Why RISC-V doesn't (yet) support KVM [LWN.net]

Why RISC-V doesn't (yet) support KVM

Posted May 20, 2021 21:02 UTC (Thu) by pbonzini (subscriber, #60935)
Parent article: Why RISC-V doesn't (yet) support KVM

https://lwn.net/ml/linux-kernel/a49a7142-104e-fdaa-4a6a-6... provides additional context as to how feature creep is delaying the freezing of the hypervisor specification.

to post comments

Why RISC-V doesn't (yet) support KVM

Posted May 20, 2021 21:36 UTC (Thu) by pdp7pdp7 (subscriber, #96517) [Link] (4 responses)

pbonzini: I am confused about your comment [1] that some people may believe it could be beneficial to slow down merging of RISC-V support upstream. I am trying to understand who would benefit in that scenario. Is the theoretical benefit that it could reduce the RISC-V maintainer workload?

[1] https://lore.kernel.org/linux-riscv/daa30135-8757-8d33-a9...

Why RISC-V doesn't (yet) support KVM

Posted May 20, 2021 22:25 UTC (Thu) by pbonzini (subscriber, #60935) [Link] (3 responses)

The patch acceptance policy can become an incentive to game the system:

1. Company X don't have hardware ready that implements an extension

2. Company X slows down ratification of the extension

3. Competitors must ship out of tree code to support the extension in Linux

The conflict of interest is outside Linux, which is what Paul Welmsley alluded to with respect to speeding up the process of ratification, but it affects Linux and the maintainers are caught in the crossfire.

Looking at the history of the hypervisor extension specification[1], the last significant change was done about one year ago (May 15, 2020). There is no justification at all for the specification not to be frozen yet.

[1] https://github.com/riscv/riscv-isa-manual/commits/master/...

Why RISC-V doesn't (yet) support KVM

Posted May 21, 2021 0:09 UTC (Fri) by yootis (subscriber, #4762) [Link] (2 responses)

So which "company X" is slowing the system down now?

Why RISC-V doesn't (yet) support KVM

Posted May 21, 2021 9:11 UTC (Fri) by farnz (subscriber, #17727) [Link] (1 responses)

My understanding is that the slowdown is not a specific company this time - it's getting the PLIC and CLIC specifications close enough to done that the RISC-V Foundation can be confident that the Hypervisor extension will not need to be modified to support direct delivery of interrupts to guest IRQ handlers without hypervisor involvement.

Why RISC-V doesn't (yet) support KVM

Posted May 21, 2021 12:08 UTC (Fri) by pbonzini (subscriber, #60935) [Link]

Which is completely useless, since you can always make a subextension.