Re: [PATCH 0/4 v2] kvm: rework KVM mmu_shrink() code

[Marcelo Tosatti <mtosatti@redhat.com>]

On Mon, Aug 23, 2010 at 07:11:11PM +0800, Xiaotian Feng wrote:
> On Mon, Aug 23, 2010 at 6:27 PM, Avi Kivity <avi@redhat.com> wrote:
> >  On 08/23/2010 01:22 PM, Avi Kivity wrote:
> >>
> >>
> >> I see a lot of soft lockups with this patchset:
> >
> > This is running the emulator.flat test case, with shadow paging.  This test
> > triggers a lot (millions) of mmu mode switches.
> >
> 
> Does following patch fix your issue?
> 
> Latest kvm mmu_shrink code rework makes kernel changes
> kvm->arch.n_used_mmu_pages/
> kvm->arch.n_max_mmu_pages at kvm_mmu_free_page/kvm_mmu_alloc_page,
> which is called
> by kvm_mmu_commit_zap_page. So the kvm->arch.n_used_mmu_pages or
> kvm_mmu_available_pages(vcpu->kvm) is unchanged after kvm_mmu_commit_zap_page(),
> This caused kvm_mmu_change_mmu_pages/__kvm_mmu_free_some_pages looping forever.
> Moving kvm_mmu_commit_zap_page would make the while loop performs as normal.
> 
> ---
> diff --git a/arch/x86/kvm/mmu.c b/arch/x86/kvm/mmu.c
> index f52a965..7e09a21 100644
> --- a/arch/x86/kvm/mmu.c
> +++ b/arch/x86/kvm/mmu.c
> @@ -1726,8 +1726,8 @@ void kvm_mmu_change_mmu_pages(struct kvm *kvm,
> unsigned int goal_nr_mmu_pages)
>                                             struct kvm_mmu_page, link);
>                         kvm_mmu_prepare_zap_page(kvm, page,
>                                                                &invalid_list);
> +                       kvm_mmu_commit_zap_page(kvm, &invalid_list);
>                 }
> -               kvm_mmu_commit_zap_page(kvm, &invalid_list);
>                 goal_nr_mmu_pages = kvm->arch.n_used_mmu_pages;
>         }
> 
> @@ -2976,9 +2976,9 @@ void __kvm_mmu_free_some_pages(struct kvm_vcpu *vcpu)
>                 sp = container_of(vcpu->kvm->arch.active_mmu_pages.prev,
>                                   struct kvm_mmu_page, link);
>                 kvm_mmu_prepare_zap_page(vcpu->kvm, sp, &invalid_list);
> +               kvm_mmu_commit_zap_page(vcpu->kvm, &invalid_list);
>                 ++vcpu->kvm->stat.mmu_recycled;
>         }
> -       kvm_mmu_commit_zap_page(vcpu->kvm, &invalid_list);
>  }
> 
>  int kvm_mmu_page_fault(struct kvm_vcpu *vcpu, gva_t cr2, u32 error_code)

Please resend with a signed-off-by, and proper subject for the patch.

Thanks