net: Optimize hard_start_xmit() return checking
[safe/jmp/linux-2.6] / net / core / netpoll.c
index 496b062..0b4d0d3 100644 (file)
@@ -9,7 +9,7 @@
  * Copyright (C) 2002  Red Hat, Inc.
  */
 
-#include <linux/smp_lock.h>
+#include <linux/moduleparam.h>
 #include <linux/netdevice.h>
 #include <linux/etherdevice.h>
 #include <linux/string.h>
@@ -25,6 +25,7 @@
 #include <net/tcp.h>
 #include <net/udp.h>
 #include <asm/unaligned.h>
+#include <trace/events/napi.h>
 
 /*
  * We maintain a small pool of fully-sized skbs, to make sure the
@@ -50,6 +51,9 @@ static atomic_t trapped;
 static void zap_completion_queue(void);
 static void arp_reply(struct sk_buff *skb);
 
+static unsigned int carrier_timeout = 4;
+module_param(carrier_timeout, uint, 0644);
+
 static void queue_process(struct work_struct *work)
 {
        struct netpoll_info *npinfo =
@@ -59,24 +63,29 @@ static void queue_process(struct work_struct *work)
 
        while ((skb = skb_dequeue(&npinfo->txq))) {
                struct net_device *dev = skb->dev;
+               const struct net_device_ops *ops = dev->netdev_ops;
+               struct netdev_queue *txq;
 
                if (!netif_device_present(dev) || !netif_running(dev)) {
                        __kfree_skb(skb);
                        continue;
                }
 
+               txq = netdev_get_tx_queue(dev, skb_get_queue_mapping(skb));
+
                local_irq_save(flags);
-               netif_tx_lock(dev);
-               if (netif_queue_stopped(dev) ||
-                   dev->hard_start_xmit(skb, dev) != NETDEV_TX_OK) {
+               __netif_tx_lock(txq, smp_processor_id());
+               if (netif_tx_queue_stopped(txq) ||
+                   netif_tx_queue_frozen(txq) ||
+                   ops->ndo_start_xmit(skb, dev) != NETDEV_TX_OK) {
                        skb_queue_head(&npinfo->txq, skb);
-                       netif_tx_unlock(dev);
+                       __netif_tx_unlock(txq);
                        local_irq_restore(flags);
 
                        schedule_delayed_work(&npinfo->tx_work, HZ/10);
                        return;
                }
-               netif_tx_unlock(dev);
+               __netif_tx_unlock(txq);
                local_irq_restore(flags);
        }
 }
@@ -86,7 +95,7 @@ static __sum16 checksum_udp(struct sk_buff *skb, struct udphdr *uh,
 {
        __wsum psum;
 
-       if (uh->check == 0 || skb->ip_summed == CHECKSUM_UNNECESSARY)
+       if (uh->check == 0 || skb_csum_unnecessary(skb))
                return 0;
 
        psum = csum_tcpudp_nofold(saddr, daddr, ulen, IPPROTO_UDP, 0);
@@ -116,51 +125,77 @@ static __sum16 checksum_udp(struct sk_buff *skb, struct udphdr *uh,
  * network adapter, forcing superfluous retries and possibly timeouts.
  * Thus, we set our budget to greater than 1.
  */
-static void poll_napi(struct netpoll *np)
+static int poll_one_napi(struct netpoll_info *npinfo,
+                        struct napi_struct *napi, int budget)
 {
-       struct netpoll_info *npinfo = np->dev->npinfo;
-       int budget = 16;
+       int work;
 
-       if (test_bit(__LINK_STATE_RX_SCHED, &np->dev->state) &&
-           npinfo->poll_owner != smp_processor_id() &&
-           spin_trylock(&npinfo->poll_lock)) {
-               npinfo->rx_flags |= NETPOLL_RX_DROP;
-               atomic_inc(&trapped);
+       /* net_rx_action's ->poll() invocations and our's are
+        * synchronized by this test which is only made while
+        * holding the napi->poll_lock.
+        */
+       if (!test_bit(NAPI_STATE_SCHED, &napi->state))
+               return budget;
 
-               np->dev->poll(np->dev, &budget);
+       npinfo->rx_flags |= NETPOLL_RX_DROP;
+       atomic_inc(&trapped);
+       set_bit(NAPI_STATE_NPSVC, &napi->state);
 
