1 /* sunvnet.c: Sun LDOM Virtual Network Driver.
3 * Copyright (C) 2007, 2008 David S. Miller <davem@davemloft.net>
6 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
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>
24 #define DRV_MODULE_NAME "sunvnet"
25 #define DRV_MODULE_VERSION "1.0"
26 #define DRV_MODULE_RELDATE "June 25, 2007"
28 static char version[] =
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);
35 /* Heuristic for the number of times to exponentially backoff and
36 * retry sending an LDC trigger when EAGAIN is encountered
38 #define VNET_MAX_RETRIES 10
40 /* Ordered from largest major to lowest */
41 static struct vio_version vnet_versions[] = {
42 { .major = 1, .minor = 0 },
45 static inline u32 vnet_tx_dring_avail(struct vio_dring_state *dr)
47 return vio_dring_avail(dr, VNET_TX_RING_SIZE);
50 static int vnet_handle_unknown(struct vnet_port *port, void *arg)
52 struct vio_msg_tag *pkt = arg;
54 pr_err("Received unknown msg [%02x:%02x:%04x:%08x]\n",
55 pkt->type, pkt->stype, pkt->stype_env, pkt->sid);
56 pr_err("Resetting connection\n");
58 ldc_disconnect(port->vio.lp);
63 static int vnet_send_attr(struct vio_driver_state *vio)
65 struct vnet_port *port = to_vnet_port(vio);
66 struct net_device *dev = port->vp->dev;
67 struct vio_net_attr_info pkt;
70 memset(&pkt, 0, sizeof(pkt));
71 pkt.tag.type = VIO_TYPE_CTRL;
72 pkt.tag.stype = VIO_SUBTYPE_INFO;
73 pkt.tag.stype_env = VIO_ATTR_INFO;
74 pkt.tag.sid = vio_send_sid(vio);
75 pkt.xfer_mode = VIO_DRING_MODE;
76 pkt.addr_type = VNET_ADDR_ETHERMAC;
78 for (i = 0; i < 6; i++)
79 pkt.addr |= (u64)dev->dev_addr[i] << ((5 - i) * 8);
80 pkt.mtu = ETH_FRAME_LEN;
82 viodbg(HS, "SEND NET ATTR xmode[0x%x] atype[0x%x] addr[%llx] "
83 "ackfreq[%u] mtu[%llu]\n",
84 pkt.xfer_mode, pkt.addr_type,
85 (unsigned long long) pkt.addr,
87 (unsigned long long) pkt.mtu);
89 return vio_ldc_send(vio, &pkt, sizeof(pkt));
92 static int handle_attr_info(struct vio_driver_state *vio,
93 struct vio_net_attr_info *pkt)
95 viodbg(HS, "GOT NET ATTR INFO xmode[0x%x] atype[0x%x] addr[%llx] "
96 "ackfreq[%u] mtu[%llu]\n",
97 pkt->xfer_mode, pkt->addr_type,
98 (unsigned long long) pkt->addr,
100 (unsigned long long) pkt->mtu);
102 pkt->tag.sid = vio_send_sid(vio);
104 if (pkt->xfer_mode != VIO_DRING_MODE ||
105 pkt->addr_type != VNET_ADDR_ETHERMAC ||
106 pkt->mtu != ETH_FRAME_LEN) {
107 viodbg(HS, "SEND NET ATTR NACK\n");
109 pkt->tag.stype = VIO_SUBTYPE_NACK;
111 (void) vio_ldc_send(vio, pkt, sizeof(*pkt));
115 viodbg(HS, "SEND NET ATTR ACK\n");
117 pkt->tag.stype = VIO_SUBTYPE_ACK;
119 return vio_ldc_send(vio, pkt, sizeof(*pkt));
124 static int handle_attr_ack(struct vio_driver_state *vio,
125 struct vio_net_attr_info *pkt)
127 viodbg(HS, "GOT NET ATTR ACK\n");
132 static int handle_attr_nack(struct vio_driver_state *vio,
133 struct vio_net_attr_info *pkt)
135 viodbg(HS, "GOT NET ATTR NACK\n");
140 static int vnet_handle_attr(struct vio_driver_state *vio, void *arg)
142 struct vio_net_attr_info *pkt = arg;
144 switch (pkt->tag.stype) {
145 case VIO_SUBTYPE_INFO:
146 return handle_attr_info(vio, pkt);
148 case VIO_SUBTYPE_ACK:
149 return handle_attr_ack(vio, pkt);
151 case VIO_SUBTYPE_NACK:
152 return handle_attr_nack(vio, pkt);
159 static void vnet_handshake_complete(struct vio_driver_state *vio)
161 struct vio_dring_state *dr;
163 dr = &vio->drings[VIO_DRIVER_RX_RING];
164 dr->snd_nxt = dr->rcv_nxt = 1;
166 dr = &vio->drings[VIO_DRIVER_TX_RING];
167 dr->snd_nxt = dr->rcv_nxt = 1;
170 /* The hypervisor interface that implements copying to/from imported
171 * memory from another domain requires that copies are done to 8-byte
172 * aligned buffers, and that the lengths of such copies are also 8-byte
175 * So we align skb->data to an 8-byte multiple and pad-out the data
176 * area so we can round the copy length up to the next multiple of
179 * The transmitter puts the actual start of the packet 6 bytes into
180 * the buffer it sends over, so that the IP headers after the ethernet
181 * header are aligned properly. These 6 bytes are not in the descriptor
182 * length, they are simply implied. This offset is represented using
183 * the VNET_PACKET_SKIP macro.
