I have installed Suse Enterprise 10 RC 2.5 x64 edition on a
dual proc EM64T machine that I have @ work. Here is what I am booting. The
chipset does not have any IOMMU engine in it. The processor has Intel® VT-x
extensions enabled.
- I boot to the Xen kernel
option in GRUB.
- I have installed the same
image as a VM (is that ok) using the Xen management console in the control
panel. I assume this is the user kernel and the Xen kernel itself is dom0.
- I have a specialized PCIe
device that appears in the config space of the kernel (when I do lspci –vv).
- Of course, it does not appear
in the VM kernel. In fact, lspci returns a null string.
- I would like to assign the
device directly to the VM kernel.
- What is the procedure I need
to do to recompile the kernel (both dom0 and VM kernel) to make this
happen?.
- Essentially, the driver
running in the VM should be able to access the device using its config
space, issue MMIO requests as well as program the device with DMA
requests. The DMA requests should obviously be using host physical
addresses.
- How do the VM kernels know
about host physical addresses?.
- Can they
make a hypercall to get a HPA from a VM GPA?.
- Alternatively, I could have a
proxy in the dom0 kernel that programs the device with DMA regions and
then does double buffering from its regions to the VM GPA.
- Even then,
I am assuming that the hypercalls would have to be implemented.
Another question. Since the Xen in the above Linux
distribution is using Intel® VT-x extensions, how do I dump the VMCS for
the user image when I boot the Xen kernel?. Is there a virtual device in the
/dev filesystem that points to the user image’s memory dump (kind of like
/dev/kmem for the VM GPA).
Thanks,
RK
