On Sun, 22 Aug 2021 16:25:23 +0100, Marc Zyngier wrote: > KVM hacking on the Apple M1 SoC has shown that kvmtool (and other > VMMs) make pretty poor use of the IPA space parameter (read: do not > use it and just pass 0). This results in a guest that cannot boot > (recent kernels will just send the VMM packing), and in general means > we don't benefit from smaller page tables at stage-2. > > This series does three things: > - It switches kvmtool away from the default 40bit, allowing large VMs > to be created (I have booted a 4TB VM) > - It reduces the requested IPA space to be as small as possible > - It tells the user why the VM cannot boot when the IPA space required > exceeds that of the HW > > [...] Applied to kvmtool (master), thanks! [1/3] kvmtool: Abstract KVM_VM_TYPE into a weak function https://git.kernel.org/will/kvmtool/c/4250819de93b [2/3] kvmtool: arm64: Use the maximum supported IPA size when creating the VM https://git.kernel.org/will/kvmtool/c/066b5c06c4e3 [3/3] kvmtool: arm64: Configure VM with the minimal required IPA space https://git.kernel.org/will/kvmtool/c/bdb86d0c0c95 Cheers, -- Will https://fixes.arm64.dev https://next.arm64.dev https://will.arm64.dev _______________________________________________ kvmarm mailing list kvmarm@xxxxxxxxxxxxxxxxxxxxx https://lists.cs.columbia.edu/mailman/listinfo/kvmarm