1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
|
From: Alexandru Avadanii <Alexandru.Avadanii@enea.com>
Date: Tue, 11 Jul 2017 19:06:47 +0200
Subject: [PATCH] lib.sh: AArch64: virt-install: Use virtio-net-pci
AArch64 virt tools (i.e. libvirt) default to "virtio-mmio", instead of
"virtio-net-pci", at least before libvirt 3.x (see [1]).
Without PCI bus info, we can't really enforce a specific order for
the guest ethernet devices.
This leads to a mismatch between cloud-init, who expects "pxe"
network to reside on "eth0", while in fact it ends up on "eth3".
As a consequence, we have no connectivity to the guest, as DHCP
is only configured by cloud-init on "eth0", and also the boot
sequence waits ~5 minutes for "eth0" to come up, before eventually
giving up.
Moreover, predictable network interface naming is out of the question
with virtio-mmio, as there is no bus information to rely on.
Therefore we will enforce "virtio-net-pci" on AArch64, instead of
the default "virtio" (which translates into "virtio-mmio" currently).
FIXME: Apparently, both legacy and modern virtio (0.9 and 1.0) are
active, which leads to triggering bug[2].
For now, throw some sed magic at it and disable legacy mode.
[1] https://www.redhat.com/archives/libvir-list/2016-August/msg00931.html
[2] https://bugzilla.redhat.com/show_bug.cgi?id=1370005
Signed-off-by: Alexandru Avadanii <Alexandru.Avadanii@enea.com>
---
mcp/scripts/lib.sh | 20 ++++++++++++++++----
1 file changed, 16 insertions(+), 4 deletions(-)
diff --git a/mcp/scripts/lib.sh b/mcp/scripts/lib.sh
index 862fdcc..7c4c46f 100644
--- a/mcp/scripts/lib.sh
+++ b/mcp/scripts/lib.sh
@@ -58,24 +58,36 @@ create_vms() {
# AArch64: prepare arch specific arguments
local virt_extra_args=""
+ local virt_pxe_net_model="virtio"
+ local virt_net_model="${virt_pxe_net_model}"
if [ "$(uname -i)" = "aarch64" ]; then
# No Cirrus VGA on AArch64, use vga std
virt_extra_args="$virt_extra_args --video=vga"
+ # AArch64 networking defaults to virtio-mmio before libvirt 3.x
+ virt_net_model="virtio-net-pci"
fi
# create vms with specified options
for node in "${vnodes[@]}"; do
virt-install --name ${node} --ram ${vnodes_ram[$node]} --vcpus ${vnodes_vcpus[$node]} --cpu host --accelerate \
- --network network:pxe,model=virtio \
- --network network:mgmt,model=virtio \
- --network network:internal,model=virtio \
- --network network:public,model=virtio \
+ --network network:pxe,model=${virt_pxe_net_model} \
+ --network network:mgmt,model=${virt_net_model} \
+ --network network:internal,model=${virt_net_model} \
+ --network network:public,model=${virt_net_model} \
--disk path=$(pwd)/images/mcp_${node}.qcow2,format=qcow2,bus=virtio,cache=none,io=native \
--os-type linux --os-variant none \
--boot hd --vnc --console pty --autostart --noreboot \
--disk path=$(pwd)/images/mcp_${node}.iso,device=cdrom \
--noautoconsole \
${virt_extra_args}
+
+ # NOTE(armband): Disable legacy virtio (0.9) in favor of modern virtio (1.0)
+ # http://blog.vmsplice.net/2011/04/how-to-pass-qemu-command-line-options.html
+ # <qemu:commandline>
+ # <qemu:arg value='-global'/>
+ # <qemu:arg value='virtio-pci.disable-legacy=on'/>
+ # </qemu:commandline>
+ EDITOR="sed -i -e \"s|\(<domain.*\)>|\1 xmlns:qemu='http://libvirt.org/schemas/domain/qemu/1.0'>|\" -e \"s|\(</domain>\)|<qemu:commandline><qemu:arg value='-global'/><qemu:arg value='virtio-pci.disable-legacy=on'/></qemu:commandline>\1|\"" virsh edit ${node}
done
}
|