185 static struct sk_buff *alloc_and_align_skb(struct net_device *dev,
188 struct sk_buff *skb = netdev_alloc_skb(dev, len+VNET_PACKET_SKIP+8+8);
189 unsigned long addr, off;
194 addr = (unsigned long) skb->data;
195 off = ((addr + 7UL) & ~7UL) - addr;
197 skb_reserve(skb, off);
202 static int vnet_rx_one(struct vnet_port *port, unsigned int len,
203 struct ldc_trans_cookie *cookies, int ncookies)
205 struct net_device *dev = port->vp->dev;
206 unsigned int copy_len;
211 if (unlikely(len < ETH_ZLEN || len > ETH_FRAME_LEN)) {
212 dev->stats.rx_length_errors++;
216 skb = alloc_and_align_skb(dev, len);
218 if (unlikely(!skb)) {
219 dev->stats.rx_missed_errors++;
223 copy_len = (len + VNET_PACKET_SKIP + 7U) & ~7U;
224 skb_put(skb, copy_len);
225 err = ldc_copy(port->vio.lp, LDC_COPY_IN,
226 skb->data, copy_len, 0,
228 if (unlikely(err < 0)) {
229 dev->stats.rx_frame_errors++;
233 skb_pull(skb, VNET_PACKET_SKIP);
235 skb->protocol = eth_type_trans(skb, dev);
237 dev->stats.rx_packets++;
238 dev->stats.rx_bytes += len;
248 dev->stats.rx_dropped++;
252 static int vnet_send_ack(struct vnet_port *port, struct vio_dring_state *dr,
253 u32 start, u32 end, u8 vio_dring_state)
255 struct vio_dring_data hdr = {
257 .type = VIO_TYPE_DATA,
258 .stype = VIO_SUBTYPE_ACK,
259 .stype_env = VIO_DRING_DATA,
260 .sid = vio_send_sid(&port->vio),
262 .dring_ident = dr->ident,
265 .state = vio_dring_state,
270 hdr.seq = dr->snd_nxt;
273 err = vio_ldc_send(&port->vio, &hdr, sizeof(hdr));
279 if ((delay <<= 1) > 128)
281 if (retries++ > VNET_MAX_RETRIES) {
282 pr_info("ECONNRESET %x:%x:%x:%x:%x:%x\n",
283 port->raddr[0], port->raddr[1],
284 port->raddr[2], port->raddr[3],
285 port->raddr[4], port->raddr[5]);
288 } while (err == -EAGAIN);
293 static u32 next_idx(u32 idx, struct vio_dring_state *dr)
295 if (++idx == dr->num_entries)
300 static u32 prev_idx(u32 idx, struct vio_dring_state *dr)
303 idx = dr->num_entries - 1;
310 static struct vio_net_desc *get_rx_desc(struct vnet_port *port,
311 struct vio_dring_state *dr,
314 struct vio_net_desc *desc = port->vio.desc_buf;
317 err = ldc_get_dring_entry(port->vio.lp, desc, dr->entry_size,
318 (index * dr->entry_size),
319 dr->cookies, dr->ncookies);
326 static int put_rx_desc(struct vnet_port *port,
327 struct vio_dring_state *dr,
328 struct vio_net_desc *desc,
333 err = ldc_put_dring_entry(port->vio.lp, desc, dr->entry_size,
334 (index * dr->entry_size),
335 dr->cookies, dr->ncookies);
342 static int vnet_walk_rx_one(struct vnet_port *port,
343 struct vio_dring_state *dr,
344 u32 index, int *needs_ack)
346 struct vio_net_desc *desc = get_rx_desc(port, dr, index);
347 struct vio_driver_state *vio = &port->vio;
351 return PTR_ERR(desc);
353 viodbg(DATA, "vio_walk_rx_one desc[%02x:%02x:%08x:%08x:%llx:%llx]\n",
354 desc->hdr.state, desc->hdr.ack,
355 desc->size, desc->ncookies,
356 desc->cookies[0].cookie_addr,
357 desc->cookies[0].cookie_size);
359 if (desc->hdr.state != VIO_DESC_READY)
361 err = vnet_rx_one(port, desc->size, desc->cookies, desc->ncookies);
362 if (err == -ECONNRESET)
