sgi-xp: enable XPNET to handle more than 64 partitions
authorDean Nelson <dcn@sgi.com>
Wed, 30 Jul 2008 05:34:12 +0000 (22:34 -0700)
committerLinus Torvalds <torvalds@linux-foundation.org>
Wed, 30 Jul 2008 16:41:50 +0000 (09:41 -0700)
Enable XPNET to support more than 64 partitions.

Signed-off-by: Dean Nelson <dcn@sgi.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
drivers/misc/sgi-xp/xpnet.c

index f9356ba..c5f59a6 100644 (file)
@@ -21,7 +21,6 @@
  */
 
 #include <linux/module.h>
-#include <linux/types.h>
 #include <linux/kernel.h>
 #include <linux/init.h>
 #include <linux/ioport.h>
@@ -32,8 +31,6 @@
 #include <linux/mii.h>
 #include <linux/smp.h>
 #include <linux/string.h>
-#include <asm/sn/io.h>
-#include <asm/sn/sn_sal.h>
 #include <asm/atomic.h>
 #include "xp.h"
 
@@ -104,7 +101,6 @@ struct xpnet_message {
  * then be released.
  */
 struct xpnet_pending_msg {
-       struct list_head free_list;
        struct sk_buff *skb;
        atomic_t use_count;
 };
@@ -120,7 +116,7 @@ struct net_device *xpnet_device;
  * When we are notified of other partitions activating, we add them to
  * our bitmask of partitions to which we broadcast.
  */
-static u64 xpnet_broadcast_partitions;
+static unsigned long *xpnet_broadcast_partitions;
 /* protect above */
 static DEFINE_SPINLOCK(xpnet_broadcast_lock);
 
@@ -140,16 +136,13 @@ static DEFINE_SPINLOCK(xpnet_broadcast_lock);
 #define XPNET_DEF_MTU (0x8000UL)
 
 /*
- * The partition id is encapsulated in the MAC address.  The following
- * define locates the octet the partid is in.
+ * The partid is encapsulated in the MAC address beginning in the following
+ * octet and it consists of two octets.
  */
-#define XPNET_PARTID_OCTET     1
-#define XPNET_LICENSE_OCTET    2
+#define XPNET_PARTID_OCTET     2
+
+/* Define the XPNET debug device structures to be used with dev_dbg() et al */
 
-/*
- * Define the XPNET debug device structure that is to be used with dev_dbg(),
- * dev_err(), dev_warn(), and dev_info().
- */
 struct device_driver xpnet_dbg_name = {
        .name = "xpnet"
 };
@@ -231,7 +224,7 @@ xpnet_receive(short partid, int channel, struct xpnet_message *msg)
                                          (void *)msg->buf_pa, msg->size);
 
                ret = xp_remote_memcpy((void *)((u64)skb->data &
-                                               ~(L1_CACHE_BYTES - 1)),
+                                               ~(L1_CACHE_BYTES - 1)),
                                       (void *)msg->buf_pa, msg->size);
 
                if (ret != xpSuccess) {
@@ -283,8 +276,6 @@ static void
 xpnet_connection_activity(enum xp_retval reason, short partid, int channel,
                          void *data, void *key)
 {
-       long bp;
-
        DBUG_ON(partid < 0 || partid >= xp_max_npartitions);
        DBUG_ON(channel != XPC_NET_CHANNEL);
 
@@ -297,31 +288,28 @@ xpnet_connection_activity(enum xp_retval reason, short partid, int channel,
 
        case xpConnected:       /* connection completed to a partition */
                spin_lock_bh(&xpnet_broadcast_lock);
-               xpnet_broadcast_partitions |= 1UL << (partid - 1);
-               bp = xpnet_broadcast_partitions;
+               __set_bit(partid, xpnet_broadcast_partitions);
                spin_unlock_bh(&xpnet_broadcast_lock);
 
                netif_carrier_on(xpnet_device);
 
-               dev_dbg(xpnet, "%s connection created to partition %d; "
-                       "xpnet_broadcast_partitions=0x%lx\n",
-                       xpnet_device->name, partid, bp);
+               dev_dbg(xpnet, "%s connected to partition %d\n",
+                       xpnet_device->name, partid);
                break;
 
        default:
                spin_lock_bh(&xpnet_broadcast_lock);
-               xpnet_broadcast_partitions &= ~(1UL << (partid - 1));
-               bp = xpnet_broadcast_partitions;
+               __clear_bit(partid, xpnet_broadcast_partitions);
                spin_unlock_bh(&xpnet_broadcast_lock);
 
