On Wed, May 13, 2015 at 07:12:32PM +0200, Paolo Bonzini wrote: > > > On 13/05/2015 19:04, Daniel P. Berrange wrote: > > I'm thinking perhaps a better approach could be for the crypto related > > APIs to call qcrypto_init() on an as-needed basis. The downside would > > be that this could delay the point at which the user sees a gnutls > > initialization failure to only after QEMU has been running for a while, > > instead of being upfront at startup. The plus side is obviously that > > we'd not need to update every binary program main() method. > > > > I notice though that QEMU does not make use of pthread_once() for > > global initializers. Is there any particular reason for this ? With > > this crypto code it is not safe to rely on being single threaded, > > since the crypto code can be invoked from I/O threads as well as > > the main event loop. So ideally I would use a pthread_once() instead > > of having a static 'bool is_initialized' protected by a pthread_mutex > > Yes---though you'd use a GOnce instead for portability. > > In fact you would likely keep both, and use a GOnce from qcrypto_init. > Then you can always call qcrypto_init from main if that helps making > early failures clearer.
Ok, I'll go with GOnce and keep an init call in vl.c for clear error reporting. Regards, Daniel -- |: http://berrange.com -o- http://www.flickr.com/photos/dberrange/ :| |: http://libvirt.org -o- http://virt-manager.org :| |: http://autobuild.org -o- http://search.cpan.org/~danberr/ :| |: http://entangle-photo.org -o- http://live.gnome.org/gtk-vnc :|