364 desc->hdr.state = VIO_DESC_DONE;
365 err = put_rx_desc(port, dr, desc, index);
368 *needs_ack = desc->hdr.ack;
372 static int vnet_walk_rx(struct vnet_port *port, struct vio_dring_state *dr,
375 struct vio_driver_state *vio = &port->vio;
376 int ack_start = -1, ack_end = -1;
378 end = (end == (u32) -1) ? prev_idx(start, dr) : next_idx(end, dr);
380 viodbg(DATA, "vnet_walk_rx start[%08x] end[%08x]\n", start, end);
382 while (start != end) {
383 int ack = 0, err = vnet_walk_rx_one(port, dr, start, &ack);
384 if (err == -ECONNRESET)
391 start = next_idx(start, dr);
392 if (ack && start != end) {
393 err = vnet_send_ack(port, dr, ack_start, ack_end,
395 if (err == -ECONNRESET)
400 if (unlikely(ack_start == -1))
401 ack_start = ack_end = prev_idx(start, dr);
402 return vnet_send_ack(port, dr, ack_start, ack_end, VIO_DRING_STOPPED);
405 static int vnet_rx(struct vnet_port *port, void *msgbuf)
407 struct vio_dring_data *pkt = msgbuf;
408 struct vio_dring_state *dr = &port->vio.drings[VIO_DRIVER_RX_RING];
409 struct vio_driver_state *vio = &port->vio;
411 viodbg(DATA, "vnet_rx stype_env[%04x] seq[%016llx] rcv_nxt[%016llx]\n",
412 pkt->tag.stype_env, pkt->seq, dr->rcv_nxt);
414 if (unlikely(pkt->tag.stype_env != VIO_DRING_DATA))
416 if (unlikely(pkt->seq != dr->rcv_nxt)) {
417 pr_err("RX out of sequence seq[0x%llx] rcv_nxt[0x%llx]\n",
418 pkt->seq, dr->rcv_nxt);
424 /* XXX Validate pkt->start_idx and pkt->end_idx XXX */
426 return vnet_walk_rx(port, dr, pkt->start_idx, pkt->end_idx);
429 static int idx_is_pending(struct vio_dring_state *dr, u32 end)
434 while (idx != dr->prod) {
439 idx = next_idx(idx, dr);
444 static int vnet_ack(struct vnet_port *port, void *msgbuf)
446 struct vio_dring_state *dr = &port->vio.drings[VIO_DRIVER_TX_RING];
447 struct vio_dring_data *pkt = msgbuf;
448 struct net_device *dev;
452 if (unlikely(pkt->tag.stype_env != VIO_DRING_DATA))
456 if (unlikely(!idx_is_pending(dr, end)))
459 dr->cons = next_idx(end, dr);
463 if (unlikely(netif_queue_stopped(dev) &&
464 vnet_tx_dring_avail(dr) >= VNET_TX_WAKEUP_THRESH(dr)))
470 static int vnet_nack(struct vnet_port *port, void *msgbuf)
472 /* XXX just reset or similar XXX */
476 static int handle_mcast(struct vnet_port *port, void *msgbuf)
478 struct vio_net_mcast_info *pkt = msgbuf;
480 if (pkt->tag.stype != VIO_SUBTYPE_ACK)
481 pr_err("%s: Got unexpected MCAST reply [%02x:%02x:%04x:%08x]\n",
491 static void maybe_tx_wakeup(unsigned long param)
493 struct vnet *vp = (struct vnet *)param;
494 struct net_device *dev = vp->dev;
497 if (likely(netif_queue_stopped(dev))) {
498 struct vnet_port *port;
501 list_for_each_entry(port, &vp->port_list, list) {
502 struct vio_dring_state *dr;
504 dr = &port->vio.drings[VIO_DRIVER_TX_RING];
505 if (vnet_tx_dring_avail(dr) <
506 VNET_TX_WAKEUP_THRESH(dr)) {
512 netif_wake_queue(dev);
514 netif_tx_unlock(dev);
517 static void vnet_event(void *arg, int event)
519 struct vnet_port *port = arg;
520 struct vio_driver_state *vio = &port->vio;
524 spin_lock_irqsave(&vio->lock, flags);
526 if (unlikely(event == LDC_EVENT_RESET ||