-               if (bp == 0)
+               if (bitmap_empty((unsigned long *)xpnet_broadcast_partitions,
+                                xp_max_npartitions)) {
                        netif_carrier_off(xpnet_device);
+               }
 
-               dev_dbg(xpnet, "%s disconnected from partition %d; "
-                       "xpnet_broadcast_partitions=0x%lx\n",
-                       xpnet_device->name, partid, bp);
+               dev_dbg(xpnet, "%s disconnected from partition %d\n",
+                       xpnet_device->name, partid);
                break;
-
        }
 }
 
@@ -424,36 +412,72 @@ xpnet_send_completed(enum xp_retval reason, short partid, int channel,
        }
 }
 
+static void
+xpnet_send(struct sk_buff *skb, struct xpnet_pending_msg *queued_msg,
+          u64 start_addr, u64 end_addr, u16 embedded_bytes, int dest_partid)
+{
+       u8 msg_buffer[XPNET_MSG_SIZE];
+       struct xpnet_message *msg = (struct xpnet_message *)&msg_buffer;
+       enum xp_retval ret;
+
+       msg->embedded_bytes = embedded_bytes;
+       if (unlikely(embedded_bytes != 0)) {
+               msg->version = XPNET_VERSION_EMBED;
+               dev_dbg(xpnet, "calling memcpy(0x%p, 0x%p, 0x%lx)\n",
+                       &msg->data, skb->data, (size_t)embedded_bytes);
+               skb_copy_from_linear_data(skb, &msg->data,
+                                         (size_t)embedded_bytes);
+       } else {
+               msg->version = XPNET_VERSION;
+       }
+       msg->magic = XPNET_MAGIC;
+       msg->size = end_addr - start_addr;
+       msg->leadin_ignore = (u64)skb->data - start_addr;
+       msg->tailout_ignore = end_addr - (u64)skb_tail_pointer(skb);
+       msg->buf_pa = __pa(start_addr);
+
+       dev_dbg(xpnet, "sending XPC message to %d:%d\n"
+               KERN_DEBUG "msg->buf_pa=0x%lx, msg->size=%u, "
+               "msg->leadin_ignore=%u, msg->tailout_ignore=%u\n",
+               dest_partid, XPC_NET_CHANNEL, msg->buf_pa, msg->size,
+               msg->leadin_ignore, msg->tailout_ignore);
+
+       atomic_inc(&queued_msg->use_count);
+
+       ret = xpc_send_notify(dest_partid, XPC_NET_CHANNEL, XPC_NOWAIT, msg,
+                             XPNET_MSG_SIZE, xpnet_send_completed, queued_msg);
+       if (unlikely(ret != xpSuccess))
+               atomic_dec(&queued_msg->use_count);
+}
+
 /*
  * Network layer has formatted a packet (skb) and is ready to place it
  * "on the wire".  Prepare and send an xpnet_message to all partitions
  * which have connected with us and are targets of this packet.
  *
  * MAC-NOTE:  For the XPNET driver, the MAC address contains the
- * destination partition_id.  If the destination partition id word
- * is 0xff, this packet is to broadcast to all partitions.
+ * destination partid.  If the destination partid octets are 0xffff,
+ * this packet is to be broadcast to all connected partitions.
  */
 static int
 xpnet_dev_hard_start_xmit(struct sk_buff *skb, struct net_device *dev)
 {
        struct xpnet_pending_msg *queued_msg;
-       enum xp_retval ret;
-       u8 msg_buffer[XPNET_MSG_SIZE];
-       struct xpnet_message *msg = (struct xpnet_message *)&msg_buffer[0];
        u64 start_addr, end_addr;
-       long dp;
-       u8 second_mac_octet;
        short dest_partid;
-       struct xpnet_dev_private *priv;
-       u16 embedded_bytes;
-
-       priv = (struct xpnet_dev_private *)dev->priv;
+       struct xpnet_dev_private *priv = (struct xpnet_dev_private *)dev->priv;
+       u16 embedded_bytes = 0;
 
        dev_dbg(xpnet, ">skb->head=0x%p skb->data=0x%p skb->tail=0x%p "
                "skb->end=0x%p skb->len=%d\n", (void *)skb->head,
                (void *)skb->data, skb_tail_pointer(skb), skb_end_pointer(skb),
                skb->len);
 
