020 };
1021
1022 static struct vnet * __devinit vnet_new(const u64 *local_mac)
1023 {
1024         struct net_device *dev;
1025         struct vnet *vp;
1026         int err, i;
1027
1028         dev = alloc_etherdev(sizeof(*vp));
1029         if (!dev)
1030                 return ERR_PTR(-ENOMEM);
1031
1032         for (i = 0; i < ETH_ALEN; i++)
1033                 dev->dev_addr[i] = (*local_mac >> (5 - i) * 8) & 0xff;
1034
1035         memcpy(dev->perm_addr, dev->dev_addr, dev->addr_len);
1036
1037         vp = netdev_priv(dev);
1038
1039         spin_lock_init(&vp->lock);
1040         vp->dev = dev;
1041
1042         INIT_LIST_HEAD(&vp->port_list);
1043         for (i = 0; i < VNET_PORT_HASH_SIZE; i++)
1044                 INIT_HLIST_HEAD(&vp->port_hash[i]);
1045         INIT_LIST_HEAD(&vp->list);
1046         vp->local_mac = *local_mac;
1047
1048         dev->netdev_ops = &vnet_ops;
1049         dev->ethtool_ops = &vnet_ethtool_ops;
1050         dev->watchdog_timeo = VNET_TX_TIMEOUT;
1051
1052         err = register_netdev(dev);
1053         if (err) {
1054                 pr_err("Cannot register net device, aborting\n");
1055                 goto err_out_free_dev;
1056         }
1057
1058         netdev_info(dev, "Sun LDOM vnet %pM\n", dev->dev_addr);
1059
1060         list_add(&vp->list, &vnet_list);
1061
1062         return vp;
1063
1064 err_out_free_dev:
1065         free_netdev(dev);
1066
1067         return ERR_PTR(err);
1068 }
1069
1070 static struct vnet * __devinit vnet_find_or_create(const u64 *local_mac)
1071 {
1072         struct vnet *iter, *vp;
1073
1074         mutex_lock(&vnet_list_mutex);
1075         vp = NULL;
1076         list_for_each_entry(iter, &vnet_list, list) {
1077                 if (iter->local_mac == *local_mac) {
1078                         vp = iter;
1079                         break;
1080                 }
1081         }
1082         if (!vp)
1083                 vp = vnet_new(local_mac);
1084         mutex_unlock(&vnet_list_mutex);
1085
1086         return vp;
1087 }
1088
1089 static const char *local_mac_prop = "local-mac-address";
1090
1091 static struct vnet * __devinit vnet_find_parent(struct mdesc_handle *hp,
1092                                                 u64 port_node)
1093 {
1094         const u64 *local_mac = NULL;
1095         u64 a;
1096
1097         mdesc_for_each_arc(a, hp, port_node, MDESC_ARC_TYPE_BACK) {
1098                 u64 target = mdesc_arc_target(hp, a);
1099                 const char *name;
1100
1101                 name = mdesc_get_property(hp, target, "name", NULL);
1102                 if (!name || strcmp(name, "network"))
1103                         continue;
1104
1105                 local_mac = mdesc_get_property(hp, target,
1106                                                local_mac_prop, NULL);
1107                 if (local_mac)
1108                         break;
1109         }
1110         if (!local_mac)
1111                 return ERR_PTR(-ENODEV);
1112
1113         return vnet_find_or_create(local_mac);
1114 }
1115
1116 static struct ldc_channel_config vnet_ldc_cfg = {
1117         .event          = vnet_event,
1118         .mtu            = 64,
1119         .mode           = LDC_MODE_UNRELIABLE,
1120 };
1121
1122 static struct vio_driver_ops vnet_vio_ops = {
1123         .send_attr              = vnet_send_attr,
1124         .handle_attr            = vnet_handle_attr,
1125         .handshake_complete     = vnet_handshake_complete,
1126 };
1127
1128 static void __devinit print_version(void)
1129 {
1130         printk_once(KERN_INFO "%s", version);
1131 }
1132
1133 const char *remote_macaddr_prop = "remote-mac-address";
1134
1135 static int __devinit vnet_port_probe(struct vio_dev *vdev,
1136                                      const struct vio_device_id *id)
1137 {
1138         struct mdesc_handle *hp;
1139         struct vnet_port *port;
1140         unsigned long flags;
1141         struct vnet *vp;
1142         const u64 *rmac;
1143         int len, i, err, switch_port;
1144
1145         print_version();
1146
1147         hp = mdesc_grab();