527 event == LDC_EVENT_UP)) {
528 vio_link_state_change(vio, event);
529 spin_unlock_irqrestore(&vio->lock, flags);
531 if (event == LDC_EVENT_RESET)
536 if (unlikely(event != LDC_EVENT_DATA_READY)) {
537 pr_warning("Unexpected LDC event %d\n", event);
538 spin_unlock_irqrestore(&vio->lock, flags);
545 struct vio_msg_tag tag;
549 err = ldc_read(vio->lp, &msgbuf, sizeof(msgbuf));
550 if (unlikely(err < 0)) {
551 if (err == -ECONNRESET)
557 viodbg(DATA, "TAG [%02x:%02x:%04x:%08x]\n",
560 msgbuf.tag.stype_env,
562 err = vio_validate_sid(vio, &msgbuf.tag);
566 if (likely(msgbuf.tag.type == VIO_TYPE_DATA)) {
567 if (msgbuf.tag.stype == VIO_SUBTYPE_INFO) {
568 err = vnet_rx(port, &msgbuf);
569 } else if (msgbuf.tag.stype == VIO_SUBTYPE_ACK) {
570 err = vnet_ack(port, &msgbuf);
573 } else if (msgbuf.tag.stype == VIO_SUBTYPE_NACK) {
574 err = vnet_nack(port, &msgbuf);
576 } else if (msgbuf.tag.type == VIO_TYPE_CTRL) {
577 if (msgbuf.tag.stype_env == VNET_MCAST_INFO)
578 err = handle_mcast(port, &msgbuf);
580 err = vio_control_pkt_engine(vio, &msgbuf);
584 err = vnet_handle_unknown(port, &msgbuf);
586 if (err == -ECONNRESET)
589 spin_unlock(&vio->lock);
590 /* Kick off a tasklet to wake the queue. We cannot call
591 * maybe_tx_wakeup directly here because we could deadlock on
592 * netif_tx_lock() with dev_watchdog()
594 if (unlikely(tx_wakeup && err != -ECONNRESET))
595 tasklet_schedule(&port->vp->vnet_tx_wakeup);
597 local_irq_restore(flags);
600 static int __vnet_tx_trigger(struct vnet_port *port)
602 struct vio_dring_state *dr = &port->vio.drings[VIO_DRIVER_TX_RING];
603 struct vio_dring_data hdr = {
605 .type = VIO_TYPE_DATA,
606 .stype = VIO_SUBTYPE_INFO,
607 .stype_env = VIO_DRING_DATA,
608 .sid = vio_send_sid(&port->vio),
610 .dring_ident = dr->ident,
611 .start_idx = dr->prod,
617 hdr.seq = dr->snd_nxt;
620 err = vio_ldc_send(&port->vio, &hdr, sizeof(hdr));
626 if ((delay <<= 1) > 128)
628 if (retries++ > VNET_MAX_RETRIES)
630 } while (err == -EAGAIN);
635 static inline bool port_is_up(struct vnet_port *vnet)
637 struct vio_driver_state *vio = &vnet->vio;
639 return !!(vio->hs_state & VIO_HS_COMPLETE);
642 struct vnet_port *__tx_port_find(struct vnet *vp, struct sk_buff *skb)
644 unsigned int hash = vnet_hashfn(skb->data);
645 struct hlist_head *hp = &vp->port_hash[hash];
646 struct vnet_port *port;
648 hlist_for_each_entry(port, hp, hash) {
649 if (!port_is_up(port))
651 if (ether_addr_equal(port->raddr, skb->data))
654 list_for_each_entry(port, &vp->port_list, list) {
655 if (!port->switch_port)
657 if (!port_is_up(port))
664 struct vnet_port *tx_port_find(struct vnet *vp, struct sk_buff *skb)
666 struct vnet_port *ret;
669 spin_lock_irqsave(&vp->lock, flags);
670 ret = __tx_port_find(vp, skb);
671 spin_unlock_irqrestore(&vp->lock, flags);
676 static int vnet_start_xmit(struct sk_buff *skb, struct net_device *dev)
678 struct vnet *vp = netdev_priv(dev);
679 struct vnet_port *port = tx_port_find(vp, skb);
680 struct vio_dring_state *dr;
681 struct vio_net_desc *d;
690 spin_lock_irqsave(&port->vio.lock, flags);
