[SPARC64]: Add Sun LDOM virtual network driver.
authorDavid S. Miller <davem@sunset.davemloft.net>
Tue, 10 Jul 2007 05:23:51 +0000 (22:23 -0700)
committerDavid S. Miller <davem@sunset.davemloft.net>
Mon, 16 Jul 2007 11:03:23 +0000 (04:03 -0700)
Signed-off-by: David S. Miller <davem@davemloft.net>
drivers/net/Kconfig
drivers/net/Makefile
drivers/net/sunvnet.c [new file with mode: 0644]
drivers/net/sunvnet.h [new file with mode: 0644]

index d17d64eb7065e251e7b6f09f6060a9671b271e03..7903f9c7839e43aaa0feba4e255aa8a859e27750 100644 (file)
@@ -604,6 +604,12 @@ config CASSINI
          Support for the Sun Cassini chip, aka Sun GigaSwift Ethernet. See also
          <http://www.sun.com/products-n-solutions/hardware/docs/pdf/817-4341-10.pdf>
 
+config SUNVNET
+       tristate "Sun Virtual Network support"
+       depends on SUN_LDOMS
+       help
+         Support for virtual network devices under Sun Logical Domains.
+
 config NET_VENDOR_3COM
        bool "3COM cards"
        depends on ISA || EISA || MCA || PCI
index c26b8674213cf33ac6c15fbd7a489b08ef44f550..b95b1b237a2626469d060e817794a969e25c68cf 100644 (file)
@@ -34,6 +34,7 @@ obj-$(CONFIG_SUNBMAC) += sunbmac.o
 obj-$(CONFIG_MYRI_SBUS) += myri_sbus.o
 obj-$(CONFIG_SUNGEM) += sungem.o sungem_phy.o
 obj-$(CONFIG_CASSINI) += cassini.o
+obj-$(CONFIG_SUNVNET) += sunvnet.o
 
 obj-$(CONFIG_MACE) += mace.o
 obj-$(CONFIG_BMAC) += bmac.o
diff --git a/drivers/net/sunvnet.c b/drivers/net/sunvnet.c
new file mode 100644 (file)
index 0000000..d764e4c
--- /dev/null
@@ -0,0 +1,1156 @@
+/* sunvnet.c: Sun LDOM Virtual Network Driver.
+ *
+ * Copyright (C) 2007 David S. Miller <davem@davemloft.net>
+ */
+
+#include <linux/module.h>
+#include <linux/kernel.h>
+#include <linux/types.h>
+#include <linux/slab.h>
+#include <linux/delay.h>
+#include <linux/init.h>
+#include <linux/netdevice.h>
+#include <linux/ethtool.h>
+#include <linux/etherdevice.h>
+
+#include <asm/vio.h>
+#include <asm/ldc.h>
+
+#include "sunvnet.h"
+
+#define DRV_MODULE_NAME                "sunvnet"
+#define PFX DRV_MODULE_NAME    ": "
+#define DRV_MODULE_VERSION     "1.0"
+#define DRV_MODULE_RELDATE     "June 25, 2007"
+
+static char version[] __devinitdata =
+       DRV_MODULE_NAME ".c:v" DRV_MODULE_VERSION " (" DRV_MODULE_RELDATE ")\n";
+MODULE_AUTHOR("David S. Miller (davem@davemloft.net)");
+MODULE_DESCRIPTION("Sun LDOM virtual network driver");
+MODULE_LICENSE("GPL");
+MODULE_VERSION(DRV_MODULE_VERSION);
+
+/* Ordered from largest major to lowest */
+static struct vio_version vnet_versions[] = {
+       { .major = 1, .minor = 0 },
+};
+
+static inline u32 vnet_tx_dring_avail(struct vio_dring_state *dr)
+{
+       return vio_dring_avail(dr, VNET_TX_RING_SIZE);
+}
+
+static int vnet_handle_unknown(struct vnet_port *port, void *arg)
+{
+       struct vio_msg_tag *pkt = arg;
+
+       printk(KERN_ERR PFX "Received unknown msg [%02x:%02x:%04x:%08x]\n",
+              pkt->type, pkt->stype, pkt->stype_env, pkt->sid);
+       printk(KERN_ERR PFX "Resetting connection.\n");
+
+       ldc_disconnect(port->vio.lp);
+
+       return -ECONNRESET;
+}
+
+static int vnet_send_attr(struct vio_driver_state *vio)
+{
+       struct vnet_port *port = to_vnet_port(vio);
+       struct net_device *dev = port->vp->dev;
+       struct vio_net_attr_info pkt;
+       int i;
+
+       memset(&pkt, 0, sizeof(pkt));
+       pkt.tag.type = VIO_TYPE_CTRL;
+       pkt.tag.stype = VIO_SUBTYPE_INFO;
+       pkt.tag.stype_env = VIO_ATTR_INFO;
+       pkt.tag.sid = vio_send_sid(vio);
+       pkt.xfer_mode = VIO_DRING_MODE;
+       pkt.addr_type = VNET_ADDR_ETHERMAC;
+       pkt.ack_freq = 0;
+       for (i = 0; i < 6; i++)
+               pkt.addr |= (u64)dev->dev_addr[i] << ((5 - i) * 8);
+       pkt.mtu = ETH_FRAME_LEN;
+
+       viodbg(HS, "SEND NET ATTR xmode[0x%x] atype[0x%x] addr[%llx] "
+              "ackfreq[%u] mtu[%llu]\n",
+              pkt.xfer_mode, pkt.addr_type,
+              (unsigned long long) pkt.addr,
+              pkt.ack_freq,
+              (unsigned long long) pkt.mtu);
+
+       return vio_ldc_send(vio, &pkt, sizeof(pkt));
+}
+
+static int handle_attr_info(struct vio_driver_state *vio,
+                           struct vio_net_attr_info *pkt)
+{
+       viodbg(HS, "GOT NET ATTR INFO xmode[0x%x] atype[0x%x] addr[%llx] "
+              "ackfreq[%u] mtu[%llu]\n",
+              pkt->xfer_mode, pkt->addr_type,
+              (unsigned long long) pkt->addr,
+              pkt->ack_freq,
+              (unsigned long long) pkt->mtu);
+
+       pkt->tag.sid = vio_send_sid(vio);
+
