Re: [RFC PATCH 5/5] virtio: Extend virtio-net PMD to support container environment

On 2016/01/28 15:15, Xie, Huawei wrote:

On 1/28/2016 10:48 AM, Tetsuya Mukawa wrote:

quoted

I measured it, and seems it takes 0.35 seconds in my environment.
This will be done only once when the port is initialized. Probably it's
not so heady.

There are 256 x 32 loop of pci scan. That is too long if we dynamically
start/tear down the container, otherwise it is ok. Some people are
struggling reducing the VM booting time from seconds to milliseconds to
compete with container technology. Let us consider if we could optimize
this.
For example, QEMU supports specifying bus/dev for a device in its
commandline, so could we assign fixed bus for virtio-net and ivshm
device? And for piix3, is it on bus 0/1?

OK, I understand the necessity. Let's consider it.
So far, the users doesn't need to specify pci address on QEMU command
line and DPDK vdev option.
But, let's change this, then we can remove this looping.

Probably specifying pci address on vdev option will not be mandatory.
if not specified, just using default value is nice.
I will fix like above in next release.

Tetsuya