Before Commit c3e5e415bc1e6 ("KVM: X86: Change kvm_sync_page()
to return true when remote flush is needed"), the return value
of kvm_sync_page() indicates whether the page is synced, and
kvm_mmu_get_page() would rebuild page when the sync fails.
But now, kvm_sync_page() returns false when the page is
synced and no tlb flushing is required, which leads to
rebuild page in kvm_mmu_get_page(). So return the return
value of mmu->sync_page() directly and check it in
kvm_mmu_get_page(). If the sync fails, the page will be
zapped and the invalid_list is not empty, so set flush as
true is accepted in mmu_sync_children().
Fixes: c3e5e415bc1e6 ("KVM: X86: Change kvm_sync_page() to return true when remote flush is needed")
Signed-off-by: Hou Wenlong <[email protected]>
---
arch/x86/kvm/mmu/mmu.c | 16 ++++++++--------
1 file changed, 8 insertions(+), 8 deletions(-)
diff --git a/arch/x86/kvm/mmu/mmu.c b/arch/x86/kvm/mmu/mmu.c
index 3b8da8b0745e..8efd165ee27c 100644
--- a/arch/x86/kvm/mmu/mmu.c
+++ b/arch/x86/kvm/mmu/mmu.c
@@ -1866,17 +1866,14 @@ static void kvm_mmu_commit_zap_page(struct kvm *kvm,
&(_kvm)->arch.mmu_page_hash[kvm_page_table_hashfn(_gfn)]) \
if ((_sp)->gfn != (_gfn) || (_sp)->role.direct) {} else
-static bool kvm_sync_page(struct kvm_vcpu *vcpu, struct kvm_mmu_page *sp,
+static int kvm_sync_page(struct kvm_vcpu *vcpu, struct kvm_mmu_page *sp,
struct list_head *invalid_list)
{
int ret = vcpu->arch.mmu->sync_page(vcpu, sp);
- if (ret < 0) {
+ if (ret < 0)
kvm_mmu_prepare_zap_page(vcpu->kvm, sp, invalid_list);
- return false;
- }
-
- return !!ret;
+ return ret;
}
static bool kvm_mmu_remote_flush_or_zap(struct kvm *kvm,
@@ -2039,6 +2036,7 @@ static struct kvm_mmu_page *kvm_mmu_get_page(struct kvm_vcpu *vcpu,
struct hlist_head *sp_list;
unsigned quadrant;
struct kvm_mmu_page *sp;
+ int ret;
int collisions = 0;
LIST_HEAD(invalid_list);
@@ -2091,11 +2089,13 @@ static struct kvm_mmu_page *kvm_mmu_get_page(struct kvm_vcpu *vcpu,
* If the sync fails, the page is zapped. If so, break
* in order to rebuild it.
*/
- if (!kvm_sync_page(vcpu, sp, &invalid_list))
+ ret = kvm_sync_page(vcpu, sp, &invalid_list);
+ if (ret < 0)
break;
WARN_ON(!list_empty(&invalid_list));
- kvm_flush_remote_tlbs(vcpu->kvm);
+ if (ret > 0)
+ kvm_flush_remote_tlbs(vcpu->kvm);
}
__clear_sp_write_flooding_count(sp);
--
2.31.1
On Tue, Mar 15, 2022, Hou Wenlong wrote:
> Before Commit c3e5e415bc1e6 ("KVM: X86: Change kvm_sync_page()
> to return true when remote flush is needed"), the return value
> of kvm_sync_page() indicates whether the page is synced, and
> kvm_mmu_get_page() would rebuild page when the sync fails.
> But now, kvm_sync_page() returns false when the page is
> synced and no tlb flushing is required, which leads to
> rebuild page in kvm_mmu_get_page(). So return the return
> value of mmu->sync_page() directly and check it in
> kvm_mmu_get_page(). If the sync fails, the page will be
> zapped and the invalid_list is not empty, so set flush as
> true is accepted in mmu_sync_children().
>
> Fixes: c3e5e415bc1e6 ("KVM: X86: Change kvm_sync_page() to return true when remote flush is needed")
> Signed-off-by: Hou Wenlong <[email protected]>
> ---
> arch/x86/kvm/mmu/mmu.c | 16 ++++++++--------
> 1 file changed, 8 insertions(+), 8 deletions(-)
>
> diff --git a/arch/x86/kvm/mmu/mmu.c b/arch/x86/kvm/mmu/mmu.c
> index 3b8da8b0745e..8efd165ee27c 100644
> --- a/arch/x86/kvm/mmu/mmu.c
> +++ b/arch/x86/kvm/mmu/mmu.c
> @@ -1866,17 +1866,14 @@ static void kvm_mmu_commit_zap_page(struct kvm *kvm,
> &(_kvm)->arch.mmu_page_hash[kvm_page_table_hashfn(_gfn)]) \
> if ((_sp)->gfn != (_gfn) || (_sp)->role.direct) {} else
>
> -static bool kvm_sync_page(struct kvm_vcpu *vcpu, struct kvm_mmu_page *sp,
> +static int kvm_sync_page(struct kvm_vcpu *vcpu, struct kvm_mmu_page *sp,
> struct list_head *invalid_list)
> {
> int ret = vcpu->arch.mmu->sync_page(vcpu, sp);
>
> - if (ret < 0) {
> + if (ret < 0)
> kvm_mmu_prepare_zap_page(vcpu->kvm, sp, invalid_list);
> - return false;
> - }
> -
> - return !!ret;
> + return ret;
Hrm, this creates an oddity in mmu_sync_children(), which does a logical-OR of
the result into a boolean. It doesn't actually change the functionality since
kvm_mmu_remote_flush_or_zap() will prioritize invalid_list, but it's weird.
