2011-05-11 02:42:34

by Xiao Guangrong

Subject: [PATCH v2] KVM: MMU: optimize pte write path if don't have protected sp

Simply return from kvm_mmu_pte_write path if no shadow page is
write-protected, then we can avoid to walk all shadow pages and hold
mmu-lock

Signed-off-by: Xiao Guangrong <[email protected]>
---
arch/x86/include/asm/kvm_host.h | 1 +
arch/x86/kvm/mmu.c | 12 +++++++++++-
2 files changed, 12 insertions(+), 1 deletions(-)

diff --git a/arch/x86/include/asm/kvm_host.h b/arch/x86/include/asm/kvm_host.h
index d2ac8e2..d2e5fb8 100644
--- a/arch/x86/include/asm/kvm_host.h
+++ b/arch/x86/include/asm/kvm_host.h
@@ -442,6 +442,7 @@ struct kvm_arch {
unsigned int n_requested_mmu_pages;
unsigned int n_max_mmu_pages;
atomic_t invlpg_counter;
+ atomic_t indirect_shadow_pages;
struct hlist_head mmu_page_hash[KVM_NUM_MMU_PAGES];
/*
* Hash table of struct kvm_mmu_page.
diff --git a/arch/x86/kvm/mmu.c b/arch/x86/kvm/mmu.c
index 2841805..7e6117dc 100644
--- a/arch/x86/kvm/mmu.c
+++ b/arch/x86/kvm/mmu.c
@@ -1038,8 +1038,10 @@ static void kvm_mmu_free_page(struct kvm *kvm, struct kvm_mmu_page *sp)
hlist_del(&sp->hash_link);
list_del(&sp->link);
free_page((unsigned long)sp->spt);
- if (!sp->role.direct)
+ if (!sp->role.direct) {
free_page((unsigned long)sp->gfns);
+ atomic_dec(&kvm->arch.indirect_shadow_pages);
+ }
kmem_cache_free(mmu_page_header_cache, sp);
kvm_mod_used_mmu_pages(kvm, -1);
}
@@ -1536,6 +1538,7 @@ static struct kvm_mmu_page *kvm_mmu_get_page(struct kvm_vcpu *vcpu,
kvm_sync_pages(vcpu, gfn);

account_shadowed(vcpu->kvm, gfn);
+ atomic_inc(&vcpu->kvm->arch.indirect_shadow_pages);
}
if (shadow_trap_nonpresent_pte != shadow_notrap_nonpresent_pte)
vcpu->arch.mmu.prefetch_page(vcpu, sp);
@@ -3233,6 +3236,13 @@ void kvm_mmu_pte_write(struct kvm_vcpu *vcpu, gpa_t gpa,
int level, npte, invlpg_counter, r, flooded = 0;
bool remote_flush, local_flush, zap_page;

+ /*
+ * If we don't have indirect shadow pages, it means no page is
+ * write-protected, so we can simply exit.
+ */
+ if (!atomic_read(&vcpu->kvm->arch.indirect_shadow_pages))
+ return;
+
zap_page = remote_flush = local_flush = false;
offset = offset_in_page(gpa);

--
1.7.4.4

2011-05-11 16:17:09

by Avi Kivity

Subject: Re: [PATCH v2] KVM: MMU: optimize pte write path if don't have protected sp

On 05/11/2011 05:44 AM, Xiao Guangrong wrote:
> Simply return from kvm_mmu_pte_write path if no shadow page is
> write-protected, then we can avoid to walk all shadow pages and hold
> mmu-lock
>
> @@ -1038,8 +1038,10 @@ static void kvm_mmu_free_page(struct kvm *kvm, struct kvm_mmu_page *sp)
> hlist_del(&sp->hash_link);
> list_del(&sp->link);
> free_page((unsigned long)sp->spt);
> - if (!sp->role.direct)
> + if (!sp->role.direct) {
> free_page((unsigned long)sp->gfns);
> + atomic_dec(&kvm->arch.indirect_shadow_pages);
> + }
> kmem_cache_free(mmu_page_header_cache, sp);
> kvm_mod_used_mmu_pages(kvm, -1);
> }
> @@ -1536,6 +1538,7 @@ static struct kvm_mmu_page *kvm_mmu_get_page(struct kvm_vcpu *vcpu,
> kvm_sync_pages(vcpu, gfn);
>
> account_shadowed(vcpu->kvm, gfn);
> + atomic_inc(&vcpu->kvm->arch.indirect_shadow_pages);
> }

Better in account_shadowed()/unaccount_shadowed(), no?


--
error compiling committee.c: too many arguments to function

2011-05-11 15:44:18

by Xiao Guangrong

Subject: Re: [PATCH v2] KVM: MMU: optimize pte write path if don't have protected sp

On 05/11/2011 07:28 PM, Avi Kivity wrote:
> On 05/11/2011 05:44 AM, Xiao Guangrong wrote:
>> Simply return from kvm_mmu_pte_write path if no shadow page is
>> write-protected, then we can avoid to walk all shadow pages and hold
>> mmu-lock
>>
>> @@ -1038,8 +1038,10 @@ static void kvm_mmu_free_page(struct kvm *kvm, struct kvm_mmu_page *sp)
>> hlist_del(&sp->hash_link);
>> list_del(&sp->link);
>> free_page((unsigned long)sp->spt);
>> - if (!sp->role.direct)
>> + if (!sp->role.direct) {
>> free_page((unsigned long)sp->gfns);
>> + atomic_dec(&kvm->arch.indirect_shadow_pages);
>> + }
>> kmem_cache_free(mmu_page_header_cache, sp);
>> kvm_mod_used_mmu_pages(kvm, -1);
>> }
>> @@ -1536,6 +1538,7 @@ static struct kvm_mmu_page *kvm_mmu_get_page(struct kvm_vcpu *vcpu,
>> kvm_sync_pages(vcpu, gfn);
>>
>> account_shadowed(vcpu->kvm, gfn);
>> + atomic_inc(&vcpu->kvm->arch.indirect_shadow_pages);
>> }
>
> Better in account_shadowed()/unaccount_shadowed(), no?
>

Yes, will fix. thanks for your reminder!