Palacios Public Git Repository

To checkout Palacios execute

  git clone http://v3vee.org/palacios/palacios.web/palacios.git
This will give you the master branch. You probably want the devel branch or one of the release branches. To switch to the devel branch, simply execute
  cd palacios
  git checkout --track -b devel origin/devel
The other branches are similar.


pull packet sending procedure in handle_pkt_tx out of critical section to avoid poten...
Lei Xia [Tue, 24 Apr 2012 19:27:57 +0000 (14:27 -0500)]
palacios/src/devices/lnx_virtio_nic.c

index 400cb0c..4e5f449 100644 (file)
@@ -266,64 +266,70 @@ static int handle_pkt_tx(struct guest_info * core,
                         struct virtio_net_state * virtio_state,
                         int quote)
 {
-    struct virtio_queue *q = &(virtio_state->tx_vq);
+    struct virtio_queue * q;
     int txed = 0, left = 0;
     unsigned long flags;
 
+    q = &(virtio_state->tx_vq);
     if (!q->ring_avail_addr) {
        return -1;
     }
 
-    flags = v3_lock_irqsave(virtio_state->tx_lock);
-    while (q->cur_avail_idx != q->avail->index) {
-       struct virtio_net_hdr_mrg_rxbuf * hdr = NULL;
+    while (1) {
        struct vring_desc * hdr_desc = NULL;
        addr_t hdr_addr = 0;
-       uint16_t desc_idx = q->avail->ring[q->cur_avail_idx % q->queue_size];
-       int desc_cnt = get_desc_count(q, desc_idx);
+       uint16_t desc_idx, tmp_idx;
+       int desc_cnt;
+       
+       flags = v3_lock_irqsave(virtio_state->tx_lock);
+
+       if(q->cur_avail_idx == q->avail->index ||
+           (quote > 0 && txed >= quote)) {
+           left = (q->cur_avail_idx != q->avail->index);
+           break;
+       }
+       
+       desc_idx = q->avail->ring[q->cur_avail_idx % q->queue_size];
+       tmp_idx = q->cur_avail_idx ++;
+       
+       v3_unlock_irqrestore(virtio_state->tx_lock, flags);
 
+       desc_cnt = get_desc_count(q, desc_idx);
        if(desc_cnt != 2){
            PrintError("VNIC: merged rx buffer not supported, desc_cnt %d\n", desc_cnt);
-           goto exit_error;
        }
 
        hdr_desc = &(q->desc[desc_idx]);
-       if (v3_gpa_to_hva(core, hdr_desc->addr_gpa, &(hdr_addr)) == -1) {
-           PrintError("Could not translate block header address\n");
-           goto exit_error;
-       }
+       if (v3_gpa_to_hva(core, hdr_desc->addr_gpa, &(hdr_addr)) != -1) {
+           struct virtio_net_hdr_mrg_rxbuf * hdr;
+           struct vring_desc * buf_desc;
 
-       hdr = (struct virtio_net_hdr_mrg_rxbuf *)hdr_addr;
-       desc_idx = hdr_desc->next;
+           hdr = (struct virtio_net_hdr_mrg_rxbuf *)hdr_addr;
+           desc_idx = hdr_desc->next;
 
-       V3_Net_Print(2, "Virtio NIC: TX hdr count : %d\n", hdr->num_buffers);
-
-       /* here we assumed that one ethernet pkt is not splitted into multiple buffer */        
-       struct vring_desc * buf_desc = &(q->desc[desc_idx]);
-       if (tx_one_pkt(core, virtio_state, buf_desc) == -1) {
-           PrintError("Virtio NIC: Fails to send packet\n");
-       }
-       if(buf_desc->next & VIRTIO_NEXT_FLAG){
-           PrintError("Virtio NIC: TX more buffer need to read\n");
+           /* here we assumed that one ethernet pkt is not splitted into multiple buffer */    
+           buf_desc = &(q->desc[desc_idx]);
+           if (tx_one_pkt(core, virtio_state, buf_desc) == -1) {
+               PrintError("Virtio NIC: Fails to send packet\n");
+           }
+       } else {
+           PrintError("Could not translate block header address\n");
        }
+
+       flags = v3_lock_irqsave(virtio_state->tx_lock);
        
        q->used->ring[q->used->index % q->queue_size].id = 
-           q->avail->ring[q->cur_avail_idx % q->queue_size];
+           q->avail->ring[tmp_idx % q->queue_size];
        
-       q->used->ring[q->used->index % q->queue_size].length = 
-           buf_desc->length; /* What do we set this to???? */
+       //q->used->ring[q->used->index % q->queue_size].length = buf_desc->length; /* What do we set this to???? */
        
        q->used->index ++;
-       q->cur_avail_idx ++;
        
-       if(++txed >= quote && quote > 0){
-           left = (q->cur_avail_idx != q->avail->index);
-           break;
-       }
+       v3_unlock_irqrestore(virtio_state->tx_lock, flags);
+
+       txed ++;
     }
-    
-    v3_unlock_irqrestore(virtio_state->tx_lock, flags);
-    
+        
     if (txed && !(q->avail->flags & VIRTIO_NO_IRQ_FLAG)) {
        v3_pci_raise_irq(virtio_state->virtio_dev->pci_bus, 
                         0, virtio_state->pci_dev);
@@ -331,16 +337,7 @@ static int handle_pkt_tx(struct guest_info * core,
        virtio_state->stats.rx_interrupts ++;
     }
 
-    if(txed > 0) {
-       V3_Net_Print(2, "Virtio Handle TX: txed pkts: %d, left %d\n", txed, left);
-    }
-
     return left;
-    
- exit_error:
-    
-    v3_unlock_irqrestore(virtio_state->tx_lock, flags);
-    return -1;
 }
 
 
@@ -842,6 +839,7 @@ static int register_dev(struct virtio_dev_state * virtio,
     return 0;
 }
 
+#if 0
 #define RATE_UPPER_THRESHOLD 10  /* 10000 pkts per second, around 100Mbits */
 #define RATE_LOWER_THRESHOLD 1
 #define PROFILE_PERIOD 10000 /*us*/
@@ -909,7 +907,7 @@ static void virtio_nic_timer(struct guest_info * core,
 static struct v3_timer_ops timer_ops = {
     .update_timer = virtio_nic_timer,
 };
-
+#endif
 
 static int connect_fn(struct v3_vm_info * info, 
                      void * frontend_data, 
@@ -930,8 +928,7 @@ static int connect_fn(struct v3_vm_info * info,
     net_state->tx_notify = 1;
     net_state->rx_notify = 1;
        
-    net_state->timer = v3_add_timer(&(info->cores[0]),
-                                &timer_ops,net_state);
+    //net_state->timer = v3_add_timer(&(info->cores[0]), &timer_ops,net_state);
 
     ops->recv = virtio_rx;
     ops->poll = virtio_poll;