Re: [Qemu-devel] [PATCH v4 00/32] Dynamic module loading for block drivers

[Paolo Bonzini]

On 19/07/2016 15:47, Colin Lord wrote:
> However, the protocol drivers do tend to see noticeable performance
> benefits. Most of them get modularized in the first three patches of
> this series. This includes the iscsi, glusterfs, ssh, curl, and rbd
> drivers.

Indeed, but... protocol drivers don't have probes, do they?  There is a
distinction between "formats" and "protocols", and only formats need to
be probed for.

> A minimal configuration excluding all of these drivers takes
> around 3 ms to reach main(), and each of these drivers respectively add
> times of approximately 0.2, 0.9, 1.2, 2.5, and a massive time of 41.6 ms
> to the time to main. So some of these aren't even a huge benefit by
> themselves but they do add up, and also it makes sense to modularize all
> of the protocol drivers if we're already modularizing one.

All this I entirely agree with.

Paolo