bitkeeper revision 1.382 (3f27e9e1QG4aZMHik9lRSaa177YIHg)

dev.c, skbuff.h, interrupt.h:
  Fix network receive path to use a softirq to avoid deadlock situations.

diff --git a/xen/include/xeno/interrupt.h b/xen/include/xeno/interrupt.h
index 96cc5779059d63548744080f92dcbf5aacc3a7dd..c37e4efd73e1f4a99adad5ba7cff39c0d30b1fa7 100644 (file)
--- a/xen/include/xeno/interrupt.h
+++ b/xen/include/xeno/interrupt.h
@@ -45,6 +45,7 @@ enum {
 enum
 {
        HI_SOFTIRQ=0,
+       NET_RX_SOFTIRQ,
        AC_TIMER_SOFTIRQ,
        TASKLET_SOFTIRQ
 };

diff --git a/xen/include/xeno/skbuff.h b/xen/include/xeno/skbuff.h
index 80154bce7ba1d513a808e8bd7df672541798e358..30d62155eade121518d68537225c2a519009a8a7 100644 (file)
--- a/xen/include/xeno/skbuff.h
+++ b/xen/include/xeno/skbuff.h
@@ -235,6 +235,31 @@ static inline void __skb_queue_head(struct sk_buff_head *list, struct sk_buff *n
     prev->next = newsk;
 }
 
+/**
+ *      __skb_queue_tail - queue a buffer at the list tail
+ *      @list: list to use
+ *      @newsk: buffer to queue
+ *
+ *      Queue a buffer at the end of a list. This function takes no locks
+ *      and you must therefore hold required locks before calling it.
+ *
+ *      A buffer cannot be placed on two lists at the same time.
+ */ 
+
+static inline void __skb_queue_tail(struct sk_buff_head *list, struct sk_buff *newsk)
+{
+        struct sk_buff *prev, *next;
+
+        newsk->list = list;
+        list->qlen++;
+        next = (struct sk_buff *)list;
+        prev = next->prev;
+        newsk->next = next;
+        newsk->prev = prev;
+        next->prev = newsk;
+        prev->next = newsk;
+}
+
 /**
  *     __skb_dequeue - remove from the head of the queue
  *     @list: list to dequeue from

diff --git a/xen/net/dev.c b/xen/net/dev.c
index 5248a42607f45b47e0e68a6af2d566437fc4f0e5..f31219734f6863f78ae3296dd7c8001996f6d360 100644 (file)
--- a/xen/net/dev.c
+++ b/xen/net/dev.c
@@ -51,6 +51,8 @@
 #define TX_RING_ADD(_i,_j) (((_i)+(_j)) & (TX_RING_SIZE-1))
 #define RX_RING_ADD(_i,_j) (((_i)+(_j)) & (RX_RING_SIZE-1))
 
+static struct sk_buff_head rx_skb_queue[NR_CPUS] __cacheline_aligned;
+
 static void make_tx_response(net_vif_t *vif, 
                              unsigned short id, 
                              unsigned char  st);
@@ -604,39 +606,68 @@ void deliver_packet(struct sk_buff *skb, net_vif_t *vif)
 
 int netif_rx(struct sk_buff *skb)
 {
-    int offset, this_cpu = smp_processor_id();
+    int this_cpu = smp_processor_id();
+    struct sk_buff_head *q = &rx_skb_queue[this_cpu];
     unsigned long flags;
 
+    /* This oughtn't to happen, really! */
+    if ( unlikely(skb_queue_len(q) > 100) )
+    {
+        printk("Congestion -- packet dropped!!\n");
+        return NET_RX_DROP;
+    }
+
     local_irq_save(flags);
+    __skb_queue_tail(q, skb);
+    local_irq_restore(flags);
 
-    ASSERT(skb->skb_type == SKB_ZERO_COPY);
+    __cpu_raise_softirq(this_cpu, NET_RX_SOFTIRQ);
 
-    /*
-     * Offset will include 16 bytes padding from dev_alloc_skb, 14 bytes for 
-     * ethernet header, plus any other alignment padding added by the driver.
-     */
-    offset = (int)skb->data & ~PAGE_MASK; 
-    skb->head = (u8 *)map_domain_mem(((skb->pf - frame_table) << PAGE_SHIFT));
-    skb->data = skb->nh.raw = skb->head + offset;
-    skb->tail = skb->data + skb->len;
-    skb_push(skb, ETH_HLEN);
-    skb->mac.raw = skb->data;
+    return NET_RX_SUCCESS;
+}
 