+       if (skb->data[0] == 0x33) {
+               dev_kfree_skb(skb);
+               return 0;       /* nothing needed to be done */
+       }
+
        /*
         * The xpnet_pending_msg tracks how many outstanding
         * xpc_send_notifies are relying on this skb.  When none
@@ -465,7 +489,6 @@ xpnet_dev_hard_start_xmit(struct sk_buff *skb, struct net_device *dev)
                         "packet\n", sizeof(struct xpnet_pending_msg));
 
                priv->stats.tx_errors++;
-
                return -ENOMEM;
        }
 
@@ -474,7 +497,6 @@ xpnet_dev_hard_start_xmit(struct sk_buff *skb, struct net_device *dev)
        end_addr = L1_CACHE_ALIGN((u64)skb_tail_pointer(skb));
 
        /* calculate how many bytes to embed in the XPC message */
-       embedded_bytes = 0;
        if (unlikely(skb->len <= XPNET_MSG_DATA_MAX)) {
                /* skb->data does fit so embed */
                embedded_bytes = skb->len;
@@ -490,78 +512,28 @@ xpnet_dev_hard_start_xmit(struct sk_buff *skb, struct net_device *dev)
        atomic_set(&queued_msg->use_count, 1);
        queued_msg->skb = skb;
 
-       second_mac_octet = skb->data[XPNET_PARTID_OCTET];
-       if (second_mac_octet == 0xff) {
+       if (skb->data[0] == 0xff) {
                /* we are being asked to broadcast to all partitions */
-               dp = xpnet_broadcast_partitions;
-       } else if (second_mac_octet != 0) {
-               dp = xpnet_broadcast_partitions &
-                   (1UL << (second_mac_octet - 1));
-       } else {
-               /* 0 is an invalid partid.  Ignore */
-               dp = 0;
-       }
-       dev_dbg(xpnet, "destination Partitions mask (dp) = 0x%lx\n", dp);
-
-       /*
-        * If we wanted to allow promiscuous mode to work like an
-        * unswitched network, this would be a good point to OR in a
-        * mask of partitions which should be receiving all packets.
-        */
-
-       /*
-        * Main send loop.
-        */
-       for (dest_partid = 0; dp && dest_partid < xp_max_npartitions;
-            dest_partid++) {
+               for_each_bit(dest_partid, xpnet_broadcast_partitions,
+                            xp_max_npartitions) {
 
-               if (!(dp & (1UL << (dest_partid - 1)))) {
-                       /* not destined for this partition */
-                       continue;
+                       xpnet_send(skb, queued_msg, start_addr, end_addr,
+                                  embedded_bytes, dest_partid);
                }
+       } else {
+               dest_partid = (short)skb->data[XPNET_PARTID_OCTET + 1];
+               dest_partid |= (short)skb->data[XPNET_PARTID_OCTET + 0] << 8;
 
-               /* remove this partition from the destinations mask */
-               dp &= ~(1UL << (dest_partid - 1));
-
-               /* found a partition to send to */
+               if (dest_partid >= 0 &&
+                   dest_partid < xp_max_npartitions &&
+                   test_bit(dest_partid, xpnet_broadcast_partitions) != 0) {
 
