[NETDEV] Add notion of link state

Add ability for network devices to flag link up/down state to the
networking core.

Autobooting code will now wait for link-up before attempting DHCP.

IPoIB reflects the Infiniband link state as the network device link state
(which is not strictly correct; we also need a succesful IPoIB IPv4
broadcast group join), but is probably more informative.

13 files changed, 110 insertions, 2 deletions

diff --git a/src/arch/i386/drivers/net/undinet.c b/src/arch/i386/drivers/net/undinet.c
index 07bec560..512c60e9 100644
--- a/src/arch/i386/drivers/net/undinet.c
+++ b/src/arch/i386/drivers/net/undinet.c
@@ -708,6 +708,9 @@ int undinet_probe ( struct undi_device *undi ) {
 		undinic->hacks |= UNDI_HACK_EB54;
 	}
 
+	/* Mark as link up; we don't handle link state */
+	netdev_link_up ( netdev );
+
 	/* Register network device */
 	if ( ( rc = register_netdev ( netdev ) ) != 0 )
 		goto err_register;
diff --git a/src/drivers/net/e1000/e1000.c b/src/drivers/net/e1000/e1000.c
index 739217cf..a9aa508a 100644
--- a/src/drivers/net/e1000/e1000.c
+++ b/src/drivers/net/e1000/e1000.c
@@ -876,6 +876,9 @@ e1000_probe ( struct pci_device *pdev,
 	
 	e1000_get_hw_control ( adapter );
 
+	/* Mark as link up; we don't yet handle link state */
+	netdev_link_up ( netdev );
+
 	if ( ( err = register_netdev ( netdev ) ) != 0)
 		goto err_register;
 		
diff --git a/src/drivers/net/ipoib.c b/src/drivers/net/ipoib.c
index 3b915bf0..e3baa14f 100644
--- a/src/drivers/net/ipoib.c
+++ b/src/drivers/net/ipoib.c
@@ -471,6 +471,12 @@ static int ipoib_transmit ( struct net_device *netdev,
 	}
 	iob_pull ( iobuf, ( sizeof ( *ipoib_pshdr ) ) );
 
+	/* Attempting transmission while link is down will put the
+	 * queue pair into an error state, so don't try it.
+	 */
+	if ( ! ibdev->link_up )
+		return -ENETUNREACH;
+
 	/* Construct address vector */
 	memset ( &av, 0, sizeof ( av ) );
 	av.qkey = IB_GLOBAL_QKEY;
@@ -790,6 +796,10 @@ static int ipoib_join_broadcast_group ( struct ipoib_device *ipoib ) {
 		return rc;
 	}
 
+	/* We will set link up on the network device when we receive
+	 * the broadcast join response.
+	 */
+
 	return 0;
 }
 
@@ -907,16 +917,24 @@ static struct net_device_operations ipoib_operations = {
  */
 static void ipoib_set_ib_params ( struct ipoib_device *ipoib ) {
 	struct ib_device *ibdev = ipoib->ibdev;
+	struct net_device *netdev = ipoib->netdev;
 	struct ipoib_mac *mac;
 
 	/* Calculate GID portion of MAC address based on port GID */
-	mac = ( ( struct ipoib_mac * ) ipoib->netdev->ll_addr );
+	mac = ( ( struct ipoib_mac * ) netdev->ll_addr );
 	memcpy ( &mac->gid, &ibdev->port_gid, sizeof ( mac->gid ) );
 
 	/* Calculate broadcast GID based on partition key */
 	memcpy ( &ipoib->broadcast_gid, &ipv4_broadcast_gid,
 		 sizeof ( ipoib->broadcast_gid ) );
 	ipoib->broadcast_gid.u.words[2] = htons ( ibdev->pkey );
+
+	/* Set net device link state to reflect Infiniband link state */
+	if ( ibdev->link_up ) {
+		netdev_link_up ( netdev );
+	} else {
+		netdev_link_down ( netdev );
+	}
 }
 
 /**
diff --git a/src/drivers/net/legacy.c b/src/drivers/net/legacy.c
index 32460adb..cbec3cf5 100644
--- a/src/drivers/net/legacy.c
+++ b/src/drivers/net/legacy.c
@@ -112,6 +112,9 @@ int legacy_probe ( void *hwdev,
 	 */
 	dev->desc.irq = nic.irqno;
 