-               atomic_dec(&trapped);
-               npinfo->rx_flags &= ~NETPOLL_RX_DROP;
-               spin_unlock(&npinfo->poll_lock);
-       }
+       work = napi->poll(napi, budget);
+       trace_napi_poll(napi);
+
+       clear_bit(NAPI_STATE_NPSVC, &napi->state);
+       atomic_dec(&trapped);
+       npinfo->rx_flags &= ~NETPOLL_RX_DROP;
+
+       return budget - work;
 }
 
-static void service_arp_queue(struct netpoll_info *npi)
+static void poll_napi(struct net_device *dev)
 {
-       struct sk_buff *skb;
+       struct napi_struct *napi;
+       int budget = 16;
 
-       if (unlikely(!npi))
-               return;
+       list_for_each_entry(napi, &dev->napi_list, dev_list) {
+               if (napi->poll_owner != smp_processor_id() &&
+                   spin_trylock(&napi->poll_lock)) {
+                       budget = poll_one_napi(dev->npinfo, napi, budget);
+                       spin_unlock(&napi->poll_lock);
+
+                       if (!budget)
+                               break;
+               }
+       }
+}
 
-       skb = skb_dequeue(&npi->arp_tx);
+static void service_arp_queue(struct netpoll_info *npi)
+{
+       if (npi) {
+               struct sk_buff *skb;
 
-       while (skb != NULL) {
-               arp_reply(skb);
-               skb = skb_dequeue(&npi->arp_tx);
+               while ((skb = skb_dequeue(&npi->arp_tx)))
+                       arp_reply(skb);
        }
 }
 
 void netpoll_poll(struct netpoll *np)
 {
-       if (!np->dev || !netif_running(np->dev) || !np->dev->poll_controller)
+       struct net_device *dev = np->dev;
+       const struct net_device_ops *ops;
+
+       if (!dev || !netif_running(dev))
+               return;
+
+       ops = dev->netdev_ops;
+       if (!ops->ndo_poll_controller)
                return;
 
        /* Process pending work on NIC */
-       np->dev->poll_controller(np->dev);
-       if (np->dev->poll)
-               poll_napi(np);
+       ops->ndo_poll_controller(dev);
+
+       poll_napi(dev);
 
-       service_arp_queue(np->dev->npinfo);
+       service_arp_queue(dev->npinfo);
 
        zap_completion_queue();
 }
@@ -197,10 +232,12 @@ static void zap_completion_queue(void)
                while (clist != NULL) {
                        struct sk_buff *skb = clist;
                        clist = clist->next;
-                       if (skb->destructor)
+                       if (skb->destructor) {
+                               atomic_inc(&skb->users);
                                dev_kfree_skb_any(skb); /* put this one back */
-                       else
+                       } else {
                                __kfree_skb(skb);
+                       }
                }
        }
 
@@ -233,11 +270,23 @@ repeat:
        return skb;
 }
 
+static int netpoll_owner_active(struct net_device *dev)
+{
+       struct napi_struct *napi;
+
+       list_for_each_entry(napi, &dev->napi_list, dev_list) {
+               if (napi->poll_owner == smp_processor_id())
+                       return 1;
+       }
+       return 0;
+}
+
 static void netpoll_send_skb(struct netpoll *np, struct sk_buff *skb)
 {
        int status = NETDEV_TX_BUSY;
        unsigned long tries;
        struct net_device *dev = np->dev;
+       const struct net_device_ops *ops = dev->netdev_ops;
        struct netpoll_info *npinfo = np->dev->npinfo;
 
        if (!npinfo || !netif_running(dev) || !netif_device_present(dev)) {
@@ -246,28 +295,39 @@ static void netpoll_send_skb(struct netpoll *np, struct sk_buff *skb)
        }
 
        /* don't get messages out of order, and no recursion */
-       if (skb_queue_len(&npinfo->txq) == 0 &&
-                   npinfo->poll_owner != smp_processor_id()) {
+       if (skb_queue_len(&npinfo->txq) == 0 && !netpoll_owner_active(dev)) {
+               struct netdev_queue *txq;
                unsigned long flags;
 
+               txq = netdev_get_tx_queue(dev, skb_get_queue_mapping(skb));
+
                local_irq_save(flags);
-               if (netif_tx_trylock(dev)) {
-                       /* try until next clock tick */
-                       for (tries = jiffies_to_usecs(1)/USEC_PER_POLL;
-                                       tries > 0; --tries) {
-                               if (!netif_queue_stopped(dev))
-                                       status = dev->hard_start_xmit(skb, dev);
+               /* try until next clock tick */
+               for (tries = jiffies_to_usecs(1)/USEC_PER_POLL;
+                    tries > 0; --tries) {
+                       if (__netif_tx_trylock(txq)) {
+                               if (!netif_tx_queue_stopped(txq)) {
+                                       status = ops->ndo_start_xmit(skb, dev);
+                                       if (status == NETDEV_TX_OK)
+                                               txq_trans_update(txq);
+                               }
+                               __netif_tx_unlock(txq);
 
