virtio: net refill on out-of-memory

If we run out of memory, use keventd to fill the buffer.  There's a
report of this happening: "Page allocation failures in guest",
Message-ID: <20090713115158.0a4892b0@mjolnir.ossman.eu>

Signed-off-by: Rusty Russell <rusty@rustcorp.com.au>
Signed-off-by: David S. Miller <davem@davemloft.net>

diff --git a/drivers/net/virtio_net.c b/drivers/net/virtio_net.c
index 2a6e81d..bbedf03 100644 (file)
--- a/drivers/net/virtio_net.c
+++ b/drivers/net/virtio_net.c
@@ -70,6 +70,9 @@ struct virtnet_info
        struct sk_buff_head recv;
        struct sk_buff_head send;
 
+       /* Work struct for refilling if we run low on memory. */
+       struct delayed_work refill;
+
        /* Chain pages by the private ptr. */
        struct page *pages;
 };
@@ -273,19 +276,22 @@ drop:
        dev_kfree_skb(skb);
 }
 
-static void try_fill_recv_maxbufs(struct virtnet_info *vi)
+static bool try_fill_recv_maxbufs(struct virtnet_info *vi, gfp_t gfp)
 {
        struct sk_buff *skb;
        struct scatterlist sg[2+MAX_SKB_FRAGS];
        int num, err, i;
+       bool oom = false;
 
        sg_init_table(sg, 2+MAX_SKB_FRAGS);
        for (;;) {
                struct virtio_net_hdr *hdr;
 
                skb = netdev_alloc_skb(vi->dev, MAX_PACKET_LEN + NET_IP_ALIGN);
-               if (unlikely(!skb))
+               if (unlikely(!skb)) {
+                       oom = true;
                        break;
+               }
 
                skb_reserve(skb, NET_IP_ALIGN);
                skb_put(skb, MAX_PACKET_LEN);
@@ -296,7 +302,7 @@ static void try_fill_recv_maxbufs(struct virtnet_info *vi)
                if (vi->big_packets) {
                        for (i = 0; i < MAX_SKB_FRAGS; i++) {
                                skb_frag_t *f = &skb_shinfo(skb)->frags[i];
-                               f->page = get_a_page(vi, GFP_ATOMIC);
+                               f->page = get_a_page(vi, gfp);
                                if (!f->page)
                                        break;
 
@@ -325,31 +331,35 @@ static void try_fill_recv_maxbufs(struct virtnet_info *vi)
        if (unlikely(vi->num > vi->max))
                vi->max = vi->num;
        vi->rvq->vq_ops->kick(vi->rvq);
+       return !oom;
 }
 
-static void try_fill_recv(struct virtnet_info *vi)
+/* Returns false if we couldn't fill entirely (OOM). */
+static bool try_fill_recv(struct virtnet_info *vi, gfp_t gfp)
 {
        struct sk_buff *skb;
        struct scatterlist sg[1];
        int err;
+       bool oom = false;
 
-       if (!vi->mergeable_rx_bufs) {
-               try_fill_recv_maxbufs(vi);
-               return;
-       }
+       if (!vi->mergeable_rx_bufs)
+               return try_fill_recv_maxbufs(vi, gfp);
 
        for (;;) {
                skb_frag_t *f;
 
                skb = netdev_alloc_skb(vi->dev, GOOD_COPY_LEN + NET_IP_ALIGN);
-               if (unlikely(!skb))
+               if (unlikely(!skb)) {
+                       oom = true;
                        break;
+               }
 
                skb_reserve(skb, NET_IP_ALIGN);
 
                f = &skb_shinfo(skb)->frags[0];
-               f->page = get_a_page(vi, GFP_ATOMIC);
+               f->page = get_a_page(vi, gfp);
                if (!f->page) {
+                       oom = true;
                        kfree_skb(skb);
                        break;
                }
@@ -373,6 +383,7 @@ static void try_fill_recv(struct virtnet_info *vi)
        if (unlikely(vi->num > vi->max))
                vi->max = vi->num;
        vi->rvq->vq_ops->kick(vi->rvq);
+       return !oom;
 }
 
 static void skb_recv_done(struct virtqueue *rvq)
@@ -385,6 +396,23 @@ static void skb_recv_done(struct virtqueue *rvq)
        }
 }
 
+static void refill_work(struct work_struct *work)
+{
+       struct virtnet_info *vi;
+       bool still_empty;
+
+       vi = container_of(work, struct virtnet_info, refill.work);
+       napi_disable(&vi->napi);
+       try_fill_recv(vi, GFP_KERNEL);
+       still_empty = (vi->num == 0);
+       napi_enable(&vi->napi);
+
+       /* In theory, this can happen: if we don't get any buffers in
+        * we will *never* try to fill again. */
+       if (still_empty)
+               schedule_delayed_work(&vi->refill, HZ/2);
+}
+
 static int virtnet_poll(struct napi_struct *napi, int budget)
 {
        struct virtnet_info *vi = container_of(napi, struct virtnet_info, napi);
@@ -400,10 +428,10 @@ again:
                received++;
        }
 
-       /* FIXME: If we oom and completely run out of inbufs, we need
-        * to start a timer trying to fill more. */
-       if (vi->num < vi->max / 2)
-               try_fill_recv(vi);
+       if (vi->num < vi->max / 2) {
+               if (!try_fill_recv(vi, GFP_ATOMIC))
+                       schedule_delayed_work(&vi->refill, 0);
+       }
 
        /* Out of packets? */
        if (received < budget) {
@@ -893,6 +921,7 @@ static int virtnet_probe(struct virtio_device *vdev)
        vi->vdev = vdev;
        vdev->priv = vi;
        vi->pages = NULL;
+       INIT_DELAYED_WORK(&vi->refill, refill_work);
 
        /* If they give us a callback when all buffers are done, we don't need
         * the timer. */
@@ -941,7 +970,7 @@ static int virtnet_probe(struct virtio_device *vdev)
        }
 
        /* Last of all, set up some receive buffers. */
-       try_fill_recv(vi);
+       try_fill_recv(vi, GFP_KERNEL);
 
        /* If we didn't even get one input buffer, we're useless. */
        if (vi->num == 0) {
@@ -958,6 +987,7 @@ static int virtnet_probe(struct virtio_device *vdev)
 
 unregister:
        unregister_netdev(dev);
+       cancel_delayed_work_sync(&vi->refill);
 free_vqs:
        vdev->config->del_vqs(vdev);
 free:
@@ -986,6 +1016,7 @@ static void virtnet_remove(struct virtio_device *vdev)
        BUG_ON(vi->num != 0);
 
        unregister_netdev(vi->dev);
+       cancel_delayed_work_sync(&vi->refill);
 
        vdev->config->del_vqs(vi->vdev);