[Qemu-devel] Passthrough and mmio overhead?

[Qemu-devel] Passthrough and mmio overhead?

[Etienne Martineau]

Hi all,

My current understanding of Qemu-kvm is such that whenever Guest VM
establish a passthrough to a pci devices, the following happens:

Setup:
- vt-d setup
- Host mmap the device's BAR via /sysfs.
- Host maintain a 'ptr' to that mmap'ed area.
- Host create a mmio slot for the Guest.
- The mmio slot 'address' is part of the Guest VM pci addr space.

Runtime:
- Guest access the mmio slot and pagefault to the host
- Host resolve the offset, dereference 'ptr' and gets the value
- Host resume the pagefault and return the value to the guest

Now the question is that if Guest vm keeps doing MMIO, will it keep page
faulting on the host?

Thanks,
Etienne