-    netdev_rx_stat[this_cpu].total++;
+static void net_rx_action(struct softirq_action *h)
+{
+    int offset, this_cpu = smp_processor_id();
+    struct sk_buff_head *q = &rx_skb_queue[this_cpu];
+    struct sk_buff *skb;
 
-    if ( skb->dst_vif == NULL )
-        skb->dst_vif = net_get_target_vif(skb->data, skb->len, skb->src_vif);
-        
-    if ( !VIF_LOCAL(skb->dst_vif) )
-        skb->dst_vif = find_vif_by_id(0);
+    local_irq_disable();
+    
+    while ( (skb = __skb_dequeue(q)) != NULL )
+    {
+        ASSERT(skb->skb_type == SKB_ZERO_COPY);
 
-    deliver_packet(skb, skb->dst_vif);
-    put_vif(skb->dst_vif);
+        /*
+         * Offset will include 16 bytes padding from dev_alloc_skb, 14 bytes
+         * for ethernet header, plus any other alignment padding added by the
+         * driver.
+         */
+        offset = (int)skb->data & ~PAGE_MASK; 
+        skb->head = (u8 *)map_domain_mem(((skb->pf - frame_table) << 
+                                          PAGE_SHIFT));
+        skb->data = skb->nh.raw = skb->head + offset;
+        skb->tail = skb->data + skb->len;
+        skb_push(skb, ETH_HLEN);
+        skb->mac.raw = skb->data;
+        
+        netdev_rx_stat[this_cpu].total++;
+        
+        if ( skb->dst_vif == NULL )
+            skb->dst_vif = net_get_target_vif(
+                skb->data, skb->len, skb->src_vif);
+        
+        if ( !VIF_LOCAL(skb->dst_vif) )
+            skb->dst_vif = find_vif_by_id(0);
+        
+        deliver_packet(skb, skb->dst_vif);
+        put_vif(skb->dst_vif);
+        
+        unmap_domain_mem(skb->head);
+        kfree_skb(skb);
+    }
 
-    unmap_domain_mem(skb->head);
-    kfree_skb(skb);
-    local_irq_restore(flags);
-    return NET_RX_SUCCESS;
+    local_irq_enable();
 }
 
 
@@ -1770,7 +1801,7 @@ long do_net_update(void)
     net_vif_t *vif;
     net_idx_t *shared_idxs;
     unsigned int i, j, idx;
-    struct sk_buff *skb, *interdom_skb = NULL;
+    struct sk_buff *skb;
     tx_req_entry_t tx;
     rx_req_entry_t rx;
     unsigned long pte_pfn, buf_pfn;
@@ -1874,11 +1905,7 @@ long do_net_update(void)
                 skb->len = tx.size - ETH_HLEN;
                 unmap_domain_mem(skb->head);
 
-                /*
-                 * We must defer netif_rx until we have released the current
-                 * domain's page_lock, or we may deadlock on SMP.
-                 */
-                interdom_skb = skb;
+                netif_rx(skb);
 
                 make_tx_response(vif, tx.id, RING_STATUS_OK);
             }
@@ -1907,11 +1934,6 @@ long do_net_update(void)
         tx_unmap_and_continue:
             unmap_domain_mem(g_data);
             spin_unlock_irq(&current->page_lock);
-            if ( interdom_skb != NULL )
-            {
-                (void)netif_rx(interdom_skb);
-                interdom_skb = NULL;
-            }
         }
 
         vif->tx_req_cons = i;
@@ -2069,7 +2091,7 @@ static void make_rx_response(net_vif_t     *vif,
 
 int setup_network_devices(void)
 {
-    int ret;
+    int i, ret;
     extern char opt_ifname[];
     struct net_device *dev = dev_get_by_name(opt_ifname);
 
@@ -2088,6 +2110,10 @@ int setup_network_devices(void)
     printk("Device %s opened and ready for use.\n", opt_ifname);
     the_dev = dev;
 
+    for ( i = 0; i < smp_num_cpus; i++ )
+        skb_queue_head_init(&rx_skb_queue[i]);
+
+    open_softirq(NET_RX_SOFTIRQ, net_rx_action, NULL);
     tasklet_enable(&net_tx_tasklet);
 
     return 1;