2018-07-24 08:18:24

by Tianyu Lan

Subject: [PATCH] KVM/MMU: Combine flushing remote tlb in mmu_set_spte()

mmu_set_spte() flushes remote tlbs for drop_parent_pte/drop_spte()
and set_spte() separately. This may introduce redundant flush. This
patch is to combine these flushes and check flush request after
calling set_spte().

Signed-off-by: Lan Tianyu <[email protected]>
---
arch/x86/kvm/mmu.c | 7 ++++---
1 file changed, 4 insertions(+), 3 deletions(-)

diff --git a/arch/x86/kvm/mmu.c b/arch/x86/kvm/mmu.c
index 22a7984..8f21632 100644
--- a/arch/x86/kvm/mmu.c
+++ b/arch/x86/kvm/mmu.c
@@ -2901,6 +2901,7 @@ static int mmu_set_spte(struct kvm_vcpu *vcpu, u64 *sptep, unsigned pte_access,
int rmap_count;
int set_spte_ret;
int ret = RET_PF_RETRY;
+ bool flush = false;

pgprintk("%s: spte %llx write_fault %d gfn %llx\n", __func__,
*sptep, write_fault, gfn);
@@ -2917,12 +2918,12 @@ static int mmu_set_spte(struct kvm_vcpu *vcpu, u64 *sptep, unsigned pte_access,

child = page_header(pte & PT64_BASE_ADDR_MASK);
drop_parent_pte(child, sptep);
- kvm_flush_remote_tlbs(vcpu->kvm);
+ flush = true;
} else if (pfn != spte_to_pfn(*sptep)) {
pgprintk("hfn old %llx new %llx\n",
spte_to_pfn(*sptep), pfn);
drop_spte(vcpu->kvm, sptep);
- kvm_flush_remote_tlbs(vcpu->kvm);
+ flush = true;
} else
was_rmapped = 1;
}
@@ -2934,7 +2935,7 @@ static int mmu_set_spte(struct kvm_vcpu *vcpu, u64 *sptep, unsigned pte_access,
ret = RET_PF_EMULATE;
kvm_make_request(KVM_REQ_TLB_FLUSH, vcpu);
}
- if (set_spte_ret & SET_SPTE_NEED_REMOTE_TLB_FLUSH)
+ if (set_spte_ret & SET_SPTE_NEED_REMOTE_TLB_FLUSH || flush)
kvm_flush_remote_tlbs(vcpu->kvm);

if (unlikely(is_mmio_spte(*sptep)))
--
2.7.4

2018-07-24 14:37:11

by Paolo Bonzini

Subject: Re: [PATCH] KVM/MMU: Combine flushing remote tlb in mmu_set_spte()

On 24/07/2018 10:17, Tianyu Lan wrote:
> mmu_set_spte() flushes remote tlbs for drop_parent_pte/drop_spte()
> and set_spte() separately. This may introduce redundant flush. This
> patch is to combine these flushes and check flush request after
> calling set_spte().
>
> Signed-off-by: Lan Tianyu <[email protected]>

Looks good, but I'd like a second opinion. Guangrong, Junaid, can you
review this?