692 dr = &port->vio.drings[VIO_DRIVER_TX_RING];
693 if (unlikely(vnet_tx_dring_avail(dr) < 2)) {
694 if (!netif_queue_stopped(dev)) {
695 netif_stop_queue(dev);
697 /* This is a hard error, log it. */
698 netdev_err(dev, "BUG! Tx Ring full when queue awake!\n");
699 dev->stats.tx_errors++;
701 spin_unlock_irqrestore(&port->vio.lock, flags);
702 return NETDEV_TX_BUSY;
705 d = vio_dring_cur(dr);
707 tx_buf = port->tx_bufs[dr->prod].buf;
708 skb_copy_from_linear_data(skb, tx_buf + VNET_PACKET_SKIP, skb->len);
711 if (len < ETH_ZLEN) {
713 memset(tx_buf+VNET_PACKET_SKIP+skb->len, 0, len - skb->len);
716 /* We don't rely on the ACKs to free the skb in vnet_start_xmit(),
717 * thus it is safe to not set VIO_ACK_ENABLE for each transmission:
718 * the protocol itself does not require it as long as the peer
719 * sends a VIO_SUBTYPE_ACK for VIO_DRING_STOPPED.
721 * An ACK for every packet in the ring is expensive as the
722 * sending of LDC messages is slow and affects performance.
724 d->hdr.ack = VIO_ACK_DISABLE;
726 d->ncookies = port->tx_bufs[dr->prod].ncookies;
727 for (i = 0; i < d->ncookies; i++)
728 d->cookies[i] = port->tx_bufs[dr->prod].cookies[i];
730 /* This has to be a non-SMP write barrier because we are writing
731 * to memory which is shared with the peer LDOM.
735 d->hdr.state = VIO_DESC_READY;
737 err = __vnet_tx_trigger(port);
738 if (unlikely(err < 0)) {
739 netdev_info(dev, "TX trigger error %d\n", err);
740 d->hdr.state = VIO_DESC_FREE;
741 dev->stats.tx_carrier_errors++;
742 goto out_dropped_unlock;
745 dev->stats.tx_packets++;
746 dev->stats.tx_bytes += skb->len;
748 dr->prod = (dr->prod + 1) & (VNET_TX_RING_SIZE - 1);
749 if (unlikely(vnet_tx_dring_avail(dr) < 2)) {
750 netif_stop_queue(dev);
751 if (vnet_tx_dring_avail(dr) > VNET_TX_WAKEUP_THRESH(dr))
752 netif_wake_queue(dev);
755 spin_unlock_irqrestore(&port->vio.lock, flags);
762 spin_unlock_irqrestore(&port->vio.lock, flags);
766 dev->stats.tx_dropped++;
770 static void vnet_tx_timeout(struct net_device *dev)
772 /* XXX Implement me XXX */
775 static int vnet_open(struct net_device *dev)
777 netif_carrier_on(dev);
778 netif_start_queue(dev);
783 static int vnet_close(struct net_device *dev)
785 netif_stop_queue(dev);
786 netif_carrier_off(dev);
791 static struct vnet_mcast_entry *__vnet_mc_find(struct vnet *vp, u8 *addr)
793 struct vnet_mcast_entry *m;
795 for (m = vp->mcast_list; m; m = m->next) {
796 if (ether_addr_equal(m->addr, addr))
802 static void __update_mc_list(struct vnet *vp, struct net_device *dev)
804 struct netdev_hw_addr *ha;
806 netdev_for_each_mc_addr(ha, dev) {
807 struct vnet_mcast_entry *m;
809 m = __vnet_mc_find(vp, ha->addr);
816 m = kzalloc(sizeof(*m), GFP_ATOMIC);
819 memcpy(m->addr, ha->addr, ETH_ALEN);
822 m->next = vp->mcast_list;
828 static void __send_mc_list(struct vnet *vp, struct vnet_port *port)
830 struct vio_net_mcast_info info;
831 struct vnet_mcast_entry *m, **pp;
834 memset(&info, 0, sizeof(info));
836 info.tag.type = VIO_TYPE_CTRL;
837 info.tag.stype = VIO_SUBTYPE_INFO;
838 info.tag.stype_env = VNET_MCAST_INFO;