+       if (pkt->xfer_mode != VIO_DRING_MODE ||
+           pkt->addr_type != VNET_ADDR_ETHERMAC ||
+           pkt->mtu != ETH_FRAME_LEN) {
+               viodbg(HS, "SEND NET ATTR NACK\n");
+
+               pkt->tag.stype = VIO_SUBTYPE_NACK;
+
+               (void) vio_ldc_send(vio, pkt, sizeof(*pkt));
+
+               return -ECONNRESET;
+       } else {
+               viodbg(HS, "SEND NET ATTR ACK\n");
+
+               pkt->tag.stype = VIO_SUBTYPE_ACK;
+
+               return vio_ldc_send(vio, pkt, sizeof(*pkt));
+       }
+
+}
+
+static int handle_attr_ack(struct vio_driver_state *vio,
+                          struct vio_net_attr_info *pkt)
+{
+       viodbg(HS, "GOT NET ATTR ACK\n");
+
+       return 0;
+}
+
+static int handle_attr_nack(struct vio_driver_state *vio,
+                           struct vio_net_attr_info *pkt)
+{
+       viodbg(HS, "GOT NET ATTR NACK\n");
+
+       return -ECONNRESET;
+}
+
+static int vnet_handle_attr(struct vio_driver_state *vio, void *arg)
+{
+       struct vio_net_attr_info *pkt = arg;
+
+       switch (pkt->tag.stype) {
+       case VIO_SUBTYPE_INFO:
+               return handle_attr_info(vio, pkt);
+
+       case VIO_SUBTYPE_ACK:
+               return handle_attr_ack(vio, pkt);
+
+       case VIO_SUBTYPE_NACK:
+               return handle_attr_nack(vio, pkt);
+
+       default:
+               return -ECONNRESET;
+       }
+}
+
+static void vnet_handshake_complete(struct vio_driver_state *vio)
+{
+       struct vio_dring_state *dr;
+
+       dr = &vio->drings[VIO_DRIVER_RX_RING];
+       dr->snd_nxt = dr->rcv_nxt = 1;
+
+       dr = &vio->drings[VIO_DRIVER_TX_RING];
+       dr->snd_nxt = dr->rcv_nxt = 1;
+}
+
+/* The hypervisor interface that implements copying to/from imported
+ * memory from another domain requires that copies are done to 8-byte
+ * aligned buffers, and that the lengths of such copies are also 8-byte
+ * multiples.
+ *
+ * So we align skb->data to an 8-byte multiple and pad-out the data
+ * area so we can round the copy length up to the next multiple of
+ * 8 for the copy.
+ *
+ * The transmitter puts the actual start of the packet 6 bytes into
+ * the buffer it sends over, so that the IP headers after the ethernet
+ * header are aligned properly.  These 6 bytes are not in the descriptor
+ * length, they are simply implied.  This offset is represented using
+ * the VNET_PACKET_SKIP macro.
+ */
+static struct sk_buff *alloc_and_align_skb(struct net_device *dev,
+                                          unsigned int len)
+{
+       struct sk_buff *skb = netdev_alloc_skb(dev, len+VNET_PACKET_SKIP+8+8);
+       unsigned long addr, off;
+
+       if (unlikely(!skb))
+               return NULL;
+
+       addr = (unsigned long) skb->data;
+       off = ((addr + 7UL) & ~7UL) - addr;
+       if (off)
+               skb_reserve(skb, off);
+
+       return skb;
+}
+
+static int vnet_rx_one(struct vnet_port *port, unsigned int len,
+                      struct ldc_trans_cookie *cookies, int ncookies)
+{
+       struct net_device *dev = port->vp->dev;
+       unsigned int copy_len;
+       struct sk_buff *skb;
+       int err;
+
+       err = -EMSGSIZE;
+       if (unlikely(len < ETH_ZLEN || len > ETH_FRAME_LEN)) {
+               dev->stats.rx_length_errors++;
+               goto out_dropped;
+       }
+
+       skb = alloc_and_align_skb(dev, len);
+       err = -ENOMEM;
+       if (unlikely(!skb)) {
+               dev->stats.rx_missed_errors++;
+               goto out_dropped;
+       }
+
+       copy_len = (len + VNET_PACKET_SKIP + 7U) & ~7U;
+       skb_put(skb, copy_len);
+       err = ldc_copy(port->vio.lp, LDC_COPY_IN,
+                      skb->data, copy_len, 0,
+                      cookies, ncookies);
+       if (unlikely(err < 0)) {
+               dev->stats.rx_frame_errors++;
+               goto out_free_skb;
+       }
+
+       skb_pull(skb, VNET_PACKET_SKIP);
+       skb_trim(skb, len);
+       skb->protocol = eth_type_trans(skb, dev);
+
+       dev->stats.rx_packets++;
+       dev->stats.rx_bytes += len;
+
+       netif_rx(skb);
+
+       return 0;
+
+out_free_skb:
+       kfree_skb(skb);
+
+out_dropped:
+       dev->stats.rx_dropped++;
+       return err;
+}
+
+static int vnet_send_ack(struct vnet_port *port, struct vio_dring_state *dr,
+                        u32 start, u32 end, u8 vio_dring_state)
+{
+       struct vio_dring_data hdr = {
+               .tag = {
+                       .type           = VIO_TYPE_DATA,
+                       .stype          = VIO_SUBTYPE_ACK,