Thanks,

Paolo

> ---
> arch/x86/kvm/mmu.c | 7 ++++---
> 1 file changed, 4 insertions(+), 3 deletions(-)
>
> diff --git a/arch/x86/kvm/mmu.c b/arch/x86/kvm/mmu.c
> index 22a7984..8f21632 100644
> --- a/arch/x86/kvm/mmu.c
> +++ b/arch/x86/kvm/mmu.c
> @@ -2901,6 +2901,7 @@ static int mmu_set_spte(struct kvm_vcpu *vcpu, u64 *sptep, unsigned pte_access,
> int rmap_count;
> int set_spte_ret;
> int ret = RET_PF_RETRY;
> + bool flush = false;
>
> pgprintk("%s: spte %llx write_fault %d gfn %llx\n", __func__,
> *sptep, write_fault, gfn);
> @@ -2917,12 +2918,12 @@ static int mmu_set_spte(struct kvm_vcpu *vcpu, u64 *sptep, unsigned pte_access,
>
> child = page_header(pte & PT64_BASE_ADDR_MASK);
> drop_parent_pte(child, sptep);
> - kvm_flush_remote_tlbs(vcpu->kvm);
> + flush = true;
> } else if (pfn != spte_to_pfn(*sptep)) {
> pgprintk("hfn old %llx new %llx\n",
> spte_to_pfn(*sptep), pfn);
> drop_spte(vcpu->kvm, sptep);
> - kvm_flush_remote_tlbs(vcpu->kvm);
> + flush = true;
> } else
> was_rmapped = 1;
> }
> @@ -2934,7 +2935,7 @@ static int mmu_set_spte(struct kvm_vcpu *vcpu, u64 *sptep, unsigned pte_access,
> ret = RET_PF_EMULATE;
> kvm_make_request(KVM_REQ_TLB_FLUSH, vcpu);
> }
> - if (set_spte_ret & SET_SPTE_NEED_REMOTE_TLB_FLUSH)
> + if (set_spte_ret & SET_SPTE_NEED_REMOTE_TLB_FLUSH || flush)
> kvm_flush_remote_tlbs(vcpu->kvm);
>
> if (unlikely(is_mmio_spte(*sptep)))
>

2018-07-25 02:27:55

by Junaid Shahid

Subject: Re: [PATCH] KVM/MMU: Combine flushing remote tlb in mmu_set_spte()

On 07/24/2018 07:35 AM, Paolo Bonzini wrote:
> On 24/07/2018 10:17, Tianyu Lan wrote:
>> mmu_set_spte() flushes remote tlbs for drop_parent_pte/drop_spte()
>> and set_spte() separately. This may introduce redundant flush. This
>> patch is to combine these flushes and check flush request after
>> calling set_spte().
>>
>> Signed-off-by: Lan Tianyu <[email protected]>
>
> Looks good, but I'd like a second opinion. Guangrong, Junaid, can you
> review this?
>
> Thanks,
>
> Paolo
>
>> ---
>> arch/x86/kvm/mmu.c | 7 ++++---
>> 1 file changed, 4 insertions(+), 3 deletions(-)
>>
>> diff --git a/arch/x86/kvm/mmu.c b/arch/x86/kvm/mmu.c
>> index 22a7984..8f21632 100644
>> --- a/arch/x86/kvm/mmu.c
>> +++ b/arch/x86/kvm/mmu.c
>> @@ -2901,6 +2901,7 @@ static int mmu_set_spte(struct kvm_vcpu *vcpu, u64 *sptep, unsigned pte_access,
>> int rmap_count;
>> int set_spte_ret;
>> int ret = RET_PF_RETRY;
>> + bool flush = false;
>>
>> pgprintk("%s: spte %llx write_fault %d gfn %llx\n", __func__,
>> *sptep, write_fault, gfn);
>> @@ -2917,12 +2918,12 @@ static int mmu_set_spte(struct kvm_vcpu *vcpu, u64 *sptep, unsigned pte_access,
>>
>> child = page_header(pte & PT64_BASE_ADDR_MASK);
>> drop_parent_pte(child, sptep);
>> - kvm_flush_remote_tlbs(vcpu->kvm);
>> + flush = true;
>> } else if (pfn != spte_to_pfn(*sptep)) {
>> pgprintk("hfn old %llx new %llx\n",
>> spte_to_pfn(*sptep), pfn);
>> drop_spte(vcpu->kvm, sptep);
>> - kvm_flush_remote_tlbs(vcpu->kvm);
>> + flush = true;
>> } else
>> was_rmapped = 1;
>> }
>> @@ -2934,7 +2935,7 @@ static int mmu_set_spte(struct kvm_vcpu *vcpu, u64 *sptep, unsigned pte_access,
>> ret = RET_PF_EMULATE;
>> kvm_make_request(KVM_REQ_TLB_FLUSH, vcpu);
>> }
>> - if (set_spte_ret & SET_SPTE_NEED_REMOTE_TLB_FLUSH)
>> + if (set_spte_ret & SET_SPTE_NEED_REMOTE_TLB_FLUSH || flush)
>> kvm_flush_remote_tlbs(vcpu->kvm);
>>
>> if (unlikely(is_mmio_spte(*sptep)))
>>
>

Reviewed-by: Junaid Shahid <[email protected]>

2018-07-25 02:36:34

by Tianyu Lan

Subject: Re: [PATCH] KVM/MMU: Combine flushing remote tlb in mmu_set_spte()

Hi Junaid:
Thanks for your review.

