On 14/05/18 20:01, Marc Orr wrote:
> The kvm struct has been bloating. For example, it's tens of kilo-bytes
> for x86, which turns out to be a large amount of memory to allocate
> contiguously via kzalloc. Thus, this patch does the following:
> 1. Uses architecture-specific routines to allocat the kvm struct via
> kzalloc for x86 and also for arm when has_tbe() is true.
> 2. Introduces a baseline allocator for the kvm struct that uses valloc.
> This allocator can be used by an architecture by defining the
> __KVM_VALLOC_ARCH_VM macro before including include/linux/kvm_host.h.
> 3. Finally, continue to default to kalloc for all other situations, as
> many architectures have taken a dependence on kalloc.
>
> Signed-off-by: Marc Orr <marcorr@xxxxxxxxxx>
> Change-Id: Ide392e51991142d1325e06351231ac685e11d820
> ---
> arch/arm/include/asm/kvm_host.h | 4 ++++
> arch/arm/kvm/guest.c | 16 ++++++++++++++++
> arch/arm64/include/asm/kvm_host.h | 4 ++++
> arch/arm64/kvm/guest.c | 16 ++++++++++++++++
> arch/x86/kvm/svm.c | 4 ++--
> arch/x86/kvm/vmx.c | 4 ++--
> include/linux/kvm_host.h | 13 +++++++++++++
> 7 files changed, 57 insertions(+), 4 deletions(-)
>
> diff --git a/arch/arm/include/asm/kvm_host.h b/arch/arm/include/asm/kvm_host.h
> index c7c28c885a19..f38fd60b4d4d 100644
> --- a/arch/arm/include/asm/kvm_host.h
> +++ b/arch/arm/include/asm/kvm_host.h
> @@ -318,4 +318,8 @@ static inline bool kvm_arm_harden_branch_predictor(void)
> static inline void kvm_vcpu_load_sysregs(struct kvm_vcpu *vcpu) {}
> static inline void kvm_vcpu_put_sysregs(struct kvm_vcpu *vcpu) {}
>
> +#define __KVM_HAVE_ARCH_VM_ALLOC
> +struct kvm *kvm_arch_alloc_vm(void);
> +void kvm_arch_free_vm(struct kvm *kvm);
> +
> #endif /* __ARM_KVM_HOST_H__ */
> diff --git a/arch/arm/kvm/guest.c b/arch/arm/kvm/guest.c
> index a18f33edc471..2b88b745f5a3 100644
> --- a/arch/arm/kvm/guest.c
> +++ b/arch/arm/kvm/guest.c
> @@ -42,6 +42,22 @@ struct kvm_stats_debugfs_item debugfs_entries[] = {
> { NULL }
> };
>
> +struct kvm *kvm_arch_alloc_vm(void)
> +{
> + if (!has_vhe())
> + return kzalloc(sizeof(struct kvm), GFP_KERNEL);
> +
> + return vzalloc(sizeof(struct kvm));
This doesn't make much sense.
Either you pick the patch I provided which makes the allocation a common
path between arm and arm64 by sticking it in common code (and then the
has_vhe()) test makes sense), or you make it per-architecture, and the
whole has_vhe() for 32bit ARM is just churn (to the point where you
shouldn't define __KVM_HAVE_ARCH_VM_ALLOC at all for that architecture).
Thanks,
M.
--
Jazz is not dead. It just smells funny...