+                       .stype_env      = VIO_DRING_DATA,
+                       .sid            = vio_send_sid(&port->vio),
+               },
+               .dring_ident            = dr->ident,
+               .start_idx              = start,
+               .end_idx                = end,
+               .state                  = vio_dring_state,
+       };
+       int err, delay;
+
+       hdr.seq = dr->snd_nxt;
+       delay = 1;
+       do {
+               err = vio_ldc_send(&port->vio, &hdr, sizeof(hdr));
+               if (err > 0) {
+                       dr->snd_nxt++;
+                       break;
+               }
+               udelay(delay);
+               if ((delay <<= 1) > 128)
+                       delay = 128;
+       } while (err == -EAGAIN);
+
+       return err;
+}
+
+static u32 next_idx(u32 idx, struct vio_dring_state *dr)
+{
+       if (++idx == dr->num_entries)
+               idx = 0;
+       return idx;
+}
+
+static u32 prev_idx(u32 idx, struct vio_dring_state *dr)
+{
+       if (idx == 0)
+               idx = dr->num_entries - 1;
+       else
+               idx--;
+
+       return idx;
+}
+
+static struct vio_net_desc *get_rx_desc(struct vnet_port *port,
+                                       struct vio_dring_state *dr,
+                                       u32 index)
+{
+       struct vio_net_desc *desc = port->vio.desc_buf;
+       int err;
+
+       err = ldc_get_dring_entry(port->vio.lp, desc, dr->entry_size,
+                                 (index * dr->entry_size),
+                                 dr->cookies, dr->ncookies);
+       if (err < 0)
+               return ERR_PTR(err);
+
+       return desc;
+}
+
+static int put_rx_desc(struct vnet_port *port,
+                      struct vio_dring_state *dr,
+                      struct vio_net_desc *desc,
+                      u32 index)
+{
+       int err;
+
+       err = ldc_put_dring_entry(port->vio.lp, desc, dr->entry_size,
+                                 (index * dr->entry_size),
+                                 dr->cookies, dr->ncookies);
+       if (err < 0)
+               return err;
+
+       return 0;
+}
+
+static int vnet_walk_rx_one(struct vnet_port *port,
+                           struct vio_dring_state *dr,
+                           u32 index, int *needs_ack)
+{
+       struct vio_net_desc *desc = get_rx_desc(port, dr, index);
+       struct vio_driver_state *vio = &port->vio;
+       int err;
+
+       if (IS_ERR(desc))
+               return PTR_ERR(desc);
+
+       viodbg(DATA, "vio_walk_rx_one desc[%02x:%02x:%08x:%08x:%lx:%lx]\n",
+              desc->hdr.state, desc->hdr.ack,
+              desc->size, desc->ncookies,
+              desc->cookies[0].cookie_addr,
+              desc->cookies[0].cookie_size);
+
+       if (desc->hdr.state != VIO_DESC_READY)
+               return 1;
+       err = vnet_rx_one(port, desc->size, desc->cookies, desc->ncookies);
+       if (err == -ECONNRESET)
+               return err;
+       desc->hdr.state = VIO_DESC_DONE;
+       err = put_rx_desc(port, dr, desc, index);
+       if (err < 0)
+               return err;
+       *needs_ack = desc->hdr.ack;
+       return 0;
+}
+
+static int vnet_walk_rx(struct vnet_port *port, struct vio_dring_state *dr,
+                       u32 start, u32 end)
+{
+       struct vio_driver_state *vio = &port->vio;
+       int ack_start = -1, ack_end = -1;
+
+       end = (end == (u32) -1) ? prev_idx(start, dr) : next_idx(end, dr);
+
+       viodbg(DATA, "vnet_walk_rx start[%08x] end[%08x]\n", start, end);
+
+       while (start != end) {
+               int ack = 0, err = vnet_walk_rx_one(port, dr, start, &ack);
+               if (err == -ECONNRESET)
+                       return err;
+               if (err != 0)
+                       break;
+               if (ack_start == -1)
+                       ack_start = start;
+               ack_end = start;
+               start = next_idx(start, dr);
+               if (ack && start != end) {
+                       err = vnet_send_ack(port, dr, ack_start, ack_end,
+                                           VIO_DRING_ACTIVE);
+                       if (err == -ECONNRESET)
+                               return err;
+                       ack_start = -1;
+               }
+       }
+       if (unlikely(ack_start == -1))
+               ack_start = ack_end = prev_idx(start, dr);
+       return vnet_send_ack(port, dr, ack_start, ack_end, VIO_DRING_STOPPED);
+}
+
+static int vnet_rx(struct vnet_port *port, void *msgbuf)
+{
+       struct vio_dring_data *pkt = msgbuf;