                                if (status == NETDEV_TX_OK)
                                        break;
 
-                               /* tickle device maybe there is some cleanup */
-                               netpoll_poll(np);
-
-                               udelay(USEC_PER_POLL);
                        }
-                       netif_tx_unlock(dev);
+
+                       /* tickle device maybe there is some cleanup */
+                       netpoll_poll(np);
+
+                       udelay(USEC_PER_POLL);
                }
+
+               WARN_ONCE(!irqs_disabled(),
+                       "netpoll_send_skb(): %s enabled interrupts in poll (%pF)\n",
+                       dev->name, ops->ndo_start_xmit);
+
                local_irq_restore(flags);
        }
 
@@ -293,18 +353,20 @@ void netpoll_send_udp(struct netpoll *np, const char *msg, int len)
        if (!skb)
                return;
 
-       memcpy(skb->data, msg, len);
+       skb_copy_to_linear_data(skb, msg, len);
        skb->len += len;
 
-       skb->h.uh = udph = (struct udphdr *) skb_push(skb, sizeof(*udph));
+       skb_push(skb, sizeof(*udph));
+       skb_reset_transport_header(skb);
+       udph = udp_hdr(skb);
        udph->source = htons(np->local_port);
        udph->dest = htons(np->remote_port);
        udph->len = htons(udp_len);
        udph->check = 0;
-       udph->check = csum_tcpudp_magic(htonl(np->local_ip),
-                                       htonl(np->remote_ip),
+       udph->check = csum_tcpudp_magic(np->local_ip,
+                                       np->remote_ip,
                                        udp_len, IPPROTO_UDP,
-                                       csum_partial((unsigned char *)udph, udp_len, 0));
+                                       csum_partial(udph, udp_len, 0));
        if (udph->check == 0)
                udph->check = CSUM_MANGLED_0;
 
@@ -321,15 +383,15 @@ void netpoll_send_udp(struct netpoll *np, const char *msg, int len)
        iph->ttl      = 64;
        iph->protocol = IPPROTO_UDP;
        iph->check    = 0;
-       put_unaligned(htonl(np->local_ip), &(iph->saddr));
-       put_unaligned(htonl(np->remote_ip), &(iph->daddr));
+       put_unaligned(np->local_ip, &(iph->saddr));
+       put_unaligned(np->remote_ip, &(iph->daddr));
        iph->check    = ip_fast_csum((unsigned char *)iph, iph->ihl);
 
        eth = (struct ethhdr *) skb_push(skb, ETH_HLEN);
        skb_reset_mac_header(skb);
        skb->protocol = eth->h_proto = htons(ETH_P_IP);
-       memcpy(eth->h_source, np->local_mac, 6);
-       memcpy(eth->h_dest, np->remote_mac, 6);
+       memcpy(eth->h_source, np->dev->dev_addr, ETH_ALEN);
+       memcpy(eth->h_dest, np->remote_mac, ETH_ALEN);
 
        skb->dev = np->dev;
 
@@ -356,13 +418,11 @@ static void arp_reply(struct sk_buff *skb)
        if (skb->dev->flags & IFF_NOARP)
                return;
 
-       if (!pskb_may_pull(skb, (sizeof(struct arphdr) +
-                                (2 * skb->dev->addr_len) +
-                                (2 * sizeof(u32)))))
+       if (!pskb_may_pull(skb, arp_hdr_len(skb->dev)))
                return;
 
        skb_reset_network_header(skb);
-       skb->h.raw = skb->data;
+       skb_reset_transport_header(skb);
        arp = arp_hdr(skb);
 
        if ((arp->ar_hrd != htons(ARPHRD_ETHER) &&
@@ -382,11 +442,12 @@ static void arp_reply(struct sk_buff *skb)
        memcpy(&tip, arp_ptr, 4);
 
        /* Should we ignore arp? */
-       if (tip != htonl(np->local_ip) || LOOPBACK(tip) || MULTICAST(tip))
+       if (tip != np->local_ip ||
+           ipv4_is_loopback(tip) || ipv4_is_multicast(tip))
                return;
 
