aOn Wed, 26 Jan 2022 13:21:17 +0000, Mark Brown <broonie@xxxxxxxxxx> wrote: > > [1 <text/plain; us-ascii (7bit)>] > On Wed, Jan 26, 2022 at 01:05:01PM +0000, Marc Zyngier wrote: > > Mark Brown <broonie@xxxxxxxxxx> wrote: > > > > > So you now only test whether it is possible to create a virtual GICv3, > > > > but don't actually create it. How does this work? > > > > Oh, that's rather obscure in the API - so the file descriptor returned > > > if the test flag is specified can't actually be used? > > > No file descriptor is returned at all from the kernel, so you should > > always get -1 as populated by kvm_create_device(). > > > See virt/kvm/kvm_main.c::kvm_ioctl_create_device(). > > Ugh, right - the descriptor is returned separately to the return code > but is then mapped onto the return code by the library since the library > just aborts on error in the non-test case and in the test case the kernel > returns an inverted boolean which gets passed straight through. Like I > say the API seems a bit confusing here. That's a very... "kind" way of putting it. M. -- Without deviation from the norm, progress is not possible.