+       struct vio_dring_state *dr = &port->vio.drings[VIO_DRIVER_RX_RING];
+       struct vio_driver_state *vio = &port->vio;
+
+       viodbg(DATA, "vnet_rx stype_env[%04x] seq[%016lx] rcv_nxt[%016lx]\n",
+              pkt->tag.stype_env, pkt->seq, dr->rcv_nxt);
+
+       if (unlikely(pkt->tag.stype_env != VIO_DRING_DATA))
+               return 0;
+       if (unlikely(pkt->seq != dr->rcv_nxt)) {
+               printk(KERN_ERR PFX "RX out of sequence seq[0x%lx] "
+                      "rcv_nxt[0x%lx]\n", pkt->seq, dr->rcv_nxt);
+               return 0;
+       }
+
+       dr->rcv_nxt++;
+
+       /* XXX Validate pkt->start_idx and pkt->end_idx XXX */
+
+       return vnet_walk_rx(port, dr, pkt->start_idx, pkt->end_idx);
+}
+
+static int idx_is_pending(struct vio_dring_state *dr, u32 end)
+{
+       u32 idx = dr->cons;
+       int found = 0;
+
+       while (idx != dr->prod) {
+               if (idx == end) {
+                       found = 1;
+                       break;
+               }
+               idx = next_idx(idx, dr);
+       }
+       return found;
+}
+
+static int vnet_ack(struct vnet_port *port, void *msgbuf)
+{
+       struct vio_dring_state *dr = &port->vio.drings[VIO_DRIVER_TX_RING];
+       struct vio_dring_data *pkt = msgbuf;
+       struct net_device *dev;
+       struct vnet *vp;
+       u32 end;
+
+       if (unlikely(pkt->tag.stype_env != VIO_DRING_DATA))
+               return 0;
+
+       end = pkt->end_idx;
+       if (unlikely(!idx_is_pending(dr, end)))
+               return 0;
+
+       dr->cons = next_idx(end, dr);
+
+       vp = port->vp;
+       dev = vp->dev;
+       if (unlikely(netif_queue_stopped(dev) &&
+                    vnet_tx_dring_avail(dr) >= VNET_TX_WAKEUP_THRESH(dr)))
+               return 1;
+
+       return 0;
+}
+
+static int vnet_nack(struct vnet_port *port, void *msgbuf)
+{
+       /* XXX just reset or similar XXX */
+       return 0;
+}
+
+static void maybe_tx_wakeup(struct vnet *vp)
+{
+       struct net_device *dev = vp->dev;
+
+       netif_tx_lock(dev);
+       if (likely(netif_queue_stopped(dev))) {
+               struct vnet_port *port;
+               int wake = 1;
+
+               list_for_each_entry(port, &vp->port_list, list) {
+                       struct vio_dring_state *dr;
+
+                       dr = &port->vio.drings[VIO_DRIVER_TX_RING];
+                       if (vnet_tx_dring_avail(dr) <
+                           VNET_TX_WAKEUP_THRESH(dr)) {
+                               wake = 0;
+                               break;
+                       }
+               }
+               if (wake)
+                       netif_wake_queue(dev);
+       }
+       netif_tx_unlock(dev);
+}
+
+static void vnet_event(void *arg, int event)
+{
+       struct vnet_port *port = arg;
+       struct vio_driver_state *vio = &port->vio;
+       unsigned long flags;
+       int tx_wakeup, err;
+
+       spin_lock_irqsave(&vio->lock, flags);
+
+       if (unlikely(event == LDC_EVENT_RESET ||
+                    event == LDC_EVENT_UP)) {
+               vio_link_state_change(vio, event);
+               spin_unlock_irqrestore(&vio->lock, flags);
+
+               return;
+       }
+
+       if (unlikely(event != LDC_EVENT_DATA_READY)) {
+               printk(KERN_WARNING PFX "Unexpected LDC event %d\n", event);
+               spin_unlock_irqrestore(&vio->lock, flags);
+               return;
+       }
+
+       tx_wakeup = err = 0;
+       while (1) {
+               union {
+                       struct vio_msg_tag tag;
+                       u64 raw[8];
+               } msgbuf;
+
+               err = ldc_read(vio->lp, &msgbuf, sizeof(msgbuf));
+               if (unlikely(err < 0)) {
+                       if (err == -ECONNRESET)
+                               vio_conn_reset(vio);
+                       break;
+               }
+               if (err == 0)
+                       break;
+               viodbg(DATA, "TAG [%02x:%02x:%04x:%08x]\n",
+                      msgbuf.tag.type,
+                      msgbuf.tag.stype,
+                      msgbuf.tag.stype_env,
+                      msgbuf.tag.sid);
+               err = vio_validate_sid(vio, &msgbuf.tag);
+               if (err < 0)
+                       break;
+
+               if (likely(msgbuf.tag.type == VIO_TYPE_DATA)) {
+                       if (msgbuf.tag.stype == VIO_SUBTYPE_INFO) {
+                               err = vnet_rx(port, &msgbuf);
+                       } else if (msgbuf.tag.stype == VIO_SUBTYPE_ACK) {