839 info.tag.sid = vio_send_sid(&port->vio);
843 for (m = vp->mcast_list; m; m = m->next) {
847 memcpy(&info.mcast_addr[n_addrs * ETH_ALEN],
849 if (++n_addrs == VNET_NUM_MCAST) {
850 info.count = n_addrs;
852 (void) vio_ldc_send(&port->vio, &info,
858 info.count = n_addrs;
859 (void) vio_ldc_send(&port->vio, &info, sizeof(info));
865 pp = &vp->mcast_list;
866 while ((m = *pp) != NULL) {
873 memcpy(&info.mcast_addr[n_addrs * ETH_ALEN],
875 if (++n_addrs == VNET_NUM_MCAST) {
876 info.count = n_addrs;
877 (void) vio_ldc_send(&port->vio, &info,
886 info.count = n_addrs;
887 (void) vio_ldc_send(&port->vio, &info, sizeof(info));
891 static void vnet_set_rx_mode(struct net_device *dev)
893 struct vnet *vp = netdev_priv(dev);
894 struct vnet_port *port;
897 spin_lock_irqsave(&vp->lock, flags);
898 if (!list_empty(&vp->port_list)) {
899 port = list_entry(vp->port_list.next, struct vnet_port, list);
901 if (port->switch_port) {
902 __update_mc_list(vp, dev);
903 __send_mc_list(vp, port);
906 spin_unlock_irqrestore(&vp->lock, flags);
909 static int vnet_change_mtu(struct net_device *dev, int new_mtu)
911 if (new_mtu != ETH_DATA_LEN)
918 static int vnet_set_mac_addr(struct net_device *dev, void *p)
923 static void vnet_get_drvinfo(struct net_device *dev,
924 struct ethtool_drvinfo *info)
926 strlcpy(info->driver, DRV_MODULE_NAME, sizeof(info->driver));
927 strlcpy(info->version, DRV_MODULE_VERSION, sizeof(info->version));
930 static u32 vnet_get_msglevel(struct net_device *dev)
932 struct vnet *vp = netdev_priv(dev);
933 return vp->msg_enable;
936 static void vnet_set_msglevel(struct net_device *dev, u32 value)
938 struct vnet *vp = netdev_priv(dev);
939 vp->msg_enable = value;
942 static const struct ethtool_ops vnet_ethtool_ops = {
943 .get_drvinfo = vnet_get_drvinfo,
944 .get_msglevel = vnet_get_msglevel,
945 .set_msglevel = vnet_set_msglevel,
946 .get_link = ethtool_op_get_link,
949 static void vnet_port_free_tx_bufs(struct vnet_port *port)
951 struct vio_dring_state *dr;
954 dr = &port->vio.drings[VIO_DRIVER_TX_RING];
956 ldc_free_exp_dring(port->vio.lp, dr->base,
957 (dr->entry_size * dr->num_entries),
958 dr->cookies, dr->ncookies);
966 for (i = 0; i < VNET_TX_RING_SIZE; i++) {
967 void *buf = port->tx_bufs[i].buf;
972 ldc_unmap(port->vio.lp,
973 port->tx_bufs[i].cookies,
974 port->tx_bufs[i].ncookies);
977 port->tx_bufs[i].buf = NULL;
981 static int vnet_port_alloc_tx_bufs(struct vnet_port *port)
983 struct vio_dring_state *dr;
985 int i, err, ncookies;
988 for (i = 0; i < VNET_TX_RING_SIZE; i++) {
989 void *buf = kzalloc(ETH_FRAME_LEN + 8, GFP_KERNEL);
990 int map_len = (ETH_FRAME_LEN + 7) & ~7;
997 if ((unsigned long)buf & (8UL - 1)) {
998 pr_err("TX buffer misaligned\n");
1003 err = ldc_map_single(port->vio.lp, buf, map_len,
1004 port->tx_bufs[i].cookies, 2,
1012 port->tx_bufs[i].buf = buf;
1013 port->tx_bufs[i].ncookies = err;
1016 dr = &port->vio.drings[VIO_DRIVER_TX_RING];
1018 len = (VNET_TX_RING_SIZE *
1019 (sizeof(struct vio_net_desc) +
1020 (sizeof(struct ldc_trans_cookie) * 2)));
1022 ncookies = VIO_MAX_RING_COOKIES;