-       size = sizeof(struct arphdr) + 2 * (skb->dev->addr_len + 4);
-       send_skb = find_skb(np, size + LL_RESERVED_SPACE(np->dev),
+       size = arp_hdr_len(skb->dev);
+       send_skb = find_skb(np, size + LL_ALLOCATED_SPACE(np->dev),
                            LL_RESERVED_SPACE(np->dev));
 
        if (!send_skb)
@@ -398,11 +459,9 @@ static void arp_reply(struct sk_buff *skb)
        send_skb->protocol = htons(ETH_P_ARP);
 
        /* Fill the device header for the ARP frame */
-
-       if (np->dev->hard_header &&
-           np->dev->hard_header(send_skb, skb->dev, ptype,
-                                sha, np->local_mac,
-                                send_skb->len) < 0) {
+       if (dev_hard_header(send_skb, skb->dev, ptype,
+                           sha, np->dev->dev_addr,
+                           send_skb->len) < 0) {
                kfree_skb(send_skb);
                return;
        }
@@ -492,9 +551,9 @@ int __netpoll_rx(struct sk_buff *skb)
                goto out;
        if (checksum_udp(skb, uh, ulen, iph->saddr, iph->daddr))
                goto out;
-       if (np->local_ip && np->local_ip != ntohl(iph->daddr))
+       if (np->local_ip && np->local_ip != iph->daddr)
                goto out;
-       if (np->remote_ip && np->remote_ip != ntohl(iph->saddr))
+       if (np->remote_ip && np->remote_ip != iph->saddr)
                goto out;
        if (np->local_port && np->local_port != ntohs(uh->dest))
                goto out;
@@ -515,6 +574,22 @@ out:
        return 0;
 }
 
+void netpoll_print_options(struct netpoll *np)
+{
+       printk(KERN_INFO "%s: local port %d\n",
+                        np->name, np->local_port);
+       printk(KERN_INFO "%s: local IP %pI4\n",
+                        np->name, &np->local_ip);
+       printk(KERN_INFO "%s: interface %s\n",
+                        np->name, np->dev_name);
+       printk(KERN_INFO "%s: remote port %d\n",
+                        np->name, np->remote_port);
+       printk(KERN_INFO "%s: remote IP %pI4\n",
+                        np->name, &np->remote_ip);
+       printk(KERN_INFO "%s: remote ethernet address %pM\n",
+                        np->name, np->remote_mac);
+}
+
 int netpoll_parse_options(struct netpoll *np, char *opt)
 {
        char *cur=opt, *delim;
@@ -527,17 +602,13 @@ int netpoll_parse_options(struct netpoll *np, char *opt)
                cur = delim;
        }
        cur++;
-       printk(KERN_INFO "%s: local port %d\n", np->name, np->local_port);
 
        if (*cur != '/') {
                if ((delim = strchr(cur, '/')) == NULL)
                        goto parse_failed;
                *delim = 0;
-               np->local_ip = ntohl(in_aton(cur));
+               np->local_ip = in_aton(cur);
                cur = delim;
-
-               printk(KERN_INFO "%s: local IP %d.%d.%d.%d\n",
-                      np->name, HIPQUAD(np->local_ip));
        }
        cur++;
 
@@ -551,8 +622,6 @@ int netpoll_parse_options(struct netpoll *np, char *opt)
        }
        cur++;
 
-       printk(KERN_INFO "%s: interface %s\n", np->name, np->dev_name);
-
        if (*cur != '@') {
                /* dst port */
                if ((delim = strchr(cur, '@')) == NULL)
@@ -562,18 +631,14 @@ int netpoll_parse_options(struct netpoll *np, char *opt)
                cur = delim;
        }
        cur++;
-       printk(KERN_INFO "%s: remote port %d\n", np->name, np->remote_port);
 
        /* dst ip */
        if ((delim = strchr(cur, '/')) == NULL)
                goto parse_failed;
        *delim = 0;
-       np->remote_ip = ntohl(in_aton(cur));
+       np->remote_ip = in_aton(cur);
        cur = delim + 1;
 
-       printk(KERN_INFO "%s: remote IP %d.%d.%d.%d\n",
-              np->name, HIPQUAD(np->remote_ip));
-
        if (*cur != 0) {
                /* MAC address */
                if ((delim = strchr(cur, ':')) == NULL)
@@ -604,15 +669,7 @@ int netpoll_parse_options(struct netpoll *np, char *opt)
                np->remote_mac[5] = simple_strtol(cur, NULL, 16);
        }
 
