[dpdk-dev] [PATCH 15/15] vfio: change VFIO init to be extendable

[Burakov, Anatoly]

> > Could there be a situation where we need to know if a particular VFIO
> > subsystem is enabled? Do you think it's worth adding (e.g.
> > vfio_enabled |= VFIO_PCI_ENABLED or something)? (I don't imply it is
> > necessary, just asking)
> 
> Well... the semantics are quite tricky here.
> 
> There are: vfio-pci, vfio-platform and vfio-amba in the Linux Kernel. I ignore
> vfio-amba because I don't know any considerable device supported by this.
> 
> With respect to the VFIO, if you have both PCI and SoC infra enabled, you
> have to check for both vfio-pci and vfio-platform.
> 
> If only PCI is there you have to check only for vfio-pci.
> But, if you check for vfio-platform as well and it is not there then the EAL init
> would fail without a reason (because you don't need vfio-platform). Vice-
> versa for SoC.
> 
> This patch improves the check to cover those situations. Later on, I'll just
> append something like:
> 
> if (!internal_config.no_soc) {
> 	soc_vfio_enable();
> 	vfio_enabled |= soc_vfio_is_enabled();
> }
> 
> and the EAL init would be happy... I hope ;).
> 
> The thing is that we don't know which devices will bind to VFIO and we don't
> know which of those devices would use vfio-pci and vfio-platform.
> So, if both PCI and SoC are enabled and only vfio-platform is there we cannot
> say here "failed because vfio-pci is missing".

OK fair enough :)

I haven't yet tested the patchset itself, I'll report back if I get any issues.

Thanks,
Anatoly