What about checking invalid_list directly and keeping the boolean return? Compile
tested only.
From: Sean Christopherson <[email protected]>
Date: Thu, 24 Mar 2022 13:07:32 -0700
Subject: [PATCH] KVM: x86/mmu: Fix shadow reuse when unsync sp doesn't need
TLB flush
Use invalid_list to detect if synchronizing an unsync shadow page failed
and resulted in the page being zapped. Since commit c3e5e415bc1e ("KVM:
X86: Change kvm_sync_page() to return true when remote flush is needed"),
kvm_sync_page() returns whether or not a TLB flush is required, it doesn't
provide any indication as to whether or not the sync was successful. If
the sync is successful but doesn't require a TLB flush, checking the
TLB flush result will cause KVM to unnecessarily rebuild the shadow page.
Fixes: c3e5e415bc1e6 ("KVM: X86: Change kvm_sync_page() to return true when remote flush is needed")
Cc: [email protected]
Reported-by: Hou Wenlong <[email protected]>
Signed-off-by: Sean Christopherson <[email protected]>
---
arch/x86/kvm/mmu/mmu.c | 15 +++++++++------
1 file changed, 9 insertions(+), 6 deletions(-)
diff --git a/arch/x86/kvm/mmu/mmu.c b/arch/x86/kvm/mmu/mmu.c
index 1361eb4599b4..b6350fec1b11 100644
--- a/arch/x86/kvm/mmu/mmu.c
+++ b/arch/x86/kvm/mmu/mmu.c
@@ -2086,16 +2086,19 @@ static struct kvm_mmu_page *kvm_mmu_get_page(struct kvm_vcpu *vcpu,
* This way the validity of the mapping is ensured, but the
* overhead of write protection is not incurred until the
* guest invalidates the TLB mapping. This allows multiple
- * SPs for a single gfn to be unsync.
- *
+ * SPs for a single gfn to be unsync. kvm_sync_page()
+ * returns true if a TLB flush is needed to ensure the
+ * guest sees the synchronized shadow page.
+ */
+ if (kvm_sync_page(vcpu, sp, &invalid_list))
+ kvm_flush_remote_tlbs(vcpu->kvm);
+
+ /*
* If the sync fails, the page is zapped. If so, break
* in order to rebuild it.
*/
- if (!kvm_sync_page(vcpu, sp, &invalid_list))
+ if (!list_empty(&invalid_list))
break;
-
- WARN_ON(!list_empty(&invalid_list));
- kvm_flush_remote_tlbs(vcpu->kvm);
}
__clear_sp_write_flooding_count(sp);
base-commit: 9b6a3be37eacee49a659232e86019e733597c045
--
On 3/24/22 21:17, Sean Christopherson wrote:
>> +static int kvm_sync_page(struct kvm_vcpu *vcpu, struct kvm_mmu_page *sp,
>> struct list_head *invalid_list)
>> {
>> int ret = vcpu->arch.mmu->sync_page(vcpu, sp);
>>
>> - if (ret < 0) {
>> + if (ret < 0)
>> kvm_mmu_prepare_zap_page(vcpu->kvm, sp, invalid_list);
>> - return false;
>> - }
>> -
>> - return !!ret;
>> + return ret;
> Hrm, this creates an oddity in mmu_sync_children(), which does a logical-OR of
> the result into a boolean. It doesn't actually change the functionality since
> kvm_mmu_remote_flush_or_zap() will prioritize invalid_list, but it's weird.
>
> What about checking invalid_list directly and keeping the boolean return? Compile
> tested only.
It's even better to check
flush |= kvm_sync_page(vcpu, sp, &invalid_list) > 0;
in mmu_sync_children. If the returned value is <0, then the page is
added to invalid_list and there is no need to set flush = true, just
like there is no need to call kvm_flush_remote_tlbs() in kvm_mmu_get_page().
Paolo
> Before Commit c3e5e415bc1e6 ("KVM: X86: Change kvm_sync_page()
> to return true when remote flush is needed"), the return value
> of kvm_sync_page() indicates whether the page is synced, and
> kvm_mmu_get_page() would rebuild page when the sync fails.
> But now, kvm_sync_page() returns false when the page is
> synced and no tlb flushing is required, which leads to
> rebuild page in kvm_mmu_get_page(). So return the return
> value of mmu->sync_page() directly and check it in
> kvm_mmu_get_page(). If the sync fails, the page will be
> zapped and the invalid_list is not empty, so set flush as
> true is accepted in mmu_sync_children().
>
> Fixes: c3e5e415bc1e6 ("KVM: X86: Change kvm_sync_page() to return true when remote flush is needed")
> Signed-off-by: Hou Wenlong <[email protected]>
Queued and Cc'ed to stable@, thanks.
Paolo