+	/* Mark as link up; legacy devices don't handle link state */
+	netdev_link_up ( netdev );
+
 	if ( ( rc = register_netdev ( netdev ) ) != 0 )
 		goto err_register;
 
diff --git a/src/drivers/net/mtnic.c b/src/drivers/net/mtnic.c
index 536fcb8d..57507065 100755
--- a/src/drivers/net/mtnic.c
+++ b/src/drivers/net/mtnic.c
@@ -1731,6 +1731,9 @@ mtnic_probe(struct pci_device *pci,
                 mac = mac >> 8;
 	}
 
+	/* Mark as link up; we don't yet handle link state */
+	netdev_link_up ( dev );
+
 	if (register_netdev(dev)) {
 		eprintf("Netdev registration failed\n");
 		return MTNIC_ERROR;
diff --git a/src/drivers/net/natsemi.c b/src/drivers/net/natsemi.c
index 98a5ff84..028b905c 100644
--- a/src/drivers/net/natsemi.c
+++ b/src/drivers/net/natsemi.c
@@ -205,6 +205,9 @@ static int natsemi_probe (struct pci_device *pci,
 		last = last1;
 	}
 
+	/* Mark as link up; we don't yet handle link state */
+	netdev_link_up ( netdev );
+
 	if ((rc = register_netdev (netdev)) != 0)
 		goto err_register_netdev;
 
diff --git a/src/drivers/net/pnic.c b/src/drivers/net/pnic.c
index b431ec52..c7f08670 100644
--- a/src/drivers/net/pnic.c
+++ b/src/drivers/net/pnic.c
@@ -250,6 +250,9 @@ static int pnic_probe ( struct pci_device *pci,
 	status = pnic_command ( pnic, PNIC_CMD_READ_MAC, NULL, 0,
 				netdev->ll_addr, ETH_ALEN, NULL );
 
+	/* Mark as link up; PNIC has no concept of link state */
+	netdev_link_up ( netdev );
+
 	/* Register network device */
 	if ( ( rc = register_netdev ( netdev ) ) != 0 )
 		goto err;
diff --git a/src/drivers/net/rtl8139.c b/src/drivers/net/rtl8139.c
index c432884c..509047a9 100644
--- a/src/drivers/net/rtl8139.c
+++ b/src/drivers/net/rtl8139.c
@@ -518,6 +518,9 @@ static int rtl_probe ( struct pci_device *pci,
 	rtl_reset ( netdev );
 	rtl_init_eeprom ( netdev );
 	nvs_read ( &rtl->eeprom.nvs, EE_MAC, netdev->ll_addr, ETH_ALEN );
+
+	/* Mark as link up; we don't yet handle link state */
+	netdev_link_up ( netdev );
 	
 	/* Register network device */
 	if ( ( rc = register_netdev ( netdev ) ) != 0 )
diff --git a/src/include/gpxe/errfile.h b/src/include/gpxe/errfile.h
index ae8b1480..011ff1f1 100644
--- a/src/include/gpxe/errfile.h
+++ b/src/include/gpxe/errfile.h
@@ -151,6 +151,7 @@
 #define ERRFILE_uri_test	      ( ERRFILE_OTHER | 0x000b0000 )
 #define ERRFILE_ibft		      ( ERRFILE_OTHER | 0x000c0000 )
 #define ERRFILE_tls		      ( ERRFILE_OTHER | 0x000d0000 )
+#define ERRFILE_ifmgmt		      ( ERRFILE_OTHER | 0x000e0000 )
 
 /** @} */
 
diff --git a/src/include/gpxe/netdevice.h b/src/include/gpxe/netdevice.h
index d8cb84d0..1ef648e1 100644
--- a/src/include/gpxe/netdevice.h
+++ b/src/include/gpxe/netdevice.h
@@ -254,6 +254,9 @@ struct net_device {
 /** Network device is open */
 #define NETDEV_OPEN 0x0001
 
+/** Network device has link */
+#define NETDEV_LINK_UP 0x0002
+
 /** Declare a link-layer protocol */
 #define __ll_protocol  __table ( struct ll_protocol, ll_protocols, 01 )
 
@@ -352,6 +355,37 @@ netdev_settings ( struct net_device *netdev ) {
 	return &netdev->settings.settings;
 }
 