-               msg->embedded_bytes = embedded_bytes;
-               if (unlikely(embedded_bytes != 0)) {
-                       msg->version = XPNET_VERSION_EMBED;
-                       dev_dbg(xpnet, "calling memcpy(0x%p, 0x%p, 0x%lx)\n",
-                               &msg->data, skb->data, (size_t)embedded_bytes);
-                       skb_copy_from_linear_data(skb, &msg->data,
-                                                 (size_t)embedded_bytes);
-               } else {
-                       msg->version = XPNET_VERSION;
-               }
-               msg->magic = XPNET_MAGIC;
-               msg->size = end_addr - start_addr;
-               msg->leadin_ignore = (u64)skb->data - start_addr;
-               msg->tailout_ignore = end_addr - (u64)skb_tail_pointer(skb);
-               msg->buf_pa = __pa(start_addr);
-
-               dev_dbg(xpnet, "sending XPC message to %d:%d\n"
-                       KERN_DEBUG "msg->buf_pa=0x%lx, msg->size=%u, "
-                       "msg->leadin_ignore=%u, msg->tailout_ignore=%u\n",
-                       dest_partid, XPC_NET_CHANNEL, msg->buf_pa, msg->size,
-                       msg->leadin_ignore, msg->tailout_ignore);
-
-               atomic_inc(&queued_msg->use_count);
-
-               ret = xpc_send_notify(dest_partid, XPC_NET_CHANNEL, XPC_NOWAIT,
-                                     &msg, sizeof(msg) + embedded_bytes - 1,
-                                     xpnet_send_completed, queued_msg);
-               if (unlikely(ret != xpSuccess)) {
-                       atomic_dec(&queued_msg->use_count);
-                       continue;
+                       xpnet_send(skb, queued_msg, start_addr, end_addr,
+                                  embedded_bytes, dest_partid);
                }
        }
 
        if (atomic_dec_return(&queued_msg->use_count) == 0) {
-               dev_dbg(xpnet, "no partitions to receive packet destined for "
-                       "%d\n", dest_partid);
-
                dev_kfree_skb(skb);
                kfree(queued_msg);
        }
@@ -589,23 +561,28 @@ xpnet_dev_tx_timeout(struct net_device *dev)
 static int __init
 xpnet_init(void)
 {
-       int i;
-       u32 license_num;
-       int result = -ENOMEM;
+       int result;
 
-       if (!ia64_platform_is("sn2"))
+       if (!is_shub() && !is_uv())
                return -ENODEV;
 
        dev_info(xpnet, "registering network device %s\n", XPNET_DEVICE_NAME);
 
+       xpnet_broadcast_partitions = kzalloc(BITS_TO_LONGS(xp_max_npartitions) *
+                                            sizeof(long), GFP_KERNEL);
+       if (xpnet_broadcast_partitions == NULL)
+               return -ENOMEM;
+
        /*
         * use ether_setup() to init the majority of our device
         * structure and then override the necessary pieces.
         */
        xpnet_device = alloc_netdev(sizeof(struct xpnet_dev_private),
                                    XPNET_DEVICE_NAME, ether_setup);
-       if (xpnet_device == NULL)
+       if (xpnet_device == NULL) {
+               kfree(xpnet_broadcast_partitions);
                return -ENOMEM;
+       }
 
        netif_carrier_off(xpnet_device);
 
@@ -623,14 +600,10 @@ xpnet_init(void)
         * MAC addresses.  We chose the first octet of the MAC to be unlikely
         * to collide with any vendor's officially issued MAC.
         */
-       xpnet_device->dev_addr[0] = 0xfe;
-       xpnet_device->dev_addr[XPNET_PARTID_OCTET] = sn_partition_id;
-       license_num = sn_partition_serial_number_val();
-       for (i = 3; i >= 0; i--) {
-               xpnet_device->dev_addr[XPNET_LICENSE_OCTET + i] =
-                   license_num & 0xff;
-               license_num = license_num >> 8;
-       }
+       xpnet_device->dev_addr[0] = 0x02;     /* locally administered, no OUI */
+
+       xpnet_device->dev_addr[XPNET_PARTID_OCTET + 1] = sn_partition_id;
+       xpnet_device->dev_addr[XPNET_PARTID_OCTET + 0] = (sn_partition_id >> 8);
 
        /*
         * ether_setup() sets this to a multicast device.  We are
@@ -646,8 +619,10 @@ xpnet_init(void)
        xpnet_device->features = NETIF_F_NO_CSUM;
 
        result = register_netdev(xpnet_device);
-       if (result != 0)
+       if (result != 0) {
                free_netdev(xpnet_device);
+               kfree(xpnet_broadcast_partitions);
+       }
 
        return result;
 }
@@ -661,8 +636,8 @@ xpnet_exit(void)
                 xpnet_device[0].name);
 
        unregister_netdev(xpnet_device);
-
        free_netdev(xpnet_device);
+       kfree(xpnet_broadcast_partitions);
 }
 
 module_exit(xpnet_exit);