On 10/09/2020 12.33, Huacai Chen wrote: > MIPS defines two kvm types: > > #define KVM_VM_MIPS_TE 0 > #define KVM_VM_MIPS_VZ 1 > > In Documentation/virt/kvm/api.rst it is said that "You probably want to > use 0 as machine type", which implies that type 0 be the "automatic" or > "default" type. And, in user-space libvirt use the null-machine (with > type 0) to detect the kvm capability, which returns "KVM not supported" > on a VZ platform. > > I try to fix it in QEMU but it is ugly: > https://lists.nongnu.org/archive/html/qemu-devel/2020-08/msg05629.html > > And Thomas Huth suggests me to change the definition of kvm type: > https://lists.nongnu.org/archive/html/qemu-devel/2020-09/msg03281.html > > So I define like this: > > #define KVM_VM_MIPS_AUTO 0 > #define KVM_VM_MIPS_VZ 1 > #define KVM_VM_MIPS_TE 2 > > Since VZ and TE cannot co-exists, using type 0 on a TE platform will > still return success (so old user-space tools have no problems on new > kernels); the advantage is that using type 0 on a VZ platform will not > return failure. So, the only problem is "new user-space tools use type > 2 on old kernels", but if we treat this as a kernel bug, we can backport > this patch to old stable kernels. > > Cc: stable@xxxxxxxxxxxxxxx > Signed-off-by: Huacai Chen <chenhc@xxxxxxxxxx> > --- > arch/mips/kvm/mips.c | 2 ++ > include/uapi/linux/kvm.h | 5 +++-- > 2 files changed, 5 insertions(+), 2 deletions(-) > > diff --git a/arch/mips/kvm/mips.c b/arch/mips/kvm/mips.c > index d7ba3f9..9efeb67 100644 > --- a/arch/mips/kvm/mips.c > +++ b/arch/mips/kvm/mips.c > @@ -138,6 +138,8 @@ extern void kvm_init_loongson_ipi(struct kvm *kvm); > int kvm_arch_init_vm(struct kvm *kvm, unsigned long type) > { > switch (type) { > + case KVM_VM_MIPS_AUTO: > + break; > #ifdef CONFIG_KVM_MIPS_VZ > case KVM_VM_MIPS_VZ: > #else > diff --git a/include/uapi/linux/kvm.h b/include/uapi/linux/kvm.h > index 29ba8e8..cfc1ae2 100644 > --- a/include/uapi/linux/kvm.h > +++ b/include/uapi/linux/kvm.h > @@ -790,9 +790,10 @@ struct kvm_ppc_resize_hpt { > #define KVM_VM_PPC_HV 1 > #define KVM_VM_PPC_PR 2 > > -/* on MIPS, 0 forces trap & emulate, 1 forces VZ ASE */ > -#define KVM_VM_MIPS_TE 0 > +/* on MIPS, 0 indicates auto, 1 forces VZ ASE, 2 forces trap & emulate */ > +#define KVM_VM_MIPS_AUTO 0 > #define KVM_VM_MIPS_VZ 1 > +#define KVM_VM_MIPS_TE 2 > > #define KVM_S390_SIE_PAGE_OFFSET 1 Thanks, I think that's the right way to go if we don't want mips to behave completely different compared to the other architectures (e.g. powerpc is also using 0 as "automatic" type in arch/powerpc/kvm/powerpc.c). Reviewed-by: Thomas Huth <thuth@xxxxxxxxxx>
