On Thu, Apr 29, 2010 at 12:29:40PM -0700, Tom Lyon wrote: > I think the plan for V4 will be to take things entirely out of the UIO > framework, and instead have a driver which supports user mode use of > "well-behaved" PCI devices. I really don't think that we should create another uio-subsystem for the kernel. The current one can be easily extended to fit your and the virtualization needs. > So, I will go outside UIO because: > 1 - it doesn't allow reads and writes to sub-drivers, just irqcontrol Can you elaborate on that? What do you mean by sub-drivers? > 2 - it doesn't have ioctls You already added ioctls. This is not a big deal. > 3 - it has its own interrupt model which doesn't use eventfds Can be extended using the ioctl. > 4 - it's ugly doing the new stuff and maintaining backwards compat. Not really. In the ioctls you will add to enable msi/msi-x you can pass the eventfds to userspace. Thats not ugly at all. Its just an addition. Here is the approach I would take: * Add an ioctl to the uio-framework and add some basic functionality like querying for some device capabilities like memory regions to mmap or io regions. * Add ioctls to uio-pci-generic to ask for interrupt capabilities of the pci device (old style irqs/msi/msi-x, how many vectors and so on) and ioctls to enable them and get an eventfd back for each vector. Thats a good extension and backwards compatible as well in my eyes. What do you think? Joerg -- To unsubscribe from this list: send the line "unsubscribe kvm" in the body of a message to majordomo@xxxxxxxxxxxxxxx More majordomo info at http://vger.kernel.org/majordomo-info.html
