On Mon, 2019-02-25 at 11:17 -0800, Matthew Garrett wrote: > On Mon, Feb 25, 2019 at 7:36 AM James Bottomley > <James.Bottomley@xxxxxxxxxxxxxxxxxxxxx> wrote: > > > The virtio driver performs discovery via virtio, which crosvm > > > implements already for all of its supported devices. This > > > substantially reduces the amount of TPM-specific code compared to > > > your suggestions, and lowers the barrier to entry for > > > implementing TPM support in other hypervisors which I hope we > > > agree is beneficial. > > > > Well, that's somewhat misleading: The reason we already have two > > hypervisor specific drivers already is because every hypervisor has > > a different virtual discovery mechanism. You didn't find the other > > two hypervisor drivers remotely useful, so why would another > > hypervisor find yours useful? > > The existing hypervisor drivers expose hypervisor-specific details. > This proposed driver provides an abstract interface that is usable by > other hypervisors. It allows building a VM that exposes TPM > functionality without requiring additional hardware emulation, > reducing the hypervisor attack surface. Well, that depends whether you think a virtio bus is an abstract concept or a hypervisor specific detail. There are currently four major hypervisors: xen, kvm, hyper-v and ESX. Of those, only one implements virtio: kvm. I agree virtio is a standard and certainly a slew of minor hypervisors implement it because they need paravirt support on Linux so they piggyback off kvm, but I don't see any of the other major hypervisors jumping on the bandwagon. I certainly agree our lives would be easier if all the major hypervisor vendors would just agree a single paravirt driver standard. > > > For me as a hypervisor implementer, what advantages do you see > > > that would make me decide to implement TPM-specific virtual > > > hardware emulation in the form of TIS rather than simply > > > leveraging a virtio driver like for other virtual devices? > > > > So your argument is that for every device we have in the Linux > > kernel, we should have the N hypervisor paravirt variants for the > > same thing? I assure you that's not going to fly because paravirt > > drivers would then outnumber real drivers by an order of magnitude. > > Well, no - in general there's no need to have more than one virtio > driver for any /class/ of hardware. For various unfortunate accidents > of history we've ended up with multiple cases where we have > hypervisor-specific drivers. Fully agree, that's why I'm doing so now. > Using the more generic virtio > infrastructure reduces the need for that, since any hypervisor should > be able to implement the backend (eg, in this case it'd be very easy > to add support for this driver to qemu, I certainly agree there ... is there a plan for this? > which would allow the use of TPMs without needing to enable a whole > bunch of additional qemu features). This isn't a discussion we'd be > having if we'd pushed back more strongly against hypervisor-specific > solutions in the past. I'm still looking for the pragmatic use case. I think yours is attack surface reduction, because the virtio discovery and operation is less code and therefore more secure than physical hardware discovery and operation? I'm not entirely sure I buy that because the TPM communication interface is pretty simple and it's fairly deep down in the kernel internal stack making it difficult to exploit. James
