[libvirt PATCH] Adds e1000e/vmxnet3 Vnet_hdr suuport

Laine Stump laine at redhat.com
Fri Aug 7 23:32:29 UTC 2020


On 8/7/20 3:46 PM, Patrick J. Magauran wrote:
> On Fri, 2020-08-07 at 17:16 +0100, Daniel P. Berrangé wrote:
>> On Thu, Aug 06, 2020 at 10:53:32PM -0400, Patrick Magauran wrote:
>>> Libvirt bases its decision about whether to apply the vnet_hdr flag
>>> to the tap interface on whether or not the selected model is
>>> VirtIO. Originally, VirtIO was the only model to support the
>>> vnet_hdr in QEMU; however, the e1000e & vmxnet3 adapters also
>>> support it(seemingly from introduction based on commits). This
>>> passes the whole packet to the host, reducing emulation overhead
>>> and improving performance.
>>>
>>> Signed-off-by: Patrick Magauran <patmagauran.j at gmail.com>
>>> ---
>>>   src/conf/domain_conf.c    | 8 +++++++-
>>>   src/conf/domain_conf.h    | 1 +
>>>   src/libvirt_private.syms  | 1 +
>>>   src/qemu/qemu_interface.c | 8 ++++----
>>>   4 files changed, 13 insertions(+), 5 deletions(-)
>>>
>>> diff --git a/src/conf/domain_conf.c b/src/conf/domain_conf.c
>>> index 69e0439e7e..cb184110f7 100644
>>> --- a/src/conf/domain_conf.c
>>> +++ b/src/conf/domain_conf.c
>>> @@ -30981,7 +30981,13 @@ virDomainNetIsVirtioModel(const
>>> virDomainNetDef *net)
>>>               net->model == VIR_DOMAIN_NET_MODEL_VIRTIO_TRANSITIONAL
>>> ||
>>>               net->model ==
>>> VIR_DOMAIN_NET_MODEL_VIRTIO_NON_TRANSITIONAL);
>>>   }
>>> -
>>> +bool
>>> +virDomainNetIsVnetCompatModel(const virDomainNetDef *net)
>>> +{
>>> +    return (virDomainNetIsVirtioModel(net) ||
>>> +            net->model == VIR_DOMAIN_NET_MODEL_E1000E ||
>>> +            net->model == VIR_DOMAIN_NET_MODEL_VMXNET3);
>>> +}
>> Should really live in the qemu_inteface.c file,  with
>> a qemuOnterfaceIsVnetCompatModel name, since the decision
>> is inherantly tied to QEMU, and inappropriate for other
>> virt drivers.
>>
>>
>> Regards,
>> Daniel
> I would think the qemu_interface.c file is more for the actual setup
> functionality rather than simple helper functions. Would the
> qemu_domain.c file make for sense?
>

If it's network interface specific and also qemu specific, then it 
should go into qemu_interface.c. That file didn't use to exist (all the 
functions there were just included in qemu_command.c) but was created as 
a place to split out all the interface-specific stuff (not just the 
toplevel "setup this kind of interface" functions).




More information about the libvir-list mailing list