+                               err = vnet_ack(port, &msgbuf);
+                               if (err > 0)
+                                       tx_wakeup |= err;
+                       } else if (msgbuf.tag.stype == VIO_SUBTYPE_NACK) {
+                               err = vnet_nack(port, &msgbuf);
+                       }
+               } else if (msgbuf.tag.type == VIO_TYPE_CTRL) {
+                       err = vio_control_pkt_engine(vio, &msgbuf);
+                       if (err)
+                               break;
+               } else {
+                       err = vnet_handle_unknown(port, &msgbuf);
+               }
+               if (err == -ECONNRESET)
+                       break;
+       }
+       spin_unlock(&vio->lock);
+       if (unlikely(tx_wakeup && err != -ECONNRESET))
+               maybe_tx_wakeup(port->vp);
+       local_irq_restore(flags);
+}
+
+static int __vnet_tx_trigger(struct vnet_port *port)
+{
+       struct vio_dring_state *dr = &port->vio.drings[VIO_DRIVER_TX_RING];
+       struct vio_dring_data hdr = {
+               .tag = {
+                       .type           = VIO_TYPE_DATA,
+                       .stype          = VIO_SUBTYPE_INFO,
+                       .stype_env      = VIO_DRING_DATA,
+                       .sid            = vio_send_sid(&port->vio),
+               },
+               .dring_ident            = dr->ident,
+               .start_idx              = dr->prod,
+               .end_idx                = (u32) -1,
+       };
+       int err, delay;
+
+       hdr.seq = dr->snd_nxt;
+       delay = 1;
+       do {
+               err = vio_ldc_send(&port->vio, &hdr, sizeof(hdr));
+               if (err > 0) {
+                       dr->snd_nxt++;
+                       break;
+               }
+               udelay(delay);
+               if ((delay <<= 1) > 128)
+                       delay = 128;
+       } while (err == -EAGAIN);
+
+       return err;
+}
+
+struct vnet_port *__tx_port_find(struct vnet *vp, struct sk_buff *skb)
+{
+       unsigned int hash = vnet_hashfn(skb->data);
+       struct hlist_head *hp = &vp->port_hash[hash];
+       struct hlist_node *n;
+       struct vnet_port *port;
+
+       hlist_for_each_entry(port, n, hp, hash) {
+               if (!compare_ether_addr(port->raddr, skb->data))
+                       return port;
+       }
+       port = NULL;
+       if (!list_empty(&vp->port_list))
+               port = list_entry(vp->port_list.next, struct vnet_port, list);
+
+       return port;
+}
+
+struct vnet_port *tx_port_find(struct vnet *vp, struct sk_buff *skb)
+{
+       struct vnet_port *ret;
+       unsigned long flags;
+
+       spin_lock_irqsave(&vp->lock, flags);
+       ret = __tx_port_find(vp, skb);
+       spin_unlock_irqrestore(&vp->lock, flags);
+
+       return ret;
+}
+
+static int vnet_start_xmit(struct sk_buff *skb, struct net_device *dev)
+{
+       struct vnet *vp = netdev_priv(dev);
+       struct vnet_port *port = tx_port_find(vp, skb);
+       struct vio_dring_state *dr;
+       struct vio_net_desc *d;
+       unsigned long flags;
+       unsigned int len;
+       void *tx_buf;
+       int i, err;
+
+       if (unlikely(!port))
+               goto out_dropped;
+
+       spin_lock_irqsave(&port->vio.lock, flags);
+
+       dr = &port->vio.drings[VIO_DRIVER_TX_RING];
+       if (unlikely(vnet_tx_dring_avail(dr) < 2)) {
+               if (!netif_queue_stopped(dev)) {
+                       netif_stop_queue(dev);
+
+                       /* This is a hard error, log it. */
+                       printk(KERN_ERR PFX "%s: BUG! Tx Ring full when "
+                              "queue awake!\n", dev->name);
+                       dev->stats.tx_errors++;
+               }
+               spin_unlock_irqrestore(&port->vio.lock, flags);
+               return NETDEV_TX_BUSY;
+       }
+
+       d = vio_dring_cur(dr);
+
+       tx_buf = port->tx_bufs[dr->prod].buf;
+       skb_copy_from_linear_data(skb, tx_buf + VNET_PACKET_SKIP, skb->len);
+
+       len = skb->len;
+       if (len < ETH_ZLEN) {
+               len = ETH_ZLEN;
+               memset(tx_buf+VNET_PACKET_SKIP+skb->len, 0, len - skb->len);
+       }
+
+       d->hdr.ack = VIO_ACK_ENABLE;
+       d->size = len;
+       d->ncookies = port->tx_bufs[dr->prod].ncookies;
+       for (i = 0; i < d->ncookies; i++)
+               d->cookies[i] = port->tx_bufs[dr->prod].cookies[i];
+
+       /* This has to be a non-SMP write barrier because we are writing
+        * to memory which is shared with the peer LDOM.