-       printk(KERN_INFO "%s: remote ethernet address "
-              "%02x:%02x:%02x:%02x:%02x:%02x\n",
-              np->name,
-              np->remote_mac[0],
-              np->remote_mac[1],
-              np->remote_mac[2],
-              np->remote_mac[3],
-              np->remote_mac[4],
-              np->remote_mac[5]);
+       netpoll_print_options(np);
 
        return 0;
 
@@ -631,7 +688,7 @@ int netpoll_setup(struct netpoll *np)
        int err;
 
        if (np->dev_name)
-               ndev = dev_get_by_name(np->dev_name);
+               ndev = dev_get_by_name(&init_net, np->dev_name);
        if (!ndev) {
                printk(KERN_ERR "%s: %s doesn't exist, aborting.\n",
                       np->name, np->dev_name);
@@ -648,8 +705,6 @@ int netpoll_setup(struct netpoll *np)
 
                npinfo->rx_flags = 0;
                npinfo->rx_np = NULL;
-               spin_lock_init(&npinfo->poll_lock);
-               npinfo->poll_owner = -1;
 
                spin_lock_init(&npinfo->rx_lock);
                skb_queue_head_init(&npinfo->arp_tx);
@@ -662,7 +717,7 @@ int netpoll_setup(struct netpoll *np)
                atomic_inc(&npinfo->refcnt);
        }
 
-       if (!ndev->poll_controller) {
+       if (!ndev->netdev_ops->ndo_poll_controller) {
                printk(KERN_ERR "%s: %s doesn't support polling, aborting.\n",
                       np->name, np->dev_name);
                err = -ENOTSUPP;
@@ -686,7 +741,7 @@ int netpoll_setup(struct netpoll *np)
                }
 
                atleast = jiffies + HZ/10;
-               atmost = jiffies + 4*HZ;
+               atmost = jiffies + carrier_timeout * HZ;
                while (!netif_carrier_ok(ndev)) {
                        if (time_after(jiffies, atmost)) {
                                printk(KERN_NOTICE
@@ -694,7 +749,7 @@ int netpoll_setup(struct netpoll *np)
                                       np->name);
                                break;
                        }
-                       cond_resched();
+                       msleep(1);
                }
 
                /* If carrier appears to come up instantly, we don't
@@ -710,9 +765,6 @@ int netpoll_setup(struct netpoll *np)
                }
        }
 
-       if (is_zero_ether_addr(np->local_mac) && ndev->dev_addr)
-               memcpy(np->local_mac, ndev->dev_addr, 6);
-
        if (!np->local_ip) {
                rcu_read_lock();
                in_dev = __in_dev_get_rcu(ndev);
@@ -725,10 +777,9 @@ int netpoll_setup(struct netpoll *np)
                        goto release;
                }
 
-               np->local_ip = ntohl(in_dev->ifa_list->ifa_local);
+               np->local_ip = in_dev->ifa_list->ifa_local;
                rcu_read_unlock();
-               printk(KERN_INFO "%s: local IP %d.%d.%d.%d\n",
-                      np->name, HIPQUAD(np->local_ip));
+               printk(KERN_INFO "%s: local IP %pI4\n", np->name, &np->local_ip);
        }
 
        if (np->rx_hook) {
@@ -779,14 +830,15 @@ void netpoll_cleanup(struct netpoll *np)
                                spin_unlock_irqrestore(&npinfo->rx_lock, flags);
                        }
 
-                       np->dev->npinfo = NULL;
                        if (atomic_dec_and_test(&npinfo->refcnt)) {
                                skb_queue_purge(&npinfo->arp_tx);
                                skb_queue_purge(&npinfo->txq);
                                cancel_rearming_delayed_work(&npinfo->tx_work);
-                               flush_scheduled_work();
 
+                               /* clean after last, unfinished work */
+                               __skb_queue_purge(&npinfo->txq);
                                kfree(npinfo);
+                               np->dev->npinfo = NULL;
                        }
                }
 
@@ -811,6 +863,7 @@ void netpoll_set_trap(int trap)
 
 EXPORT_SYMBOL(netpoll_set_trap);
 EXPORT_SYMBOL(netpoll_trap);
+EXPORT_SYMBOL(netpoll_print_options);
 EXPORT_SYMBOL(netpoll_parse_options);
 EXPORT_SYMBOL(netpoll_setup);
 EXPORT_SYMBOL(netpoll_cleanup);