[Top][All Lists]
[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]
Re: [Qemu-devel] e1000, virtio_net: Check link status in can_receive
From: |
Mark McLoughlin |
Subject: |
Re: [Qemu-devel] e1000, virtio_net: Check link status in can_receive |
Date: |
Tue, 14 Apr 2009 09:18:09 +0100 |
On Sun, 2009-04-12 at 05:17 -0400, Yan Vugenfirer wrote:
> From: Yan Vugenfirer <address@hidden>
> Date: Sun, 5 Apr 2009 10:20:53 -0400
> Subject: [PATCH 1/1] e1000, virtio_net: Check link status in can_receive
>
> Fixing the bug of 100% cpu usage by qemu after using "set_link <NIC> down"
> monitor command. The fix is for virtio_net and for e1000 emulations.
>
> ---
> hw/e1000.c | 3 ++-
> hw/virtio-net.c | 3 ++-
> 2 files changed, 4 insertions(+), 2 deletions(-)
>
> diff --git a/hw/e1000.c b/hw/e1000.c
> index 1644201..36878d9 100644
> --- a/hw/e1000.c
> +++ b/hw/e1000.c
> @@ -590,7 +590,8 @@ e1000_can_receive(void *opaque)
> {
> E1000State *s = opaque;
>
> - return (s->mac_reg[RCTL] & E1000_RCTL_EN);
> + return (s->mac_reg[RCTL] & E1000_RCTL_EN) &&
> + (s->mac_reg[STATUS] & E1000_STATUS_LU);
> }
>
> static void
> diff --git a/hw/virtio-net.c b/hw/virtio-net.c
> index ad55bb7..dcd18c1 100644
> --- a/hw/virtio-net.c
> +++ b/hw/virtio-net.c
> @@ -259,7 +259,8 @@ static void virtio_net_handle_rx(VirtIODevice *vdev,
> VirtQueue *vq)
> static int do_virtio_net_can_receive(VirtIONet *n, int bufsize)
> {
> if (!virtio_queue_ready(n->rx_vq) ||
> - !(n->vdev.status & VIRTIO_CONFIG_S_DRIVER_OK))
> + !(n->vdev.status & VIRTIO_CONFIG_S_DRIVER_OK) ||
> + !(n->status & VIRTIO_NET_S_LINK_UP))
> return 0;
>
> if (virtio_queue_empty(n->rx_vq) ||
Firstly, I think this "100% CPU" bug is specific to kvm - we have some
code in kvm's tree to buffer tap packets which causes this, I think. I
have some qemu patches in the works to remove this difference between
the trees.
Your patch does make some sense, but I think what we really want is for
qemu_send_packet() to drop the tap packet when virtio/e1000 is down.
This is the behaviour we have in the other direction. Does the patch
below fix the problem for you?
Thanks,
Mark.
diff --git a/qemu/net.c b/qemu/net.c
index 4d07905..95fd808 100644
--- a/qemu/net.c
+++ b/qemu/net.c
@@ -409,8 +409,10 @@ int qemu_send_packet(VLANClientState *vc1, const uint8_t
*buf, int size)
hex_dump(stdout, buf, size);
#endif
for(vc = vlan->first_client; vc != NULL; vc = vc->next) {
- if (vc != vc1 && !vc->link_down) {
- if (!vc->fd_can_read || vc->fd_can_read(vc->opaque)) {
+ if (vc != vc1) {
+ if (vc->link_down)
+ ret = 0;
+ else if (!vc->fd_can_read || vc->fd_can_read(vc->opaque)) {
vc->fd_read(vc->opaque, buf, size);
ret = 0;
}