+        */
+       wmb();
+
+       d->hdr.state = VIO_DESC_READY;
+
+       err = __vnet_tx_trigger(port);
+       if (unlikely(err < 0)) {
+               printk(KERN_INFO PFX "%s: TX trigger error %d\n",
+                      dev->name, err);
+               d->hdr.state = VIO_DESC_FREE;
+               dev->stats.tx_carrier_errors++;
+               goto out_dropped_unlock;
+       }
+
+       dev->stats.tx_packets++;
+       dev->stats.tx_bytes += skb->len;
+
+       dr->prod = (dr->prod + 1) & (VNET_TX_RING_SIZE - 1);
+       if (unlikely(vnet_tx_dring_avail(dr) < 2)) {
+               netif_stop_queue(dev);
+               if (vnet_tx_dring_avail(dr) > VNET_TX_WAKEUP_THRESH(dr))
+                       netif_wake_queue(dev);
+       }
+
+       spin_unlock_irqrestore(&port->vio.lock, flags);
+
+       dev_kfree_skb(skb);
+
+       dev->trans_start = jiffies;
+       return NETDEV_TX_OK;
+
+out_dropped_unlock:
+       spin_unlock_irqrestore(&port->vio.lock, flags);
+
+out_dropped:
+       dev_kfree_skb(skb);
+       dev->stats.tx_dropped++;
+       return NETDEV_TX_OK;
+}
+
+static void vnet_tx_timeout(struct net_device *dev)
+{
+       /* XXX Implement me XXX */
+}
+
+static int vnet_open(struct net_device *dev)
+{
+       netif_carrier_on(dev);
+       netif_start_queue(dev);
+
+       return 0;
+}
+
+static int vnet_close(struct net_device *dev)
+{
+       netif_stop_queue(dev);
+       netif_carrier_off(dev);
+
+       return 0;
+}
+
+static void vnet_set_rx_mode(struct net_device *dev)
+{
+       /* XXX Implement multicast support XXX */
+}
+
+static int vnet_change_mtu(struct net_device *dev, int new_mtu)
+{
+       if (new_mtu != ETH_DATA_LEN)
+               return -EINVAL;
+
+       dev->mtu = new_mtu;
+       return 0;
+}
+
+static int vnet_set_mac_addr(struct net_device *dev, void *p)
+{
+       return -EINVAL;
+}
+
+static void vnet_get_drvinfo(struct net_device *dev,
+                            struct ethtool_drvinfo *info)
+{
+       strcpy(info->driver, DRV_MODULE_NAME);
+       strcpy(info->version, DRV_MODULE_VERSION);
+}
+
+static u32 vnet_get_msglevel(struct net_device *dev)
+{
+       struct vnet *vp = netdev_priv(dev);
+       return vp->msg_enable;
+}
+
+static void vnet_set_msglevel(struct net_device *dev, u32 value)
+{
+       struct vnet *vp = netdev_priv(dev);
+       vp->msg_enable = value;
+}
+
+static const struct ethtool_ops vnet_ethtool_ops = {
+       .get_drvinfo            = vnet_get_drvinfo,
+       .get_msglevel           = vnet_get_msglevel,
+       .set_msglevel           = vnet_set_msglevel,
+       .get_link               = ethtool_op_get_link,
+       .get_perm_addr          = ethtool_op_get_perm_addr,
+};
+
+static void vnet_port_free_tx_bufs(struct vnet_port *port)
+{
+       struct vio_dring_state *dr;
+       int i;
+
+       dr = &port->vio.drings[VIO_DRIVER_TX_RING];
+       if (dr->base) {
+               ldc_free_exp_dring(port->vio.lp, dr->base,
+                                  (dr->entry_size * dr->num_entries),
+                                  dr->cookies, dr->ncookies);
+               dr->base = NULL;
+               dr->entry_size = 0;
+               dr->num_entries = 0;
+               dr->pending = 0;
+               dr->ncookies = 0;
+       }
+
+       for (i = 0; i < VNET_TX_RING_SIZE; i++) {
+               void *buf = port->tx_bufs[i].buf;
+
+               if (!buf)
+                       continue;
+
+               ldc_unmap(port->vio.lp,
+                         port->tx_bufs[i].cookies,
+                         port->tx_bufs[i].ncookies);
+
+               kfree(buf);
+               port->tx_bufs[i].buf = NULL;
+       }
+}
+
+static int __devinit vnet_port_alloc_tx_bufs(struct vnet_port *port)
+{
+       struct vio_dring_state *dr;
+       unsigned long len;
+       int i, err, ncookies;
+       void *dring;
+
+       for (i = 0; i < VNET_TX_RING_SIZE; i++) {
+               void *buf = kzalloc(ETH_FRAME_LEN + 8, GFP_KERNEL);
+               int map_len = (ETH_FRAME_LEN + 7) & ~7;
+
+               err = -ENOMEM;
+               if (!buf) {
+                       printk(KERN_ERR "TX buffer allocation failure\n");
+                       goto err_out;
+               }
+               err = -EFAULT;
+               if ((unsigned long)buf & (8UL - 1)) {
+                       printk(KERN_ERR "TX buffer misaligned\n");
+                       kfree(buf);
+                       goto err_out;
+               }
+
+               err = ldc_map_single(port->vio.lp, buf, map_len,
+                                    port->tx_bufs[i].cookies, 2,
+                                    (LDC_MAP_SHADOW |
+                                     LDC_MAP_DIRECT |
+                                     LDC_MAP_RW));
+               if (err < 0) {
+                       kfree(buf);
+                       goto err_out;
+               }
+               port->tx_bufs[i].buf = buf;
+               port->tx_bufs[i].ncookies = err;
+       }
+
+       dr = &port->vio.drings[VIO_DRIVER_TX_RING];
+
+       len = (VNET_TX_RING_SIZE *
+              (sizeof(struct vio_net_desc) +
+               (sizeof(struct ldc_trans_cookie) * 2)));
+
+       ncookies = VIO_MAX_RING_COOKIES;
+       dring = ldc_alloc_exp_dring(port->vio.lp, len,
