Re: [PATCH v3 1/3] uio: add ioctl support

["Michael S. Tsirkin" <mst@redhat.com>]

On Mon, Oct 05, 2015 at 01:36:35PM +0300, Vlad Zolotarov wrote:
> >And then there's the issue of why we even need this, why not just
> >write a whole new driver for this, like the previous driver did (which
> >also used ioctls, yes, I didn't have the chance to object to that before
> >everyone else did...)
> 
> Which "previous driver" do u refer here?
> IMHO writing something instead of UIO (not just uio_pci_generic) seems like
> an overkill for solving this issue. Supporting MSI-X interrupts seem like a
> very beneficial feature for uio_pci_generic and it's really not _THAT_
> complicated API - just look at VFIO for a comparison... ;)

Except most things VFIO does is actually there for security.
Which, for a device that can do DMA and isn't even behind an IOMMU,
sounds like a pretty big deal actually.

> uio_pci_generic is clearly missing this important feature. And creating
> another user space driver infrastructure just to add it seems extremely
> unjustified.

uio_pci_generic was always intended to be used with extremely simple
devices which don't do DMA, or where someone else has set up the IOMMU
(like kvm does with CONFIG_KVM_DEVICE_ASSIGNMENT).

We need to be much more careful with MSI if there's no IOMMU.

-- 
MST