+/**
+ * Mark network device as having link up
+ *
+ * @v netdev		Network device
+ */
+static inline __attribute__ (( always_inline )) void
+netdev_link_up ( struct net_device *netdev ) {
+	netdev->state |= NETDEV_LINK_UP;
+}
+
+/**
+ * Mark network device as having link down
+ *
+ * @v netdev		Network device
+ */
+static inline __attribute__ (( always_inline )) void
+netdev_link_down ( struct net_device *netdev ) {
+	netdev->state &= ~NETDEV_LINK_UP;
+}
+
+/**
+ * Check link state of network device
+ *
+ * @v netdev		Network device
+ * @ret link_up		Link is up
+ */
+static inline __attribute__ (( always_inline )) int
+netdev_link_ok ( struct net_device *netdev ) {
+	return ( netdev->state & NETDEV_LINK_UP );
+}
+
 extern int netdev_tx ( struct net_device *netdev, struct io_buffer *iobuf );
 extern void netdev_tx_complete_err ( struct net_device *netdev,
 				 struct io_buffer *iobuf, int rc );
diff --git a/src/include/usr/ifmgmt.h b/src/include/usr/ifmgmt.h
index c7d35da8..7b49d349 100644
--- a/src/include/usr/ifmgmt.h
+++ b/src/include/usr/ifmgmt.h
@@ -12,5 +12,6 @@ struct net_device;
 extern int ifopen ( struct net_device *netdev );
 extern void ifclose ( struct net_device *netdev );
 extern void ifstat ( struct net_device *netdev );
+extern int iflinkwait ( struct net_device *netdev, unsigned int max_wait_ms );
 
 #endif /* _USR_IFMGMT_H */
diff --git a/src/usr/autoboot.c b/src/usr/autoboot.c
index c1a61ec0..cff6e95d 100644
--- a/src/usr/autoboot.c
+++ b/src/usr/autoboot.c
@@ -38,6 +38,9 @@
  *
  */
 
+/** Time to wait for link-up */
+#define LINK_WAIT_MS 15000
+
 /**
  * Identify the boot network device
  *
@@ -136,6 +139,14 @@ static int netboot ( struct net_device *netdev ) {
 		return rc;
 	ifstat ( netdev );
 
+	/* Wait for link-up */
+	printf ( "Waiting for link-up on %s...", netdev->name );
+	if ( ( rc = iflinkwait ( netdev, LINK_WAIT_MS ) ) != 0 ) {
+		printf ( " no link detected\n" );
+		return rc;
+	}
+	printf ( " ok\n" );
+
 	/* Configure device via DHCP */
 	if ( ( rc = dhcp ( netdev ) ) != 0 )
 		return rc;
diff --git a/src/usr/ifmgmt.c b/src/usr/ifmgmt.c
index 5f4323de..9c88ab53 100644
--- a/src/usr/ifmgmt.c
+++ b/src/usr/ifmgmt.c
@@ -18,8 +18,11 @@
 
 #include <string.h>
 #include <stdio.h>
+#include <unistd.h>
+#include <errno.h>
 #include <gpxe/netdevice.h>
 #include <gpxe/device.h>
+#include <gpxe/process.h>
 #include <usr/ifmgmt.h>
 
 /** @file
@@ -61,9 +64,28 @@ void ifclose ( struct net_device *netdev ) {
  * @v netdev		Network device
  */
 void ifstat ( struct net_device *netdev ) {
-	printf ( "%s: %s on %s (%s) TX:%d TXE:%d RX:%d RXE:%d\n",
+	printf ( "%s: %s on %s (%s)\n"
+		 "  [Link:%s, TX:%d TXE:%d RX:%d RXE:%d]\n",
 		 netdev->name, netdev_hwaddr ( netdev ), netdev->dev->name,
 		 ( ( netdev->state & NETDEV_OPEN ) ? "open" : "closed" ),
+		 ( netdev_link_ok ( netdev ) ? "up" : "down" ),
 		 netdev->stats.tx_ok, netdev->stats.tx_err,
 		 netdev->stats.rx_ok, netdev->stats.rx_err );
 }
+
+/**
+ * Wait for link-up
+ *
+ * @v netdev		Network device
+ * @v max_wait_ms	Maximum time to wait, in ms
+ */
+int iflinkwait ( struct net_device *netdev, unsigned int max_wait_ms ) {
+	while ( 1 ) {
+		if ( netdev_link_ok ( netdev ) )
+			return 0;
+		if ( max_wait_ms-- == 0 )
+			return -ETIMEDOUT;
+		step();
+		mdelay ( 1 );
+	}
+}