Re: [RFC PATCH 3/6] KVM: Define kvm_flush_remote_tlbs_range

[David Matlack <dmatlack@google.com>]

On Mon, Jan 09, 2023 at 09:53:44PM +0000, Raghavendra Rao Ananta wrote:
> Define kvm_flush_remote_tlbs_range() to limit the TLB flush only
> to a certain range of addresses. Replace this with the existing
> call to kvm_flush_remote_tlbs() in the MMU notifier path.
> Architectures such as arm64 can define this to flush only the
> necessary addresses, instead of the entire range.
> 
> Signed-off-by: Raghavendra Rao Ananta <rananta@google.com>
> ---
>  arch/arm64/kvm/mmu.c     | 10 ++++++++++
>  include/linux/kvm_host.h |  1 +
>  virt/kvm/kvm_main.c      |  7 ++++++-
>  3 files changed, 17 insertions(+), 1 deletion(-)
> 
> diff --git a/arch/arm64/kvm/mmu.c b/arch/arm64/kvm/mmu.c
> index 39d9a334efb57..70f76bc909c5d 100644
> --- a/arch/arm64/kvm/mmu.c
> +++ b/arch/arm64/kvm/mmu.c
> @@ -91,6 +91,16 @@ void kvm_flush_remote_tlbs(struct kvm *kvm)
>  	kvm_call_hyp(__kvm_tlb_flush_vmid, &kvm->arch.mmu);
>  }
>  
> +void kvm_flush_remote_tlbs_range(struct kvm *kvm, unsigned long start, unsigned long end)
> +{
> +	struct kvm_s2_mmu *mmu = &kvm->arch.mmu;
> +
> +	if (system_supports_tlb_range())
> +		kvm_call_hyp(__kvm_tlb_flush_range_vmid_ipa, mmu, start, end, 0);
> +	else
> +		kvm_flush_remote_tlbs(kvm);
> +}
> +
>  static bool kvm_is_device_pfn(unsigned long pfn)
>  {
>  	return !pfn_is_map_memory(pfn);
> diff --git a/include/linux/kvm_host.h b/include/linux/kvm_host.h
> index f51eb9419bfc3..a76cede9dc3bb 100644
> --- a/include/linux/kvm_host.h
> +++ b/include/linux/kvm_host.h
> @@ -1359,6 +1359,7 @@ int kvm_vcpu_yield_to(struct kvm_vcpu *target);
>  void kvm_vcpu_on_spin(struct kvm_vcpu *vcpu, bool usermode_vcpu_not_eligible);
>  
>  void kvm_flush_remote_tlbs(struct kvm *kvm);
> +void kvm_flush_remote_tlbs_range(struct kvm *kvm, unsigned long start, unsigned long end);
>  
>  #ifdef KVM_ARCH_NR_OBJS_PER_MEMORY_CACHE
>  int kvm_mmu_topup_memory_cache(struct kvm_mmu_memory_cache *mc, int min);
> diff --git a/virt/kvm/kvm_main.c b/virt/kvm/kvm_main.c
> index 03e6a38094c17..f538ecc984f5b 100644
> --- a/virt/kvm/kvm_main.c
> +++ b/virt/kvm/kvm_main.c
> @@ -376,6 +376,11 @@ void kvm_flush_remote_tlbs(struct kvm *kvm)
>  		++kvm->stat.generic.remote_tlb_flush;
>  }
>  EXPORT_SYMBOL_GPL(kvm_flush_remote_tlbs);
> +
> +void kvm_flush_remote_tlbs_range(struct kvm *kvm, unsigned long start, unsigned long end)

It's ambiguous what start/end represent. Case in point,
__kvm_handle_hva_range() is passing in HVAs but then patch 4 passes in
GFNs.

Probably kvm_flush_tlbs_range() should accept GFN and there can be a
helper wrapper that does the HVA-to-GFN conversion.

> +{
> +	kvm_flush_remote_tlbs(kvm);
> +}

FYI I also proposed a common kvm_flush_remote_tlbs() in my Common MMU
series [1].

Could I interest you in grabbing patches 29-33 from that series, which
has the same end result (common kvm_flush_remote_tlbs_range()) but also
hooks up the KVM/x86 range-based flushing, and folding them into this
series?

[1] https://lore.kernel.org/kvm/20221208193857.4090582-33-dmatlack@google.com/

>  #endif
>  
>  static void kvm_flush_shadow_all(struct kvm *kvm)
> @@ -637,7 +642,7 @@ static __always_inline int __kvm_handle_hva_range(struct kvm *kvm,
>  	}
>  
>  	if (range->flush_on_ret && ret)
> -		kvm_flush_remote_tlbs(kvm);
> +		kvm_flush_remote_tlbs_range(kvm, range->start, range->end - 1);
>  
>  	if (locked) {
>  		KVM_MMU_UNLOCK(kvm);
> -- 
> 2.39.0.314.g84b9a713c41-goog
>