Re: [PATCH V3 4/13] KVM/MMU: Flush tlb directly in the kvm_handle_hva_range()

[Paolo Bonzini]

On 27/09/2018 05:49, Tianyu Lan wrote:
> This patch is to flush tlb directly in the kvm_handle_hva_range()
> when range flush is available.
> 
> Signed-off-by: Lan Tianyu <tianyu....@microsoft.com>
> ---
>  arch/x86/kvm/mmu.c | 7 +++++++
>  1 file changed, 7 insertions(+)
> 
> diff --git a/arch/x86/kvm/mmu.c b/arch/x86/kvm/mmu.c
> index d10d8423e8d6..877edae0401f 100644
> --- a/arch/x86/kvm/mmu.c
> +++ b/arch/x86/kvm/mmu.c
> @@ -1888,6 +1888,13 @@ static int kvm_handle_hva_range(struct kvm *kvm,
>                                                &iterator)
>                               ret |= handler(kvm, iterator.rmap, memslot,
>                                              iterator.gfn, iterator.level, 
> data);
> +
> +                     if (ret && kvm_available_flush_tlb_with_range()) {
> +                             kvm_flush_remote_tlbs_with_address(kvm,
> +                                             gfn_start,
> +                                             gfn_end - gfn_start);
> +                             ret = 0;
> +                     }
>               }
>       }
>  
> 


Not all callers need a TLB flush (in particular kvm_test_age_hva)
require a flush.  My suggestion is to rewrite kvm_test_age_hva like this:

index 4705a7f4169e..f72364a0ef9c 100644
--- a/arch/x86/kvm/mmu.c
+++ b/arch/x86/kvm/mmu.c
@@ -1898,12 +1898,13 @@ static int kvm_test_age_rmapp(struct kvm *kvm,
                              struct kvm_memory_slot *slot, gfn_t gfn,
                              int level, unsigned long data)
 {
+       bool *result = (bool *)data;
        u64 *sptep;
        struct rmap_iterator iter;

        for_each_rmap_spte(rmap_head, &iter, sptep)
                if (is_accessed_spte(*sptep))
-                       return 1;
+                       *result = true;
        return 0;
 }

@@ -1929,7 +1930,10 @@ int kvm_age_hva(struct kvm *kvm,

 int kvm_test_age_hva(struct kvm *kvm, unsigned long hva)
 {
-       return kvm_handle_hva(kvm, hva, 0, kvm_test_age_rmapp);
+       bool result = false;
+       kvm_handle_hva(kvm, hva, (unsigned long) &result,
+                       kvm_test_age_rmapp);
+       return result;
 }

 #ifdef MMU_DEBUG

and move the flush from kvm_set_pte_rmapp to
kvm_mmu_notifier_change_pte, making kvm_set_spte_hva return an int;
otherwise, it will flush twice.  For non-x86 architectures just grep for
"set_spte_hva" and make the various kvm_set_spte_hva implementations
return false.

Paolo
_______________________________________________
devel mailing list
de...@linuxdriverproject.org
http://driverdev.linuxdriverproject.org/mailman/listinfo/driverdev-devel