On 7/25/2018 10:26 AM, Junaid Shahid wrote:
> On 07/24/2018 07:35 AM, Paolo Bonzini wrote:
>> On 24/07/2018 10:17, Tianyu Lan wrote:
>>> mmu_set_spte() flushes remote tlbs for drop_parent_pte/drop_spte()
>>> and set_spte() separately. This may introduce redundant flush. This
>>> patch is to combine these flushes and check flush request after
>>> calling set_spte().
>>>
>>> Signed-off-by: Lan Tianyu <[email protected]>
>>
>> Looks good, but I'd like a second opinion. Guangrong, Junaid, can you
>> review this?
>>
>> Thanks,
>>
>> Paolo
>>
>>> ---
>>> arch/x86/kvm/mmu.c | 7 ++++---
>>> 1 file changed, 4 insertions(+), 3 deletions(-)
>>>
>>> diff --git a/arch/x86/kvm/mmu.c b/arch/x86/kvm/mmu.c
>>> index 22a7984..8f21632 100644
>>> --- a/arch/x86/kvm/mmu.c
>>> +++ b/arch/x86/kvm/mmu.c
>>> @@ -2901,6 +2901,7 @@ static int mmu_set_spte(struct kvm_vcpu *vcpu, u64 *sptep, unsigned pte_access,
>>> int rmap_count;
>>> int set_spte_ret;
>>> int ret = RET_PF_RETRY;
>>> + bool flush = false;
>>>
>>> pgprintk("%s: spte %llx write_fault %d gfn %llx\n", __func__,
>>> *sptep, write_fault, gfn);
>>> @@ -2917,12 +2918,12 @@ static int mmu_set_spte(struct kvm_vcpu *vcpu, u64 *sptep, unsigned pte_access,
>>>
>>> child = page_header(pte & PT64_BASE_ADDR_MASK);
>>> drop_parent_pte(child, sptep);
>>> - kvm_flush_remote_tlbs(vcpu->kvm);
>>> + flush = true;
>>> } else if (pfn != spte_to_pfn(*sptep)) {
>>> pgprintk("hfn old %llx new %llx\n",
>>> spte_to_pfn(*sptep), pfn);
>>> drop_spte(vcpu->kvm, sptep);
>>> - kvm_flush_remote_tlbs(vcpu->kvm);
>>> + flush = true;
>>> } else
>>> was_rmapped = 1;
>>> }
>>> @@ -2934,7 +2935,7 @@ static int mmu_set_spte(struct kvm_vcpu *vcpu, u64 *sptep, unsigned pte_access,
>>> ret = RET_PF_EMULATE;
>>> kvm_make_request(KVM_REQ_TLB_FLUSH, vcpu);
>>> }
>>> - if (set_spte_ret & SET_SPTE_NEED_REMOTE_TLB_FLUSH)
>>> + if (set_spte_ret & SET_SPTE_NEED_REMOTE_TLB_FLUSH || flush)
>>> kvm_flush_remote_tlbs(vcpu->kvm);
>>>
>>> if (unlikely(is_mmio_spte(*sptep)))
>>>
>>
>
> Reviewed-by: Junaid Shahid <[email protected]>
>

2018-07-25 06:24:17

by Xiao Guangrong

Subject: Re: [PATCH] KVM/MMU: Combine flushing remote tlb in mmu_set_spte()

On 07/24/2018 10:35 PM, Paolo Bonzini wrote:
> On 24/07/2018 10:17, Tianyu Lan wrote:
>> mmu_set_spte() flushes remote tlbs for drop_parent_pte/drop_spte()
>> and set_spte() separately. This may introduce redundant flush. This
>> patch is to combine these flushes and check flush request after
>> calling set_spte().
>>
>> Signed-off-by: Lan Tianyu <[email protected]>
>
> Looks good, but I'd like a second opinion. Guangrong, Junaid, can you
> review this?
>

It looks good to me.

Reviewed-by: Xiao Guangrong <[email protected]>

BTW, the @intel box is not accessible to me now. ;)