1023 dring = ldc_alloc_exp_dring(port->vio.lp, len,
1024 dr->cookies, &ncookies,
1028 if (IS_ERR(dring)) {
1029 err = PTR_ERR(dring);
1034 dr->entry_size = (sizeof(struct vio_net_desc) +
1035 (sizeof(struct ldc_trans_cookie) * 2));
1036 dr->num_entries = VNET_TX_RING_SIZE;
1037 dr->prod = dr->cons = 0;
1038 dr->pending = VNET_TX_RING_SIZE;
1039 dr->ncookies = ncookies;
1044 vnet_port_free_tx_bufs(port);
1049 static LIST_HEAD(vnet_list);
1050 static DEFINE_MUTEX(vnet_list_mutex);
1052 static const struct net_device_ops vnet_ops = {
1053 .ndo_open = vnet_open,
1054 .ndo_stop = vnet_close,
1055 .ndo_set_rx_mode = vnet_set_rx_mode,
1056 .ndo_set_mac_address = vnet_set_mac_addr,
1057 .ndo_validate_addr = eth_validate_addr,
1058 .ndo_tx_timeout = vnet_tx_timeout,
1059 .ndo_change_mtu = vnet_change_mtu,
1060 .ndo_start_xmit = vnet_start_xmit,
1063 static struct vnet *vnet_new(const u64 *local_mac)
1065 struct net_device *dev;
1069 dev = alloc_etherdev(sizeof(*vp));
1071 return ERR_PTR(-ENOMEM);
1073 for (i = 0; i < ETH_ALEN; i++)
1074 dev->dev_addr[i] = (*local_mac >> (5 - i) * 8) & 0xff;
1076 vp = netdev_priv(dev);
1078 spin_lock_init(&vp->lock);
1079 tasklet_init(&vp->vnet_tx_wakeup, maybe_tx_wakeup, (unsigned long)vp);
1082 INIT_LIST_HEAD(&vp->port_list);
1083 for (i = 0; i < VNET_PORT_HASH_SIZE; i++)
1084 INIT_HLIST_HEAD(&vp->port_hash[i]);
1085 INIT_LIST_HEAD(&vp->list);
1086 vp->local_mac = *local_mac;
1088 dev->netdev_ops = &vnet_ops;
1089 dev->ethtool_ops = &vnet_ethtool_ops;
1090 dev->watchdog_timeo = VNET_TX_TIMEOUT;
1092 err = register_netdev(dev);
1094 pr_err("Cannot register net device, aborting\n");
1095 goto err_out_free_dev;
1098 netdev_info(dev, "Sun LDOM vnet %pM\n", dev->dev_addr);
1100 list_add(&vp->list, &vnet_list);
1107 return ERR_PTR(err);
1110 static struct vnet *vnet_find_or_create(const u64 *local_mac)
1112 struct vnet *iter, *vp;
1114 mutex_lock(&vnet_list_mutex);
1116 list_for_each_entry(iter, &vnet_list, list) {
1117 if (iter->local_mac == *local_mac) {
1123 vp = vnet_new(local_mac);
1124 mutex_unlock(&vnet_list_mutex);
1129 static void vnet_cleanup(void)
1132 struct net_device *dev;
1134 mutex_lock(&vnet_list_mutex);
1135 while (!list_empty(&vnet_list)) {
1136 vp = list_first_entry(&vnet_list, struct vnet, list);
1137 list_del(&vp->list);
1139 tasklet_kill(&vp->vnet_tx_wakeup);
1140 /* vio_unregister_driver() should have cleaned up port_list */
1141 BUG_ON(!list_empty(&vp->port_list));
1142 unregister_netdev(dev);
1145 mutex_unlock(&vnet_list_mutex);
1148 static const char *local_mac_prop = "local-mac-address";
1150 static struct vnet *vnet_find_parent(struct mdesc_handle *hp,
1153 const u64 *local_mac = NULL;
1156 mdesc_for_each_arc(a, hp, port_node, MDESC_ARC_TYPE_BACK) {
1157 u64 target = mdesc_arc_target(hp, a);
1160 name = mdesc_get_property(hp, target, "name", NULL);
1161 if (!name || strcmp(name, "network"))
1164 local_mac = mdesc_get_property(hp, target,
1165 local_mac_prop, NULL);
1170 return ERR_PTR(-ENODEV);
1172 return vnet_find_or_create(local_mac);
