[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]
Re: [Qemu-devel] Is network backend netmap worth keeping?
From: |
Markus Armbruster |
Subject: |
Re: [Qemu-devel] Is network backend netmap worth keeping? |
Date: |
Fri, 13 Sep 2019 15:04:05 +0200 |
User-agent: |
Gnus/5.13 (Gnus v5.13) Emacs/26.2 (gnu/linux) |
Giuseppe Lettieri <address@hidden> writes:
> Hi all,
>
> I have been thinking of the submodule suggestion and I have also
> prepared a patch for it (attached). However, I am not sure about what
> we want to achieve with it. In particular, I am not sure that the
> option is useful for end users. The problem is that netmap, unlike
> capstone and slirp, is a library plus a kernel module. If netmap is
> not installed in the system, the --enable-netmap=git option will
> successfully build qemu with netmap support, but then -netdev netmap
> will fail at runtime.
Yes.
What happens when I build with --enable-netmap=system on host A, then
run the resulting binary on some host B that doesn't have netmap
installed?
> On the other end, if netmap is installed in the
> system, using the =git option may cause qemu to be built against a
> different, possibly incompatible version.
Yes. We default to netmap=system, though. If you break things by
passing arcane arguments to configure, you get to keep the pieces :)
> If the option is only useful for developers to check that some qemu
> change does not break anything, then probably it should be enabled in
> some other, less visible way. What do you think?
I think an --enable-netmap patterned after --enable-capstone and
--enable-slirp has sufficiently low visibility as long as the default is
sane.
We clearly want configure to pick netmap=system when the system provides
netmap.
What shall configure pick when the system doesn't provide it? If you
think netmap=git is too dangerous for general audience, consider
disabling netmap then. Experts can still compile-test with
--enable-netmap=git. Our CI certainly should.