+                                   dr->cookies, &ncookies,
+                                   (LDC_MAP_SHADOW |
+                                    LDC_MAP_DIRECT |
+                                    LDC_MAP_RW));
+       if (IS_ERR(dring)) {
+               err = PTR_ERR(dring);
+               goto err_out;
+       }
+
+       dr->base = dring;
+       dr->entry_size = (sizeof(struct vio_net_desc) +
+                         (sizeof(struct ldc_trans_cookie) * 2));
+       dr->num_entries = VNET_TX_RING_SIZE;
+       dr->prod = dr->cons = 0;
+       dr->pending = VNET_TX_RING_SIZE;
+       dr->ncookies = ncookies;
+
+       return 0;
+
+err_out:
+       vnet_port_free_tx_bufs(port);
+
+       return err;
+}
+
+static struct ldc_channel_config vnet_ldc_cfg = {
+       .event          = vnet_event,
+       .mtu            = 64,
+       .mode           = LDC_MODE_UNRELIABLE,
+};
+
+static struct vio_driver_ops vnet_vio_ops = {
+       .send_attr              = vnet_send_attr,
+       .handle_attr            = vnet_handle_attr,
+       .handshake_complete     = vnet_handshake_complete,
+};
+
+const char *remote_macaddr_prop = "remote-mac-address";
+
+static int __devinit vnet_port_probe(struct vio_dev *vdev,
+                                    const struct vio_device_id *id)
+{
+       struct mdesc_node *endp;
+       struct vnet_port *port;
+       unsigned long flags;
+       struct vnet *vp;
+       const u64 *rmac;
+       int len, i, err, switch_port;
+
+       vp = dev_get_drvdata(vdev->dev.parent);
+       if (!vp) {
+               printk(KERN_ERR PFX "Cannot find port parent vnet.\n");
+               return -ENODEV;
+       }
+
+       rmac = md_get_property(vdev->mp, remote_macaddr_prop, &len);
+       if (!rmac) {
+               printk(KERN_ERR PFX "Port lacks %s property.\n",
+                      remote_macaddr_prop);
+               return -ENODEV;
+       }
+
+       endp = vio_find_endpoint(vdev);
+       if (!endp) {
+               printk(KERN_ERR PFX "Port lacks channel-endpoint.\n");
+               return -ENODEV;
+       }
+
+       port = kzalloc(sizeof(*port), GFP_KERNEL);
+       if (!port) {
+               printk(KERN_ERR PFX "Cannot allocate vnet_port.\n");
+               return -ENOMEM;
+       }
+
+       for (i = 0; i < ETH_ALEN; i++)
+               port->raddr[i] = (*rmac >> (5 - i) * 8) & 0xff;
+
+       port->vp = vp;
+
+       err = vio_driver_init(&port->vio, vdev, VDEV_NETWORK, endp,
+                             vnet_versions, ARRAY_SIZE(vnet_versions),
+                             &vnet_vio_ops, vp->dev->name);
+       if (err)
+               goto err_out_free_port;
+
+       err = vio_ldc_alloc(&port->vio, &vnet_ldc_cfg, port);
+       if (err)
+               goto err_out_free_port;
+
+       err = vnet_port_alloc_tx_bufs(port);
+       if (err)
+               goto err_out_free_ldc;
+
+       INIT_HLIST_NODE(&port->hash);
+       INIT_LIST_HEAD(&port->list);
+
+       switch_port = 0;
+       if (md_get_property(vdev->mp, "switch-port", NULL) != NULL)
+               switch_port = 1;
+
+       spin_lock_irqsave(&vp->lock, flags);
+       if (switch_port)
+               list_add(&port->list, &vp->port_list);
+       else
+               list_add_tail(&port->list, &vp->port_list);
+       hlist_add_head(&port->hash, &vp->port_hash[vnet_hashfn(port->raddr)]);
+       spin_unlock_irqrestore(&vp->lock, flags);
+
+       dev_set_drvdata(&vdev->dev, port);
+
+       printk(KERN_INFO "%s: PORT ( remote-mac ", vp->dev->name);
+       for (i = 0; i < 6; i++)
+               printk("%2.2x%c", port->raddr[i], i == 5 ? ' ' : ':');
+       if (switch_port)
+               printk("switch-port ");
+       printk(")\n");
+
+       vio_port_up(&port->vio);
+
+       return 0;
+
+err_out_free_ldc:
+       vio_ldc_free(&port->vio);
+
+err_out_free_port:
+       kfree(port);
+
+       return err;
+}
+
+static int vnet_port_remove(struct vio_dev *vdev)
+{
+       struct vnet_port *port = dev_get_drvdata(&vdev->dev);
+
+       if (port) {
+               struct vnet *vp = port->vp;
+               unsigned long flags;
+
+               del_timer_sync(&port->vio.timer);
+
+               spin_lock_irqsave(&vp->lock, flags);
+               list_del(&port->list);
+               hlist_del(&port->hash);
+               spin_unlock_irqrestore(&vp->lock, flags);
+
+               vnet_port_free_tx_bufs(port);
+               vio_ldc_free(&port->vio);
+
+               dev_set_drvdata(&vdev->dev, NULL);
+
+               kfree(port);
+       }
+       return 0;
+}
+
+static struct vio_device_id vnet_port_match[] = {
+       {
+               .type = "vnet-port",
+       },
+       {},
+};
+MODULE_DEVICE_TABLE(vio, vnet_match);
+
+static struct vio_driver vnet_port_driver = {
+       .id_table       = vnet_port_match,
+       .probe          = vnet_port_probe,
+       .remove         = vnet_port_remove,
+       .driver         = {
+               .name   = "vnet_port",
+               .owner  = THIS_MODULE,
+       }
+};
+
+const char *local_mac_prop = "local-mac-address";
+
+static int __devinit vnet_probe(struct vio_dev *vdev,
+                               const struct vio_device_id *id)
+{
+       static int vnet_version_printed;