1148
1149         vp = vnet_find_parent(hp, vdev->mp);
1150         if (IS_ERR(vp)) {
1151                 pr_err("Cannot find port parent vnet\n");
1152                 err = PTR_ERR(vp);
1153                 goto err_out_put_mdesc;
1154         }
1155
1156         rmac = mdesc_get_property(hp, vdev->mp, remote_macaddr_prop, &len);
1157         err = -ENODEV;
1158         if (!rmac) {
1159                 pr_err("Port lacks %s property\n", remote_macaddr_prop);
1160                 goto err_out_put_mdesc;
1161         }
1162
1163         port = kzalloc(sizeof(*port), GFP_KERNEL);
1164         err = -ENOMEM;
1165         if (!port)
1166                 goto err_out_put_mdesc;
1167
1168         for (i = 0; i < ETH_ALEN; i++)
1169                 port->raddr[i] = (*rmac >> (5 - i) * 8) & 0xff;
1170
1171         port->vp = vp;
1172
1173         err = vio_driver_init(&port->vio, vdev, VDEV_NETWORK,
1174                               vnet_versions, ARRAY_SIZE(vnet_versions),
1175                               &vnet_vio_ops, vp->dev->name);
1176         if (err)
1177                 goto err_out_free_port;
1178
1179         err = vio_ldc_alloc(&port->vio, &vnet_ldc_cfg, port);
1180         if (err)
1181                 goto err_out_free_port;
1182
1183         err = vnet_port_alloc_tx_bufs(port);
1184         if (err)
1185                 goto err_out_free_ldc;
1186
1187         INIT_HLIST_NODE(&port->hash);
1188         INIT_LIST_HEAD(&port->list);
1189
1190         switch_port = 0;
1191         if (mdesc_get_property(hp, vdev->mp, "switch-port", NULL) != NULL)
1192                 switch_port = 1;
1193         port->switch_port = switch_port;
1194
1195         spin_lock_irqsave(&vp->lock, flags);
1196         if (switch_port)
1197                 list_add(&port->list, &vp->port_list);
1198         else
1199                 list_add_tail(&port->list, &vp->port_list);
1200         hlist_add_head(&port->hash, &vp->port_hash[vnet_hashfn(port->raddr)]);
1201         spin_unlock_irqrestore(&vp->lock, flags);
1202
1203         dev_set_drvdata(&vdev->dev, port);
1204
1205         pr_info("%s: PORT ( remote-mac %pM%s )\n",
1206                 vp->dev->name, port->raddr, switch_port ? " switch-port" : "");
1207
1208         vio_port_up(&port->vio);
1209
1210         mdesc_release(hp);
1211
1212         return 0;
1213
1214 err_out_free_ldc:
1215         vio_ldc_free(&port->vio);
1216
1217 err_out_free_port:
1218         kfree(port);
1219
1220 err_out_put_mdesc:
1221         mdesc_release(hp);
1222         return err;
1223 }
1224
1225 static int vnet_port_remove(struct vio_dev *vdev)
1226 {
1227         struct vnet_port *port = dev_get_drvdata(&vdev->dev);
1228
1229         if (port) {
1230                 struct vnet *vp = port->vp;
1231                 unsigned long flags;
1232
1233                 del_timer_sync(&port->vio.timer);
1234
1235                 spin_lock_irqsave(&vp->lock, flags);
1236                 list_del(&port->list);
1237                 hlist_del(&port->hash);
1238                 spin_unlock_irqrestore(&vp->lock, flags);
1239
1240                 vnet_port_free_tx_bufs(port);
1241                 vio_ldc_free(&port->vio);
1242
1243                 dev_set_drvdata(&vdev->dev, NULL);
1244
1245                 kfree(port);
1246         }
1247         return 0;
1248 }
1249
1250 static const struct vio_device_id vnet_port_match[] = {
1251         {
1252                 .type = "vnet-port",
1253         },
1254         {},
1255 };
1256 MODULE_DEVICE_TABLE(vio, vnet_port_match);
1257
1258 static struct vio_driver vnet_port_driver = {
1259         .id_table       = vnet_port_match,
1260         .probe          = vnet_port_probe,
1261         .remove         = vnet_port_remove,
1262         .name           = "vnet_port",
1263 };
1264
1265 static int __init vnet_init(void)
1266 {
1267         return vio_register_driver(&vnet_port_driver);
1268 }
1269
1270 static void __exit vnet_exit(void)
1271 {
1272         vio_unregister_driver(&vnet_port_driver);
1273 }
1274
1275 module_init(vnet_init);
1276 module_exit(vnet_exit);