Re: [Qemu-devel] [Bug 1286253] [NEW] virtio-net acceleration features not set when plugged into backend dynamically

[Mario Smarduch]

I don't have an x86 setup handy, but looking through the code 'virtio-net-pci' 
will
work but  not 'virtio-net-device'. My understanding was  the goal is for any
backend to plug into any transport  on any platform. I'll try  an x86 test with 
 
'virtio-net-device' option and  see what happens. 

Yes I did check the options on host after device is plugged and guest after 
features
are probed, only the MAC feature is set.

Thanks,
- Mario

-----Original Message-----
From: address@hidden [mailto:address@hidden On Behalf Of Michael S. Tsirkin
Sent: Monday, March 03, 2014 4:49 AM
To: Stefan Hajnoczi
Cc: Bug 1286253; address@hidden
Subject: Re: [Qemu-devel] [Bug 1286253] [NEW] virtio-net acceleration features 
not set when plugged into backend dynamically

On Mon, Mar 03, 2014 at 01:37:46PM +0100, Stefan Hajnoczi wrote:
> On Fri, Feb 28, 2014 at 05:40:19PM -0000, Mario Smarduch wrote:
> > When using indpendent transport and backend in this case virtio-net- 
> > device transport, none of the acceleration features are set after 
> > guest probes the transport the backend is plugged into. For 
> > virtio-net this leads to low throughput/performance.  This holds 
> > true for  virtio-mmio, PCI transports and most likely for others as 
> > well (CCW, S390) and other backends
> > 
> > Command to run:
> > ./qemu-system-arm -enable-kvm -smp 2 -kernel zImage -dtb ./guest-a15.dtb -m 
> > 512 -M vexpress-a15 -cpu cortex-a15 -nographic \
> >       -append "root=/dev/vda rw console=ttyAMA0 rootwait" -drive 
> > if=none,file=/mnt/gauss.root,id=vm1 \
> >       -device virtio-blk-device,drive=vm1 -netdev 
> > type=tap,id=net0,ifname=tap0 \
> >       -device virtio-net-device,netdev=net0,mac="52:54:00:12:34:58"
> > 
> > For x86 same virtio command for network.
> 
> Can you explain in more detail?
> 
> The command-line looks sane.  The virtio-net-device instance should 
> figure out the tap supports offloads.
> 
> Did you try adding a printf to virtio_net_get_features() to see why 
> the offload features are not being detected?
> 
> Stefan

IIUC mmio does not set any feature bits, so you get the fallback behaviour:

$ git grep DEFINE_VIRTIO_NET_FEATURES
hw/s390x/s390-virtio-bus.c: DEFINE_VIRTIO_NET_FEATURES(VirtIOS390Device, 
host_featur
hw/s390x/virtio-ccw.c:    DEFINE_VIRTIO_NET_FEATURES(VirtioCcwDevice, 
host_features[0])
hw/virtio/virtio-pci.c:    DEFINE_VIRTIO_NET_FEATURES(VirtIOPCIProxy, 
host_features),
include/hw/virtio/virtio-net.h:#define DEFINE_VIRTIO_NET_FEATURES(_state, 
_field) \

But it should work for pci, and certainly does for x86.


--
MST