+       struct net_device *dev;
+       struct vnet *vp;
+       const u64 *mac;
+       int err, i, len;
+
+       if (vnet_version_printed++ == 0)
+               printk(KERN_INFO "%s", version);
+
+       mac = md_get_property(vdev->mp, local_mac_prop, &len);
+       if (!mac) {
+               printk(KERN_ERR PFX "vnet lacks %s property.\n",
+                      local_mac_prop);
+               err = -ENODEV;
+               goto err_out;
+       }
+
+       dev = alloc_etherdev(sizeof(*vp));
+       if (!dev) {
+               printk(KERN_ERR PFX "Etherdev alloc failed, aborting.\n");
+               err = -ENOMEM;
+               goto err_out;
+       }
+
+       for (i = 0; i < ETH_ALEN; i++)
+               dev->dev_addr[i] = (*mac >> (5 - i) * 8) & 0xff;
+
+       memcpy(dev->perm_addr, dev->dev_addr, dev->addr_len);
+
+       SET_NETDEV_DEV(dev, &vdev->dev);
+
+       vp = netdev_priv(dev);
+
+       spin_lock_init(&vp->lock);
+       vp->dev = dev;
+       vp->vdev = vdev;
+
+       INIT_LIST_HEAD(&vp->port_list);
+       for (i = 0; i < VNET_PORT_HASH_SIZE; i++)
+               INIT_HLIST_HEAD(&vp->port_hash[i]);
+
+       dev->open = vnet_open;
+       dev->stop = vnet_close;
+       dev->set_multicast_list = vnet_set_rx_mode;
+       dev->set_mac_address = vnet_set_mac_addr;
+       dev->tx_timeout = vnet_tx_timeout;
+       dev->ethtool_ops = &vnet_ethtool_ops;
+       dev->watchdog_timeo = VNET_TX_TIMEOUT;
+       dev->change_mtu = vnet_change_mtu;
+       dev->hard_start_xmit = vnet_start_xmit;
+
+       err = register_netdev(dev);
+       if (err) {
+               printk(KERN_ERR PFX "Cannot register net device, "
+                      "aborting.\n");
+               goto err_out_free_dev;
+       }
+
+       printk(KERN_INFO "%s: Sun LDOM vnet ", dev->name);
+
+       for (i = 0; i < 6; i++)
+               printk("%2.2x%c", dev->dev_addr[i], i == 5 ? '\n' : ':');
+
+       dev_set_drvdata(&vdev->dev, vp);
+
+       return 0;
+
+err_out_free_dev:
+       free_netdev(dev);
+
+err_out:
+       return err;
+}
+
+static int vnet_remove(struct vio_dev *vdev)
+{
+
+       struct vnet *vp = dev_get_drvdata(&vdev->dev);
+
+       if (vp) {
+               /* XXX unregister port, or at least check XXX */
+               unregister_netdevice(vp->dev);
+               dev_set_drvdata(&vdev->dev, NULL);
+       }
+       return 0;
+}
+
+static struct vio_device_id vnet_match[] = {
+       {
+               .type = "network",
+       },
+       {},
+};
+MODULE_DEVICE_TABLE(vio, vnet_match);
+
+static struct vio_driver vnet_driver = {
+       .id_table       = vnet_match,
+       .probe          = vnet_probe,
+       .remove         = vnet_remove,
+       .driver         = {
+               .name   = "vnet",
+               .owner  = THIS_MODULE,
+       }
+};
+
+static int __init vnet_init(void)
+{
+       int err = vio_register_driver(&vnet_driver);
+
+       if (!err) {
+               err = vio_register_driver(&vnet_port_driver);
+               if (err)
+                       vio_unregister_driver(&vnet_driver);
+       }
+
+       return err;
+}
+
+static void __exit vnet_exit(void)
+{
+       vio_unregister_driver(&vnet_port_driver);
+       vio_unregister_driver(&vnet_driver);
+}
+
+module_init(vnet_init);
+module_exit(vnet_exit);
diff --git a/drivers/net/sunvnet.h b/drivers/net/sunvnet.h
new file mode 100644 (file)
index 0000000..1c88730
--- /dev/null
@@ -0,0 +1,70 @@
+#ifndef _SUNVNET_H
+#define _SUNVNET_H
+
+#define DESC_NCOOKIES(entry_size)      \
+       ((entry_size) - sizeof(struct vio_net_desc))
+
+/* length of time before we decide the hardware is borked,
+ * and dev->tx_timeout() should be called to fix the problem
+ */
+#define VNET_TX_TIMEOUT                        (5 * HZ)
+
+#define VNET_TX_RING_SIZE              512
+#define VNET_TX_WAKEUP_THRESH(dr)      ((dr)->pending / 4)
+
+/* VNET packets are sent in buffers with the first 6 bytes skipped
+ * so that after the ethernet header the IPv4/IPv6 headers are aligned
+ * properly.
+ */
+#define VNET_PACKET_SKIP               6
+
+struct vnet_tx_entry {
+       void                    *buf;
+       unsigned int            ncookies;
+       struct ldc_trans_cookie cookies[2];
+};
+
+struct vnet;
+struct vnet_port {
+       struct vio_driver_state vio;
+
+       struct hlist_node       hash;
+       u8                      raddr[ETH_ALEN];
+
+       struct vnet             *vp;
+
+       struct vnet_tx_entry    tx_bufs[VNET_TX_RING_SIZE];
+
+       struct list_head        list;
+};
+
+static inline struct vnet_port *to_vnet_port(struct vio_driver_state *vio)
+{
+       return container_of(vio, struct vnet_port, vio);
+}
+
+#define VNET_PORT_HASH_SIZE    16
+#define VNET_PORT_HASH_MASK    (VNET_PORT_HASH_SIZE - 1)
+
+static inline unsigned int vnet_hashfn(u8 *mac)
+{
+       unsigned int val = mac[4] ^ mac[5];
+
+       return val & (VNET_PORT_HASH_MASK);
+}
+
+struct vnet {
+       /* Protects port_list and port_hash.  */
+       spinlock_t              lock;
+
+       struct net_device       *dev;
+
+       u32                     msg_enable;
+       struct vio_dev          *vdev;
+
+       struct list_head        port_list;
+
+       struct hlist_head       port_hash[VNET_PORT_HASH_SIZE];
+};
+
+#endif /* _SUNVNET_H */