1175 static struct ldc_channel_config vnet_ldc_cfg = {
1176 .event = vnet_event,
1178 .mode = LDC_MODE_UNRELIABLE,
1181 static struct vio_driver_ops vnet_vio_ops = {
1182 .send_attr = vnet_send_attr,
1183 .handle_attr = vnet_handle_attr,
1184 .handshake_complete = vnet_handshake_complete,
1187 static void print_version(void)
1189 printk_once(KERN_INFO "%s", version);
1192 const char *remote_macaddr_prop = "remote-mac-address";
1194 static int vnet_port_probe(struct vio_dev *vdev, const struct vio_device_id *id)
1196 struct mdesc_handle *hp;
1197 struct vnet_port *port;
1198 unsigned long flags;
1201 int len, i, err, switch_port;
1207 vp = vnet_find_parent(hp, vdev->mp);
1209 pr_err("Cannot find port parent vnet\n");
1211 goto err_out_put_mdesc;
1214 rmac = mdesc_get_property(hp, vdev->mp, remote_macaddr_prop, &len);
1217 pr_err("Port lacks %s property\n", remote_macaddr_prop);
1218 goto err_out_put_mdesc;
1221 port = kzalloc(sizeof(*port), GFP_KERNEL);
1224 goto err_out_put_mdesc;
1226 for (i = 0; i < ETH_ALEN; i++)
1227 port->raddr[i] = (*rmac >> (5 - i) * 8) & 0xff;
1231 err = vio_driver_init(&port->vio, vdev, VDEV_NETWORK,
1232 vnet_versions, ARRAY_SIZE(vnet_versions),
1233 &vnet_vio_ops, vp->dev->name);
1235 goto err_out_free_port;
1237 err = vio_ldc_alloc(&port->vio, &vnet_ldc_cfg, port);
1239 goto err_out_free_port;
1241 err = vnet_port_alloc_tx_bufs(port);
1243 goto err_out_free_ldc;
1245 INIT_HLIST_NODE(&port->hash);
1246 INIT_LIST_HEAD(&port->list);
1249 if (mdesc_get_property(hp, vdev->mp, "switch-port", NULL) != NULL)
1251 port->switch_port = switch_port;
1253 spin_lock_irqsave(&vp->lock, flags);
1255 list_add(&port->list, &vp->port_list);
1257 list_add_tail(&port->list, &vp->port_list);
1258 hlist_add_head(&port->hash, &vp->port_hash[vnet_hashfn(port->raddr)]);
1259 spin_unlock_irqrestore(&vp->lock, flags);
1261 dev_set_drvdata(&vdev->dev, port);
1263 pr_info("%s: PORT ( remote-mac %pM%s )\n",
1264 vp->dev->name, port->raddr, switch_port ? " switch-port" : "");
1266 vio_port_up(&port->vio);
1273 vio_ldc_free(&port->vio);
1283 static int vnet_port_remove(struct vio_dev *vdev)
1285 struct vnet_port *port = dev_get_drvdata(&vdev->dev);
1288 struct vnet *vp = port->vp;
1289 unsigned long flags;
1291 del_timer_sync(&port->vio.timer);
1293 spin_lock_irqsave(&vp->lock, flags);
1294 list_del(&port->list);
1295 hlist_del(&port->hash);
1296 spin_unlock_irqrestore(&vp->lock, flags);
1298 vnet_port_free_tx_bufs(port);
1299 vio_ldc_free(&port->vio);
1301 dev_set_drvdata(&vdev->dev, NULL);
1309 static const struct vio_device_id vnet_port_match[] = {
1311 .type = "vnet-port",
1315 MODULE_DEVICE_TABLE(vio, vnet_port_match);
1317 static struct vio_driver vnet_port_driver = {
1318 .id_table = vnet_port_match,
1319 .probe = vnet_port_probe,
1320 .remove = vnet_port_remove,
1321 .name = "vnet_port",
1324 static int __init vnet_init(void)
1326 return vio_register_driver(&vnet_port_driver);
1329 static void __exit vnet_exit(void)
1331 vio_unregister_driver(&vnet_port_driver);
1335 module_init(vnet_init);
1336 module_exit(vnet_exit);